diff options
64 files changed, 7846 insertions, 2874 deletions
diff --git a/Jenkinsfile b/Jenkinsfile index 4bdbddf..a345bc0 100644 --- a/Jenkinsfile +++ b/Jenkinsfile @@ -14,8 +14,8 @@ pipeline { sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\ source `pwd`/setup_environment &&\ make -j' - }, "9.1" : { - sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ + }, "10.1" : { + sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/10.1_env_setup.sh &&\ source `pwd`/setup_environment &&\ make -j' } @@ -24,7 +24,7 @@ pipeline { stage('simulations-build'){ steps{ sh 'rm -rf gpgpu-sim_simulations' - sh 'git clone [email protected]:gpgpu-sim/gpgpu-sim_simulations.git && \ + sh 'git clone [email protected]:purdue-aalp/gpgpu-sim_simulations.git && \ cd gpgpu-sim_simulations && \ git pull && \ ln -s /home/tgrogers-raid/a/common/data_dirs benchmarks/' @@ -34,7 +34,7 @@ pipeline { source ./benchmarks/src/setup_environment && \ make -j -C ./benchmarks/src rodinia_2.0-ft sdk-4.2 && \ make -C ./benchmarks/src data' - sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ + sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/10.1_env_setup.sh &&\ source `pwd`/setup_environment &&\ cd gpgpu-sim_simulations && \ source ./benchmarks/src/setup_environment && \ @@ -47,71 +47,58 @@ pipeline { parallel "4.2-rodinia": { sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\ source `pwd`/setup_environment &&\ - ./gpgpu-sim_simulations/util/job_launching/run_simulations.py -B rodinia_2.0-ft -C GTX480,GTX480-PTXPLUS -N regress-$$ && \ - PLOTDIR="jenkins/${JOB_NAME}/${BUILD_NUMBER}/4.2-rodinia" && ssh [email protected] mkdir -p /home/dynamo/a/tgrogers/website/gpgpu-sim-plots/$PLOTDIR && \ - ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -N regress-$$ -s stats-per-app-4.2.csv && \ - ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c stats-per-app-4.2.csv -p [email protected]:~/website/gpgpu-sim-plots/$PLOTDIR -w https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/$PLOTDIR -n $PLOTDIR ' - }, "9.1-functest": { - sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ + export PTXAS_CUDA_INSTALL_PATH=/home/tgrogers-raid/a/common/cuda-10.1 &&\ + ./gpgpu-sim_simulations/util/job_launching/run_simulations.py -B rodinia_2.0-ft -C TITANV-PTXPLUS -N regress-$$ && \ + PLOTDIR="jenkins/${JOB_NAME}/${BUILD_NUMBER}/4.2" && ssh [email protected] mkdir -p /home/dynamo/a/tgrogers/website/gpgpu-sim-plots/$PLOTDIR && \ + ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -N regress-$$ -s stats-per-app-4.2.csv' + }, "10.1-functest": { + sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/10.1_env_setup.sh &&\ source `pwd`/setup_environment &&\ ./gpgpu-sim_simulations/util/job_launching/run_simulations.py -B rodinia_2.0-ft,sdk-4.2 -C TITANV -N regress-$$ && \ - PLOTDIR="jenkins/${JOB_NAME}/${BUILD_NUMBER}/9.1-rodinia" && ssh [email protected] mkdir -p /home/dynamo/a/tgrogers/website/gpgpu-sim-plots/$PLOTDIR && \ - ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -s stats-per-app-9.1.csv -N regress-$$ && \ - ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c stats-per-app-9.1.csv -p [email protected]:~/website/gpgpu-sim-plots/$PLOTDIR -w https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/$PLOTDIR -n $PLOTDIR' + PLOTDIR="jenkins/${JOB_NAME}/${BUILD_NUMBER}/10.1" && ssh [email protected] mkdir -p /home/dynamo/a/tgrogers/website/gpgpu-sim-plots/$PLOTDIR && \ + ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -s stats-per-app-10.1.csv -N regress-$$' } } } - stage('4.2-correlate'){ + stage('correlate-delta-and-archive') { steps { - sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\ - source `pwd`/setup_environment &&\ - PLOTDIR="jenkins/${JOB_NAME}" &&\ - ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B rodinia_2.0-ft -C GTX480,GTX480-PTXPLUS > stats-per-kernel-4.2.csv &&\ - ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh stats-per-kernel-4.2.csv $PLOTDIR ${BUILD_NUMBER}' - } - } - stage('9.1-correlate'){ - steps { - sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ - source `pwd`/setup_environment &&\ - PLOTDIR="jenkins/${JOB_NAME}" &&\ - ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B rodinia_2.0-ft,sdk-4.2 -C TITANV > stats-per-kernel-9.1.csv &&\ - ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh stats-per-kernel-9.1.csv $PLOTDIR ${BUILD_NUMBER}' - } - } - stage('archive-and-delta') { - steps { - sh 'rm -rf gpgpu-sim-results-repo' + sh './gpgpu-sim_simulations/run_hw/get_hw_data.sh' + sh 'rm -rf ./gpgpu-sim_simulations/util/plotting/correl-html && rm -rf gpgpu-sim-results-repo && rm -rf ./gpgpu-sim_simulations/util/plotting/htmls' sh 'git clone [email protected]:purdue-aalp/gpgpu-sim-results-repo.git' sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\ - ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B rodinia_2.0-ft -C GTX480-PTXPLUS > stats-per-kernel-4.2-ptxplus.csv &&\ - ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B rodinia_2.0-ft -C GTX480 > stats-per-kernel-4.2-ptx.csv' - sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ - ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B rodinia_2.0-ft,sdk-4.2 -C TITANV > stats-per-kernel-9.1-titanx.csv' - sh './gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/jenkins/quick-regress/AALP/gpgpu-sim_distribution/dev-purdue-integration/stats-per-app-4.2.csv,./stats-per-app-4.2.csv -R > per-app-merge-4.2.csv' - sh './gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/jenkins/quick-regress/AALP/gpgpu-sim_distribution/dev-purdue-integration/stats-per-app-9.1.csv,./stats-per-app-9.1.csv -R > per-app-merge-9.1.csv' + ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B rodinia_2.0-ft -C TITANV-PTXPLUS -A > stats-per-kernel-4.2-ptxplus.csv' + sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/10.1_env_setup.sh &&\ + ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B rodinia_2.0-ft,sdk-4.2 -C TITANV -A > stats-per-kernel-10.1.csv' + sh 'if [ ! -d ./gpgpu-sim-results-repo/${JOB_NAME} ]; then mkdir -p ./gpgpu-sim-results-repo/${JOB_NAME}/ ; cp ./gpgpu-sim-results-repo/purdue-aalp/gpgpu-sim_distribution/dev/* ./gpgpu-sim-results-repo/${JOB_NAME}/ ; fi' + sh './gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/${JOB_NAME}/stats-per-app-4.2.csv,./stats-per-app-4.2.csv -R > per-app-merge-4.2.csv' + sh './gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/${JOB_NAME}/stats-per-app-10.1.csv,./stats-per-app-10.1.csv -R > per-app-merge-10.1.csv' sh 'PLOTDIR="jenkins/${JOB_NAME}" &&\ - ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c per-app-merge-4.2.csv -p [email protected]:~/website/gpgpu-sim-plots/$PLOTDIR/deltas -w https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/$PLOTDIR -n $PLOTDIR/deltas &&\ - ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c per-app-merge-9.1.csv -p [email protected]:~/website/gpgpu-sim-plots/$PLOTDIR/deltas -w https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/$PLOTDIR/deltas -n $PLOTDIR/deltas &&\ - ./gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/jenkins/quick-regress/AALP/gpgpu-sim_distribution/dev-purdue-integration/stats-per-kernel-4.2-ptx.csv,./stats-per-kernel-4.2-ptx.csv -R > per-kernel-merge-4.2-ptx.csv &&\ - ./gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/jenkins/quick-regress/AALP/gpgpu-sim_distribution/dev-purdue-integration/stats-per-kernel-4.2-ptxplus.csv,./stats-per-kernel-4.2-ptxplus.csv -R > per-kernel-merge-4.2-ptxplus.csv &&\ - ./gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/jenkins/quick-regress/AALP/gpgpu-sim_distribution/dev-purdue-integration/stats-per-kernel-9.1-titanx.csv,./stats-per-kernel-9.1-titanx.csv -R > per-kernel-merge-9.1-titanx.csv &&\ - ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh per-kernel-merge-4.2-ptx.csv $PLOTDIR ${BUILD_NUMBER} &&\ - ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh per-kernel-merge-4.2-ptxplus.csv $PLOTDIR ${BUILD_NUMBER} &&\ - ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh per-kernel-merge-9.1-titanx.csv $PLOTDIR ${BUILD_NUMBER} &&\ - mkdir -p ./jenkins/quick-regress/${JOB_NAME}/ && cp stats-per-*.csv ./jenkins/quick-regress/${JOB_NAME}/ &&\ + ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c per-app-merge-4.2.csv -P cuda-4.2 &&\ + ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c per-app-merge-10.1.csv -P cuda-10.1 &&\ + ./gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/${JOB_NAME}/stats-per-kernel-4.2-ptxplus.csv,./stats-per-kernel-4.2-ptxplus.csv -R > per-kernel-merge-4.2-ptxplus.csv &&\ + ./gpgpu-sim_simulations/util/plotting/merge-stats.py -c ./gpgpu-sim-results-repo/${JOB_NAME}/stats-per-kernel-10.1.csv,./stats-per-kernel-10.1.csv -R > per-kernel-merge-10.1.csv &&\ + ./gpgpu-sim_simulations/util/plotting/plot-correlation.py -c per-kernel-merge-4.2-ptxplus.csv -p cuda-4.2 | grep "Correl=" | tee correl.4.2.txt &&\ + ./gpgpu-sim_simulations/util/plotting/plot-correlation.py -c per-kernel-merge-10.1.csv -p cuda-10.1 | grep "Correl=" | tee correl.10.1.txt &&\ + mkdir -p ./gpgpu-sim-results-repo/${JOB_NAME}/ && cp stats-per-*.csv ./gpgpu-sim-results-repo/${JOB_NAME}/ &&\ cd ./gpgpu-sim-results-repo &&\ - git diff --quiet && git diff --staged --quiet || git commit -am "Jenkins automated checkin ${BUILD_NUMBER}" &&\ + git diff --quiet && git diff --staged --quiet || git commit -am "Jenkins automated checkin ${JOB_NAME} Build:${BUILD_NUMBER}" &&\ git push' + + sh 'PLOTDIR="/home/dynamo/a/tgrogers/website/gpgpu-sim-plots/jenkins/${JOB_NAME}" &&\ + ssh [email protected] mkdir -p $PLOTDIR/${BUILD_NUMBER} && \ + scp ./gpgpu-sim_simulations/util/plotting/correl-html/* [email protected]:$PLOTDIR/${BUILD_NUMBER} &&\ + scp ./gpgpu-sim_simulations/util/plotting/htmls/* [email protected]:$PLOTDIR/${BUILD_NUMBER} &&\ + ssh [email protected] "cd $PLOTDIR && rm -rf latest && cp -r ${BUILD_NUMBER} latest"' } } } post { success { - emailext body: "See ${BUILD_URL}", + emailext body:'''${SCRIPT, template="groovy-html.success.template"}''', recipientProviders: [[$class: 'CulpritsRecipientProvider'], [$class: 'RequesterRecipientProvider']], subject: "[AALP Jenkins] Build #${BUILD_NUMBER} - Success!", + attachmentsPattern: 'correl.*.txt', to: '[email protected]' } failure { @@ -164,6 +164,8 @@ $(SIM_LIB_DIR)/libcudart.so: makedirs $(LIBS) cudalib if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.0; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.1 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.1; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.2 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.2; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.10.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.10.0; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.10.1 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.10.1; fi $(SIM_LIB_DIR)/libcudart.dylib: makedirs $(LIBS) cudalib g++ -dynamiclib -Wl,-headerpad_max_install_names,-undefined,dynamic_lookup,-compatibility_version,1.1,-current_version,1.1\ diff --git a/cuobjdump_to_ptxplus/Makefile b/cuobjdump_to_ptxplus/Makefile index 242a066..e95136a 100644 --- a/cuobjdump_to_ptxplus/Makefile +++ b/cuobjdump_to_ptxplus/Makefile @@ -27,11 +27,11 @@ $(OUTPUT_DIR)/cuobjdump_to_ptxplus: $(OUTPUT_DIR)/cuobjdumpInst.o $(OUTPUT_DIR)/ ${LD} ${LDFLAGS} -o $@ $(OUTPUT_DIR)/cuobjdumpInst.o $(OUTPUT_DIR)/cuobjdumpInstList.o $(OUTPUT_DIR)/cuobjdump_to_ptxplus.o $(PTX_PARSER_OBJECTS) $(SASS_PARSER_OBJECTS) $(ELF_PARSER_OBJECTS) $(HEADER_PARSER_OBJECTS) -$(OUTPUT_DIR)/lex.ptx_.c : ../src/cuda-sim/ptx.l - ${LEX} ${LEXFLAGS} -o$(OUTPUT_DIR)/lex.ptx_.c ../src/cuda-sim/ptx.l +$(OUTPUT_DIR)/lex.ptx_.c : ptx.l + ${LEX} ${LEXFLAGS} -o$(OUTPUT_DIR)/lex.ptx_.c ptx.l -$(OUTPUT_DIR)/ptx.tab.c : ../src/cuda-sim/ptx.y - ${YACC} ${YFLAGS} --name-prefix=ptx_ -v ../src/cuda-sim/ptx.y --file-prefix=$(OUTPUT_DIR)/ptx +$(OUTPUT_DIR)/ptx.tab.c : ptx.y + ${YACC} ${YFLAGS} --name-prefix=ptx_ -v ptx.y --file-prefix=$(OUTPUT_DIR)/ptx $(OUTPUT_DIR)/ptx.tab.h :$(OUTPUT_DIR)/ptx.tab.c diff --git a/cuobjdump_to_ptxplus/ptx.l b/cuobjdump_to_ptxplus/ptx.l new file mode 100644 index 0000000..3232361 --- /dev/null +++ b/cuobjdump_to_ptxplus/ptx.l @@ -0,0 +1,471 @@ +/* +Copyright (c) 2009-2011, Tor M. Aamodt +The University of British Columbia +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +Redistributions of source code must retain the above copyright notice, this +list of conditions and the following disclaimer. +Redistributions in binary form must reproduce the above copyright notice, this +list of conditions and the following disclaimer in the documentation and/or +other materials provided with the distribution. +Neither the name of The University of British Columbia nor the names of its +contributors may be used to endorse or promote products derived from this +software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND +ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +%option nounput +%option noyywrap +%option yylineno +%option prefix="ptx_" +%{ +#include "opcodes.h" +#include "ptx.tab.h" +#include <string.h> + +#define LINEBUF_SIZE (64*1024) +char linebuf[LINEBUF_SIZE]; +unsigned col = 0; +#define TC col+=strlen(ptx_text); +#define CHECK_UNSIGNED \ + if( yytext[strlen(yytext)-1]=='U' ) { \ + printf("GPGPU-Sim: ERROR ** U modifier not implemented\n"); \ + abort(); \ + } +int ptx_error( const char *s ); +%} + +%s IN_STRING +%s IN_COMMENT +%x IN_FUNC_DECL +%x NOT_OPCODE +%% + +abs TC; ptx_lval.int_value = ABS_OP; return OPCODE; +add TC; ptx_lval.int_value = ADD_OP; return OPCODE; +addp TC; ptx_lval.int_value = ADDP_OP; return OPCODE; +addc TC; ptx_lval.int_value = ADDC_OP; return OPCODE; +and TC; ptx_lval.int_value = AND_OP; return OPCODE; +andn TC; ptx_lval.int_value = ANDN_OP; return OPCODE; +atom TC; ptx_lval.int_value = ATOM_OP; return OPCODE; +bar.warp TC; ptx_lval.int_value = NOP_OP; return OPCODE; +bar TC; ptx_lval.int_value = BAR_OP; return OPCODE; +bfe TC; ptx_lval.int_value = BFE_OP; return OPCODE; +bfi TC; ptx_lval.int_value = BFI_OP; return OPCODE; +bfind TC; ptx_lval.int_value = BFIND_OP; return OPCODE; +bra TC; ptx_lval.int_value = BRA_OP; return OPCODE; +brx TC; ptx_lval.int_value = BRX_OP; return OPCODE; +brev TC; ptx_lval.int_value = BREV_OP; return OPCODE; +brkpt TC; ptx_lval.int_value = BRKPT_OP; return OPCODE; + +wmma TC; ptx_lval.int_value = MMA_OP; return OPCODE; +wmma\.load TC; ptx_lval.int_value = MMA_LD_OP; return OPCODE; +wmma\.store TC; ptx_lval.int_value = MMA_ST_OP; return OPCODE; + +call TC; BEGIN(NOT_OPCODE); ptx_lval.int_value = CALL_OP; return OPCODE; // blocking opcode token in case the callee has the same name as an opcode +callp TC; BEGIN(NOT_OPCODE); ptx_lval.int_value = CALLP_OP; return OPCODE; +clz TC; ptx_lval.int_value = CLZ_OP; return OPCODE; +cnot TC; ptx_lval.int_value = CNOT_OP; return OPCODE; +cos TC; ptx_lval.int_value = COS_OP; return OPCODE; +cvt TC; ptx_lval.int_value = CVT_OP; return OPCODE; +cvta TC; ptx_lval.int_value = CVTA_OP; return OPCODE; +div TC; ptx_lval.int_value = DIV_OP; return OPCODE; +dp4a TC; ptx_lval.int_value = DP4A_OP; return OPCODE; +ex2 TC; ptx_lval.int_value = EX2_OP; return OPCODE; +exit TC; ptx_lval.int_value = EXIT_OP; return OPCODE; +fma TC; ptx_lval.int_value = FMA_OP; return OPCODE; +isspacep TC; ptx_lval.int_value = ISSPACEP_OP; return OPCODE; +ld TC; ptx_lval.int_value = LD_OP; return OPCODE; +ld.volatile TC; ptx_lval.int_value = LD_OP; return OPCODE; +ldu TC; ptx_lval.int_value = LDU_OP; return OPCODE; +lg2 TC; ptx_lval.int_value = LG2_OP; return OPCODE; +mad24 TC; ptx_lval.int_value = MAD24_OP; return OPCODE; +mad TC; ptx_lval.int_value = MAD_OP; return OPCODE; +madc TC; ptx_lval.int_value = MADC_OP; return OPCODE; +madp TC; ptx_lval.int_value = MADP_OP; return OPCODE; +max TC; ptx_lval.int_value = MAX_OP; return OPCODE; +membar TC; ptx_lval.int_value = MEMBAR_OP; return OPCODE; +min TC; ptx_lval.int_value = MIN_OP; return OPCODE; +mov TC; ptx_lval.int_value = MOV_OP; return OPCODE; +mul24 TC; ptx_lval.int_value = MUL24_OP; return OPCODE; +mul TC; ptx_lval.int_value = MUL_OP; return OPCODE; +neg TC; ptx_lval.int_value = NEG_OP; return OPCODE; +nandn TC; ptx_lval.int_value = NANDN_OP; return OPCODE; +norn TC; ptx_lval.int_value = NORN_OP; return OPCODE; +not TC; ptx_lval.int_value = NOT_OP; return OPCODE; +or TC; ptx_lval.int_value = OR_OP; return OPCODE; +orn TC; ptx_lval.int_value = ORN_OP; return OPCODE; +pmevent TC; ptx_lval.int_value = PMEVENT_OP; return OPCODE; +popc TC; ptx_lval.int_value = POPC_OP; return OPCODE; +prefetch TC; ptx_lval.int_value = PREFETCH_OP; return OPCODE; +prefetchu TC; ptx_lval.int_value = PREFETCHU_OP; return OPCODE; +prmt TC; ptx_lval.int_value = PRMT_OP; return OPCODE; +rcp TC; ptx_lval.int_value = RCP_OP; return OPCODE; +red TC; ptx_lval.int_value = RED_OP; return OPCODE; +rem TC; ptx_lval.int_value = REM_OP; return OPCODE; +ret TC; ptx_lval.int_value = RET_OP; return OPCODE; +retp TC; ptx_lval.int_value = RETP_OP; return OPCODE; +rsqrt TC; ptx_lval.int_value = RSQRT_OP; return OPCODE; +sad TC; ptx_lval.int_value = SAD_OP; return OPCODE; +selp TC; ptx_lval.int_value = SELP_OP; return OPCODE; +setp TC; ptx_lval.int_value = SETP_OP; return OPCODE; +set TC; ptx_lval.int_value = SET_OP; return OPCODE; +shfl TC; ptx_lval.int_value = SHFL_OP; return OPCODE; +shl TC; ptx_lval.int_value = SHL_OP; return OPCODE; +shr TC; ptx_lval.int_value = SHR_OP; return OPCODE; +sin TC; ptx_lval.int_value = SIN_OP; return OPCODE; +slct TC; ptx_lval.int_value = SLCT_OP; return OPCODE; +sqrt TC; ptx_lval.int_value = SQRT_OP; return OPCODE; +sst TC; ptx_lval.int_value = SST_OP; return OPCODE; +ssy TC; ptx_lval.int_value = SSY_OP; return OPCODE; +st TC; ptx_lval.int_value = ST_OP; return OPCODE; +st.volatile TC; ptx_lval.int_value = ST_OP; return OPCODE; +sub TC; ptx_lval.int_value = SUB_OP; return OPCODE; +subc TC; ptx_lval.int_value = SUBC_OP; return OPCODE; +suld TC; ptx_lval.int_value = SULD_OP; return OPCODE; +sured TC; ptx_lval.int_value = SURED_OP; return OPCODE; +surst TC; ptx_lval.int_value = SUST_OP; return OPCODE; +suq TC; ptx_lval.int_value = SUQ_OP; return OPCODE; +tex TC; BEGIN(NOT_OPCODE); ptx_lval.int_value = TEX_OP; return OPCODE; +txq TC; ptx_lval.int_value = TEX_OP; return OPCODE; +trap TC; ptx_lval.int_value = TRAP_OP; return OPCODE; +vabsdiff TC; ptx_lval.int_value = VABSDIFF_OP; return OPCODE; +vadd TC; ptx_lval.int_value = VADD_OP; return OPCODE; +vmad TC; ptx_lval.int_value = VMAD_OP; return OPCODE; +vmax TC; ptx_lval.int_value = VMAX_OP; return OPCODE; +vmin TC; ptx_lval.int_value = VMIN_OP; return OPCODE; +vset TC; ptx_lval.int_value = VSET_OP; return OPCODE; +vshl TC; ptx_lval.int_value = VSHL_OP; return OPCODE; +vshr TC; ptx_lval.int_value = VSHR_OP; return OPCODE; +vsub TC; ptx_lval.int_value = VSUB_OP; return OPCODE; +vote TC; ptx_lval.int_value = VOTE_OP; return OPCODE; +xor TC; ptx_lval.int_value = XOR_OP; return OPCODE; +nop TC; ptx_lval.int_value = NOP_OP; return OPCODE; +break TC; ptx_lval.int_value = BREAK_OP; return OPCODE; +breakaddr TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE; + +"CPTX_END" printf("ENDING CUSTOM PTX.\n"); BEGIN(IN_COMMENT); + +<INITIAL,NOT_OPCODE,IN_INST,IN_FUNC_DECL>{ +\.a\.sync TC; ptx_lval.int_value = LOAD_A; return WMMA_DIRECTIVE; +\.b\.sync TC; ptx_lval.int_value = LOAD_B; return WMMA_DIRECTIVE; +\.c\.sync TC; ptx_lval.int_value = LOAD_C; return WMMA_DIRECTIVE; +\.d\.sync TC; ptx_lval.int_value = STORE_D; return WMMA_DIRECTIVE; +\.mma\.sync TC;ptx_lval.int_value=MMA; return WMMA_DIRECTIVE; + +\.row TC; ptx_lval.int_value = ROW; return LAYOUT; +\.col TC; ptx_lval.int_value = COL; return LAYOUT; +\.m16n16k16 TC; ptx_lval.int_value = M16N16K16; return CONFIGURATION; +\.f4e TC; return PRMT_F4E_MODE; +\.b4e TC; return PRMT_B4E_MODE; +\.rc8 TC; return PRMT_RC8_MODE; +\.ecl TC; return PRMT_ECL_MODE; +\.ecr TC; return PRMT_ECR_MODE; +\.rc16 TC; return PRMT_RC16_MODE; + +\.align TC; return ALIGN_DIRECTIVE; +\.branchtargets TC; return BRANCHTARGETS_DIRECTIVE; +\.byte TC; return BYTE_DIRECTIVE; /* not in PTX 2.1 */ +\.callprototype TC; return CALLPROTOTYPE_DIRECTIVE; +\.calltargets TC; return CALLTARGETS_DIRECTIVE; +\.const\[[0-9]+\] TC; ptx_lval.int_value = atoi(yytext+7); return CONST_DIRECTIVE; +\.const TC; ptx_lval.int_value = 0; return CONST_DIRECTIVE; +\.entry TC; return ENTRY_DIRECTIVE; +\.extern TC; return EXTERN_DIRECTIVE; +\.file TC; BEGIN(INITIAL); return FILE_DIRECTIVE; +\.func TC; BEGIN(IN_FUNC_DECL); return FUNC_DIRECTIVE; // blocking opcode parsing in case the function has the same name as an opcode (e.g. sin(), cos()) +\.global TC; return GLOBAL_DIRECTIVE; +\.global.volatile TC; return GLOBAL_DIRECTIVE; //TODO: fix this! +\.local TC; return LOCAL_DIRECTIVE; +\.loc TC; return LOC_DIRECTIVE; +\.maxnctapersm TC; return MAXNCTAPERSM_DIRECTIVE; +\.maxnreg TC; return MAXNNREG_DIRECTIVE; +\.maxntid TC; return MAXNTID_DIRECTIVE; +\.minnctapersm TC; return MINNCTAPERSM_DIRECTIVE; +\.param TC; return PARAM_DIRECTIVE; +\.pragma TC; return PRAGMA_DIRECTIVE; +\.reg TC; return REG_DIRECTIVE; +\.reqntid TC; return REQNTID_DIRECTIVE; +\.section TC; return SECTION_DIRECTIVE; +\.shared TC; return SHARED_DIRECTIVE; +\.sreg TC; return SREG_DIRECTIVE; +\.sstarr TC; return SSTARR_DIRECTIVE; +\.struct TC; return STRUCT_DIRECTIVE; +\.surf TC; return SURF_DIRECTIVE; /* not in PTX 2.1 */ +\.target TC; return TARGET_DIRECTIVE; +\.tex TC; BEGIN(NOT_OPCODE); return TEX_DIRECTIVE; +\.union TC; return UNION_DIRECTIVE; /* not in PTX 2.1 */ +\.version TC; return VERSION_DIRECTIVE; +\.visible TC; return VISIBLE_DIRECTIVE; +\.weak TC; return WEAK_DIRECTIVE; +\.address_size TC; return ADDRESS_SIZE_DIRECTIVE; +\.weak TC; return WEAK_DIRECTIVE; + +\.constptr TC; return CONSTPTR_DIRECTIVE; /* Ptx plus directive for pointer to constant memory */ +\.ptr TC; return PTR_DIRECTIVE; /* Added for new OpenCL genrated code */ + +"%clock" TC; ptx_lval.int_value = CLOCK_REG; return SPECIAL_REGISTER; +"%halfclock" TC; ptx_lval.int_value = HALFCLOCK_ID; return SPECIAL_REGISTER; +"%clock64" TC; ptx_lval.int_value = CLOCK64_REG; return SPECIAL_REGISTER; +"%ctaid" TC; ptx_lval.int_value = CTAID_REG; return SPECIAL_REGISTER; +"%envreg"[0-9]+ TC; sscanf(yytext+7,"%u",&ptx_lval.int_value); ptx_lval.int_value<<=16; ptx_lval.int_value += ENVREG_REG; return SPECIAL_REGISTER; +"%gridid" TC; ptx_lval.int_value = GRIDID_REG; return SPECIAL_REGISTER; +"%laneid" TC; ptx_lval.int_value = LANEID_REG; return SPECIAL_REGISTER; +"%lanemask_eq" TC; ptx_lval.int_value = LANEMASK_EQ_REG; return SPECIAL_REGISTER; +"%lanemask_le" TC; ptx_lval.int_value = LANEMASK_LE_REG; return SPECIAL_REGISTER; +"%lanemask_lt" TC; ptx_lval.int_value = LANEMASK_LT_REG; return SPECIAL_REGISTER; +"%lanemask_ge" TC; ptx_lval.int_value = LANEMASK_GE_REG; return SPECIAL_REGISTER; +"%lanemask_gt" TC; ptx_lval.int_value = LANEMASK_GT_REG; return SPECIAL_REGISTER; +"%nctaid" TC; ptx_lval.int_value = NCTAID_REG; return SPECIAL_REGISTER; +"%ntid" TC; ptx_lval.int_value = NTID_REG; return SPECIAL_REGISTER; +"%nsmid" TC; ptx_lval.int_value = NSMID_REG; return SPECIAL_REGISTER; +"%nwarpid" TC; ptx_lval.int_value = NWARPID_REG; return SPECIAL_REGISTER; +"%pm"[0-3] TC; sscanf(yytext+3,"%u",&ptx_lval.int_value); ptx_lval.int_value<<=16; ptx_lval.int_value += PM_REG; return SPECIAL_REGISTER; +"%smid" TC; ptx_lval.int_value = SMID_REG; return SPECIAL_REGISTER; +"%tid" TC; ptx_lval.int_value = TID_REG; return SPECIAL_REGISTER; +"%warpid" TC; ptx_lval.int_value = WARPID_REG; return SPECIAL_REGISTER; +"WARP_SZ" TC; ptx_lval.int_value = WARPSZ_REG; return SPECIAL_REGISTER; + +[a-zA-Z_][a-zA-Z0-9_$]* TC; ptx_lval.string_value = strdup(yytext); return IDENTIFIER; +[$%][a-zA-Z0-9_$]+ TC; ptx_lval.string_value = strdup(yytext); return IDENTIFIER; + +[0-9]+\.[0-9]+ TC; sscanf(yytext,"%lf", &ptx_lval.double_value); return DOUBLE_OPERAND; + +0[xX][0-9a-fA-F]+U? TC; CHECK_UNSIGNED; sscanf(yytext,"%x", &ptx_lval.int_value); return INT_OPERAND; +0[0-7]+U? TC; printf("GPGPU-Sim: ERROR ** parsing octal not (yet) implemented\n"); abort(); return INT_OPERAND; +0[bB][01]+U? TC; printf("GPGPU-Sim: ERROR ** parsing binary not (yet) implemented\n"); abort(); return INT_OPERAND; +[-]?[0-9]+U? TC; CHECK_UNSIGNED; ptx_lval.int_value = atoi(yytext); return INT_OPERAND; + +0[fF][0-9a-fA-F]{8} TC; sscanf(yytext+2,"%x", (unsigned*)(void*)&ptx_lval.float_value); return FLOAT_OPERAND; +0[dD][0-9a-fA-F]{16} TC; sscanf(yytext+2,"%Lx", (unsigned long long*)(void*)&ptx_lval.double_value); return DOUBLE_OPERAND; + +\.s8 TC; return S8_TYPE; +\.s16 TC; return S16_TYPE; +\.s32 TC; return S32_TYPE; +\.s64 TC; return S64_TYPE; +\.u8 TC; return U8_TYPE; +\.u16 TC; return U16_TYPE; +\.u32 TC; return U32_TYPE; +\.u64 TC; return U64_TYPE; +\.f16 TC; return F16_TYPE; +\.f16x2 TC; return F16_TYPE; /* TODO: figure out what this should really be */ +\.f32 TC; return F32_TYPE; +\.f64 TC; return F64_TYPE; +\.ff64 TC; return FF64_TYPE; +\.b8 TC; return B8_TYPE; +\.b16 TC; return B16_TYPE; +\.b32 TC; return B32_TYPE; +\.b64 TC; return B64_TYPE; +\.bb64 TC; return BB64_TYPE; +\.bb128 TC; return BB128_TYPE; +\.pred TC; return PRED_TYPE; + +\.texref TC; BEGIN(NOT_OPCODE); return TEXREF_TYPE; +\.samplerref TC; return SAMPLERREF_TYPE; +\.surfref TC; return SURFREF_TYPE; + +\.v2 TC; return V2_TYPE; +\.v3 TC; return V3_TYPE; +\.v4 TC; return V4_TYPE; + +\.half TC; return HALF_OPTION; /* ptxplus */ +\.cc TC; return EXTP_OPTION; /* extended precision option */ + +\.equ TC; return EQU_OPTION; +\.neu TC; return NEU_OPTION; +\.ltu TC; return LTU_OPTION; +\.leu TC; return LEU_OPTION; +\.gtu TC; return GTU_OPTION; +\.geu TC; return GEU_OPTION; +\.num TC; return NUM_OPTION; +\.nan TC; return NAN_OPTION; + +\.sat TC; return SAT_OPTION; + +\.eq TC; return EQ_OPTION; +\.ne TC; return NE_OPTION; +\.lt TC; return LT_OPTION; +\.le TC; return LE_OPTION; +\.gt TC; return GT_OPTION; +\.ge TC; return GE_OPTION; +\.cf TC; return CF_OPTION; +\.sf TC; return SF_OPTION; +\.nsf TC; return NSF_OPTION; + +\.lo TC; return LO_OPTION; +\.ls TC; return LS_OPTION; +\.hi TC; return HI_OPTION; +\.hs TC; return HS_OPTION; + + +\.rni TC; return RNI_OPTION; +\.rzi TC; return RZI_OPTION; +\.rmi TC; return RMI_OPTION; +\.rpi TC; return RPI_OPTION; + +\.rn TC; return RN_OPTION; +\.rz TC; return RZ_OPTION; +\.rm TC; return RM_OPTION; +\.rp TC; return RP_OPTION; + +\.ftz TC; return FTZ_OPTION; + +\.neg TC; return NEG_OPTION; + +\.wide TC; return WIDE_OPTION; +\.uni TC; return UNI_OPTION; + +\.sync TC; return SYNC_OPTION; +\.arrive TC; return ARRIVE_OPTION; +\.red TC; return RED_OPTION; + + +\.approx TC; return APPROX_OPTION; +\.full TC; return FULL_OPTION; + +\.any TC; return ANY_OPTION; +\.all TC; return ALL_OPTION; +\.ballot TC; return BALLOT_OPTION; +\.gl TC; return GLOBAL_OPTION; +\.cta TC; return CTA_OPTION; +\.sys TC; return SYS_OPTION; + +\.exit TC; return EXIT_OPTION; + +\.abs TC; return ABS_OPTION; + +\.to TC; return TO_OPTION; + +\.ca TC; return CA_OPTION; +\.cg TC; return CG_OPTION; +\.cs TC; return CS_OPTION; +\.lu TC; return LU_OPTION; +\.cv TC; return CV_OPTION; + +\.wb TC; return WB_OPTION; +\.wt TC; return WT_OPTION; + +\.nc TC; return NC_OPTION; + +\.up TC; return UP_OPTION; +\.down TC; return DOWN_OPTION; +\.bfly TC; return BFLY_OPTION; +\.idx TC; return IDX_OPTION; + +\.popc TC; return ATOMIC_POPC; +\.and TC; return ATOMIC_AND; +\.or TC; return ATOMIC_OR; +\.xor TC; return ATOMIC_XOR; +\.cas TC; return ATOMIC_CAS; +\.exch TC; return ATOMIC_EXCH; +\.add TC; return ATOMIC_ADD; +\.inc TC; return ATOMIC_INC; +\.dec TC; return ATOMIC_DEC; +\.min TC; return ATOMIC_MIN; +\.max TC; return ATOMIC_MAX; + + + +\.1d TC; return GEOM_MODIFIER_1D; +\.2d TC; return GEOM_MODIFIER_2D; +\.3d TC; return GEOM_MODIFIER_3D; + +\.0 TC; ptx_lval.int_value = 0; return DIMENSION_MODIFIER; +\.1 TC; ptx_lval.int_value = 1; return DIMENSION_MODIFIER; +\.2 TC; ptx_lval.int_value = 2; return DIMENSION_MODIFIER; +\.x TC; ptx_lval.int_value = 0; return DIMENSION_MODIFIER; +\.y TC; ptx_lval.int_value = 1; return DIMENSION_MODIFIER; +\.z TC; ptx_lval.int_value = 2; return DIMENSION_MODIFIER; + +"-" TC; return MINUS; +"+" TC; return PLUS; +"," TC; return COMMA; +"@" TC; return PRED; +"|" TC; return PIPE; +"[" TC; return LEFT_SQUARE_BRACKET; +"]" TC; return RIGHT_SQUARE_BRACKET; +"<" TC; return LEFT_ANGLE_BRACKET; +">" TC; return RIGHT_ANGLE_BRACKET; +"(" TC; return LEFT_PAREN; +")" TC; return RIGHT_PAREN; +":" TC; BEGIN(INITIAL); return COLON; +";" TC; BEGIN(INITIAL); return SEMI_COLON; +"!" TC; return EXCLAMATION; +"=" TC; return EQUALS; +"}" TC; return RIGHT_BRACE; +\. TC; return PERIOD; +"/" TC; return BACKSLASH; + +"//"[^\n]* TC; // eat single + +\n.* col=0; strncpy(linebuf, yytext + 1, LINEBUF_SIZE); yyless( 1 ); + +" " TC; +"\t" TC; + + +} + +<IN_FUNC_DECL>"{" TC; BEGIN(INITIAL); return LEFT_BRACE; // starting a statement block (allow next token to be parsed as an opcode) +<INITIAL,NOT_OPCODE>"{" TC; return LEFT_BRACE; // starting a vector operand (next token cannot be opcode) + +<INITIAL>{ +"/*" BEGIN(IN_COMMENT); +} +<IN_COMMENT>{ +"*/" BEGIN(INITIAL); +"CPTX_BEGIN" printf("BEGINNING CUSTOM PTX.\n"); BEGIN(INITIAL); +[^C*\n]+ // eat comment in chunks +"C" // eat the lone C +"*" // eat the lone star +\n TC; +} + +<INITIAL>{ +"\"" BEGIN(IN_STRING); +} +<IN_STRING>{ +"\"" TC; BEGIN(INITIAL); return STRING; +[^\"]* TC; ptx_lval.string_value = strdup(yytext); +} + +<*>\t@@DWARF.*\n + +<INITIAL,NOT_OPCODE,IN_FUNC_DECL>. TC; ptx_error((const char*)NULL); +%% + +extern int g_error_detected; +extern const char *g_filename; + +int ptx_error( const char *s ) +{ + int i; + g_error_detected = 1; + fflush(stdout); + if( s != NULL ) + printf("%s:%u: Syntax error:\n\n", g_filename, ptx_lineno ); + printf(" %s\n", linebuf ); + printf(" "); + for( i=0; i < col-1; i++ ) { + if( linebuf[i] == '\t' ) printf("\t"); + else printf(" "); + } + + printf("^\n\n"); + fflush(stdout); + //exit(1); + return 0; +} diff --git a/cuobjdump_to_ptxplus/ptx.y b/cuobjdump_to_ptxplus/ptx.y new file mode 100644 index 0000000..45392fb --- /dev/null +++ b/cuobjdump_to_ptxplus/ptx.y @@ -0,0 +1,629 @@ +/* +Copyright (c) 2009-2011, Tor M. Aamodt +The University of British Columbia +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +Redistributions of source code must retain the above copyright notice, this +list of conditions and the following disclaimer. +Redistributions in binary form must reproduce the above copyright notice, this +list of conditions and the following disclaimer in the documentation and/or +other materials provided with the distribution. +Neither the name of The University of British Columbia nor the names of its +contributors may be used to endorse or promote products derived from this +software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND +ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +%union { + double double_value; + float float_value; + int int_value; + char * string_value; + void * ptr_value; +} + +%token <string_value> STRING +%token <int_value> OPCODE +%token <int_value> WMMA_DIRECTIVE +%token <int_value> LAYOUT +%token <int_value> CONFIGURATION +%token ALIGN_DIRECTIVE +%token BRANCHTARGETS_DIRECTIVE +%token BYTE_DIRECTIVE +%token CALLPROTOTYPE_DIRECTIVE +%token CALLTARGETS_DIRECTIVE +%token <int_value> CONST_DIRECTIVE +%token CONSTPTR_DIRECTIVE +%token PTR_DIRECTIVE +%token ENTRY_DIRECTIVE +%token EXTERN_DIRECTIVE +%token FILE_DIRECTIVE +%token FUNC_DIRECTIVE +%token GLOBAL_DIRECTIVE +%token LOCAL_DIRECTIVE +%token LOC_DIRECTIVE +%token MAXNCTAPERSM_DIRECTIVE +%token MAXNNREG_DIRECTIVE +%token MAXNTID_DIRECTIVE +%token MINNCTAPERSM_DIRECTIVE +%token PARAM_DIRECTIVE +%token PRAGMA_DIRECTIVE +%token REG_DIRECTIVE +%token REQNTID_DIRECTIVE +%token SECTION_DIRECTIVE +%token SHARED_DIRECTIVE +%token SREG_DIRECTIVE +%token SSTARR_DIRECTIVE +%token STRUCT_DIRECTIVE +%token SURF_DIRECTIVE +%token TARGET_DIRECTIVE +%token TEX_DIRECTIVE +%token UNION_DIRECTIVE +%token VERSION_DIRECTIVE +%token ADDRESS_SIZE_DIRECTIVE +%token VISIBLE_DIRECTIVE +%token WEAK_DIRECTIVE +%token <string_value> IDENTIFIER +%token <int_value> INT_OPERAND +%token <float_value> FLOAT_OPERAND +%token <double_value> DOUBLE_OPERAND +%token S8_TYPE +%token S16_TYPE +%token S32_TYPE +%token S64_TYPE +%token U8_TYPE +%token U16_TYPE +%token U32_TYPE +%token U64_TYPE +%token F16_TYPE +%token F32_TYPE +%token F64_TYPE +%token FF64_TYPE +%token B8_TYPE +%token B16_TYPE +%token B32_TYPE +%token B64_TYPE +%token BB64_TYPE +%token BB128_TYPE +%token PRED_TYPE +%token TEXREF_TYPE +%token SAMPLERREF_TYPE +%token SURFREF_TYPE +%token V2_TYPE +%token V3_TYPE +%token V4_TYPE +%token COMMA +%token PRED +%token HALF_OPTION +%token EXTP_OPTION +%token EQ_OPTION +%token NE_OPTION +%token LT_OPTION +%token LE_OPTION +%token GT_OPTION +%token GE_OPTION +%token LO_OPTION +%token LS_OPTION +%token HI_OPTION +%token HS_OPTION +%token EQU_OPTION +%token NEU_OPTION +%token LTU_OPTION +%token LEU_OPTION +%token GTU_OPTION +%token GEU_OPTION +%token NUM_OPTION +%token NAN_OPTION +%token CF_OPTION +%token SF_OPTION +%token NSF_OPTION +%token LEFT_SQUARE_BRACKET +%token RIGHT_SQUARE_BRACKET +%token WIDE_OPTION +%token <int_value> SPECIAL_REGISTER +%token MINUS +%token PLUS +%token COLON +%token SEMI_COLON +%token EXCLAMATION +%token PIPE +%token RIGHT_BRACE +%token LEFT_BRACE +%token EQUALS +%token PERIOD +%token BACKSLASH +%token <int_value> DIMENSION_MODIFIER +%token RN_OPTION +%token RZ_OPTION +%token RM_OPTION +%token RP_OPTION +%token RNI_OPTION +%token RZI_OPTION +%token RMI_OPTION +%token RPI_OPTION +%token UNI_OPTION +%token GEOM_MODIFIER_1D +%token GEOM_MODIFIER_2D +%token GEOM_MODIFIER_3D +%token SAT_OPTION +%token FTZ_OPTION +%token NEG_OPTION +%token SYNC_OPTION +%token RED_OPTION +%token ARRIVE_OPTION +%token ATOMIC_POPC +%token ATOMIC_AND +%token ATOMIC_OR +%token ATOMIC_XOR +%token ATOMIC_CAS +%token ATOMIC_EXCH +%token ATOMIC_ADD +%token ATOMIC_INC +%token ATOMIC_DEC +%token ATOMIC_MIN +%token ATOMIC_MAX +%token LEFT_ANGLE_BRACKET +%token RIGHT_ANGLE_BRACKET +%token LEFT_PAREN +%token RIGHT_PAREN +%token APPROX_OPTION +%token FULL_OPTION +%token ANY_OPTION +%token ALL_OPTION +%token BALLOT_OPTION +%token GLOBAL_OPTION +%token CTA_OPTION +%token SYS_OPTION +%token EXIT_OPTION +%token ABS_OPTION +%token TO_OPTION +%token CA_OPTION; +%token CG_OPTION; +%token CS_OPTION; +%token LU_OPTION; +%token CV_OPTION; +%token WB_OPTION; +%token WT_OPTION; +%token NC_OPTION; +%token UP_OPTION; +%token DOWN_OPTION; +%token BFLY_OPTION; +%token IDX_OPTION; +%token PRMT_F4E_MODE; +%token PRMT_B4E_MODE; +%token PRMT_RC8_MODE; +%token PRMT_RC16_MODE; +%token PRMT_ECL_MODE; +%token PRMT_ECR_MODE; + +%type <int_value> function_decl_header +%type <ptr_value> function_decl + +%{ + #include "ptx_parser.h" + #include <stdlib.h> + #include <string.h> + #include <math.h> + void syntax_not_implemented(); + extern int g_func_decl; + int ptx_lex(void); + int ptx_error(const char *); +%} + +%% + +input: /* empty */ + | input directive_statement + | input function_defn + | input function_decl + ; + +function_defn: function_decl { set_symtab($1); func_header(".skip"); } statement_block { end_function(); } + | function_decl { set_symtab($1); } block_spec_list { func_header(".skip"); } statement_block { end_function(); } + ; + +block_spec: MAXNTID_DIRECTIVE INT_OPERAND COMMA INT_OPERAND COMMA INT_OPERAND {func_header_info_int(".maxntid", $2); + func_header_info_int(",", $4); + func_header_info_int(",", $6); + maxnt_id($2, $4, $6);} + | MINNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".minnctapersm", $2); printf("GPGPU-Sim: Warning: .minnctapersm ignored. \n"); } + | MAXNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".maxnctapersm", $2); printf("GPGPU-Sim: Warning: .maxnctapersm ignored. \n"); } + ; + +block_spec_list: block_spec + | block_spec_list block_spec + ; + +function_decl: function_decl_header LEFT_PAREN { start_function($1); func_header_info("(");} param_entry RIGHT_PAREN {func_header_info(")");} function_ident_param { $$ = reset_symtab(); } + | function_decl_header { start_function($1); } function_ident_param { $$ = reset_symtab(); } + | function_decl_header { start_function($1); add_function_name(""); g_func_decl=0; $$ = reset_symtab(); } + ; + +function_ident_param: IDENTIFIER { add_function_name($1); } LEFT_PAREN {func_header_info("(");} param_list RIGHT_PAREN { g_func_decl=0; func_header_info(")"); } + | IDENTIFIER { add_function_name($1); g_func_decl=0; } + ; + +function_decl_header: ENTRY_DIRECTIVE { $$ = 1; g_func_decl=1; func_header(".entry"); } + | VISIBLE_DIRECTIVE ENTRY_DIRECTIVE { $$ = 1; g_func_decl=1; func_header(".entry"); } + | WEAK_DIRECTIVE ENTRY_DIRECTIVE { $$ = 1; g_func_decl=1; func_header(".entry"); } + | FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } + | VISIBLE_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } + | WEAK_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } + | EXTERN_DIRECTIVE FUNC_DIRECTIVE { $$ = 2; g_func_decl=1; func_header(".func"); } + | WEAK_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } + ; + +param_list: /*empty*/ + | param_entry { add_directive(); } + | param_list COMMA {func_header_info(",");} param_entry { add_directive(); } + +param_entry: PARAM_DIRECTIVE { add_space_spec(param_space_unclassified,0); } variable_spec ptr_spec identifier_spec { add_function_arg(); } + | REG_DIRECTIVE { add_space_spec(reg_space,0); } variable_spec identifier_spec { add_function_arg(); } + +ptr_spec: /*empty*/ + | PTR_DIRECTIVE ptr_space_spec ptr_align_spec + | PTR_DIRECTIVE ptr_align_spec + +ptr_space_spec: GLOBAL_DIRECTIVE { add_ptr_spec(global_space); } + | LOCAL_DIRECTIVE { add_ptr_spec(local_space); } + | SHARED_DIRECTIVE { add_ptr_spec(shared_space); } + | CONST_DIRECTIVE { add_ptr_spec(global_space); } + +ptr_align_spec: ALIGN_DIRECTIVE INT_OPERAND + +statement_block: LEFT_BRACE statement_list RIGHT_BRACE + +statement_list: directive_statement { add_directive(); } + | instruction_statement { add_instruction(); } + | statement_list directive_statement { add_directive(); } + | statement_list instruction_statement { add_instruction(); } + | statement_list {start_inst_group();} statement_block {end_inst_group();} + | {start_inst_group();} statement_block {end_inst_group();} + ; + +directive_statement: variable_declaration SEMI_COLON + | VERSION_DIRECTIVE DOUBLE_OPERAND { add_version_info($2, 0); } + | VERSION_DIRECTIVE DOUBLE_OPERAND PLUS { add_version_info($2,1); } + | ADDRESS_SIZE_DIRECTIVE INT_OPERAND {/*Do nothing*/} + | TARGET_DIRECTIVE IDENTIFIER COMMA IDENTIFIER { target_header2($2,$4); } + | TARGET_DIRECTIVE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER { target_header3($2,$4,$6); } + | TARGET_DIRECTIVE IDENTIFIER { target_header($2); } + | FILE_DIRECTIVE INT_OPERAND STRING { add_file($2,$3); } + | FILE_DIRECTIVE INT_OPERAND STRING COMMA INT_OPERAND COMMA INT_OPERAND { add_file($2,$3); } + | LOC_DIRECTIVE INT_OPERAND INT_OPERAND INT_OPERAND + | PRAGMA_DIRECTIVE STRING SEMI_COLON { add_pragma($2); } + | function_decl SEMI_COLON {/*Do nothing*/} + ; + +variable_declaration: variable_spec identifier_list { add_variables(); } + | variable_spec identifier_spec EQUALS initializer_list { add_variables(); } + | variable_spec identifier_spec EQUALS literal_operand { add_variables(); } + | CONSTPTR_DIRECTIVE IDENTIFIER COMMA IDENTIFIER COMMA INT_OPERAND { add_constptr($2, $4, $6); } + ; + +variable_spec: var_spec_list { set_variable_type(); } + +identifier_list: identifier_spec + | identifier_list COMMA identifier_spec; + +identifier_spec: IDENTIFIER { add_identifier($1,0,NON_ARRAY_IDENTIFIER); func_header_info($1);} + | IDENTIFIER LEFT_ANGLE_BRACKET INT_OPERAND RIGHT_ANGLE_BRACKET { func_header_info($1); func_header_info_int("<", $3); func_header_info(">"); + int i,lbase,l; + char *id = NULL; + lbase = strlen($1); + for( i=0; i < $3; i++ ) { + l = lbase + (int)log10(i+1)+10; + id = (char*) malloc(l); + snprintf(id,l,"%s%u",$1,i); + add_identifier(id,0,NON_ARRAY_IDENTIFIER); + } + free($1); + } + | IDENTIFIER LEFT_SQUARE_BRACKET RIGHT_SQUARE_BRACKET { add_identifier($1,0,ARRAY_IDENTIFIER_NO_DIM); func_header_info($1); func_header_info("["); func_header_info("]");} + | IDENTIFIER LEFT_SQUARE_BRACKET INT_OPERAND RIGHT_SQUARE_BRACKET { add_identifier($1,$3,ARRAY_IDENTIFIER); func_header_info($1); func_header_info_int("[",$3); func_header_info("]");} + ; + +var_spec_list: var_spec + | var_spec_list var_spec; + +var_spec: space_spec + | type_spec + | align_spec + | VISIBLE_DIRECTIVE + | EXTERN_DIRECTIVE { add_extern_spec(); } + | WEAK_DIRECTIVE + ; + +align_spec: ALIGN_DIRECTIVE INT_OPERAND { add_alignment_spec($2); } + +space_spec: REG_DIRECTIVE { add_space_spec(reg_space,0); } + | SREG_DIRECTIVE { add_space_spec(reg_space,0); } + | addressable_spec + ; + +addressable_spec: CONST_DIRECTIVE { add_space_spec(const_space,$1); } + | GLOBAL_DIRECTIVE { add_space_spec(global_space,0); } + | LOCAL_DIRECTIVE { add_space_spec(local_space,0); } + | PARAM_DIRECTIVE { add_space_spec(param_space_unclassified,0); } + | SHARED_DIRECTIVE { add_space_spec(shared_space,0); } + | SSTARR_DIRECTIVE { add_space_spec(sstarr_space,0); } + | SURF_DIRECTIVE { add_space_spec(surf_space,0); } + | TEX_DIRECTIVE { add_space_spec(tex_space,0); } + ; + +type_spec: scalar_type + | vector_spec scalar_type + ; + +vector_spec: V2_TYPE { add_option(V2_TYPE); func_header_info(".v2");} + | V3_TYPE { add_option(V3_TYPE); func_header_info(".v3");} + | V4_TYPE { add_option(V4_TYPE); func_header_info(".v4");} + ; + +scalar_type: S8_TYPE { add_scalar_type_spec( S8_TYPE ); } + | S16_TYPE { add_scalar_type_spec( S16_TYPE ); } + | S32_TYPE { add_scalar_type_spec( S32_TYPE ); } + | S64_TYPE { add_scalar_type_spec( S64_TYPE ); } + | U8_TYPE { add_scalar_type_spec( U8_TYPE ); } + | U16_TYPE { add_scalar_type_spec( U16_TYPE ); } + | U32_TYPE { add_scalar_type_spec( U32_TYPE ); } + | U64_TYPE { add_scalar_type_spec( U64_TYPE ); } + | F16_TYPE { add_scalar_type_spec( F16_TYPE ); } + | F32_TYPE { add_scalar_type_spec( F32_TYPE ); } + | F64_TYPE { add_scalar_type_spec( F64_TYPE ); } + | FF64_TYPE { add_scalar_type_spec( FF64_TYPE ); } + | B8_TYPE { add_scalar_type_spec( B8_TYPE ); } + | B16_TYPE { add_scalar_type_spec( B16_TYPE ); } + | B32_TYPE { add_scalar_type_spec( B32_TYPE ); } + | B64_TYPE { add_scalar_type_spec( B64_TYPE ); } + | BB64_TYPE { add_scalar_type_spec( BB64_TYPE ); } + | BB128_TYPE { add_scalar_type_spec( BB128_TYPE ); } + | PRED_TYPE { add_scalar_type_spec( PRED_TYPE ); } + | TEXREF_TYPE { add_scalar_type_spec( TEXREF_TYPE ); } + | SAMPLERREF_TYPE { add_scalar_type_spec( SAMPLERREF_TYPE ); } + | SURFREF_TYPE { add_scalar_type_spec( SURFREF_TYPE ); } + ; + +initializer_list: LEFT_BRACE literal_list RIGHT_BRACE { add_array_initializer(); } + | LEFT_BRACE initializer_list RIGHT_BRACE { syntax_not_implemented(); } + +literal_list: literal_operand + | literal_list COMMA literal_operand; + +instruction_statement: instruction SEMI_COLON + | IDENTIFIER COLON { add_label($1); } + | pred_spec instruction SEMI_COLON; + +instruction: opcode_spec LEFT_PAREN operand RIGHT_PAREN { set_return(); } COMMA operand COMMA LEFT_PAREN operand_list RIGHT_PAREN + | opcode_spec operand COMMA LEFT_PAREN operand_list RIGHT_PAREN + | opcode_spec operand COMMA LEFT_PAREN RIGHT_PAREN + | opcode_spec operand_list + | opcode_spec + ; + +opcode_spec: OPCODE { add_opcode($1); } option_list + | OPCODE { add_opcode($1); } + +pred_spec: PRED IDENTIFIER { add_pred($2,0, -1); } + | PRED EXCLAMATION IDENTIFIER { add_pred($3,1, -1); } + | PRED IDENTIFIER LT_OPTION { add_pred($2,0,1); } + | PRED IDENTIFIER EQ_OPTION { add_pred($2,0,2); } + | PRED IDENTIFIER LE_OPTION { add_pred($2,0,3); } + | PRED IDENTIFIER NE_OPTION { add_pred($2,0,5); } + | PRED IDENTIFIER GE_OPTION { add_pred($2,0,6); } + | PRED IDENTIFIER EQU_OPTION { add_pred($2,0,10); } + | PRED IDENTIFIER GTU_OPTION { add_pred($2,0,12); } + | PRED IDENTIFIER NEU_OPTION { add_pred($2,0,13); } + | PRED IDENTIFIER CF_OPTION { add_pred($2,0,17); } + | PRED IDENTIFIER SF_OPTION { add_pred($2,0,19); } + | PRED IDENTIFIER NSF_OPTION { add_pred($2,0,28); } + ; + +option_list: option + | option option_list ; + +option: type_spec + | compare_spec + | addressable_spec + | rounding_mode + | wmma_spec + | prmt_spec + | SYNC_OPTION { add_option(SYNC_OPTION); } + | ARRIVE_OPTION { add_option(ARRIVE_OPTION); } + | RED_OPTION { add_option(RED_OPTION); } + | UNI_OPTION { add_option(UNI_OPTION); } + | WIDE_OPTION { add_option(WIDE_OPTION); } + | ANY_OPTION { add_option(ANY_OPTION); } + | ALL_OPTION { add_option(ALL_OPTION); } + | BALLOT_OPTION { add_option(BALLOT_OPTION); } + | GLOBAL_OPTION { add_option(GLOBAL_OPTION); } + | CTA_OPTION { add_option(CTA_OPTION); } + | SYS_OPTION { add_option(SYS_OPTION); } + | GEOM_MODIFIER_1D { add_option(GEOM_MODIFIER_1D); } + | GEOM_MODIFIER_2D { add_option(GEOM_MODIFIER_2D); } + | GEOM_MODIFIER_3D { add_option(GEOM_MODIFIER_3D); } + | SAT_OPTION { add_option(SAT_OPTION); } + | FTZ_OPTION { add_option(FTZ_OPTION); } + | NEG_OPTION { add_option(NEG_OPTION); } + | APPROX_OPTION { add_option(APPROX_OPTION); } + | FULL_OPTION { add_option(FULL_OPTION); } + | EXIT_OPTION { add_option(EXIT_OPTION); } + | ABS_OPTION { add_option(ABS_OPTION); } + | atomic_operation_spec ; + | TO_OPTION { add_option(TO_OPTION); } + | HALF_OPTION { add_option(HALF_OPTION); } + | EXTP_OPTION { add_option(EXTP_OPTION); } + | CA_OPTION { add_option(CA_OPTION); } + | CG_OPTION { add_option(CG_OPTION); } + | CS_OPTION { add_option(CS_OPTION); } + | LU_OPTION { add_option(LU_OPTION); } + | CV_OPTION { add_option(CV_OPTION); } + | WB_OPTION { add_option(WB_OPTION); } + | WT_OPTION { add_option(WT_OPTION); } + | NC_OPTION { add_option(NC_OPTION); } + | UP_OPTION { add_option(UP_OPTION); } + | DOWN_OPTION { add_option(DOWN_OPTION); } + | BFLY_OPTION { add_option(BFLY_OPTION); } + | IDX_OPTION { add_option(IDX_OPTION); } + ; + +atomic_operation_spec: ATOMIC_AND { add_option(ATOMIC_AND); } + | ATOMIC_POPC { add_option(ATOMIC_POPC); } + | ATOMIC_OR { add_option(ATOMIC_OR); } + | ATOMIC_XOR { add_option(ATOMIC_XOR); } + | ATOMIC_CAS { add_option(ATOMIC_CAS); } + | ATOMIC_EXCH { add_option(ATOMIC_EXCH); } + | ATOMIC_ADD { add_option(ATOMIC_ADD); } + | ATOMIC_INC { add_option(ATOMIC_INC); } + | ATOMIC_DEC { add_option(ATOMIC_DEC); } + | ATOMIC_MIN { add_option(ATOMIC_MIN); } + | ATOMIC_MAX { add_option(ATOMIC_MAX); } + ; + +rounding_mode: floating_point_rounding_mode + | integer_rounding_mode; + + +floating_point_rounding_mode: RN_OPTION { add_option(RN_OPTION); } + | RZ_OPTION { add_option(RZ_OPTION); } + | RM_OPTION { add_option(RM_OPTION); } + | RP_OPTION { add_option(RP_OPTION); } + ; + +integer_rounding_mode: RNI_OPTION { add_option(RNI_OPTION); } + | RZI_OPTION { add_option(RZI_OPTION); } + | RMI_OPTION { add_option(RMI_OPTION); } + | RPI_OPTION { add_option(RPI_OPTION); } + ; + +compare_spec:EQ_OPTION { add_option(EQ_OPTION); } + | NE_OPTION { add_option(NE_OPTION); } + | LT_OPTION { add_option(LT_OPTION); } + | LE_OPTION { add_option(LE_OPTION); } + | GT_OPTION { add_option(GT_OPTION); } + | GE_OPTION { add_option(GE_OPTION); } + | LO_OPTION { add_option(LO_OPTION); } + | LS_OPTION { add_option(LS_OPTION); } + | HI_OPTION { add_option(HI_OPTION); } + | HS_OPTION { add_option(HS_OPTION); } + | EQU_OPTION { add_option(EQU_OPTION); } + | NEU_OPTION { add_option(NEU_OPTION); } + | LTU_OPTION { add_option(LTU_OPTION); } + | LEU_OPTION { add_option(LEU_OPTION); } + | GTU_OPTION { add_option(GTU_OPTION); } + | GEU_OPTION { add_option(GEU_OPTION); } + | NUM_OPTION { add_option(NUM_OPTION); } + | NAN_OPTION { add_option(NAN_OPTION); } + ; + +prmt_spec: PRMT_F4E_MODE { add_option( PRMT_F4E_MODE); } + | PRMT_B4E_MODE { add_option( PRMT_B4E_MODE); } + | PRMT_RC8_MODE { add_option( PRMT_RC8_MODE); } + | PRMT_RC16_MODE{ add_option( PRMT_RC16_MODE);} + | PRMT_ECL_MODE { add_option( PRMT_ECL_MODE); } + | PRMT_ECR_MODE { add_option( PRMT_ECR_MODE); } + ; + +wmma_spec: WMMA_DIRECTIVE LAYOUT CONFIGURATION{add_space_spec(global_space,0);add_ptr_spec(global_space); add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);} + | WMMA_DIRECTIVE LAYOUT LAYOUT CONFIGURATION{add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);add_wmma_option($4);} + ; + +vp_spec: WMMA_DIRECTIVE LAYOUT CONFIGURATION{add_space_spec(global_space,0);add_ptr_spec(global_space);add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);} + | WMMA_DIRECTIVE LAYOUT LAYOUT CONFIGURATION{add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);add_wmma_option($4);} + ; + + + +operand_list: operand + | operand COMMA operand_list; + +operand: IDENTIFIER { add_scalar_operand( $1 ); } + | EXCLAMATION IDENTIFIER { add_neg_pred_operand( $2 ); } + | MINUS IDENTIFIER { add_scalar_operand( $2 ); change_operand_neg(); } + | memory_operand + | literal_operand + | builtin_operand + | vector_operand + | MINUS vector_operand { change_operand_neg(); } + | tex_operand + | IDENTIFIER PLUS INT_OPERAND { add_address_operand($1,$3); } + | IDENTIFIER LO_OPTION { add_scalar_operand( $1 ); change_operand_lohi(1);} + | MINUS IDENTIFIER LO_OPTION { add_scalar_operand( $2 ); change_operand_lohi(1); change_operand_neg();} + | IDENTIFIER HI_OPTION { add_scalar_operand( $1 ); change_operand_lohi(2);} + | MINUS IDENTIFIER HI_OPTION { add_scalar_operand( $2 ); change_operand_lohi(2); change_operand_neg();} + | IDENTIFIER PIPE IDENTIFIER { add_2vector_operand($1,$3); change_double_operand_type(-1);} + | IDENTIFIER PIPE IDENTIFIER LO_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-1); change_operand_lohi(1);} + | IDENTIFIER PIPE IDENTIFIER HI_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-1); change_operand_lohi(2);} + | IDENTIFIER BACKSLASH IDENTIFIER { add_2vector_operand($1,$3); change_double_operand_type(-3);} + | IDENTIFIER BACKSLASH IDENTIFIER LO_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-3); change_operand_lohi(1);} + | IDENTIFIER BACKSLASH IDENTIFIER HI_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-3); change_operand_lohi(2);} + ; + +vector_operand: LEFT_BRACE IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_2vector_operand($2,$4); } + | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_3vector_operand($2,$4,$6); } + | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_4vector_operand($2,$4,$6,$8); } + | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_8vector_operand($2,$4,$6,$8,$10,$12,$14,$16); } + | LEFT_BRACE IDENTIFIER RIGHT_BRACE { add_1vector_operand($2); } + ; + +tex_operand: LEFT_SQUARE_BRACKET IDENTIFIER COMMA { add_scalar_operand($2); } + vector_operand + RIGHT_SQUARE_BRACKET + ; + +builtin_operand: SPECIAL_REGISTER DIMENSION_MODIFIER { add_builtin_operand($1,$2); } + | SPECIAL_REGISTER { add_builtin_operand($1,-1); } + ; + +memory_operand : LEFT_SQUARE_BRACKET address_expression RIGHT_SQUARE_BRACKET { add_memory_operand(); } + | IDENTIFIER LEFT_SQUARE_BRACKET address_expression RIGHT_SQUARE_BRACKET { add_memory_operand(); change_memory_addr_space($1); } + | IDENTIFIER LEFT_SQUARE_BRACKET literal_operand RIGHT_SQUARE_BRACKET { change_memory_addr_space($1); } + | IDENTIFIER LEFT_SQUARE_BRACKET twin_operand RIGHT_SQUARE_BRACKET { change_memory_addr_space($1); add_memory_operand();} + | MINUS memory_operand { change_operand_neg(); } + ; + +twin_operand : IDENTIFIER PLUS IDENTIFIER { add_double_operand($1,$3); change_double_operand_type(1); } + | IDENTIFIER PLUS IDENTIFIER LO_OPTION { add_double_operand($1,$3); change_double_operand_type(1); change_operand_lohi(1); } + | IDENTIFIER PLUS IDENTIFIER HI_OPTION { add_double_operand($1,$3); change_double_operand_type(1); change_operand_lohi(2); } + | IDENTIFIER PLUS EQUALS IDENTIFIER { add_double_operand($1,$4); change_double_operand_type(2); } + | IDENTIFIER PLUS EQUALS IDENTIFIER LO_OPTION { add_double_operand($1,$4); change_double_operand_type(2); change_operand_lohi(1); } + | IDENTIFIER PLUS EQUALS IDENTIFIER HI_OPTION { add_double_operand($1,$4); change_double_operand_type(2); change_operand_lohi(2); } + | IDENTIFIER PLUS EQUALS INT_OPERAND { add_address_operand($1,$4); change_double_operand_type(3); } + ; + +literal_operand : INT_OPERAND { add_literal_int($1); } + | FLOAT_OPERAND { add_literal_float($1); } + | DOUBLE_OPERAND { add_literal_double($1); } + ; + +address_expression: IDENTIFIER { add_address_operand($1,0); } + | IDENTIFIER LO_OPTION { add_address_operand($1,0); change_operand_lohi(1);} + | IDENTIFIER HI_OPTION { add_address_operand($1,0); change_operand_lohi(2); } + | IDENTIFIER PLUS INT_OPERAND { add_address_operand($1,$3); } + | INT_OPERAND { add_address_operand2($1); } + ; + +%% + +extern int ptx_lineno; +extern const char *g_filename; + +void syntax_not_implemented() +{ + printf("Parse error (%s:%u): this syntax is not (yet) implemented:\n",g_filename,ptx_lineno); + ptx_error(NULL); + abort(); +} diff --git a/diff.diff b/diff.diff deleted file mode 100644 index 8ab27fe..0000000 --- a/diff.diff +++ /dev/null @@ -1,173 +0,0 @@ -diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc -index 5964a4d..53436d3 100644 ---- a/libcuda/cuda_runtime_api.cc -+++ b/libcuda/cuda_runtime_api.cc -@@ -182,6 +182,8 @@ cudaError_t g_last_cudaError = cudaSuccess; - - extern stream_manager *g_stream_manager; - -+ -+ - void register_ptx_function( const char *name, function_info *impl ) - { - // no longer need this -@@ -1341,6 +1343,29 @@ std::string get_app_binary(){ - return self_exe_path; - } - -+static int get_app_cuda_version() { -+ int app_cuda_version = 0; -+ char fname[1024]; -+ snprintf(fname,1024,"_app_cuda_version_XXXXXX"); -+ int fd=mkstemp(fname); -+ close(fd); -+ std::string app_cuda_version_command = "ldd " + get_app_binary() + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + fname; -+ system(app_cuda_version_command.c_str()); -+ FILE * cmd = fopen(fname, "r"); -+ char buf[256]; -+ while (fgets(buf, sizeof(buf), cmd) != 0) { -+ std::cout << buf; -+ app_cuda_version = atoi(buf); -+ } -+ fclose(cmd); -+ if ( app_cuda_version == 0 ) { -+ printf( "Error - Cannot detect the app's CUDA version.\n" ); -+ exit(1); -+ } -+ return app_cuda_version; -+} -+ -+ - //! Call cuobjdump to extract everything (-elf -sass -ptx) - /*! - * This Function extract the whole PTX (for all the files) using cuobjdump -@@ -1366,23 +1391,10 @@ void extract_code_using_cuobjdump(){ - // Running cuobjdump using dynamic link to current process - // Needs the option '-all' to extract PTX from CDP-enabled binary - extern bool g_cdp_enabled; -- const int current_cuda = atoi(getenv("CUDA_VERSION_NUMBER")); -- if ( context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() && current_cuda > 4020 ) { -- const char* cuda_42_path = getenv("CUDA_42_INSTALL_PATH"); -- if ( NULL == cuda_42_path ) { -- printf("You are using a new version of CUDA and PTXPLUS. You are required to have CUDA SDK 4.2 and explicitly " -- "point to it via the environment variable $CUDA_42_INSTALL_PATH.\n\n"); -- exit(1); -- } else { -- command = "$CUDA_42_INSTALL_PATH/bin/cuobjdump"; -- } -- } else { -- command = "$CUDA_42_INSTALL_PATH/bin/cuobjdump"; -- } - if(!g_cdp_enabled) -- command += " -ptx -elf -sass " + app_binary + " > " + fname; -+ command = "$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass " + app_binary + " > " + fname; - else -- command += " -ptx -elf -sass -all " + app_binary + " > " + fname; -+ command = "$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass -all " + app_binary + " > " + fname; - bool parse_output = true; - int result = system(command.c_str()); - if(result) { -@@ -1765,6 +1777,7 @@ void cuobjdumpParseBinary(unsigned int handle){ - //TODO: Remove temporarily files as per configurations - } - -+ - void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) - { - #if (CUDART_VERSION < 2010) -@@ -1783,26 +1796,10 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) - // Making this a runtime variable based on the app, enables GPGPU-Sim compiled - // with a newer version of CUDA to run apps compiled with older versions of - // CUDA. This is especially useful for PTXPLUS execution. -- char fname[1024]; -- snprintf(fname,1024,"_app_cuda_version_XXXXXX"); -- int fd=mkstemp(fname); -- close(fd); -- int app_cuda_version = 0; -- std::string app_cuda_version_command = "ldd " + get_app_binary() + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + fname; -- system(app_cuda_version_command.c_str()); -- FILE * cmd = fopen(fname, "r"); -- char buf[256]; -- while (fgets(buf, sizeof(buf), cmd) != 0) { -- std::cout << buf; -- app_cuda_version = atoi(buf); -- } -- fclose(cmd); -- if ( app_cuda_version == 0 ) { -- printf( "Error - Cannot detect the app's CUDA version.\n" ); -- exit(1); -- } -+ int app_cuda_version = get_app_cuda_version(); -+ assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); - const char* filename; -- if ( app_cuda_version < 6 ) { -+#if CUDART_VERSION < 6000 - // FatBin handle from the .fatbin.c file (one of the intermediate files generated by NVCC) - typedef struct {int m; int v; const unsigned long long* d; char* f;} __fatDeviceText __attribute__ ((aligned (8))); - __fatDeviceText * fatDeviceText = (__fatDeviceText *) fatCubin; -@@ -1814,9 +1811,9 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) - char * pfatbin = (char*) fatDeviceText->d; - int offset = *((int*)(pfatbin+48)); - filename = (pfatbin+16+offset); -- } else { -+#else - filename = "default"; -- } -+#endif - - // The extracted file name is associated with a fat_cubin_handle passed - // into cudaLaunch(). Inside cudaLaunch(), the associated file name is -diff --git a/setup_environment b/setup_environment -index 8bcb9aa..e5cd4fa 100644 ---- a/setup_environment -+++ b/setup_environment -@@ -134,18 +134,26 @@ else - echo "configured without a power model."; - fi - -+export PTXAS_CUDA_INSTALL_PATH=$CUDA_INSTALL_PATH -+echo ""; -+echo "----------------------------------------------------------------------------"; -+echo "WARNING - If you only care about PTX execution, ignore this warning. GPGPU-Sim supports PTX execution in modern CUDA." - if [ $CUDA_VERSION_NUMBER -gt 4200 ]; then -- echo ""; -- echo "----------------------------------------------------------------------------"; -- echo "WARNING - If you only care about PTX execution, ignore this warning. GPGPU-Sim supports PTX execution in modern CUDA." -- echo "If you want to run PTXPLUS, download SDK 4.2 (in addition to your modern CUDA) and set \$CUDA_42_INSTALL_PATH." -- echo "The following text describes why:"; -- echo "If you are using PTXPLUS, only sm_1x is supported and it requires that the app binaries are compiled in CUDA 4.2 or less."; -- echo "New versions of CUDA tools have dropped parsing support for sm_1x - therefore to use PTXPLUS with a modern card configuration, you must"; -- echo "have both CUDA 4.2 and a modern CUDA installed. When running PTXPLUS under a CUDA version >= 4.2, the simulator will fail unless \$CUDA_42_INSTALL_PATH"; -- echo "has been set properly. You must set this manually."; -- echo "----------------------------------------------------------------------------" -+ echo "If you want to run PTXPLUS (sm_1x SASS) with a modern card configuration, the apps and simulator must be compiled with CUDA 4.2." -+ echo "You can still run a PASCAL configuration when compiling with 4.2 by setting the \$PTXAS_CUDA_INSTALL_PATH directory environment variable." -+else -+ echo "If you want to run PTXPLUS (sm_1x SASS) with a modern card configuration - set the envronment variable" -+ echo "\$PTXAS_CUDA_INSTALL_PATH to point a CUDA version compabible with your card configurations (i.e. 8+ for PASCAL, 9+ for VOLTA etc..)" -+ echo "For example: \"export \$PTXAS_CUDA_INSTALL_PATH=/usr/local/cuda-9.1\"" - fi -+echo "The following text describes why:"; -+echo "If you are using PTXPLUS, only sm_1x is supported and it requires that the app and simulator binaries are compiled in CUDA 4.2 or less."; -+echo "The simulator requires it since CUDA headers desribe struct sizes in the exec which change from gen to gen."; -+echo "The apps require 4.2 because new versions of CUDA tools have dropped parsing support for generating sm_1x"; -+echo "When running using modern config (i.e. pascal) and PTXPLUS with CUDA 4.2, the \$PTXAS_CUDA_INSTALL_PATH env variable is required to get proper register usage" -+echo "(and hence occupancy) using a version of CUDA that knows the register usage on the real card." -+echo ""; -+echo "----------------------------------------------------------------------------"; - - echo "setup_environment succeeded"; - -diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc -index 22f03b9..d8e04d4 100644 ---- a/src/cuda-sim/ptx_loader.cc -+++ b/src/cuda-sim/ptx_loader.cc -@@ -339,7 +339,7 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num - snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",g_occupancy_sm_number); - #endif - -- snprintf(commandline,1024,"$CUDA_INSTALL_PATH/bin/ptxas %s -v %s --output-file /dev/null 2> %s", -+ snprintf(commandline,1024,"$PTXAS_CUDA_INSTALL_PATH/bin/ptxas %s -v %s --output-file /dev/null 2> %s", - extra_flags, fname2, tempfile_ptxinfo); - printf("GPGPU-Sim PTX: generating ptxinfo using \"%s\"\n", commandline); - result = system(commandline); diff --git a/groovy-html.nightly.success.template b/groovy-html.nightly.success.template new file mode 100644 index 0000000..210b924 --- /dev/null +++ b/groovy-html.nightly.success.template @@ -0,0 +1,74 @@ +<STYLE> + BODY, TABLE, TD, TH, P { + font-family: Courier New, Verdana, Helvetica, sans serif; + font-size: 10px; + color: black; + } + .console { + font-family: Courier New; + } + .filesChanged { + width: 10%; + padding-left: 10px; + } + .section { + width: 100%; + border: thin black dotted; + } + .td-title-main { + color: white; + font-size: 200%; + padding-left: 5px; + font-weight: bold; + } + .td-title { + color: white; + font-size: 120%; + font-weight: bold; + padding-left: 5px; + text-transform: uppercase; + } + .td-title-tests { + font-weight: bold; + font-size: 120%; + } + .td-header-maven-module { + font-weight: bold; + font-size: 120%; + } + .td-maven-artifact { + padding-left: 5px; + } + .tr-title { + background-color: <%= (build.result == null || build.result.toString() == 'SUCCESS') ? '#27AE60' : build.result.toString() == 'FAILURE' ? '#E74C3C' : '#f4e242' %>; + } + .test { + padding-left: 20px; + } + .test-fixed { + color: #27AE60; + } + .test-failed { + color: #E74C3C; + } +</STYLE> +<BODY> + +<% + import hudson.model.* + + def JOB_NAME= build.getEnvVars()["JOB_NAME"]; + def BUILD_NUMBER= build.getEnvVars()["BUILD_NUMBER"]; + def BUILD_URL= build.getEnvVars()["BUILD_URL"]; +%> + +See ${BUILD_URL} for full build detials. + +<pre>Statistics changes since last commit: +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/ +<pre> Correlation: +<pre> 9.1: +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/titanv-cycles.cuda-9.1.nightly.html +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/titanv-cycles.cuda-9.1.nightly.apps.txt +<pre> Correlation also posted to: https://engineering.purdue.edu/tgrogers/group/correlator.html +</BODY> diff --git a/groovy-html.success.template b/groovy-html.success.template new file mode 100644 index 0000000..974cc61 --- /dev/null +++ b/groovy-html.success.template @@ -0,0 +1,77 @@ +<STYLE> + BODY, TABLE, TD, TH, P { + font-family: Courier New, Verdana, Helvetica, sans serif; + font-size: 10px; + color: black; + } + .console { + font-family: Courier New; + } + .filesChanged { + width: 10%; + padding-left: 10px; + } + .section { + width: 100%; + border: thin black dotted; + } + .td-title-main { + color: white; + font-size: 200%; + padding-left: 5px; + font-weight: bold; + } + .td-title { + color: white; + font-size: 120%; + font-weight: bold; + padding-left: 5px; + text-transform: uppercase; + } + .td-title-tests { + font-weight: bold; + font-size: 120%; + } + .td-header-maven-module { + font-weight: bold; + font-size: 120%; + } + .td-maven-artifact { + padding-left: 5px; + } + .tr-title { + background-color: <%= (build.result == null || build.result.toString() == 'SUCCESS') ? '#27AE60' : build.result.toString() == 'FAILURE' ? '#E74C3C' : '#f4e242' %>; + } + .test { + padding-left: 20px; + } + .test-fixed { + color: #27AE60; + } + .test-failed { + color: #E74C3C; + } +</STYLE> +<BODY> + +<% + import hudson.model.* + + def JOB_NAME= build.getEnvVars()["JOB_NAME"]; + def BUILD_NUMBER= build.getEnvVars()["BUILD_NUMBER"]; + def BUILD_URL= build.getEnvVars()["BUILD_URL"]; +%> + +See ${BUILD_URL} for full build detials. + +<pre>Statistics changes since last commit: +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/ +<pre> Correlation: +<pre> 4.2: +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/titanv-cycles.cuda-4.2.html +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/titanv-cycles.cuda-4.2.apps.txt +<pre> 10.1: +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/titanv-cycles.cuda-10.1.html +<pre> https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/jenkins/${JOB_NAME}/${BUILD_NUMBER}/titanv-cycles.cuda-10.1.apps.txt +<pre> Correlation also posted to: https://engineering.purdue.edu/tgrogers/group/correlator.html +</BODY> diff --git a/libcuda/Makefile b/libcuda/Makefile index 13932e2..c8ff2e3 100644 --- a/libcuda/Makefile +++ b/libcuda/Makefile @@ -111,10 +111,10 @@ $(OUTPUT_DIR)/%.o: %.cc $(CPP) $(CXXFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/%.o: %.c - $(CC) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ + $(CPP) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/%.o: $(OUTPUT_DIR)/%.c - $(CC) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ + $(CPP) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/cuobjdump_parser.c: cuobjdump.y $(YACC) $(YFLAGS) -p cuobjdump_ -o$@ $< --file-prefix=$(OUTPUT_DIR)/cuobjdump diff --git a/libcuda/cuda_api.h b/libcuda/cuda_api.h index 3808e8a..27983b4 100644 --- a/libcuda/cuda_api.h +++ b/libcuda/cuda_api.h @@ -1,5 +1,5 @@ /* - * Copyright 1993-2014 NVIDIA Corporation. All rights reserved. + * Copyright 1993-2018 NVIDIA Corporation. All rights reserved. * * NOTICE TO LICENSEE: * @@ -63,6 +63,16 @@ typedef uint64_t cuuint64_t; /** * CUDA API versioning support */ +#if defined(__CUDA_API_VERSION_INTERNAL) || defined(__DOXYGEN_ONLY__) || defined(CUDA_ENABLE_DEPRECATED) +#define __CUDA_DEPRECATED +#elif defined(_MSC_VER) +#define __CUDA_DEPRECATED __declspec(deprecated) +#elif defined(__GNUC__) +#define __CUDA_DEPRECATED __attribute__((deprecated)) +#else +#define __CUDA_DEPRECATED +#endif + #if defined(CUDA_FORCE_API_VERSION) #if (CUDA_FORCE_API_VERSION == 3010) #define __CUDA_API_VERSION 3010 @@ -70,7 +80,7 @@ typedef uint64_t cuuint64_t; #error "Unsupported value of CUDA_FORCE_API_VERSION" #endif #else - #define __CUDA_API_VERSION 8000 + #define __CUDA_API_VERSION 10010 #endif /* CUDA_FORCE_API_VERSION */ #if defined(__CUDA_API_VERSION_INTERNAL) || defined(CUDA_API_PER_THREAD_DEFAULT_STREAM) @@ -136,9 +146,19 @@ typedef uint64_t cuuint64_t; #define cuTexRefSetAddress2D cuTexRefSetAddress2D_v3 #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 4010 */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 6050 + #define cuLinkCreate cuLinkCreate_v2 + #define cuLinkAddData cuLinkAddData_v2 + #define cuLinkAddFile cuLinkAddFile_v2 +#endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 6050 */ +#if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 6050 #define cuMemHostRegister cuMemHostRegister_v2 #define cuGraphicsResourceSetMapFlags cuGraphicsResourceSetMapFlags_v2 #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 6050 */ +#if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 10010 + #define cuStreamBeginCapture __CUDA_API_PTSZ(cuStreamBeginCapture_v2) +#elif defined(__CUDA_API_PER_THREAD_DEFAULT_STREAM) + #define cuStreamBeginCapture __CUDA_API_PTSZ(cuStreamBeginCapture) +#endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 10010 */ #if !defined(__CUDA_API_VERSION_INTERNAL) #if defined(__CUDA_API_VERSION) && __CUDA_API_VERSION >= 3020 && __CUDA_API_VERSION < 4010 @@ -164,19 +184,33 @@ typedef uint64_t cuuint64_t; #define cuStreamGetPriority __CUDA_API_PTSZ(cuStreamGetPriority) #define cuStreamGetFlags __CUDA_API_PTSZ(cuStreamGetFlags) + #define cuStreamGetCtx __CUDA_API_PTSZ(cuStreamGetCtx) #define cuStreamWaitEvent __CUDA_API_PTSZ(cuStreamWaitEvent) + #define cuStreamEndCapture __CUDA_API_PTSZ(cuStreamEndCapture) + #define cuStreamIsCapturing __CUDA_API_PTSZ(cuStreamIsCapturing) + #define cuStreamGetCaptureInfo __CUDA_API_PTSZ(cuStreamGetCaptureInfo) #define cuStreamAddCallback __CUDA_API_PTSZ(cuStreamAddCallback) #define cuStreamAttachMemAsync __CUDA_API_PTSZ(cuStreamAttachMemAsync) #define cuStreamQuery __CUDA_API_PTSZ(cuStreamQuery) #define cuStreamSynchronize __CUDA_API_PTSZ(cuStreamSynchronize) #define cuEventRecord __CUDA_API_PTSZ(cuEventRecord) #define cuLaunchKernel __CUDA_API_PTSZ(cuLaunchKernel) + #define cuLaunchHostFunc __CUDA_API_PTSZ(cuLaunchHostFunc) #define cuGraphicsMapResources __CUDA_API_PTSZ(cuGraphicsMapResources) #define cuGraphicsUnmapResources __CUDA_API_PTSZ(cuGraphicsUnmapResources) #define cuStreamWriteValue32 __CUDA_API_PTSZ(cuStreamWriteValue32) #define cuStreamWaitValue32 __CUDA_API_PTSZ(cuStreamWaitValue32) + #define cuStreamWriteValue64 __CUDA_API_PTSZ(cuStreamWriteValue64) + #define cuStreamWaitValue64 __CUDA_API_PTSZ(cuStreamWaitValue64) #define cuStreamBatchMemOp __CUDA_API_PTSZ(cuStreamBatchMemOp) + + #define cuLaunchCooperativeKernel __CUDA_API_PTSZ(cuLaunchCooperativeKernel) + + #define cuSignalExternalSemaphoresAsync __CUDA_API_PTSZ(cuSignalExternalSemaphoresAsync) + #define cuWaitExternalSemaphoresAsync __CUDA_API_PTSZ(cuWaitExternalSemaphoresAsync) + + #define cuGraphLaunch __CUDA_API_PTSZ(cuGraphLaunch) #endif /** @@ -200,7 +234,7 @@ typedef uint64_t cuuint64_t; /** * CUDA API version number */ -#define CUDA_VERSION 8000 +#define CUDA_VERSION 10010 #ifdef __cplusplus extern "C" { @@ -209,7 +243,7 @@ extern "C" { /** * CUDA device pointer * CUdeviceptr is defined as an unsigned integer type whose size matches the size of a pointer on the target platform. - */ + */ #if __CUDA_API_VERSION >= 3020 #if defined(_WIN64) || defined(__LP64__) @@ -233,16 +267,23 @@ typedef struct CUstream_st *CUstream; /**< CUDA stream */ typedef struct CUgraphicsResource_st *CUgraphicsResource; /**< CUDA graphics interop resource */ typedef unsigned long long CUtexObject; /**< An opaque value that represents a CUDA texture object */ typedef unsigned long long CUsurfObject; /**< An opaque value that represents a CUDA surface object */ +typedef struct CUextMemory_st *CUexternalMemory; /**< CUDA external memory */ +typedef struct CUextSemaphore_st *CUexternalSemaphore; /**< CUDA external semaphore */ +typedef struct CUgraph_st *CUgraph; /**< CUDA graph */ +typedef struct CUgraphNode_st *CUgraphNode; /**< CUDA graph node */ +typedef struct CUgraphExec_st *CUgraphExec; /**< CUDA executable graph */ +#ifndef CU_UUID_HAS_BEEN_DEFINED +#define CU_UUID_HAS_BEEN_DEFINED typedef struct CUuuid_st { /**< CUDA definition of UUID */ char bytes[16]; } CUuuid; - +#endif #if __CUDA_API_VERSION >= 4010 /** - * CUDA IPC handle size + * CUDA IPC handle size */ #define CU_IPC_HANDLE_SIZE 64 @@ -289,7 +330,7 @@ typedef enum CUctx_flags_enum { CU_CTX_BLOCKING_SYNC = 0x04, /**< Set blocking synchronization as default scheduling * \deprecated This flag was deprecated as of CUDA 4.0 * and was replaced with ::CU_CTX_SCHED_BLOCKING_SYNC. */ - CU_CTX_SCHED_MASK = 0x07, + CU_CTX_SCHED_MASK = 0x07, CU_CTX_MAP_HOST = 0x08, /**< Support mapped pinned allocations */ CU_CTX_LMEM_RESIZE_TO_MAX = 0x10, /**< Keep local memory allocation after launch */ CU_CTX_FLAGS_MASK = 0x1f @@ -335,20 +376,26 @@ typedef enum CUevent_flags_enum { #if __CUDA_API_VERSION >= 8000 /** - * Flags for ::cuStreamWaitValue32 + * Flags for ::cuStreamWaitValue32 and ::cuStreamWaitValue64 */ typedef enum CUstreamWaitValue_flags_enum { - CU_STREAM_WAIT_VALUE_GEQ = 0x0, /**< Wait until (int32_t)(*addr - value) >= 0. Note this is a - cyclic comparison which ignores wraparound. (Default behavior.) */ + CU_STREAM_WAIT_VALUE_GEQ = 0x0, /**< Wait until (int32_t)(*addr - value) >= 0 (or int64_t for 64 bit + values). Note this is a cyclic comparison which ignores wraparound. + (Default behavior.) */ CU_STREAM_WAIT_VALUE_EQ = 0x1, /**< Wait until *addr == value. */ CU_STREAM_WAIT_VALUE_AND = 0x2, /**< Wait until (*addr & value) != 0. */ + CU_STREAM_WAIT_VALUE_NOR = 0x3, /**< Wait until ~(*addr | value) != 0. Support for this operation can be + queried with ::cuDeviceGetAttribute() and + ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR.*/ CU_STREAM_WAIT_VALUE_FLUSH = 1<<30 /**< Follow the wait operation with a flush of outstanding remote writes. This means that, if a remote write operation is guaranteed to have reached the device before the wait can be satisfied, that write is guaranteed to be visible to downstream device work. The device is permitted to reorder remote writes internally. For example, this flag would be required if two remote writes arrive in a defined order, the wait is satisfied by the - second write, and downstream work needs to observe the first write. */ + second write, and downstream work needs to observe the first write. + Support for this operation is restricted to selected platforms and can be + queried with ::CU_DEVICE_ATTRIBUTE_CAN_USE_WAIT_VALUE_FLUSH.*/ } CUstreamWaitValue_flags; /** @@ -370,6 +417,8 @@ typedef enum CUstreamWriteValue_flags_enum { typedef enum CUstreamBatchMemOpType_enum { CU_STREAM_MEM_OP_WAIT_VALUE_32 = 1, /**< Represents a ::cuStreamWaitValue32 operation */ CU_STREAM_MEM_OP_WRITE_VALUE_32 = 2, /**< Represents a ::cuStreamWriteValue32 operation */ + CU_STREAM_MEM_OP_WAIT_VALUE_64 = 4, /**< Represents a ::cuStreamWaitValue64 operation */ + CU_STREAM_MEM_OP_WRITE_VALUE_64 = 5, /**< Represents a ::cuStreamWriteValue64 operation */ CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES = 3 /**< This has the same effect as ::CU_STREAM_WAIT_VALUE_FLUSH, but as a standalone operation. */ } CUstreamBatchMemOpType; @@ -384,7 +433,7 @@ typedef union CUstreamBatchMemOpParams_union { CUdeviceptr address; union { cuuint32_t value; - cuuint64_t pad; + cuuint64_t value64; }; unsigned int flags; CUdeviceptr alias; /**< For driver internal use. Initial value is unimportant. */ @@ -394,7 +443,7 @@ typedef union CUstreamBatchMemOpParams_union { CUdeviceptr address; union { cuuint32_t value; - cuuint64_t pad; + cuuint64_t value64; }; unsigned int flags; CUdeviceptr alias; /**< For driver internal use. Initial value is unimportant. */ @@ -496,7 +545,7 @@ typedef enum CUdevice_attribute_enum { CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE = 38, /**< Size of L2 cache in bytes */ CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR = 39, /**< Maximum resident threads per multiprocessor */ CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT = 40, /**< Number of asynchronous engines */ - CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING = 41, /**< Device shares a unified address space with the host */ + CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING = 41, /**< Device shares a unified address space with the host */ CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH = 42, /**< Maximum 1D layered texture width */ CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS = 43, /**< Maximum layers in a 1D layered texture */ CU_DEVICE_ATTRIBUTE_CAN_TEX2D_GATHER = 44, /**< Deprecated, do not use. */ @@ -530,7 +579,7 @@ typedef enum CUdevice_attribute_enum { CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH = 72, /**< Maximum 2D linear texture pitch in bytes */ CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_WIDTH = 73, /**< Maximum mipmapped 2D texture width */ CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_HEIGHT = 74,/**< Maximum mipmapped 2D texture height */ - CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = 75, /**< Major compute capability version number */ + CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = 75, /**< Major compute capability version number */ CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR = 76, /**< Minor compute capability version number */ CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH = 77, /**< Maximum mipmapped 1D texture width */ CU_DEVICE_ATTRIBUTE_STREAM_PRIORITIES_SUPPORTED = 78, /**< Device supports stream priorities */ @@ -539,7 +588,7 @@ typedef enum CUdevice_attribute_enum { CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR = 81, /**< Maximum shared memory available per multiprocessor in bytes */ CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR = 82, /**< Maximum number of 32-bit registers available per multiprocessor */ CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY = 83, /**< Device can allocate managed memory on this system */ - CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD = 84, /**< Device is on a multi-GPU board */ + CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD = 84, /**< Device is on a multi-GPU board */ CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID = 85, /**< Unique id for a group of devices on the same multi-GPU board */ CU_DEVICE_ATTRIBUTE_HOST_NATIVE_ATOMIC_SUPPORTED = 86, /**< Link between the device and the host supports native atomic operations (this is a placeholder attribute, and is not supported on any current hardware)*/ CU_DEVICE_ATTRIBUTE_SINGLE_TO_DOUBLE_PRECISION_PERF_RATIO = 87, /**< Ratio of single precision performance (in floating-point operations per second) to double precision performance */ @@ -547,6 +596,16 @@ typedef enum CUdevice_attribute_enum { CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS = 89, /**< Device can coherently access managed memory concurrently with the CPU */ CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED = 90, /**< Device supports compute preemption. */ CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM = 91, /**< Device can access host registered memory at the same virtual address as the CPU */ + CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS = 92, /**< ::cuStreamBatchMemOp and related APIs are supported. */ + CU_DEVICE_ATTRIBUTE_CAN_USE_64_BIT_STREAM_MEM_OPS = 93, /**< 64-bit operations are supported in ::cuStreamBatchMemOp and related APIs. */ + CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR = 94, /**< ::CU_STREAM_WAIT_VALUE_NOR is supported. */ + CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH = 95, /**< Device supports launching cooperative kernels via ::cuLaunchCooperativeKernel */ + CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH = 96, /**< Device can participate in cooperative kernels launched via ::cuLaunchCooperativeKernelMultiDevice */ + CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN = 97, /**< Maximum optin shared memory per block */ + CU_DEVICE_ATTRIBUTE_CAN_FLUSH_REMOTE_WRITES = 98, /**< Both the ::CU_STREAM_WAIT_VALUE_FLUSH flag and the ::CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES MemOp are supported on the device. See \ref CUDA_MEMOP for additional details. */ + CU_DEVICE_ATTRIBUTE_HOST_REGISTER_SUPPORTED = 99, /**< Device supports host memory registration via ::cudaHostRegister. */ + CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES = 100, /**< Device accesses pageable memory via the host's page tables. */ + CU_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST = 101, /**< The host can directly access managed memory on the device without migration. */ CU_DEVICE_ATTRIBUTE_MAX } CUdevice_attribute; @@ -577,7 +636,8 @@ typedef enum CUpointer_attribute_enum { CU_POINTER_ATTRIBUTE_P2P_TOKENS = 5, /**< A pair of tokens for use with the nv-p2p.h Linux kernel interface */ CU_POINTER_ATTRIBUTE_SYNC_MEMOPS = 6, /**< Synchronize every synchronous memory operation initiated on this region */ CU_POINTER_ATTRIBUTE_BUFFER_ID = 7, /**< A process-wide unique ID for an allocated memory region*/ - CU_POINTER_ATTRIBUTE_IS_MANAGED = 8 /**< Indicates if the pointer points to managed memory */ + CU_POINTER_ATTRIBUTE_IS_MANAGED = 8, /**< Indicates if the pointer points to managed memory */ + CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL = 9 /**< A device ordinal of a device on which a pointer was allocated or registered */ } CUpointer_attribute; /** @@ -633,11 +693,28 @@ typedef enum CUfunction_attribute_enum { CU_FUNC_ATTRIBUTE_BINARY_VERSION = 6, /** - * The attribute to indicate whether the function has been compiled with + * The attribute to indicate whether the function has been compiled with * user specified option "-Xptxas --dlcm=ca" set . */ CU_FUNC_ATTRIBUTE_CACHE_MODE_CA = 7, + /** + * The maximum size in bytes of dynamically-allocated shared memory that can be used by + * this function. If the user-specified dynamic shared memory size is larger than this + * value, the launch will fail. + * See ::cuFuncSetAttribute + */ + CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES = 8, + + /** + * On devices where the L1 cache and shared memory use the same hardware resources, + * this sets the shared memory carveout preference, in percent of the total shared memory. + * Refer to ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR. + * This is only a hint, and the driver can choose a different ratio if required to execute the function. + * See ::cuFuncSetAttribute + */ + CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT = 9, + CU_FUNC_ATTRIBUTE_MAX } CUfunction_attribute; @@ -661,6 +738,15 @@ typedef enum CUsharedconfig_enum { } CUsharedconfig; /** + * Shared memory carveout configurations. These may be passed to ::cuFuncSetAttribute + */ +typedef enum CUshared_carveout_enum { + CU_SHAREDMEM_CARVEOUT_DEFAULT = -1, /**< No preference for shared memory or L1 (default) */ + CU_SHAREDMEM_CARVEOUT_MAX_SHARED = 100, /**< Prefer maximum available shared memory, minimum L1 cache */ + CU_SHAREDMEM_CARVEOUT_MAX_L1 = 0 /**< Prefer maximum available L1 cache, minimum shared memory */ +} CUshared_carveout; + +/** * Memory types */ typedef enum CUmemorytype_enum { @@ -838,6 +924,37 @@ typedef enum CUjit_option_enum CU_JIT_NEW_SM3X_OPT, CU_JIT_FAST_COMPILE, + /** + * Array of device symbol names that will be relocated to the corresponing + * host addresses stored in ::CU_JIT_GLOBAL_SYMBOL_ADDRESSES.\n + * Must contain ::CU_JIT_GLOBAL_SYMBOL_COUNT entries.\n + * When loding a device module, driver will relocate all encountered + * unresolved symbols to the host addresses.\n + * It is only allowed to register symbols that correspond to unresolved + * global variables.\n + * It is illegal to register the same device symbol at multiple addresses.\n + * Option type: const char **\n + * Applies to: dynamic linker only + */ + CU_JIT_GLOBAL_SYMBOL_NAMES, + + /** + * Array of host addresses that will be used to relocate corresponding + * device symbols stored in ::CU_JIT_GLOBAL_SYMBOL_NAMES.\n + * Must contain ::CU_JIT_GLOBAL_SYMBOL_COUNT entries.\n + * Option type: void **\n + * Applies to: dynamic linker only + */ + CU_JIT_GLOBAL_SYMBOL_ADDRESSES, + + /** + * Number of entries in ::CU_JIT_GLOBAL_SYMBOL_NAMES and + * ::CU_JIT_GLOBAL_SYMBOL_ADDRESSES arrays.\n + * Option type: unsigned int\n + * Applies to: dynamic linker only + */ + CU_JIT_GLOBAL_SYMBOL_COUNT, + CU_JIT_NUM_OPTIONS } CUjit_option; @@ -847,10 +964,6 @@ typedef enum CUjit_option_enum */ typedef enum CUjit_target_enum { - CU_TARGET_COMPUTE_10 = 10, /**< Compute device class 1.0 */ - CU_TARGET_COMPUTE_11 = 11, /**< Compute device class 1.1 */ - CU_TARGET_COMPUTE_12 = 12, /**< Compute device class 1.2 */ - CU_TARGET_COMPUTE_13 = 13, /**< Compute device class 1.3 */ CU_TARGET_COMPUTE_20 = 20, /**< Compute device class 2.0 */ CU_TARGET_COMPUTE_21 = 21, /**< Compute device class 2.1 */ CU_TARGET_COMPUTE_30 = 30, /**< Compute device class 3.0 */ @@ -860,9 +973,12 @@ typedef enum CUjit_target_enum CU_TARGET_COMPUTE_50 = 50, /**< Compute device class 5.0 */ CU_TARGET_COMPUTE_52 = 52, /**< Compute device class 5.2 */ CU_TARGET_COMPUTE_53 = 53, /**< Compute device class 5.3 */ - CU_TARGET_COMPUTE_60 = 60, /**< Compute device class 6.0. This must be removed for CUDA 7.0 toolkit. See bug 1518217. */ - CU_TARGET_COMPUTE_61 = 61, /**< Compute device class 6.1. This must be removed for CUDA 7.0 toolkit.*/ - CU_TARGET_COMPUTE_62 = 62 /**< Compute device class 6.2. This must be removed for CUDA 7.0 toolkit.*/ + CU_TARGET_COMPUTE_60 = 60, /**< Compute device class 6.0.*/ + CU_TARGET_COMPUTE_61 = 61, /**< Compute device class 6.1.*/ + CU_TARGET_COMPUTE_62 = 62, /**< Compute device class 6.2.*/ + CU_TARGET_COMPUTE_70 = 70, /**< Compute device class 7.0.*/ + CU_TARGET_COMPUTE_72 = 72, /**< Compute device class 7.2.*/ + CU_TARGET_COMPUTE_75 = 75 /**< Compute device class 7.5.*/ } CUjit_target; /** @@ -877,7 +993,7 @@ typedef enum CUjit_fallback_enum } CUjit_fallback; /** - * Caching modes for dlcm + * Caching modes for dlcm */ typedef enum CUjit_cacheMode_enum { @@ -969,6 +1085,7 @@ typedef enum CUlimit_enum { CU_LIMIT_MALLOC_HEAP_SIZE = 0x02, /**< GPU malloc heap size */ CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH = 0x03, /**< GPU device runtime launch synchronize depth */ CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT = 0x04, /**< GPU device runtime pending launch count */ + CU_LIMIT_MAX_L2_FETCH_GRANULARITY = 0x05, /**< A value between 0 and 128 that indicates the maximum fetch granularity of L2 (in Bytes). This is a hint */ CU_LIMIT_MAX } CUlimit; @@ -982,13 +1099,102 @@ typedef enum CUresourcetype_enum { CU_RESOURCE_TYPE_PITCH2D = 0x03 /**< Pitch 2D resource */ } CUresourcetype; +#ifdef _WIN32 +#define CUDA_CB __stdcall +#else +#define CUDA_CB +#endif + +#if __CUDA_API_VERSION >= 10000 + +/** + * CUDA host function + * \param userData Argument value passed to the function + */ +typedef void (CUDA_CB *CUhostFn)(void *userData); + +/** + * GPU kernel node parameters + */ +typedef struct CUDA_KERNEL_NODE_PARAMS_st { + CUfunction func; /**< Kernel to launch */ + unsigned int gridDimX; /**< Width of grid in blocks */ + unsigned int gridDimY; /**< Height of grid in blocks */ + unsigned int gridDimZ; /**< Depth of grid in blocks */ + unsigned int blockDimX; /**< X dimension of each thread block */ + unsigned int blockDimY; /**< Y dimension of each thread block */ + unsigned int blockDimZ; /**< Z dimension of each thread block */ + unsigned int sharedMemBytes; /**< Dynamic shared-memory size per thread block in bytes */ + void **kernelParams; /**< Array of pointers to kernel parameters */ + void **extra; /**< Extra options */ +} CUDA_KERNEL_NODE_PARAMS; + +/** + * Memset node parameters + */ +typedef struct CUDA_MEMSET_NODE_PARAMS_st { + CUdeviceptr dst; /**< Destination device pointer */ + size_t pitch; /**< Pitch of destination device pointer. Unused if height is 1 */ + unsigned int value; /**< Value to be set */ + unsigned int elementSize; /**< Size of each element in bytes. Must be 1, 2, or 4. */ + size_t width; /**< Width in bytes, of the row */ + size_t height; /**< Number of rows */ +} CUDA_MEMSET_NODE_PARAMS; + +/** + * Host node parameters + */ +typedef struct CUDA_HOST_NODE_PARAMS_st { + CUhostFn fn; /**< The function to call when the node executes */ + void* userData; /**< Argument to pass to the function */ +} CUDA_HOST_NODE_PARAMS; + +/** + * Graph node types + */ +typedef enum CUgraphNodeType_enum { + CU_GRAPH_NODE_TYPE_KERNEL = 0, /**< GPU kernel node */ + CU_GRAPH_NODE_TYPE_MEMCPY = 1, /**< Memcpy node */ + CU_GRAPH_NODE_TYPE_MEMSET = 2, /**< Memset node */ + CU_GRAPH_NODE_TYPE_HOST = 3, /**< Host (executable) node */ + CU_GRAPH_NODE_TYPE_GRAPH = 4, /**< Node which executes an embedded graph */ + CU_GRAPH_NODE_TYPE_EMPTY = 5, /**< Empty (no-op) node */ + CU_GRAPH_NODE_TYPE_COUNT +} CUgraphNodeType; + +/** + * Possible stream capture statuses returned by ::cuStreamIsCapturing + */ +typedef enum CUstreamCaptureStatus_enum { + CU_STREAM_CAPTURE_STATUS_NONE = 0, /**< Stream is not capturing */ + CU_STREAM_CAPTURE_STATUS_ACTIVE = 1, /**< Stream is actively capturing */ + CU_STREAM_CAPTURE_STATUS_INVALIDATED = 2 /**< Stream is part of a capture sequence that + has been invalidated, but not terminated */ +} CUstreamCaptureStatus; + +#endif /* __CUDA_API_VERSION >= 10000 */ + +#if __CUDA_API_VERSION >= 10010 + +/** + * Possible modes for stream capture thread interactions. For more details see + * ::cuStreamBeginCapture and ::cuThreadExchangeStreamCaptureMode + */ +typedef enum CUstreamCaptureMode_enum { + CU_STREAM_CAPTURE_MODE_GLOBAL = 0, + CU_STREAM_CAPTURE_MODE_THREAD_LOCAL = 1, + CU_STREAM_CAPTURE_MODE_RELAXED = 2 +} CUstreamCaptureMode; + +#endif /* __CUDA_API_VERSION >= 10010 */ + /** * Error codes */ typedef enum cudaError_enum { /** * The API call returned with no errors. In the case of query calls, this - * can also mean that the operation being queried is complete (see + * also means that the operation being queried is complete (see * ::cuEventQuery() and ::cuStreamQuery()). */ CUDA_SUCCESS = 0, @@ -1148,7 +1354,7 @@ typedef enum cudaError_enum { /** * This indicates that the ::CUcontext passed to the API call can - * only be bound to a single CPU thread at a time but is already + * only be bound to a single CPU thread at a time but is already * bound to a CPU thread. */ CUDA_ERROR_CONTEXT_ALREADY_IN_USE = 216, @@ -1176,6 +1382,11 @@ typedef enum cudaError_enum { CUDA_ERROR_NVLINK_UNCORRECTABLE = 220, /** + * This indicates that the PTX JIT compiler library was not found. + */ + CUDA_ERROR_JIT_COMPILER_NOT_FOUND = 221, + + /** * This indicates that the device kernel source is invalid. */ CUDA_ERROR_INVALID_SOURCE = 300, @@ -1207,6 +1418,12 @@ typedef enum cudaError_enum { CUDA_ERROR_INVALID_HANDLE = 400, /** + * This indicates that a resource required by the API call is not in a + * valid state to perform the requested operation. + */ + CUDA_ERROR_ILLEGAL_STATE = 401, + + /** * This indicates that a named symbol was not found. Examples of symbols * are global/constant variable names, texture names, and surface names. */ @@ -1255,7 +1472,7 @@ typedef enum cudaError_enum { * mode. */ CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING = 703, - + /** * This error indicates that a call to ::cuCtxEnablePeerAccess() is * trying to re-enable peer access to a context which has already @@ -1264,9 +1481,9 @@ typedef enum cudaError_enum { CUDA_ERROR_PEER_ACCESS_ALREADY_ENABLED = 704, /** - * This error indicates that ::cuCtxDisablePeerAccess() is - * trying to disable peer access which has not been enabled yet - * via ::cuCtxEnablePeerAccess(). + * This error indicates that ::cuCtxDisablePeerAccess() is + * trying to disable peer access which has not been enabled yet + * via ::cuCtxEnablePeerAccess(). */ CUDA_ERROR_PEER_ACCESS_NOT_ENABLED = 705, @@ -1285,15 +1502,15 @@ typedef enum cudaError_enum { /** * A device-side assert triggered during kernel execution. The context - * cannot be used anymore, and must be destroyed. All existing device - * memory allocations from this context are invalid and must be + * cannot be used anymore, and must be destroyed. All existing device + * memory allocations from this context are invalid and must be * reconstructed if the program is to continue using CUDA. */ CUDA_ERROR_ASSERT = 710, /** * This error indicates that the hardware resources required to enable - * peer access have been exhausted for one or more of the devices + * peer access have been exhausted for one or more of the devices * passed to ::cuCtxEnablePeerAccess(). */ CUDA_ERROR_TOO_MANY_PEERS = 711, @@ -1358,13 +1575,22 @@ typedef enum cudaError_enum { /** * An exception occurred on the device while executing a kernel. Common * causes include dereferencing an invalid device pointer and accessing - * out of bounds shared memory. + * out of bounds shared memory. Less common cases can be system specific - more + * information about these cases can be found in the system specific user guide. * This leaves the process in an inconsistent state and any further CUDA work * will return the same error. To continue using CUDA, the process must be terminated * and relaunched. */ CUDA_ERROR_LAUNCH_FAILED = 719, + /** + * This error indicates that the number of blocks launched per grid for a kernel that was + * launched via either ::cuLaunchCooperativeKernel or ::cuLaunchCooperativeKernelMultiDevice + * exceeds the maximum number of blocks as allowed by ::cuOccupancyMaxActiveBlocksPerMultiprocessor + * or ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags times the number of multiprocessors + * as specified by the device attribute ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. + */ + CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE = 720, /** * This error indicates that the attempted operation is not permitted. @@ -1378,6 +1604,86 @@ typedef enum cudaError_enum { CUDA_ERROR_NOT_SUPPORTED = 801, /** + * This error indicates that the system is not yet ready to start any CUDA + * work. To continue using CUDA, verify the system configuration is in a + * valid state and all required driver daemons are actively running. + * More information about this error can be found in the system specific + * user guide. + */ + CUDA_ERROR_SYSTEM_NOT_READY = 802, + + /** + * This error indicates that there is a mismatch between the versions of + * the display driver and the CUDA driver. Refer to the compatibility documentation + * for supported versions. + */ + CUDA_ERROR_SYSTEM_DRIVER_MISMATCH = 803, + + /** + * This error indicates that the system was upgraded to run with forward compatibility + * but the visible hardware detected by CUDA does not support this configuration. + * Refer to the compatibility documentation for the supported hardware matrix or ensure + * that only supported hardware is visible during initialization via the CUDA_VISIBLE_DEVICES + * environment variable. + */ + CUDA_ERROR_COMPAT_NOT_SUPPORTED_ON_DEVICE = 804, + + /** + * This error indicates that the operation is not permitted when + * the stream is capturing. + */ + CUDA_ERROR_STREAM_CAPTURE_UNSUPPORTED = 900, + + /** + * This error indicates that the current capture sequence on the stream + * has been invalidated due to a previous error. + */ + CUDA_ERROR_STREAM_CAPTURE_INVALIDATED = 901, + + /** + * This error indicates that the operation would have resulted in a merge + * of two independent capture sequences. + */ + CUDA_ERROR_STREAM_CAPTURE_MERGE = 902, + + /** + * This error indicates that the capture was not initiated in this stream. + */ + CUDA_ERROR_STREAM_CAPTURE_UNMATCHED = 903, + + /** + * This error indicates that the capture sequence contains a fork that was + * not joined to the primary stream. + */ + CUDA_ERROR_STREAM_CAPTURE_UNJOINED = 904, + + /** + * This error indicates that a dependency would have been created which + * crosses the capture sequence boundary. Only implicit in-stream ordering + * dependencies are allowed to cross the boundary. + */ + CUDA_ERROR_STREAM_CAPTURE_ISOLATION = 905, + + /** + * This error indicates a disallowed implicit dependency on a current capture + * sequence from cudaStreamLegacy. + */ + CUDA_ERROR_STREAM_CAPTURE_IMPLICIT = 906, + + /** + * This error indicates that the operation is not permitted on an event which + * was last recorded in a capturing stream. + */ + CUDA_ERROR_CAPTURED_EVENT = 907, + + /** + * A stream capture sequence not initiated with the ::CU_STREAM_CAPTURE_MODE_RELAXED + * argument to ::cuStreamBeginCapture was passed to ::cuStreamEndCapture in a + * different thread. + */ + CUDA_ERROR_STREAM_CAPTURE_WRONG_THREAD = 908, + + /** * This indicates that an unknown internal error has occurred. */ CUDA_ERROR_UNKNOWN = 999 @@ -1387,17 +1693,13 @@ typedef enum cudaError_enum { * P2P Attributes */ typedef enum CUdevice_P2PAttribute_enum { - CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK = 0x01, /**< A relative value indicating the performance of the link between two devices */ - CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED = 0x02, /**< P2P Access is enable */ - CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED = 0x03 /**< Atomic operation over the link supported */ + CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK = 0x01, /**< A relative value indicating the performance of the link between two devices */ + CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED = 0x02, /**< P2P Access is enable */ + CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED = 0x03, /**< Atomic operation over the link supported */ + CU_DEVICE_P2P_ATTRIBUTE_ACCESS_ACCESS_SUPPORTED = 0x04, /**< \deprecated use CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED instead */ + CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED = 0x04 /**< Accessing CUDA arrays over the link supported */ } CUdevice_P2PAttribute; -#ifdef _WIN32 -#define CUDA_CB __stdcall -#else -#define CUDA_CB -#endif - /** * CUDA stream callback * \param hStream The stream the callback was added to, as passed to ::cuStreamAddCallback. May be NULL. @@ -1633,7 +1935,7 @@ typedef struct CUDA_TEXTURE_DESC_st { CUfilter_mode mipmapFilterMode; /**< Mipmap filter mode */ float mipmapLevelBias; /**< Mipmap level bias */ float minMipmapLevelClamp; /**< Mipmap minimum level clamp */ - float maxMipmapLevelClamp; /**< Mipmap maximum level clamp */ + float maxMipmapLevelClamp; /**< Mipmap maximum level clamp */ float borderColor[4]; /**< Border Color */ int reserved[12]; } CUDA_TEXTURE_DESC; @@ -1706,9 +2008,281 @@ typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st { #endif /* __CUDA_API_VERSION >= 5000 */ +#if __CUDA_API_VERSION >= 9000 + +/** + * Kernel launch parameters + */ +typedef struct CUDA_LAUNCH_PARAMS_st { + CUfunction function; /**< Kernel to launch */ + unsigned int gridDimX; /**< Width of grid in blocks */ + unsigned int gridDimY; /**< Height of grid in blocks */ + unsigned int gridDimZ; /**< Depth of grid in blocks */ + unsigned int blockDimX; /**< X dimension of each thread block */ + unsigned int blockDimY; /**< Y dimension of each thread block */ + unsigned int blockDimZ; /**< Z dimension of each thread block */ + unsigned int sharedMemBytes; /**< Dynamic shared-memory size per thread block in bytes */ + CUstream hStream; /**< Stream identifier */ + void **kernelParams; /**< Array of pointers to kernel parameters */ +} CUDA_LAUNCH_PARAMS; + +#endif /* __CUDA_API_VERSION >= 9000 */ + +#if __CUDA_API_VERSION >= 10000 + +/** + * External memory handle types + */ +typedef enum CUexternalMemoryHandleType_enum { + /** + * Handle is an opaque file descriptor + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD = 1, + /** + * Handle is an opaque shared NT handle + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 = 2, + /** + * Handle is an opaque, globally shared handle + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, + /** + * Handle is a D3D12 heap object + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP = 4, + /** + * Handle is a D3D12 committed resource + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE = 5 +} CUexternalMemoryHandleType; + +/** + * Indicates that the external memory object is a dedicated resource + */ +#define CUDA_EXTERNAL_MEMORY_DEDICATED 0x1 + +/** + * External memory handle descriptor + */ +typedef struct CUDA_EXTERNAL_MEMORY_HANDLE_DESC_st { + /** + * Type of the handle + */ + CUexternalMemoryHandleType type; + union { + /** + * File descriptor referencing the memory object. Valid + * when type is + * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD + */ + int fd; + /** + * Win32 handle referencing the semaphore object. Valid when + * type is one of the following: + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE + * Exactly one of 'handle' and 'name' must be non-NULL. If + * type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * then 'name' must be NULL. + */ + struct { + /** + * Valid NT handle. Must be NULL if 'name' is non-NULL + */ + void *handle; + /** + * Name of a valid memory object. + * Must be NULL if 'handle' is non-NULL. + */ + const void *name; + } win32; + } handle; + /** + * Size of the memory allocation + */ + unsigned long long size; + /** + * Flags must either be zero or ::CUDA_EXTERNAL_MEMORY_DEDICATED + */ + unsigned int flags; + unsigned int reserved[16]; +} CUDA_EXTERNAL_MEMORY_HANDLE_DESC; + +/** + * External memory buffer descriptor + */ +typedef struct CUDA_EXTERNAL_MEMORY_BUFFER_DESC_st { + /** + * Offset into the memory object where the buffer's base is + */ + unsigned long long offset; + /** + * Size of the buffer + */ + unsigned long long size; + /** + * Flags reserved for future use. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; +} CUDA_EXTERNAL_MEMORY_BUFFER_DESC; + +/** + * External memory mipmap descriptor + */ +typedef struct CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC_st { + /** + * Offset into the memory object where the base level of the + * mipmap chain is. + */ + unsigned long long offset; + /** + * Format, dimension and type of base level of the mipmap chain + */ + CUDA_ARRAY3D_DESCRIPTOR arrayDesc; + /** + * Total number of levels in the mipmap chain + */ + unsigned int numLevels; + unsigned int reserved[16]; +} CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC; + +/** + * External semaphore handle types + */ +typedef enum CUexternalSemaphoreHandleType_enum { + /** + * Handle is an opaque file descriptor + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD = 1, + /** + * Handle is an opaque shared NT handle + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 = 2, + /** + * Handle is an opaque, globally shared handle + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, + /** + * Handle is a shared NT handle referencing a D3D12 fence object + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE = 4 +} CUexternalSemaphoreHandleType; + +/** + * External semaphore handle descriptor + */ +typedef struct CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC_st { + /** + * Type of the handle + */ + CUexternalSemaphoreHandleType type; + union { + /** + * File descriptor referencing the semaphore object. Valid + * when type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD + */ + int fd; + /** + * Win32 handle referencing the semaphore object. Valid when + * type is one of the following: + * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 + * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE + * Exactly one of 'handle' and 'name' must be non-NULL. If + * type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * then 'name' must be NULL. + */ + struct { + /** + * Valid NT handle. Must be NULL if 'name' is non-NULL + */ + void *handle; + /** + * Name of a valid synchronization primitive. + * Must be NULL if 'handle' is non-NULL. + */ + const void *name; + } win32; + } handle; + /** + * Flags reserved for the future. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; +} CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC; + +/** + * External semaphore signal parameters + */ +typedef struct CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS_st { + struct { + /** + * Parameters for fence objects + */ + struct { + /** + * Value of fence to be signaled + */ + unsigned long long value; + } fence; + unsigned int reserved[16]; + } params; + /** + * Flags reserved for the future. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; +} CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS; + +/** + * External semaphore wait parameters + */ +typedef struct CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS_st { + struct { + /** + * Parameters for fence objects + */ + struct { + /** + * Value of fence to be waited on + */ + unsigned long long value; + } fence; + unsigned int reserved[16]; + } params; + /** + * Flags reserved for the future. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; +} CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS; + + +#endif /* __CUDA_API_VERSION >= 10000 */ + +/** + * If set, each kernel launched as part of ::cuLaunchCooperativeKernelMultiDevice only + * waits for prior work in the stream corresponding to that GPU to complete before the + * kernel begins execution. + */ +#define CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC 0x01 + +/** + * If set, any subsequent work pushed in a stream that participated in a call to + * ::cuLaunchCooperativeKernelMultiDevice will only wait for the kernel launched on + * the GPU corresponding to that stream to complete before it begins execution. + */ +#define CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC 0x02 + /** * If set, the CUDA array is a collection of layers, where each layer is either a 1D - * or a 2D array and the Depth member of CUDA_ARRAY3D_DESCRIPTOR specifies the number + * or a 2D array and the Depth member of CUDA_ARRAY3D_DESCRIPTOR specifies the number * of layers, not the depth of a 3D array. */ #define CUDA_ARRAY3D_LAYERED 0x01 @@ -1741,10 +2315,16 @@ typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st { /** * This flag if set indicates that the CUDA * array is a DEPTH_TEXTURE. -*/ + */ #define CUDA_ARRAY3D_DEPTH_TEXTURE 0x10 /** + * This flag indicates that the CUDA array may be bound as a color target + * in an external graphics API + */ +#define CUDA_ARRAY3D_COLOR_ATTACHMENT 0x20 + +/** * Override the texref format with a format inferred from the array. * Flag for ::cuTexRefSetArray() */ @@ -1847,7 +2427,9 @@ typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st { * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::CUresult + * \sa + * ::CUresult, + * ::cudaGetErrorString */ CUresult CUDAAPI cuGetErrorString(CUresult error, const char **pStr); @@ -1866,7 +2448,9 @@ CUresult CUDAAPI cuGetErrorString(CUresult error, const char **pStr); * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::CUresult + * \sa + * ::CUresult, + * ::cudaGetErrorName */ CUresult CUDAAPI cuGetErrorName(CUresult error, const char **pStr); @@ -1897,7 +2481,9 @@ CUresult CUDAAPI cuGetErrorName(CUresult error, const char **pStr); * \return * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_VALUE, - * ::CUDA_ERROR_INVALID_DEVICE + * ::CUDA_ERROR_INVALID_DEVICE, + * ::CUDA_ERROR_SYSTEM_DRIVER_MISMATCH, + * ::CUDA_ERROR_COMPAT_NOT_SUPPORTED_ON_DEVICE * \notefnerr */ CUresult CUDAAPI cuInit(unsigned int Flags); @@ -1917,11 +2503,15 @@ CUresult CUDAAPI cuInit(unsigned int Flags); */ /** - * \brief Returns the CUDA driver version + * \brief Returns the latest CUDA version supported by driver + * + * Returns in \p *driverVersion the version of CUDA supported by + * the driver. The version is returned as + * (1000 × major + 10 × minor). For example, CUDA 9.2 + * would be represented by 9020. * - * Returns in \p *driverVersion the version number of the installed CUDA - * driver. This function automatically returns ::CUDA_ERROR_INVALID_VALUE if - * the \p driverVersion argument is NULL. + * This function automatically returns ::CUDA_ERROR_INVALID_VALUE if + * \p driverVersion is NULL. * * \param driverVersion - Returns the CUDA driver version * @@ -1929,6 +2519,10 @@ CUresult CUDAAPI cuInit(unsigned int Flags); * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_VALUE * \notefnerr + * + * \sa + * ::cudaDriverGetVersion, + * ::cudaRuntimeGetVersion */ CUresult CUDAAPI cuDriverGetVersion(int *driverVersion); @@ -1968,6 +2562,8 @@ CUresult CUDAAPI cuDriverGetVersion(int *driverVersion); * ::cuDeviceGetAttribute, * ::cuDeviceGetCount, * ::cuDeviceGetName, + * ::cuDeviceGetUuid, + * ::cuDeviceGetLuid, * ::cuDeviceTotalMem */ CUresult CUDAAPI cuDeviceGet(CUdevice *device, int ordinal); @@ -1976,7 +2572,7 @@ CUresult CUDAAPI cuDeviceGet(CUdevice *device, int ordinal); * \brief Returns the number of compute-capable devices * * Returns in \p *count the number of devices with compute capability greater - * than or equal to 1.0 that are available for execution. If there is no such + * than or equal to 2.0 that are available for execution. If there is no such * device, ::cuDeviceGetCount() returns 0. * * \param count - Returned number of compute-capable devices @@ -1992,8 +2588,11 @@ CUresult CUDAAPI cuDeviceGet(CUdevice *device, int ordinal); * \sa * ::cuDeviceGetAttribute, * ::cuDeviceGetName, + * ::cuDeviceGetUuid, + * ::cuDeviceGetLuid, * ::cuDeviceGet, - * ::cuDeviceTotalMem + * ::cuDeviceTotalMem, + * ::cudaGetDeviceCount */ CUresult CUDAAPI cuDeviceGetCount(int *count); @@ -2019,12 +2618,75 @@ CUresult CUDAAPI cuDeviceGetCount(int *count); * * \sa * ::cuDeviceGetAttribute, + * ::cuDeviceGetUuid, + * ::cuDeviceGetLuid, * ::cuDeviceGetCount, * ::cuDeviceGet, - * ::cuDeviceTotalMem + * ::cuDeviceTotalMem, + * ::cudaGetDeviceProperties */ CUresult CUDAAPI cuDeviceGetName(char *name, int len, CUdevice dev); +#if __CUDA_API_VERSION >= 9020 +/** + * \brief Return an UUID for the device + * + * Returns 16-octets identifing the device \p dev in the structure + * pointed by the \p uuid. + * + * \param uuid - Returned UUID + * \param dev - Device to get identifier string for + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_DEVICE + * \notefnerr + * + * \sa + * ::cuDeviceGetAttribute, + * ::cuDeviceGetCount, + * ::cuDeviceGetName, + * ::cuDeviceGetLuid, + * ::cuDeviceGet, + * ::cuDeviceTotalMem, + * ::cudaGetDeviceProperties + */ +CUresult CUDAAPI cuDeviceGetUuid(CUuuid *uuid, CUdevice dev); +#endif + +#if defined(_WIN32) && __CUDA_API_VERSION >= 10000 +/** + * \brief Return an LUID and device node mask for the device + * + * Return identifying information (\p luid and \p deviceNodeMask) to allow + * matching device with graphics APIs. + * + * \param luid - Returned LUID + * \param deviceNodeMask - Returned device node mask + * \param dev - Device to get identifier string for + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_DEVICE + * \notefnerr + * + * \sa + * ::cuDeviceGetAttribute, + * ::cuDeviceGetCount, + * ::cuDeviceGetName, + * ::cuDeviceGet, + * ::cuDeviceTotalMem, + * ::cudaGetDeviceProperties + */ +CUresult CUDAAPI cuDeviceGetLuid(char *luid, unsigned int *deviceNodeMask, CUdevice dev); +#endif + #if __CUDA_API_VERSION >= 3020 /** * \brief Returns the total amount of memory on the device @@ -2048,7 +2710,9 @@ CUresult CUDAAPI cuDeviceGetName(char *name, int len, CUdevice dev); * ::cuDeviceGetAttribute, * ::cuDeviceGetCount, * ::cuDeviceGetName, + * ::cuDeviceGetUuid, * ::cuDeviceGet, + * ::cudaMemGetInfo */ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -2074,15 +2738,15 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * - ::CU_DEVICE_ATTRIBUTE_MAX_PITCH: Maximum pitch in bytes allowed by the * memory copy functions that involve memory regions allocated through * ::cuMemAllocPitch(); - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH: Maximum 1D + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH: Maximum 1D * texture width; * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH: Maximum width * for a 1D texture bound to linear memory; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH: Maximum + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH: Maximum * mipmapped 1D texture width; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH: Maximum 2D + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH: Maximum 2D * texture width; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT: Maximum 2D + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT: Maximum 2D * texture height; * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH: Maximum width * for a 2D texture bound to linear memory; @@ -2090,40 +2754,40 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * for a 2D texture bound to linear memory; * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH: Maximum pitch * in bytes for a 2D texture bound to linear memory; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_WIDTH: Maximum + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_WIDTH: Maximum * mipmapped 2D texture width; * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_HEIGHT: Maximum * mipmapped 2D texture height; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH: Maximum 3D + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH: Maximum 3D * texture width; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT: Maximum 3D + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT: Maximum 3D * texture height; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH: Maximum 3D + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH: Maximum 3D * texture depth; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH_ALTERNATE: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH_ALTERNATE: * Alternate maximum 3D texture width, 0 if no alternate * maximum 3D texture size is supported; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT_ALTERNATE: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT_ALTERNATE: * Alternate maximum 3D texture height, 0 if no alternate * maximum 3D texture size is supported; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH_ALTERNATE: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH_ALTERNATE: * Alternate maximum 3D texture depth, 0 if no alternate * maximum 3D texture size is supported; * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_WIDTH: * Maximum cubemap texture width or height; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH: * Maximum 1D layered texture width; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS: * Maximum layers in a 1D layered texture; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH: * Maximum 2D layered texture width; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT: * Maximum 2D layered texture height; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS: * Maximum layers in a 2D layered texture; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_WIDTH: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_WIDTH: * Maximum cubemap layered texture width or height; - * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_LAYERS: + * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_LAYERS: * Maximum layers in a cubemap layered texture; * - ::CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_WIDTH: * Maximum 1D surface width; @@ -2189,19 +2853,20 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * - ::CU_DEVICE_ATTRIBUTE_PCI_BUS_ID: PCI bus identifier of the device; * - ::CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID: PCI device (also known as slot) identifier * of the device; + * - ::CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID: PCI domain identifier of the device * - ::CU_DEVICE_ATTRIBUTE_TCC_DRIVER: 1 if the device is using a TCC driver. TCC * is only available on Tesla hardware running Windows Vista or later; * - ::CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE: Peak memory clock frequency in kilohertz; * - ::CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH: Global memory bus width in bits; * - ::CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE: Size of L2 cache in bytes. 0 if the device doesn't have L2 cache; * - ::CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR: Maximum resident threads per multiprocessor; - * - ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING: 1 if the device shares a unified address space with + * - ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING: 1 if the device shares a unified address space with * the host, or 0 if not; * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR: Major compute capability version number; * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR: Minor compute capability version number; - * - ::CU_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED: 1 if device supports caching globals + * - ::CU_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED: 1 if device supports caching globals * in L1 cache, 0 if caching globals in L1 cache is not supported by the device; - * - ::CU_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED: 1 if device supports caching locals + * - ::CU_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED: 1 if device supports caching locals * in L1 cache, 0 if caching locals in L1 cache is not supported by the device; * - ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR: Maximum amount of * shared memory available to a multiprocessor in bytes; this amount is shared @@ -2225,6 +2890,12 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED: Device supports Compute Preemption. * - ::CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM: Device can access host registered * memory at the same virtual address as the CPU. + * - ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN: The maximum per block shared memory size + * suported on this device. This is the maximum value that can be opted into when using the cuFuncSetAttribute() call. + * For more details see ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES + * - ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES: Device accesses pageable memory via the host's + * page tables. + * - ::CU_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST: The host can directly access managed memory on the device without migration. * * \param pi - Returned device attribute value * \param attrib - Device attribute to query @@ -2242,8 +2913,11 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * \sa * ::cuDeviceGetCount, * ::cuDeviceGetName, + * ::cuDeviceGetUuid, * ::cuDeviceGet, - * ::cuDeviceTotalMem + * ::cuDeviceTotalMem, + * ::cudaDeviceGetAttribute, + * ::cudaGetDeviceProperties */ CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, CUdevice dev); @@ -2319,10 +2993,11 @@ CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, CUdevi * ::cuDeviceGetAttribute, * ::cuDeviceGetCount, * ::cuDeviceGetName, + * ::cuDeviceGetUuid, * ::cuDeviceGet, * ::cuDeviceTotalMem */ -CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevice dev); +__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevice dev); /** * \brief Returns the compute capability of the device @@ -2330,7 +3005,7 @@ CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevice dev); * \deprecated * * This function was deprecated as of CUDA 5.0 and its functionality superceded - * by ::cuDeviceGetAttribute(). + * by ::cuDeviceGetAttribute(). * * Returns in \p *major and \p *minor the major and minor revision numbers that * define the compute capability of the device \p dev. @@ -2352,10 +3027,11 @@ CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevice dev); * ::cuDeviceGetAttribute, * ::cuDeviceGetCount, * ::cuDeviceGetName, + * ::cuDeviceGetUuid, * ::cuDeviceGet, * ::cuDeviceTotalMem */ -CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *minor, CUdevice dev); +__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *minor, CUdevice dev); /** @} */ /* END CUDA_DEVICE_DEPRECATED */ @@ -2368,8 +3044,8 @@ CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *minor, CUdevice dev) * This section describes the primary context management functions of the low-level * CUDA driver application programming interface. * - * The primary context unique per device and it's shared with CUDA runtime API. - * Those functions allows seemless integration with other libraries using CUDA. + * The primary context is unique per device and shared with the CUDA runtime API. + * These functions allow integration with other libraries using CUDA. * * @{ */ @@ -2385,9 +3061,9 @@ CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *minor, CUdevice dev) * Unlike ::cuCtxCreate() the newly created context is not pushed onto the stack. * * Context creation will fail with ::CUDA_ERROR_UNKNOWN if the compute mode of - * the device is ::CU_COMPUTEMODE_PROHIBITED. The function ::cuDeviceGetAttribute() - * can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE to determine the compute mode - * of the device. + * the device is ::CU_COMPUTEMODE_PROHIBITED. The function ::cuDeviceGetAttribute() + * can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE to determine the compute mode + * of the device. * The <i>nvidia-smi</i> tool can be used to set the compute mode for * devices. Documentation for <i>nvidia-smi</i> can be obtained by passing a * -h option to it. @@ -2495,8 +3171,9 @@ CUresult CUDAAPI cuDevicePrimaryCtxRelease(CUdevice dev); * \e C > \e P, then CUDA will yield to other OS threads when waiting for * the GPU (::CU_CTX_SCHED_YIELD), otherwise CUDA will not yield while * waiting for results and actively spin on the processor (::CU_CTX_SCHED_SPIN). - * However, on low power devices like Tegra, it always defaults to - * ::CU_CTX_SCHED_BLOCKING_SYNC. + * Additionally, on Tegra devices, ::CU_CTX_SCHED_AUTO uses a heuristic based on + * the power profile of the platform and may choose ::CU_CTX_SCHED_BLOCKING_SYNC + * for low-powered devices. * * - ::CU_CTX_LMEM_RESIZE_TO_MAX: Instruct CUDA to not reduce local memory * after resizing local memory for a kernel. This can prevent thrashing by @@ -2518,7 +3195,8 @@ CUresult CUDAAPI cuDevicePrimaryCtxRelease(CUdevice dev); * \sa ::cuDevicePrimaryCtxRetain, * ::cuDevicePrimaryCtxGetState, * ::cuCtxCreate, - * ::cuCtxGetFlags + * ::cuCtxGetFlags, + * ::cudaSetDeviceFlags */ CUresult CUDAAPI cuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags); @@ -2541,8 +3219,10 @@ CUresult CUDAAPI cuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags); * ::CUDA_ERROR_INVALID_VALUE, * \notefnerr * - * \sa ::cuDevicePrimaryCtxSetFlags, - * ::cuCtxGetFlags + * \sa + * ::cuDevicePrimaryCtxSetFlags, + * ::cuCtxGetFlags, + * ::cudaGetDeviceFlags */ CUresult CUDAAPI cuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags, int *active); @@ -2579,8 +3259,8 @@ CUresult CUDAAPI cuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags, i * ::cuCtxPushCurrent, * ::cuCtxSetCacheConfig, * ::cuCtxSetLimit, - * ::cuCtxSynchronize - * + * ::cuCtxSynchronize, + * ::cudaDeviceReset */ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); @@ -2598,6 +3278,9 @@ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); * This section describes the context management functions of the low-level * CUDA driver application programming interface. * + * Please note that some functions are described in + * \ref CUDA_PRIMARY_CTX "Primary Context Management" section. + * * @{ */ @@ -2605,11 +3288,13 @@ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); /** * \brief Create a CUDA context * + * \note In most cases it is recommended to use ::cuDevicePrimaryCtxRetain. + * * Creates a new CUDA context and associates it with the calling thread. The * \p flags parameter is described below. The context is created with a usage - * count of 1 and the caller of ::cuCtxCreate() must call ::cuCtxDestroy() or - * when done using the context. If a context is already current to the thread, - * it is supplanted by the newly created context and may be restored by a subsequent + * count of 1 and the caller of ::cuCtxCreate() must call ::cuCtxDestroy() + * when done using the context. If a context is already current to the thread, + * it is supplanted by the newly created context and may be restored by a subsequent * call to ::cuCtxPopCurrent(). * * The three LSBs of the \p flags parameter can be used to control how the OS @@ -2626,23 +3311,24 @@ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); * results from the GPU. This can increase latency when waiting for the GPU, * but can increase the performance of CPU threads performing work in parallel * with the GPU. - * + * * - ::CU_CTX_SCHED_BLOCKING_SYNC: Instruct CUDA to block the CPU thread on a * synchronization primitive when waiting for the GPU to finish work. * * - ::CU_CTX_BLOCKING_SYNC: Instruct CUDA to block the CPU thread on a * synchronization primitive when waiting for the GPU to finish work. <br> * <b>Deprecated:</b> This flag was deprecated as of CUDA 4.0 and was - * replaced with ::CU_CTX_SCHED_BLOCKING_SYNC. + * replaced with ::CU_CTX_SCHED_BLOCKING_SYNC. * * - ::CU_CTX_SCHED_AUTO: The default value if the \p flags parameter is zero, * uses a heuristic based on the number of active CUDA contexts in the * process \e C and the number of logical processors in the system \e P. If - * \e C > \e P, then CUDA will yield to other OS threads when waiting for - * the GPU (::CU_CTX_SCHED_YIELD), otherwise CUDA will not yield while - * waiting for results and actively spin on the processor (::CU_CTX_SCHED_SPIN). - * However, on low power devices like Tegra, it always defaults to - * ::CU_CTX_SCHED_BLOCKING_SYNC. + * \e C > \e P, then CUDA will yield to other OS threads when waiting for + * the GPU (::CU_CTX_SCHED_YIELD), otherwise CUDA will not yield while + * waiting for results and actively spin on the processor (::CU_CTX_SCHED_SPIN). + * Additionally, on Tegra devices, ::CU_CTX_SCHED_AUTO uses a heuristic based on + * the power profile of the platform and may choose ::CU_CTX_SCHED_BLOCKING_SYNC + * for low-powered devices. * * - ::CU_CTX_MAP_HOST: Instruct CUDA to support mapped pinned allocations. * This flag must be set in order to allocate pinned host memory that is @@ -2654,10 +3340,10 @@ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); * memory usage at the cost of potentially increased memory usage. * * Context creation will fail with ::CUDA_ERROR_UNKNOWN if the compute mode of - * the device is ::CU_COMPUTEMODE_PROHIBITED. The function ::cuDeviceGetAttribute() - * can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE to determine the - * compute mode of the device. The <i>nvidia-smi</i> tool can be used to set - * the compute mode for * devices. + * the device is ::CU_COMPUTEMODE_PROHIBITED. The function ::cuDeviceGetAttribute() + * can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE to determine the + * compute mode of the device. The <i>nvidia-smi</i> tool can be used to set + * the compute mode for * devices. * Documentation for <i>nvidia-smi</i> can be obtained by passing a * -h option to it. * @@ -2700,7 +3386,7 @@ CUresult CUDAAPI cuCtxCreate(CUcontext *pctx, unsigned int flags, CUdevice dev); * It is the responsibility of the calling function to ensure that no API * call issues using \p ctx while ::cuCtxDestroy() is executing. * - * If \p ctx is current to the calling thread then \p ctx will also be + * If \p ctx is current to the calling thread then \p ctx will also be * popped from the current thread's context stack (as though ::cuCtxPopCurrent() * were called). If \p ctx is current to other threads, then \p ctx will * remain current to those threads, and attempting to access \p ctx from @@ -2769,8 +3455,8 @@ CUresult CUDAAPI cuCtxPushCurrent(CUcontext ctx); /** * \brief Pops the current CUDA context from the current CPU thread. * - * Pops the current CUDA context from the CPU thread and passes back the - * old context handle in \p *pctx. That context may then be made current + * Pops the current CUDA context from the CPU thread and passes back the + * old context handle in \p *pctx. That context may then be made current * to a different CPU thread by calling ::cuCtxPushCurrent(). * * If a context was current to the CPU thread before ::cuCtxCreate() or @@ -2808,7 +3494,7 @@ CUresult CUDAAPI cuCtxPopCurrent(CUcontext *pctx); * calling CPU thread is unbound and ::CUDA_SUCCESS is returned. * * If there exists a CUDA context stack on the calling CPU thread, this - * will replace the top of that stack with \p ctx. + * will replace the top of that stack with \p ctx. * If \p ctx is NULL then this will be equivalent to popping the top * of the calling CPU thread's CUDA context stack (or a no-op if the * calling CPU thread's CUDA context stack is empty). @@ -2822,7 +3508,11 @@ CUresult CUDAAPI cuCtxPopCurrent(CUcontext *pctx); * ::CUDA_ERROR_INVALID_CONTEXT * \notefnerr * - * \sa ::cuCtxGetCurrent, ::cuCtxCreate, ::cuCtxDestroy + * \sa + * ::cuCtxGetCurrent, + * ::cuCtxCreate, + * ::cuCtxDestroy, + * ::cudaSetDevice */ CUresult CUDAAPI cuCtxSetCurrent(CUcontext ctx); @@ -2841,7 +3531,11 @@ CUresult CUDAAPI cuCtxSetCurrent(CUcontext ctx); * ::CUDA_ERROR_NOT_INITIALIZED, * \notefnerr * - * \sa ::cuCtxSetCurrent, ::cuCtxCreate, ::cuCtxDestroy + * \sa + * ::cuCtxSetCurrent, + * ::cuCtxCreate, + * ::cuCtxDestroy, + * ::cudaGetDevice */ CUresult CUDAAPI cuCtxGetCurrent(CUcontext *pctx); #endif /* __CUDA_API_VERSION >= 4000 */ @@ -2871,7 +3565,8 @@ CUresult CUDAAPI cuCtxGetCurrent(CUcontext *pctx); * ::cuCtxPushCurrent, * ::cuCtxSetCacheConfig, * ::cuCtxSetLimit, - * ::cuCtxSynchronize + * ::cuCtxSynchronize, + * ::cudaGetDevice */ CUresult CUDAAPI cuCtxGetDevice(CUdevice *device); @@ -2899,7 +3594,8 @@ CUresult CUDAAPI cuCtxGetDevice(CUdevice *device); * ::cuCtxGetDevice * ::cuCtxGetLimit, * ::cuCtxGetSharedMemConfig, - * ::cuCtxGetStreamPriorityRange + * ::cuCtxGetStreamPriorityRange, + * ::cudaGetDeviceFlags */ CUresult CUDAAPI cuCtxGetFlags(unsigned int *flags); #endif /* __CUDA_API_VERSION >= 7000 */ @@ -2909,7 +3605,7 @@ CUresult CUDAAPI cuCtxGetFlags(unsigned int *flags); * * Blocks until the device has completed all preceding requested tasks. * ::cuCtxSynchronize() returns an error if one of the preceding tasks failed. - * If the context was created with the ::CU_CTX_SCHED_BLOCKING_SYNC flag, the + * If the context was created with the ::CU_CTX_SCHED_BLOCKING_SYNC flag, the * CPU thread will block until the GPU context has finished its work. * * \return @@ -2929,7 +3625,8 @@ CUresult CUDAAPI cuCtxGetFlags(unsigned int *flags); * ::cuCtxPopCurrent, * ::cuCtxPushCurrent, * ::cuCtxSetCacheConfig, - * ::cuCtxSetLimit + * ::cuCtxSetLimit, + * ::cudaDeviceSynchronize */ CUresult CUDAAPI cuCtxSynchronize(void); @@ -2947,50 +3644,41 @@ CUresult CUDAAPI cuCtxSynchronize(void); * discussed here. * * - ::CU_LIMIT_STACK_SIZE controls the stack size in bytes of each GPU thread. - * This limit is only applicable to devices of compute capability 2.0 and - * higher. Attempting to set this limit on devices of compute capability - * less than 2.0 will result in the error ::CUDA_ERROR_UNSUPPORTED_LIMIT - * being returned. + * Note that the CUDA driver will set the \p limit to the maximum of \p value + * and what the kernel function requires. * * - ::CU_LIMIT_PRINTF_FIFO_SIZE controls the size in bytes of the FIFO used * by the ::printf() device system call. Setting ::CU_LIMIT_PRINTF_FIFO_SIZE * must be performed before launching any kernel that uses the ::printf() * device system call, otherwise ::CUDA_ERROR_INVALID_VALUE will be returned. - * This limit is only applicable to devices of compute capability 2.0 and - * higher. Attempting to set this limit on devices of compute capability - * less than 2.0 will result in the error ::CUDA_ERROR_UNSUPPORTED_LIMIT - * being returned. * * - ::CU_LIMIT_MALLOC_HEAP_SIZE controls the size in bytes of the heap used * by the ::malloc() and ::free() device system calls. Setting * ::CU_LIMIT_MALLOC_HEAP_SIZE must be performed before launching any kernel * that uses the ::malloc() or ::free() device system calls, otherwise - * ::CUDA_ERROR_INVALID_VALUE will be returned. This limit is only applicable - * to devices of compute capability 2.0 and higher. Attempting to set this - * limit on devices of compute capability less than 2.0 will result in the - * error ::CUDA_ERROR_UNSUPPORTED_LIMIT being returned. + * ::CUDA_ERROR_INVALID_VALUE will be returned. * * - ::CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH controls the maximum nesting depth of * a grid at which a thread can safely call ::cudaDeviceSynchronize(). Setting - * this limit must be performed before any launch of a kernel that uses the + * this limit must be performed before any launch of a kernel that uses the * device runtime and calls ::cudaDeviceSynchronize() above the default sync - * depth, two levels of grids. Calls to ::cudaDeviceSynchronize() will fail - * with error code ::cudaErrorSyncDepthExceeded if the limitation is + * depth, two levels of grids. Calls to ::cudaDeviceSynchronize() will fail + * with error code ::cudaErrorSyncDepthExceeded if the limitation is * violated. This limit can be set smaller than the default or up the maximum * launch depth of 24. When setting this limit, keep in mind that additional * levels of sync depth require the driver to reserve large amounts of device - * memory which can no longer be used for user allocations. If these - * reservations of device memory fail, ::cuCtxSetLimit will return + * memory which can no longer be used for user allocations. If these + * reservations of device memory fail, ::cuCtxSetLimit will return * ::CUDA_ERROR_OUT_OF_MEMORY, and the limit can be reset to a lower value. * This limit is only applicable to devices of compute capability 3.5 and * higher. Attempting to set this limit on devices of compute capability less - * than 3.5 will result in the error ::CUDA_ERROR_UNSUPPORTED_LIMIT being + * than 3.5 will result in the error ::CUDA_ERROR_UNSUPPORTED_LIMIT being * returned. * * - ::CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT controls the maximum number of * outstanding device runtime launches that can be made from the current * context. A grid is outstanding from the point of launch up until the grid - * is known to have been completed. Device runtime launches which violate + * is known to have been completed. Device runtime launches which violate * this limitation fail and return ::cudaErrorLaunchPendingCountExceeded when * ::cudaGetLastError() is called after launch. If more pending launches than * the default (2048 launches) are needed for a module using the device @@ -3004,6 +3692,10 @@ CUresult CUDAAPI cuCtxSynchronize(void); * than 3.5 will result in the error ::CUDA_ERROR_UNSUPPORTED_LIMIT being * returned. * + * - ::CU_LIMIT_MAX_L2_FETCH_GRANULARITY controls the L2 cache fetch granularity. + * Values can range from 0B to 128B. This is purely a performance hint and + * it can be ignored or clamped depending on the platform. + * * \param limit - Limit to set * \param value - Size of limit * @@ -3011,7 +3703,8 @@ CUresult CUDAAPI cuCtxSynchronize(void); * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_VALUE, * ::CUDA_ERROR_UNSUPPORTED_LIMIT, - * ::CUDA_ERROR_OUT_OF_MEMORY + * ::CUDA_ERROR_OUT_OF_MEMORY, + * ::CUDA_ERROR_INVALID_CONTEXT * \notefnerr * * \sa ::cuCtxCreate, @@ -3024,7 +3717,8 @@ CUresult CUDAAPI cuCtxSynchronize(void); * ::cuCtxPopCurrent, * ::cuCtxPushCurrent, * ::cuCtxSetCacheConfig, - * ::cuCtxSynchronize + * ::cuCtxSynchronize, + * ::cudaDeviceSetLimit */ CUresult CUDAAPI cuCtxSetLimit(CUlimit limit, size_t value); @@ -3043,6 +3737,7 @@ CUresult CUDAAPI cuCtxSetLimit(CUlimit limit, size_t value); * child grid launches to complete. * - ::CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT: maximum number of outstanding * device runtime launches that can be made from this context. + * - ::CU_LIMIT_MAX_L2_FETCH_GRANULARITY: L2 cache fetch granularity. * * \param limit - Limit to query * \param pvalue - Returned size of limit @@ -3063,7 +3758,8 @@ CUresult CUDAAPI cuCtxSetLimit(CUlimit limit, size_t value); * ::cuCtxPushCurrent, * ::cuCtxSetCacheConfig, * ::cuCtxSetLimit, - * ::cuCtxSynchronize + * ::cuCtxSynchronize, + * ::cudaDeviceGetLimit */ CUresult CUDAAPI cuCtxGetLimit(size_t *pvalue, CUlimit limit); @@ -3106,7 +3802,8 @@ CUresult CUDAAPI cuCtxGetLimit(size_t *pvalue, CUlimit limit); * ::cuCtxSetCacheConfig, * ::cuCtxSetLimit, * ::cuCtxSynchronize, - * ::cuFuncSetCacheConfig + * ::cuFuncSetCacheConfig, + * ::cudaDeviceGetCacheConfig */ CUresult CUDAAPI cuCtxGetCacheConfig(CUfunc_cache *pconfig); @@ -3156,7 +3853,8 @@ CUresult CUDAAPI cuCtxGetCacheConfig(CUfunc_cache *pconfig); * ::cuCtxPushCurrent, * ::cuCtxSetLimit, * ::cuCtxSynchronize, - * ::cuFuncSetCacheConfig + * ::cuFuncSetCacheConfig, + * ::cudaDeviceSetCacheConfig */ CUresult CUDAAPI cuCtxSetCacheConfig(CUfunc_cache config); @@ -3165,20 +3863,20 @@ CUresult CUDAAPI cuCtxSetCacheConfig(CUfunc_cache config); * \brief Returns the current shared memory configuration for the current context. * * This function will return in \p pConfig the current size of shared memory banks - * in the current context. On devices with configurable shared memory banks, - * ::cuCtxSetSharedMemConfig can be used to change this setting, so that all - * subsequent kernel launches will by default use the new bank size. When - * ::cuCtxGetSharedMemConfig is called on devices without configurable shared + * in the current context. On devices with configurable shared memory banks, + * ::cuCtxSetSharedMemConfig can be used to change this setting, so that all + * subsequent kernel launches will by default use the new bank size. When + * ::cuCtxGetSharedMemConfig is called on devices without configurable shared * memory, it will return the fixed bank size of the hardware. * * The returned bank configurations can be either: - * - ::CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE: shared memory bank width is + * - ::CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE: shared memory bank width is * four bytes. * - ::CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE: shared memory bank width will * eight bytes. * * \param pConfig - returned shared memory configuration - * \return + * \return * ::CUDA_SUCCESS, * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, @@ -3199,6 +3897,7 @@ CUresult CUDAAPI cuCtxSetCacheConfig(CUfunc_cache config); * ::cuCtxSynchronize, * ::cuCtxGetSharedMemConfig, * ::cuFuncSetCacheConfig, + * ::cudaDeviceGetSharedMemConfig */ CUresult CUDAAPI cuCtxGetSharedMemConfig(CUsharedconfig *pConfig); @@ -3206,16 +3905,16 @@ CUresult CUDAAPI cuCtxGetSharedMemConfig(CUsharedconfig *pConfig); * \brief Sets the shared memory configuration for the current context. * * On devices with configurable shared memory banks, this function will set - * the context's shared memory bank size which is used for subsequent kernel - * launches. + * the context's shared memory bank size which is used for subsequent kernel + * launches. * * Changed the shared memory configuration between launches may insert a device * side synchronization point between those launches. * * Changing the shared memory bank size will not increase shared memory usage - * or affect occupancy of kernels, but may have major effects on performance. + * or affect occupancy of kernels, but may have major effects on performance. * Larger bank sizes will allow for greater potential bandwidth to shared memory, - * but will change what kinds of accesses to shared memory will result in bank + * but will change what kinds of accesses to shared memory will result in bank * conflicts. * * This function will do nothing on devices with fixed shared memory bank size. @@ -3251,6 +3950,7 @@ CUresult CUDAAPI cuCtxGetSharedMemConfig(CUsharedconfig *pConfig); * ::cuCtxSynchronize, * ::cuCtxGetSharedMemConfig, * ::cuFuncSetCacheConfig, + * ::cudaDeviceSetSharedMemConfig */ CUresult CUDAAPI cuCtxSetSharedMemConfig(CUsharedconfig config); #endif @@ -3276,6 +3976,7 @@ CUresult CUDAAPI cuCtxSetSharedMemConfig(CUsharedconfig config); * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, + * ::CUDA_ERROR_INVALID_VALUE, * ::CUDA_ERROR_UNKNOWN * \notefnerr * @@ -3327,7 +4028,8 @@ CUresult CUDAAPI cuCtxGetApiVersion(CUcontext ctx, unsigned int *version); * ::cuCtxGetDevice, * ::cuCtxGetFlags, * ::cuCtxSetLimit, - * ::cuCtxSynchronize + * ::cuCtxSynchronize, + * ::cudaDeviceGetStreamPriorityRange */ CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, int *greatestPriority); @@ -3384,7 +4086,7 @@ CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, int *greatestPr * ::cuCtxSetLimit, * ::cuCtxSynchronize */ -CUresult CUDAAPI cuCtxAttach(CUcontext *pctx, unsigned int flags); +__CUDA_DEPRECATED CUresult CUDAAPI cuCtxAttach(CUcontext *pctx, unsigned int flags); /** * \brief Decrement a context's usage-count @@ -3420,7 +4122,7 @@ CUresult CUDAAPI cuCtxAttach(CUcontext *pctx, unsigned int flags); * ::cuCtxSetLimit, * ::cuCtxSynchronize */ -CUresult CUDAAPI cuCtxDetach(CUcontext ctx); +__CUDA_DEPRECATED CUresult CUDAAPI cuCtxDetach(CUcontext ctx); /** @} */ /* END CUDA_CTX_DEPRECATED */ @@ -3463,7 +4165,8 @@ CUresult CUDAAPI cuCtxDetach(CUcontext ctx); * ::CUDA_ERROR_FILE_NOT_FOUND, * ::CUDA_ERROR_NO_BINARY_FOR_GPU, * ::CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND, - * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED + * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED, + * ::CUDA_ERROR_JIT_COMPILER_NOT_FOUND * \notefnerr * * \sa ::cuModuleGetFunction, @@ -3499,7 +4202,8 @@ CUresult CUDAAPI cuModuleLoad(CUmodule *module, const char *fname); * ::CUDA_ERROR_OUT_OF_MEMORY, * ::CUDA_ERROR_NO_BINARY_FOR_GPU, * ::CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND, - * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED + * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED, + * ::CUDA_ERROR_JIT_COMPILER_NOT_FOUND * \notefnerr * * \sa ::cuModuleGetFunction, @@ -3523,7 +4227,7 @@ CUresult CUDAAPI cuModuleLoadData(CUmodule *module, const void *image); * as Windows \c FindResource() to obtain the pointer. Options are passed as * an array via \p options and any corresponding parameters are passed in * \p optionValues. The number of total options is supplied via \p numOptions. - * Any outputs will be returned via \p optionValues. + * Any outputs will be returned via \p optionValues. * * \param module - Returned module * \param image - Module data to load @@ -3541,7 +4245,8 @@ CUresult CUDAAPI cuModuleLoadData(CUmodule *module, const void *image); * ::CUDA_ERROR_OUT_OF_MEMORY, * ::CUDA_ERROR_NO_BINARY_FOR_GPU, * ::CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND, - * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED + * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED, + * ::CUDA_ERROR_JIT_COMPILER_NOT_FOUND * \notefnerr * * \sa ::cuModuleGetFunction, @@ -3582,7 +4287,8 @@ CUresult CUDAAPI cuModuleLoadDataEx(CUmodule *module, const void *image, unsigne * ::CUDA_ERROR_OUT_OF_MEMORY, * ::CUDA_ERROR_NO_BINARY_FOR_GPU, * ::CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND, - * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED + * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED, + * ::CUDA_ERROR_JIT_COMPILER_NOT_FOUND * \notefnerr * * \sa ::cuModuleGetFunction, @@ -3680,7 +4386,9 @@ CUresult CUDAAPI cuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, const cha * ::cuModuleLoadData, * ::cuModuleLoadDataEx, * ::cuModuleLoadFatBinary, - * ::cuModuleUnload + * ::cuModuleUnload, + * ::cudaGetSymbolAddress, + * ::cudaGetSymbolSize */ CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, CUmodule hmod, const char *name); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -3714,7 +4422,8 @@ CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, CUmodule hm * ::cuModuleLoadData, * ::cuModuleLoadDataEx, * ::cuModuleLoadFatBinary, - * ::cuModuleUnload + * ::cuModuleUnload, + * ::cudaGetTextureReference */ CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef, CUmodule hmod, const char *name); @@ -3745,7 +4454,8 @@ CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef, CUmodule hmod, const char * ::cuModuleLoadData, * ::cuModuleLoadDataEx, * ::cuModuleLoadFatBinary, - * ::cuModuleUnload + * ::cuModuleUnload, + * ::cudaGetSurfaceReference */ CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef, CUmodule hmod, const char *name); @@ -3780,7 +4490,8 @@ CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef, CUmodule hmod, const ch * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE, - * ::CUDA_ERROR_OUT_OF_MEMORY + * ::CUDA_ERROR_OUT_OF_MEMORY, + * ::CUDA_ERROR_JIT_COMPILER_NOT_FOUND * \notefnerr * * \sa ::cuLinkAddData, @@ -3952,7 +4663,8 @@ cuLinkDestroy(CUlinkState state); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemGetInfo */ CUresult CUDAAPI cuMemGetInfo(size_t *free, size_t *total); @@ -3985,7 +4697,8 @@ CUresult CUDAAPI cuMemGetInfo(size_t *free, size_t *total); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMalloc */ CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, size_t bytesize); @@ -4046,7 +4759,8 @@ CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, size_t bytesize); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMallocPitch */ CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch, size_t WidthInBytes, size_t Height, unsigned int ElementSizeBytes); @@ -4075,7 +4789,8 @@ CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch, size_t Width * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaFree */ CUresult CUDAAPI cuMemFree(CUdeviceptr dptr); @@ -4096,6 +4811,7 @@ CUresult CUDAAPI cuMemFree(CUdeviceptr dptr); * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, + * ::CUDA_ERROR_NOT_FOUND, * ::CUDA_ERROR_INVALID_VALUE * \notefnerr * @@ -4129,7 +4845,7 @@ CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, CUdevic * Note all host memory allocated using ::cuMemHostAlloc() will automatically * be immediately accessible to all contexts on all devices which support unified * addressing (as may be queried using ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING). - * The device pointer that may be used to access this host memory from those + * The device pointer that may be used to access this host memory from those * contexts is always equal to the returned host pointer \p *pp. * See \ref CUDA_UNIFIED for additional details. * @@ -4154,7 +4870,8 @@ CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, CUdevic * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMallocHost */ CUresult CUDAAPI cuMemAllocHost(void **pp, size_t bytesize); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -4184,7 +4901,8 @@ CUresult CUDAAPI cuMemAllocHost(void **pp, size_t bytesize); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaFreeHost */ CUresult CUDAAPI cuMemFreeHost(void *p); @@ -4211,8 +4929,7 @@ CUresult CUDAAPI cuMemFreeHost(void *p); * * - ::CU_MEMHOSTALLOC_DEVICEMAP: Maps the allocation into the CUDA address * space. The device pointer to the memory may be obtained by calling - * ::cuMemHostGetDevicePointer(). This feature is available only on GPUs - * with compute capability greater than or equal to 1.1. + * ::cuMemHostGetDevicePointer(). * * - ::CU_MEMHOSTALLOC_WRITECOMBINED: Allocates the memory as write-combined * (WC). WC memory can be transferred across the PCI Express bus more @@ -4237,8 +4954,8 @@ CUresult CUDAAPI cuMemFreeHost(void *p); * Note all host memory allocated using ::cuMemHostAlloc() will automatically * be immediately accessible to all contexts on all devices which support unified * addressing (as may be queried using ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING). - * Unless the flag ::CU_MEMHOSTALLOC_WRITECOMBINED is specified, the device pointer - * that may be used to access this host memory from those contexts is always equal + * Unless the flag ::CU_MEMHOSTALLOC_WRITECOMBINED is specified, the device pointer + * that may be used to access this host memory from those contexts is always equal * to the returned host pointer \p *pp. If the flag ::CU_MEMHOSTALLOC_WRITECOMBINED * is specified, then the function ::cuMemHostGetDevicePointer() must be used * to query the device pointer, even if the context supports unified addressing. @@ -4266,7 +4983,8 @@ CUresult CUDAAPI cuMemFreeHost(void *p); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaHostAlloc */ CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize, unsigned int Flags); @@ -4319,7 +5037,8 @@ CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize, unsigned int Flags); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaHostGetDevicePointer */ CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, unsigned int Flags); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -4344,7 +5063,10 @@ CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, unsigned * ::CUDA_ERROR_INVALID_VALUE * \notefnerr * - * \sa ::cuMemAllocHost, ::cuMemHostAlloc + * \sa + * ::cuMemAllocHost, + * ::cuMemHostAlloc, + * ::cudaHostGetFlags */ CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p); @@ -4429,6 +5151,7 @@ CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p); * managed memory supporting devices on which contexts were previously created, even if * those contexts have been destroyed. These environment variables are described * in the CUDA programming guide under the "CUDA environment variables" section. + * - On ARM, managed memory is not available on discrete gpu with Drive PX-2. * * \param dptr - Returned device pointer * \param bytesize - Requested allocation size in bytes @@ -4454,7 +5177,8 @@ CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p); * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, - * ::cuDeviceGetAttribute, ::cuStreamAttachMemAsync + * ::cuDeviceGetAttribute, ::cuStreamAttachMemAsync, + * ::cudaMallocManaged */ CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, unsigned int flags); @@ -4469,7 +5193,7 @@ CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, unsigned * * \param dev - Returned device handle * - * \param pciBusId - String in one of the following forms: + * \param pciBusId - String in one of the following forms: * [domain]:[bus]:[device].[function] * [domain]:[bus]:[device] * [bus]:[device].[function] @@ -4483,7 +5207,11 @@ CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, unsigned * ::CUDA_ERROR_INVALID_DEVICE * \notefnerr * - * \sa ::cuDeviceGet, ::cuDeviceGetAttribute, ::cuDeviceGetPCIBusId + * \sa + * ::cuDeviceGet, + * ::cuDeviceGetAttribute, + * ::cuDeviceGetPCIBusId, + * ::cudaDeviceGetByPCIBusId */ CUresult CUDAAPI cuDeviceGetByPCIBusId(CUdevice *dev, const char *pciBusId); @@ -4511,65 +5239,73 @@ CUresult CUDAAPI cuDeviceGetByPCIBusId(CUdevice *dev, const char *pciBusId); * ::CUDA_ERROR_INVALID_DEVICE * \notefnerr * - * \sa ::cuDeviceGet, ::cuDeviceGetAttribute, ::cuDeviceGetByPCIBusId + * \sa + * ::cuDeviceGet, + * ::cuDeviceGetAttribute, + * ::cuDeviceGetByPCIBusId, + * ::cudaDeviceGetPCIBusId */ CUresult CUDAAPI cuDeviceGetPCIBusId(char *pciBusId, int len, CUdevice dev); /** * \brief Gets an interprocess handle for a previously allocated event * - * Takes as input a previously allocated event. This event must have been - * created with the ::CU_EVENT_INTERPROCESS and ::CU_EVENT_DISABLE_TIMING + * Takes as input a previously allocated event. This event must have been + * created with the ::CU_EVENT_INTERPROCESS and ::CU_EVENT_DISABLE_TIMING * flags set. This opaque handle may be copied into other processes and * opened with ::cuIpcOpenEventHandle to allow efficient hardware * synchronization between GPU work in different processes. * - * After the event has been opened in the importing process, - * ::cuEventRecord, ::cuEventSynchronize, ::cuStreamWaitEvent and - * ::cuEventQuery may be used in either process. Performing operations - * on the imported event after the exported event has been freed + * After the event has been opened in the importing process, + * ::cuEventRecord, ::cuEventSynchronize, ::cuStreamWaitEvent and + * ::cuEventQuery may be used in either process. Performing operations + * on the imported event after the exported event has been freed * with ::cuEventDestroy will result in undefined behavior. * - * IPC functionality is restricted to devices with support for unified - * addressing on Linux operating systems. + * IPC functionality is restricted to devices with support for unified + * addressing on Linux and Windows operating systems. + * IPC functionality on Windows is restricted to GPUs in TCC mode * * \param pHandle - Pointer to a user allocated CUipcEventHandle * in which to return the opaque event handle - * \param event - Event allocated with ::CU_EVENT_INTERPROCESS and + * \param event - Event allocated with ::CU_EVENT_INTERPROCESS and * ::CU_EVENT_DISABLE_TIMING flags. * * \return * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_HANDLE, * ::CUDA_ERROR_OUT_OF_MEMORY, - * ::CUDA_ERROR_MAP_FAILED + * ::CUDA_ERROR_MAP_FAILED, + * ::CUDA_ERROR_INVALID_VALUE * - * \sa - * ::cuEventCreate, - * ::cuEventDestroy, + * \sa + * ::cuEventCreate, + * ::cuEventDestroy, * ::cuEventSynchronize, * ::cuEventQuery, * ::cuStreamWaitEvent, * ::cuIpcOpenEventHandle, * ::cuIpcGetMemHandle, * ::cuIpcOpenMemHandle, - * ::cuIpcCloseMemHandle + * ::cuIpcCloseMemHandle, + * ::cudaIpcGetEventHandle */ CUresult CUDAAPI cuIpcGetEventHandle(CUipcEventHandle *pHandle, CUevent event); /** * \brief Opens an interprocess event handle for use in the current process * - * Opens an interprocess event handle exported from another process with - * ::cuIpcGetEventHandle. This function returns a ::CUevent that behaves like - * a locally created event with the ::CU_EVENT_DISABLE_TIMING flag specified. + * Opens an interprocess event handle exported from another process with + * ::cuIpcGetEventHandle. This function returns a ::CUevent that behaves like + * a locally created event with the ::CU_EVENT_DISABLE_TIMING flag specified. * This event must be freed with ::cuEventDestroy. * - * Performing operations on the imported event after the exported event has + * Performing operations on the imported event after the exported event has * been freed with ::cuEventDestroy will result in undefined behavior. * - * IPC functionality is restricted to devices with support for unified - * addressing on Linux operating systems. + * IPC functionality is restricted to devices with support for unified + * addressing on Linux and Windows operating systems. + * IPC functionality on Windows is restricted to GPUs in TCC mode * * \param phEvent - Returns the imported event * \param handle - Interprocess handle to open @@ -4579,18 +5315,20 @@ CUresult CUDAAPI cuIpcGetEventHandle(CUipcEventHandle *pHandle, CUevent event); * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_MAP_FAILED, * ::CUDA_ERROR_PEER_ACCESS_UNSUPPORTED, - * ::CUDA_ERROR_INVALID_HANDLE + * ::CUDA_ERROR_INVALID_HANDLE, + * ::CUDA_ERROR_INVALID_VALUE * * \sa - * ::cuEventCreate, - * ::cuEventDestroy, + * ::cuEventCreate, + * ::cuEventDestroy, * ::cuEventSynchronize, * ::cuEventQuery, * ::cuStreamWaitEvent, * ::cuIpcGetEventHandle, * ::cuIpcGetMemHandle, * ::cuIpcOpenMemHandle, - * ::cuIpcCloseMemHandle + * ::cuIpcCloseMemHandle, + * ::cudaIpcOpenEventHandle */ CUresult CUDAAPI cuIpcOpenEventHandle(CUevent *phEvent, CUipcEventHandle handle); @@ -4598,36 +5336,39 @@ CUresult CUDAAPI cuIpcOpenEventHandle(CUevent *phEvent, CUipcEventHandle handle) * \brief Gets an interprocess memory handle for an existing device memory * allocation * - * Takes a pointer to the base of an existing device memory allocation created - * with ::cuMemAlloc and exports it for use in another process. This is a + * Takes a pointer to the base of an existing device memory allocation created + * with ::cuMemAlloc and exports it for use in another process. This is a * lightweight operation and may be called multiple times on an allocation - * without adverse effects. + * without adverse effects. * * If a region of memory is freed with ::cuMemFree and a subsequent call * to ::cuMemAlloc returns memory with the same device address, * ::cuIpcGetMemHandle will return a unique handle for the - * new memory. + * new memory. * - * IPC functionality is restricted to devices with support for unified - * addressing on Linux operating systems. + * IPC functionality is restricted to devices with support for unified + * addressing on Linux and Windows operating systems. + * IPC functionality on Windows is restricted to GPUs in TCC mode * * \param pHandle - Pointer to user allocated ::CUipcMemHandle to return * the handle in. - * \param dptr - Base pointer to previously allocated device memory + * \param dptr - Base pointer to previously allocated device memory * * \returns * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_HANDLE, * ::CUDA_ERROR_OUT_OF_MEMORY, * ::CUDA_ERROR_MAP_FAILED, - * + * ::CUDA_ERROR_INVALID_VALUE + * * \sa * ::cuMemAlloc, * ::cuMemFree, * ::cuIpcGetEventHandle, * ::cuIpcOpenEventHandle, * ::cuIpcOpenMemHandle, - * ::cuIpcCloseMemHandle + * ::cuIpcCloseMemHandle, + * ::cudaIpcGetMemHandle */ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); @@ -4636,14 +5377,17 @@ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); * and returns a device pointer usable in the local process. * * Maps memory exported from another process with ::cuIpcGetMemHandle into - * the current device address space. For contexts on different devices + * the current device address space. For contexts on different devices * ::cuIpcOpenMemHandle can attempt to enable peer access between the - * devices as if the user called ::cuCtxEnablePeerAccess. This behavior is - * controlled by the ::CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS flag. + * devices as if the user called ::cuCtxEnablePeerAccess. This behavior is + * controlled by the ::CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS flag. * ::cuDeviceCanAccessPeer can determine if a mapping is possible. * + * ::cuIpcOpenMemHandle can open handles to devices that may not be visible + * in the process calling the API. + * * Contexts that may open ::CUipcMemHandles are restricted in the following way. - * ::CUipcMemHandles from each ::CUdevice in a given process may only be opened + * ::CUipcMemHandles from each ::CUdevice in a given process may only be opened * by one ::CUcontext per ::CUdevice per other process. * * Memory returned from ::cuIpcOpenMemHandle must be freed with @@ -4653,9 +5397,10 @@ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); * ::cuIpcCloseMemHandle in the importing context will result in undefined * behavior. * - * IPC functionality is restricted to devices with support for unified - * addressing on Linux operating systems. - * + * IPC functionality is restricted to devices with support for unified + * addressing on Linux and Windows operating systems. + * IPC functionality on Windows is restricted to GPUs in TCC mode + * * \param pdptr - Returned device pointer * \param handle - ::CUipcMemHandle to open * \param Flags - Flags for this operation. Must be specified as ::CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS @@ -4665,9 +5410,10 @@ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_MAP_FAILED, * ::CUDA_ERROR_INVALID_HANDLE, - * ::CUDA_ERROR_TOO_MANY_PEERS + * ::CUDA_ERROR_TOO_MANY_PEERS, + * ::CUDA_ERROR_INVALID_VALUE * - * \note No guarantees are made about the address returned in \p *pdptr. + * \note No guarantees are made about the address returned in \p *pdptr. * In particular, multiple processes may not receive the same address for the same \p handle. * * \sa @@ -4679,12 +5425,13 @@ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); * ::cuIpcCloseMemHandle, * ::cuCtxEnablePeerAccess, * ::cuDeviceCanAccessPeer, + * ::cudaIpcOpenMemHandle */ CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle, unsigned int Flags); /** * \brief Close memory mapped with ::cuIpcOpenMemHandle - * + * * Unmaps memory returnd by ::cuIpcOpenMemHandle. The original allocation * in the exporting process as well as imported mappings in other processes * will be unaffected. @@ -4692,17 +5439,18 @@ CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle, u * Any resources used to enable peer access will be freed if this is the * last mapping using them. * - * IPC functionality is restricted to devices with support for unified - * addressing on Linux operating systems. + * IPC functionality is restricted to devices with support for unified + * addressing on Linux and Windows operating systems. + * IPC functionality on Windows is restricted to GPUs in TCC mode * * \param dptr - Device pointer returned by ::cuIpcOpenMemHandle - * + * * \returns * ::CUDA_SUCCESS, * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_MAP_FAILED, * ::CUDA_ERROR_INVALID_HANDLE, - * + * ::CUDA_ERROR_INVALID_VALUE * \sa * ::cuMemAlloc, * ::cuMemFree, @@ -4710,6 +5458,7 @@ CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle, u * ::cuIpcOpenEventHandle, * ::cuIpcGetMemHandle, * ::cuIpcOpenMemHandle, + * ::cudaIpcCloseMemHandle */ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); @@ -4722,8 +5471,8 @@ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); * Page-locks the memory range specified by \p p and \p bytesize and maps it * for the device(s) as specified by \p Flags. This memory range also is added * to the same tracking mechanism as ::cuMemHostAlloc to automatically accelerate - * calls to functions such as ::cuMemcpyHtoD(). Since the memory can be accessed - * directly by the device, it can be read or written with much higher bandwidth + * calls to functions such as ::cuMemcpyHtoD(). Since the memory can be accessed + * directly by the device, it can be read or written with much higher bandwidth * than pageable memory that has not been registered. Page-locking excessive * amounts of memory may degrade system performance, since it reduces the amount * of memory available to the system for paging. As a result, this function is @@ -4741,8 +5490,7 @@ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); * * - ::CU_MEMHOSTREGISTER_DEVICEMAP: Maps the allocation into the CUDA address * space. The device pointer to the memory may be obtained by calling - * ::cuMemHostGetDevicePointer(). This feature is available only on GPUs - * with compute capability greater than or equal to 1.1. + * ::cuMemHostGetDevicePointer(). * * - ::CU_MEMHOSTREGISTER_IOMEMORY: The pointer is treated as pointing to some * I/O memory space, e.g. the PCI Express resource of a 3rd party device. @@ -4773,7 +5521,7 @@ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); * on devices that have a non-zero value for the device attribute. Note however that * such devices should access the memory using only of the two pointers and not both. * - * The memory page-locked by this function must be unregistered with + * The memory page-locked by this function must be unregistered with * ::cuMemHostUnregister(). * * \param p - Host pointer to memory to page-lock @@ -4792,7 +5540,11 @@ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); * ::CUDA_ERROR_NOT_SUPPORTED * \notefnerr * - * \sa ::cuMemHostUnregister, ::cuMemHostGetFlags, ::cuMemHostGetDevicePointer + * \sa + * ::cuMemHostUnregister, + * ::cuMemHostGetFlags, + * ::cuMemHostGetDevicePointer, + * ::cudaHostRegister */ CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, unsigned int Flags); @@ -4816,17 +5568,19 @@ CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, unsigned int Flags) * ::CUDA_ERROR_HOST_MEMORY_NOT_REGISTERED, * \notefnerr * - * \sa ::cuMemHostRegister + * \sa + * ::cuMemHostRegister, + * ::cudaHostUnregister */ CUresult CUDAAPI cuMemHostUnregister(void *p); /** * \brief Copies memory * - * Copies data between two pointers. - * \p dst and \p src are base pointers of the destination and source, respectively. + * Copies data between two pointers. + * \p dst and \p src are base pointers of the destination and source, respectively. * \p ByteCount specifies the number of bytes to copy. - * Note that this function infers the type of the transfer (host to host, host to + * Note that this function infers the type of the transfer (host to host, host to * device, device to device, or device to host) from the pointer values. This * function is only allowed in contexts which support unified addressing. * @@ -4852,7 +5606,10 @@ CUresult CUDAAPI cuMemHostUnregister(void *p); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpy, + * ::cudaMemcpyToSymbol, + * ::cudaMemcpyFromSymbol */ CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount); @@ -4860,9 +5617,9 @@ CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount); * \brief Copies device memory between two contexts * * Copies from device memory in one context to device memory in another - * context. \p dstDevice is the base device pointer of the destination memory - * and \p dstContext is the destination context. \p srcDevice is the base - * device pointer of the source memory and \p srcContext is the source pointer. + * context. \p dstDevice is the base device pointer of the destination memory + * and \p dstContext is the destination context. \p srcDevice is the base + * device pointer of the source memory and \p srcContext is the source pointer. * \p ByteCount specifies the number of bytes to copy. * * \param dstDevice - Destination device pointer @@ -4881,7 +5638,8 @@ CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount); * \note_sync * * \sa ::cuMemcpyDtoD, ::cuMemcpy3DPeer, ::cuMemcpyDtoDAsync, ::cuMemcpyPeerAsync, - * ::cuMemcpy3DPeerAsync + * ::cuMemcpy3DPeerAsync, + * ::cudaMemcpyPeer */ CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount); @@ -4917,7 +5675,9 @@ CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdev * ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpy, + * ::cudaMemcpyToSymbol */ CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount); @@ -4950,7 +5710,9 @@ CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, size_t * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpy, + * ::cudaMemcpyFromSymbol */ CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount); @@ -4983,7 +5745,10 @@ CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, size_t ByteC * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpy, + * ::cudaMemcpyToSymbol, + * ::cudaMemcpyFromSymbol */ CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount); @@ -5018,7 +5783,8 @@ CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpyToArray */ CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, CUdeviceptr srcDevice, size_t ByteCount); @@ -5055,7 +5821,8 @@ CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, CUdeviceptr sr * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpyFromArray */ CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, size_t srcOffset, size_t ByteCount); @@ -5090,7 +5857,8 @@ CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, size_t sr * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpyToArray */ CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount); @@ -5125,7 +5893,8 @@ CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset, const void *sr * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpyFromArray */ CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount); @@ -5164,7 +5933,8 @@ CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpyArrayToArray */ CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount); @@ -5209,9 +5979,9 @@ CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArr * * \par * If ::srcMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::srcDevice and ::srcPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::srcArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::srcArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -5236,9 +6006,9 @@ CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArr * * \par * If ::dstMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::dstDevice and ::dstPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::dstArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::dstArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -5325,7 +6095,10 @@ CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArr * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpy2D, + * ::cudaMemcpy2DToArray, + * ::cudaMemcpy2DFromArray */ CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy); @@ -5368,9 +6141,9 @@ CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy); * * \par * If ::srcMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::srcDevice and ::srcPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::srcArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::srcArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -5390,9 +6163,9 @@ CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy); * * \par * If ::dstMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::dstDevice and ::dstPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::dstArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::dstArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -5484,7 +6257,10 @@ CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpy2D, + * ::cudaMemcpy2DToArray, + * ::cudaMemcpy2DFromArray */ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); @@ -5535,9 +6311,9 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); * * \par * If ::srcMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::srcDevice and ::srcPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::srcArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::srcArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -5559,9 +6335,9 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); * * \par * If ::dstMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::dstDevice and ::dstPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::dstArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::dstArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -5652,7 +6428,8 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMemcpy3D */ CUresult CUDAAPI cuMemcpy3D(const CUDA_MEMCPY3D *pCopy); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -5677,17 +6454,18 @@ CUresult CUDAAPI cuMemcpy3D(const CUDA_MEMCPY3D *pCopy); * \note_sync * * \sa ::cuMemcpyDtoD, ::cuMemcpyPeer, ::cuMemcpyDtoDAsync, ::cuMemcpyPeerAsync, - * ::cuMemcpy3DPeerAsync + * ::cuMemcpy3DPeerAsync, + * ::cudaMemcpy3DPeer */ CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy); /** * \brief Copies memory asynchronously * - * Copies data between two pointers. - * \p dst and \p src are base pointers of the destination and source, respectively. + * Copies data between two pointers. + * \p dst and \p src are base pointers of the destination and source, respectively. * \p ByteCount specifies the number of bytes to copy. - * Note that this function infers the type of the transfer (host to host, host to + * Note that this function infers the type of the transfer (host to host, host to * device, device to device, or device to host) from the pointer values. This * function is only allowed in contexts which support unified addressing. * @@ -5701,7 +6479,8 @@ CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy); * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -5717,7 +6496,10 @@ CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy); * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpyAsync, + * ::cudaMemcpyToSymbolAsync, + * ::cudaMemcpyFromSymbolAsync */ CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount, CUstream hStream); @@ -5725,9 +6507,9 @@ CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCoun * \brief Copies device memory between two contexts asynchronously. * * Copies from device memory in one context to device memory in another - * context. \p dstDevice is the base device pointer of the destination memory - * and \p dstContext is the destination context. \p srcDevice is the base - * device pointer of the source memory and \p srcContext is the source pointer. + * context. \p dstDevice is the base device pointer of the destination memory + * and \p dstContext is the destination context. \p srcDevice is the base + * device pointer of the source memory and \p srcContext is the source pointer. * \p ByteCount specifies the number of bytes to copy. * * \param dstDevice - Destination device pointer @@ -5742,13 +6524,15 @@ CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCoun * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream * - * \sa ::cuMemcpyDtoD, ::cuMemcpyPeer, ::cuMemcpy3DPeer, ::cuMemcpyDtoDAsync, - * ::cuMemcpy3DPeerAsync + * \sa ::cuMemcpyDtoD, ::cuMemcpyPeer, ::cuMemcpy3DPeer, ::cuMemcpyDtoDAsync, + * ::cuMemcpy3DPeerAsync, + * ::cudaMemcpyPeerAsync */ CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount, CUstream hStream); #endif /* __CUDA_API_VERSION >= 4000 */ @@ -5771,7 +6555,8 @@ CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -5787,7 +6572,9 @@ CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpyAsync, + * ::cudaMemcpyToSymbolAsync */ CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream); @@ -5808,7 +6595,8 @@ CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, s * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -5824,7 +6612,9 @@ CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, s * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpyAsync, + * ::cudaMemcpyFromSymbolAsync */ CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream); @@ -5845,7 +6635,8 @@ CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, size_t * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -5861,7 +6652,10 @@ CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, size_t * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpyAsync, + * ::cudaMemcpyToSymbolAsync, + * ::cudaMemcpyFromSymbolAsync */ CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream); @@ -5884,7 +6678,8 @@ CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -5900,7 +6695,8 @@ CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpyToArrayAsync */ CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream); @@ -5923,7 +6719,8 @@ CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const voi * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -5939,7 +6736,8 @@ CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const voi * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpyFromArrayAsync */ CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream); @@ -5987,9 +6785,9 @@ CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOf * * \par * If ::srcMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::srcDevice and ::srcPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::srcArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::srcArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -6004,9 +6802,9 @@ CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOf * * \par * If ::dstMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::dstDevice and ::dstPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::dstArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::dstArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -6088,7 +6886,8 @@ CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOf * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -6104,7 +6903,10 @@ CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOf * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpy2DAsync, + * ::cudaMemcpy2DToArrayAsync, + * ::cudaMemcpy2DFromArrayAsync */ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); @@ -6155,9 +6957,9 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); * * \par * If ::srcMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::srcDevice and ::srcPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::srcArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::srcArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -6179,9 +6981,9 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); * * \par * If ::dstMemoryType is ::CU_MEMORYTYPE_UNIFIED, ::dstDevice and ::dstPitch - * specify the (unified virtual address space) base address of the source data - * and the bytes per row to apply. ::dstArray is ignored. - * This value may be used only if unified addressing is supported in the calling + * specify the (unified virtual address space) base address of the source data + * and the bytes per row to apply. ::dstArray is ignored. + * This value may be used only if unified addressing is supported in the calling * context. * * \par @@ -6260,7 +7062,8 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * \note_async * \note_null_stream @@ -6276,7 +7079,8 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemcpy3DAsync */ CUresult CUDAAPI cuMemcpy3DAsync(const CUDA_MEMCPY3D *pCopy, CUstream hStream); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -6303,7 +7107,8 @@ CUresult CUDAAPI cuMemcpy3DAsync(const CUDA_MEMCPY3D *pCopy, CUstream hStream); * \note_null_stream * * \sa ::cuMemcpyDtoD, ::cuMemcpyPeer, ::cuMemcpyDtoDAsync, ::cuMemcpyPeerAsync, - * ::cuMemcpy3DPeerAsync + * ::cuMemcpy3DPeerAsync, + * ::cudaMemcpy3DPeerAsync */ CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, CUstream hStream); #endif /* __CUDA_API_VERSION >= 4000 */ @@ -6339,7 +7144,8 @@ CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, CUstream h * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset */ CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, size_t N); @@ -6373,7 +7179,8 @@ CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, size_t N); * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset */ CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, size_t N); @@ -6407,7 +7214,8 @@ CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, size_t N) * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32Async + * ::cuMemsetD32Async, + * ::cudaMemset */ CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, size_t N); @@ -6446,7 +7254,8 @@ CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, size_t N); * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset2D */ CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height); @@ -6486,7 +7295,8 @@ CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, unsigned c * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset2D */ CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height); @@ -6526,7 +7336,8 @@ CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, unsigned * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset2D */ CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height); @@ -6562,7 +7373,8 @@ CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, unsigned * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemsetAsync */ CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream); @@ -6598,7 +7410,8 @@ CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemsetAsync */ CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size_t N, CUstream hStream); @@ -6633,7 +7446,8 @@ CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, - * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, ::cuMemsetD32 + * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, ::cuMemsetD32, + * ::cudaMemsetAsync */ CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream); @@ -6674,7 +7488,8 @@ CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset2DAsync */ CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height, CUstream hStream); @@ -6683,7 +7498,7 @@ CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsig * * Sets the 2D memory range of \p Width 16-bit values to the specified value * \p us. \p Height specifies the number of rows to set, and \p dstPitch - * specifies the number of bytes between each row. The \p dstDevice pointer + * specifies the number of bytes between each row. The \p dstDevice pointer * and \p dstPitch offset must be two byte aligned. This function performs * fastest when the pitch is one that has been passed back by * ::cuMemAllocPitch(). @@ -6716,7 +7531,8 @@ CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsig * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D32, ::cuMemsetD2D32Async, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset2DAsync */ CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height, CUstream hStream); @@ -6758,7 +7574,8 @@ CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsi * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, - * ::cuMemsetD32, ::cuMemsetD32Async + * ::cuMemsetD32, ::cuMemsetD32Async, + * ::cudaMemset2DAsync */ CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height, CUstream hStream); @@ -6861,7 +7678,8 @@ CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsi * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMallocArray */ CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR *pAllocateArray); @@ -6894,7 +7712,8 @@ CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR *pA * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaArrayGetInfo */ CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, CUarray hArray); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -6913,7 +7732,8 @@ CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, C * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_HANDLE, - * ::CUDA_ERROR_ARRAY_IS_MAPPED + * ::CUDA_ERROR_ARRAY_IS_MAPPED, + * ::CUDA_ERROR_CONTEXT_IS_DESTROYED * \notefnerr * * \sa ::cuArray3DCreate, ::cuArray3DGetDescriptor, ::cuArrayCreate, @@ -6925,7 +7745,8 @@ CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, C * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaFreeArray */ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); @@ -6954,22 +7775,22 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * - A 1D array is allocated if \p Height and \p Depth extents are both zero. * - A 2D array is allocated if only \p Depth extent is zero. * - A 3D array is allocated if all three extents are non-zero. - * - A 1D layered CUDA array is allocated if only \p Height is zero and the - * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The number + * - A 1D layered CUDA array is allocated if only \p Height is zero and the + * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The number * of layers is determined by the depth extent. - * - A 2D layered CUDA array is allocated if all three extents are non-zero and - * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The number + * - A 2D layered CUDA array is allocated if all three extents are non-zero and + * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The number * of layers is determined by the depth extent. * - A cubemap CUDA array is allocated if all three extents are non-zero and the - * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p Height, and - * \p Depth must be six. A cubemap is a special type of 2D layered CUDA array, - * where the six layers represent the six faces of a cube. The order of the six + * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p Height, and + * \p Depth must be six. A cubemap is a special type of 2D layered CUDA array, + * where the six layers represent the six faces of a cube. The order of the six * layers in memory is the same as that listed in ::CUarray_cubemap_face. - * - A cubemap layered CUDA array is allocated if all three extents are non-zero, - * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are set. - * \p Width must be equal to \p Height, and \p Depth must be a multiple of six. - * A cubemap layered CUDA array is a special type of 2D layered CUDA array that - * consists of a collection of cubemaps. The first six layers represent the first + * - A cubemap layered CUDA array is allocated if all three extents are non-zero, + * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are set. + * \p Width must be equal to \p Height, and \p Depth must be a multiple of six. + * A cubemap layered CUDA array is a special type of 2D layered CUDA array that + * consists of a collection of cubemaps. The first six layers represent the first * cubemap, the next six layers form the second cubemap, and so on. * * - ::Format specifies the format of the elements; ::CUarray_format is @@ -6990,11 +7811,11 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * - \p NumChannels specifies the number of packed components per CUDA array * element; it may be 1, 2, or 4; * - * - ::Flags may be set to - * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA arrays. If this flag is set, + * - ::Flags may be set to + * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA arrays. If this flag is set, * \p Depth specifies the number of layers, not the depth of a 3D array. - * - ::CUDA_ARRAY3D_SURFACE_LDST to enable surface references to be bound to the CUDA array. - * If this flag is not set, ::cuSurfRefSetArray will fail when attempting to bind the CUDA array + * - ::CUDA_ARRAY3D_SURFACE_LDST to enable surface references to be bound to the CUDA array. + * If this flag is not set, ::cuSurfRefSetArray will fail when attempting to bind the CUDA array * to a surface reference. * - ::CUDA_ARRAY3D_CUBEMAP to enable creation of cubemaps. If this flag is set, \p Width must be * equal to \p Height, and \p Depth must be six. If the ::CUDA_ARRAY3D_LAYERED flag is also set, @@ -7002,20 +7823,20 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * - ::CUDA_ARRAY3D_TEXTURE_GATHER to indicate that the CUDA array will be used for texture gather. * Texture gather can only be performed on 2D CUDA arrays. * - * \p Width, \p Height and \p Depth must meet certain size requirements as listed in the following table. - * All values are specified in elements. Note that for brevity's sake, the full name of the device attribute - * is not specified. For ex., TEXTURE1D_WIDTH refers to the device attribute + * \p Width, \p Height and \p Depth must meet certain size requirements as listed in the following table. + * All values are specified in elements. Note that for brevity's sake, the full name of the device attribute + * is not specified. For ex., TEXTURE1D_WIDTH refers to the device attribute * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH. * - * Note that 2D CUDA arrays have different size requirements if the ::CUDA_ARRAY3D_TEXTURE_GATHER flag - * is set. \p Width and \p Height must not be greater than ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH + * Note that 2D CUDA arrays have different size requirements if the ::CUDA_ARRAY3D_TEXTURE_GATHER flag + * is set. \p Width and \p Height must not be greater than ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH * and ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_HEIGHT respectively, in that case. * * <table> * <tr><td><b>CUDA array type</b></td> - * <td><b>Valid extents that must always be met<br>{(width range in elements), (height range), + * <td><b>Valid extents that must always be met<br>{(width range in elements), (height range), * (depth range)}</b></td> - * <td><b>Valid extents with CUDA_ARRAY3D_SURFACE_LDST set<br> + * <td><b>Valid extents with CUDA_ARRAY3D_SURFACE_LDST set<br> * {(width range in elements), (height range), (depth range)}</b></td></tr> * <tr><td>1D</td> * <td><small>{ (1,TEXTURE1D_WIDTH), 0, 0 }</small></td> @@ -7025,28 +7846,28 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * <td><small>{ (1,SURFACE2D_WIDTH), (1,SURFACE2D_HEIGHT), 0 }</small></td></tr> * <tr><td>3D</td> * <td><small>{ (1,TEXTURE3D_WIDTH), (1,TEXTURE3D_HEIGHT), (1,TEXTURE3D_DEPTH) } - * <br>OR<br>{ (1,TEXTURE3D_WIDTH_ALTERNATE), (1,TEXTURE3D_HEIGHT_ALTERNATE), + * <br>OR<br>{ (1,TEXTURE3D_WIDTH_ALTERNATE), (1,TEXTURE3D_HEIGHT_ALTERNATE), * (1,TEXTURE3D_DEPTH_ALTERNATE) }</small></td> - * <td><small>{ (1,SURFACE3D_WIDTH), (1,SURFACE3D_HEIGHT), + * <td><small>{ (1,SURFACE3D_WIDTH), (1,SURFACE3D_HEIGHT), * (1,SURFACE3D_DEPTH) }</small></td></tr> * <tr><td>1D Layered</td> - * <td><small>{ (1,TEXTURE1D_LAYERED_WIDTH), 0, + * <td><small>{ (1,TEXTURE1D_LAYERED_WIDTH), 0, * (1,TEXTURE1D_LAYERED_LAYERS) }</small></td> - * <td><small>{ (1,SURFACE1D_LAYERED_WIDTH), 0, + * <td><small>{ (1,SURFACE1D_LAYERED_WIDTH), 0, * (1,SURFACE1D_LAYERED_LAYERS) }</small></td></tr> * <tr><td>2D Layered</td> - * <td><small>{ (1,TEXTURE2D_LAYERED_WIDTH), (1,TEXTURE2D_LAYERED_HEIGHT), + * <td><small>{ (1,TEXTURE2D_LAYERED_WIDTH), (1,TEXTURE2D_LAYERED_HEIGHT), * (1,TEXTURE2D_LAYERED_LAYERS) }</small></td> - * <td><small>{ (1,SURFACE2D_LAYERED_WIDTH), (1,SURFACE2D_LAYERED_HEIGHT), + * <td><small>{ (1,SURFACE2D_LAYERED_WIDTH), (1,SURFACE2D_LAYERED_HEIGHT), * (1,SURFACE2D_LAYERED_LAYERS) }</small></td></tr> * <tr><td>Cubemap</td> * <td><small>{ (1,TEXTURECUBEMAP_WIDTH), (1,TEXTURECUBEMAP_WIDTH), 6 }</small></td> - * <td><small>{ (1,SURFACECUBEMAP_WIDTH), + * <td><small>{ (1,SURFACECUBEMAP_WIDTH), * (1,SURFACECUBEMAP_WIDTH), 6 }</small></td></tr> * <tr><td>Cubemap Layered</td> - * <td><small>{ (1,TEXTURECUBEMAP_LAYERED_WIDTH), (1,TEXTURECUBEMAP_LAYERED_WIDTH), + * <td><small>{ (1,TEXTURECUBEMAP_LAYERED_WIDTH), (1,TEXTURECUBEMAP_LAYERED_WIDTH), * (1,TEXTURECUBEMAP_LAYERED_LAYERS) }</small></td> - * <td><small>{ (1,SURFACECUBEMAP_LAYERED_WIDTH), (1,SURFACECUBEMAP_LAYERED_WIDTH), + * <td><small>{ (1,SURFACECUBEMAP_LAYERED_WIDTH), (1,SURFACECUBEMAP_LAYERED_WIDTH), * (1,SURFACECUBEMAP_LAYERED_LAYERS) }</small></td></tr> * </table> * @@ -7105,7 +7926,8 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaMalloc3DArray */ CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray); @@ -7129,7 +7951,8 @@ CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE, - * ::CUDA_ERROR_INVALID_HANDLE + * ::CUDA_ERROR_INVALID_HANDLE, + * ::CUDA_ERROR_CONTEXT_IS_DESTROYED * \notefnerr * * \sa ::cuArray3DCreate, ::cuArrayCreate, @@ -7141,7 +7964,8 @@ CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, - * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 + * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, + * ::cudaArrayGetInfo */ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -7175,22 +7999,22 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * - A 1D mipmapped array is allocated if \p Height and \p Depth extents are both zero. * - A 2D mipmapped array is allocated if only \p Depth extent is zero. * - A 3D mipmapped array is allocated if all three extents are non-zero. - * - A 1D layered CUDA mipmapped array is allocated if only \p Height is zero and the - * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The number + * - A 1D layered CUDA mipmapped array is allocated if only \p Height is zero and the + * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The number * of layers is determined by the depth extent. - * - A 2D layered CUDA mipmapped array is allocated if all three extents are non-zero and - * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The number + * - A 2D layered CUDA mipmapped array is allocated if all three extents are non-zero and + * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The number * of layers is determined by the depth extent. * - A cubemap CUDA mipmapped array is allocated if all three extents are non-zero and the - * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p Height, and - * \p Depth must be six. A cubemap is a special type of 2D layered CUDA array, - * where the six layers represent the six faces of a cube. The order of the six + * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p Height, and + * \p Depth must be six. A cubemap is a special type of 2D layered CUDA array, + * where the six layers represent the six faces of a cube. The order of the six * layers in memory is the same as that listed in ::CUarray_cubemap_face. - * - A cubemap layered CUDA mipmapped array is allocated if all three extents are non-zero, - * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are set. - * \p Width must be equal to \p Height, and \p Depth must be a multiple of six. - * A cubemap layered CUDA array is a special type of 2D layered CUDA array that - * consists of a collection of cubemaps. The first six layers represent the first + * - A cubemap layered CUDA mipmapped array is allocated if all three extents are non-zero, + * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are set. + * \p Width must be equal to \p Height, and \p Depth must be a multiple of six. + * A cubemap layered CUDA array is a special type of 2D layered CUDA array that + * consists of a collection of cubemaps. The first six layers represent the first * cubemap, the next six layers form the second cubemap, and so on. * * - ::Format specifies the format of the elements; ::CUarray_format is @@ -7211,11 +8035,11 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * - \p NumChannels specifies the number of packed components per CUDA array * element; it may be 1, 2, or 4; * - * - ::Flags may be set to - * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA mipmapped arrays. If this flag is set, + * - ::Flags may be set to + * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA mipmapped arrays. If this flag is set, * \p Depth specifies the number of layers, not the depth of a 3D array. * - ::CUDA_ARRAY3D_SURFACE_LDST to enable surface references to be bound to individual mipmap levels of - * the CUDA mipmapped array. If this flag is not set, ::cuSurfRefSetArray will fail when attempting to + * the CUDA mipmapped array. If this flag is not set, ::cuSurfRefSetArray will fail when attempting to * bind a mipmap level of the CUDA mipmapped array to a surface reference. * - ::CUDA_ARRAY3D_CUBEMAP to enable creation of mipmapped cubemaps. If this flag is set, \p Width must be * equal to \p Height, and \p Depth must be six. If the ::CUDA_ARRAY3D_LAYERED flag is also set, @@ -7223,34 +8047,48 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * - ::CUDA_ARRAY3D_TEXTURE_GATHER to indicate that the CUDA mipmapped array will be used for texture gather. * Texture gather can only be performed on 2D CUDA mipmapped arrays. * - * \p Width, \p Height and \p Depth must meet certain size requirements as listed in the following table. - * All values are specified in elements. Note that for brevity's sake, the full name of the device attribute - * is not specified. For ex., TEXTURE1D_MIPMAPPED_WIDTH refers to the device attribute + * \p Width, \p Height and \p Depth must meet certain size requirements as listed in the following table. + * All values are specified in elements. Note that for brevity's sake, the full name of the device attribute + * is not specified. For ex., TEXTURE1D_MIPMAPPED_WIDTH refers to the device attribute * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH. * * <table> * <tr><td><b>CUDA array type</b></td> - * <td><b>Valid extents that must always be met<br>{(width range in elements), (height range), - * (depth range)}</b></td></tr> + * <td><b>Valid extents that must always be met<br>{(width range in elements), (height range), + * (depth range)}</b></td> + * <td><b>Valid extents with CUDA_ARRAY3D_SURFACE_LDST set<br> + * {(width range in elements), (height range), (depth range)}</b></td></tr> * <tr><td>1D</td> - * <td><small>{ (1,TEXTURE1D_MIPMAPPED_WIDTH), 0, 0 }</small></td></tr> + * <td><small>{ (1,TEXTURE1D_MIPMAPPED_WIDTH), 0, 0 }</small></td> + * <td><small>{ (1,SURFACE1D_WIDTH), 0, 0 }</small></td></tr> * <tr><td>2D</td> - * <td><small>{ (1,TEXTURE2D_MIPMAPPED_WIDTH), (1,TEXTURE2D_MIPMAPPED_HEIGHT), 0 }</small></td></tr> + * <td><small>{ (1,TEXTURE2D_MIPMAPPED_WIDTH), (1,TEXTURE2D_MIPMAPPED_HEIGHT), 0 }</small></td> + * <td><small>{ (1,SURFACE2D_WIDTH), (1,SURFACE2D_HEIGHT), 0 }</small></td></tr> * <tr><td>3D</td> * <td><small>{ (1,TEXTURE3D_WIDTH), (1,TEXTURE3D_HEIGHT), (1,TEXTURE3D_DEPTH) } - * <br>OR<br>{ (1,TEXTURE3D_WIDTH_ALTERNATE), (1,TEXTURE3D_HEIGHT_ALTERNATE), - * (1,TEXTURE3D_DEPTH_ALTERNATE) }</small></td></tr> + * <br>OR<br>{ (1,TEXTURE3D_WIDTH_ALTERNATE), (1,TEXTURE3D_HEIGHT_ALTERNATE), + * (1,TEXTURE3D_DEPTH_ALTERNATE) }</small></td> + * <td><small>{ (1,SURFACE3D_WIDTH), (1,SURFACE3D_HEIGHT), + * (1,SURFACE3D_DEPTH) }</small></td></tr> * <tr><td>1D Layered</td> - * <td><small>{ (1,TEXTURE1D_LAYERED_WIDTH), 0, - * (1,TEXTURE1D_LAYERED_LAYERS) }</small></td></tr> + * <td><small>{ (1,TEXTURE1D_LAYERED_WIDTH), 0, + * (1,TEXTURE1D_LAYERED_LAYERS) }</small></td> + * <td><small>{ (1,SURFACE1D_LAYERED_WIDTH), 0, + * (1,SURFACE1D_LAYERED_LAYERS) }</small></td></tr> * <tr><td>2D Layered</td> - * <td><small>{ (1,TEXTURE2D_LAYERED_WIDTH), (1,TEXTURE2D_LAYERED_HEIGHT), - * (1,TEXTURE2D_LAYERED_LAYERS) }</small></td></tr> + * <td><small>{ (1,TEXTURE2D_LAYERED_WIDTH), (1,TEXTURE2D_LAYERED_HEIGHT), + * (1,TEXTURE2D_LAYERED_LAYERS) }</small></td> + * <td><small>{ (1,SURFACE2D_LAYERED_WIDTH), (1,SURFACE2D_LAYERED_HEIGHT), + * (1,SURFACE2D_LAYERED_LAYERS) }</small></td></tr> * <tr><td>Cubemap</td> - * <td><small>{ (1,TEXTURECUBEMAP_WIDTH), (1,TEXTURECUBEMAP_WIDTH), 6 }</small></td></tr> + * <td><small>{ (1,TEXTURECUBEMAP_WIDTH), (1,TEXTURECUBEMAP_WIDTH), 6 }</small></td> + * <td><small>{ (1,SURFACECUBEMAP_WIDTH), + * (1,SURFACECUBEMAP_WIDTH), 6 }</small></td></tr> * <tr><td>Cubemap Layered</td> - * <td><small>{ (1,TEXTURECUBEMAP_LAYERED_WIDTH), (1,TEXTURECUBEMAP_LAYERED_WIDTH), - * (1,TEXTURECUBEMAP_LAYERED_LAYERS) }</small></td></tr> + * <td><small>{ (1,TEXTURECUBEMAP_LAYERED_WIDTH), (1,TEXTURECUBEMAP_LAYERED_WIDTH), + * (1,TEXTURECUBEMAP_LAYERED_LAYERS) }</small></td> + * <td><small>{ (1,SURFACECUBEMAP_LAYERED_WIDTH), (1,SURFACECUBEMAP_LAYERED_WIDTH), + * (1,SURFACECUBEMAP_LAYERED_LAYERS) }</small></td></tr> * </table> * * @@ -7268,7 +8106,11 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * ::CUDA_ERROR_UNKNOWN * \notefnerr * - * \sa ::cuMipmappedArrayDestroy, ::cuMipmappedArrayGetLevel, ::cuArrayCreate, + * \sa + * ::cuMipmappedArrayDestroy, + * ::cuMipmappedArrayGetLevel, + * ::cuArrayCreate, + * ::cudaMallocMipmappedArray */ CUresult CUDAAPI cuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc, unsigned int numMipmapLevels); @@ -7294,7 +8136,11 @@ CUresult CUDAAPI cuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_AR * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * - * \sa ::cuMipmappedArrayCreate, ::cuMipmappedArrayDestroy, ::cuArrayCreate, + * \sa + * ::cuMipmappedArrayCreate, + * ::cuMipmappedArrayDestroy, + * ::cuArrayCreate, + * ::cudaGetMipmappedArrayLevel */ CUresult CUDAAPI cuMipmappedArrayGetLevel(CUarray *pLevelArray, CUmipmappedArray hMipmappedArray, unsigned int level); @@ -7311,10 +8157,15 @@ CUresult CUDAAPI cuMipmappedArrayGetLevel(CUarray *pLevelArray, CUmipmappedArray * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_HANDLE, - * ::CUDA_ERROR_ARRAY_IS_MAPPED + * ::CUDA_ERROR_ARRAY_IS_MAPPED, + * ::CUDA_ERROR_CONTEXT_IS_DESTROYED * \notefnerr * - * \sa ::cuMipmappedArrayCreate, ::cuMipmappedArrayGetLevel, ::cuArrayCreate, + * \sa + * ::cuMipmappedArrayCreate, + * ::cuMipmappedArrayGetLevel, + * ::cuArrayCreate, + * ::cudaFreeMipmappedArray */ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); @@ -7328,43 +8179,42 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * ___MANBRIEF___ unified addressing functions of the low-level CUDA driver * API (___CURRENT_FILE___) ___ENDMANBRIEF___ * - * This section describes the unified addressing functions of the + * This section describes the unified addressing functions of the * low-level CUDA driver application programming interface. * * @{ * * \section CUDA_UNIFIED_overview Overview * - * CUDA devices can share a unified address space with the host. + * CUDA devices can share a unified address space with the host. * For these devices there is no distinction between a device - * pointer and a host pointer -- the same pointer value may be - * used to access memory from the host program and from a kernel + * pointer and a host pointer -- the same pointer value may be + * used to access memory from the host program and from a kernel * running on the device (with exceptions enumerated below). * * \section CUDA_UNIFIED_support Supported Platforms - * - * Whether or not a device supports unified addressing may be - * queried by calling ::cuDeviceGetAttribute() with the device + * + * Whether or not a device supports unified addressing may be + * queried by calling ::cuDeviceGetAttribute() with the device * attribute ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING. * - * Unified addressing is automatically enabled in 64-bit processes - * on devices with compute capability greater than or equal to 2.0. + * Unified addressing is automatically enabled in 64-bit processes * * \section CUDA_UNIFIED_lookup Looking Up Information from Pointer Values * - * It is possible to look up information about the memory which backs a + * It is possible to look up information about the memory which backs a * pointer value. For instance, one may want to know if a pointer points - * to host or device memory. As another example, in the case of device - * memory, one may want to know on which CUDA device the memory - * resides. These properties may be queried using the function + * to host or device memory. As another example, in the case of device + * memory, one may want to know on which CUDA device the memory + * resides. These properties may be queried using the function * ::cuPointerGetAttribute() * * Since pointers are unique, it is not necessary to specify information - * about the pointers specified to the various copy functions in the + * about the pointers specified to the various copy functions in the * CUDA API. The function ::cuMemcpy() may be used to perform a copy * between two pointers, ignoring whether they point to host or device * memory (making ::cuMemcpyHtoD(), ::cuMemcpyDtoD(), and ::cuMemcpyDtoH() - * unnecessary for devices supporting unified addressing). For + * unnecessary for devices supporting unified addressing). For * multidimensional copies, the memory type ::CU_MEMORYTYPE_UNIFIED may be * used to specify that the CUDA driver should infer the location of the * pointer from its value. @@ -7373,45 +8223,45 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * * All host memory allocated in all contexts using ::cuMemAllocHost() and * ::cuMemHostAlloc() is always directly accessible from all contexts on - * all devices that support unified addressing. This is the case regardless + * all devices that support unified addressing. This is the case regardless * of whether or not the flags ::CU_MEMHOSTALLOC_PORTABLE and * ::CU_MEMHOSTALLOC_DEVICEMAP are specified. * - * The pointer value through which allocated host memory may be accessed - * in kernels on all devices that support unified addressing is the same + * The pointer value through which allocated host memory may be accessed + * in kernels on all devices that support unified addressing is the same * as the pointer value through which that memory is accessed on the host, - * so it is not necessary to call ::cuMemHostGetDevicePointer() to get the device + * so it is not necessary to call ::cuMemHostGetDevicePointer() to get the device * pointer for these allocations. - * + * * Note that this is not the case for memory allocated using the flag * ::CU_MEMHOSTALLOC_WRITECOMBINED, as discussed below. * * \section CUDA_UNIFIED_autopeerregister Automatic Registration of Peer Memory * - * Upon enabling direct access from a context that supports unified addressing - * to another peer context that supports unified addressing using - * ::cuCtxEnablePeerAccess() all memory allocated in the peer context using - * ::cuMemAlloc() and ::cuMemAllocPitch() will immediately be accessible + * Upon enabling direct access from a context that supports unified addressing + * to another peer context that supports unified addressing using + * ::cuCtxEnablePeerAccess() all memory allocated in the peer context using + * ::cuMemAlloc() and ::cuMemAllocPitch() will immediately be accessible * by the current context. The device pointer value through * which any peer memory may be accessed in the current context * is the same pointer value through which that memory may be * accessed in the peer context. * * \section CUDA_UNIFIED_exceptions Exceptions, Disjoint Addressing - * + * * Not all memory may be accessed on devices through the same pointer * value through which they are accessed on the host. These exceptions * are host memory registered using ::cuMemHostRegister() and host memory - * allocated using the flag ::CU_MEMHOSTALLOC_WRITECOMBINED. For these + * allocated using the flag ::CU_MEMHOSTALLOC_WRITECOMBINED. For these * exceptions, there exists a distinct host and device address for the * memory. The device address is guaranteed to not overlap any valid host - * pointer range and is guaranteed to have the same value across all - * contexts that support unified addressing. - * - * This device address may be queried using ::cuMemHostGetDevicePointer() - * when a context using unified addressing is current. Either the host - * or the unified device pointer value may be used to refer to this memory - * through ::cuMemcpy() and similar functions using the + * pointer range and is guaranteed to have the same value across all + * contexts that support unified addressing. + * + * This device address may be queried using ::cuMemHostGetDevicePointer() + * when a context using unified addressing is current. Either the host + * or the unified device pointer value may be used to refer to this memory + * through ::cuMemcpy() and similar functions using the * ::CU_MEMORYTYPE_UNIFIED memory type. * */ @@ -7419,69 +8269,69 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); #if __CUDA_API_VERSION >= 4000 /** * \brief Returns information about a pointer - * + * * The supported attributes are: - * - * - ::CU_POINTER_ATTRIBUTE_CONTEXT: - * - * Returns in \p *data the ::CUcontext in which \p ptr was allocated or - * registered. - * The type of \p data must be ::CUcontext *. - * + * + * - ::CU_POINTER_ATTRIBUTE_CONTEXT: + * + * Returns in \p *data the ::CUcontext in which \p ptr was allocated or + * registered. + * The type of \p data must be ::CUcontext *. + * * If \p ptr was not allocated by, mapped by, or registered with - * a ::CUcontext which uses unified virtual addressing then + * a ::CUcontext which uses unified virtual addressing then * ::CUDA_ERROR_INVALID_VALUE is returned. - * - * - ::CU_POINTER_ATTRIBUTE_MEMORY_TYPE: - * - * Returns in \p *data the physical memory type of the memory that + * + * - ::CU_POINTER_ATTRIBUTE_MEMORY_TYPE: + * + * Returns in \p *data the physical memory type of the memory that * \p ptr addresses as a ::CUmemorytype enumerated value. * The type of \p data must be unsigned int. - * - * If \p ptr addresses device memory then \p *data is set to - * ::CU_MEMORYTYPE_DEVICE. The particular ::CUdevice on which the - * memory resides is the ::CUdevice of the ::CUcontext returned by the + * + * If \p ptr addresses device memory then \p *data is set to + * ::CU_MEMORYTYPE_DEVICE. The particular ::CUdevice on which the + * memory resides is the ::CUdevice of the ::CUcontext returned by the * ::CU_POINTER_ATTRIBUTE_CONTEXT attribute of \p ptr. - * - * If \p ptr addresses host memory then \p *data is set to + * + * If \p ptr addresses host memory then \p *data is set to * ::CU_MEMORYTYPE_HOST. - * + * * If \p ptr was not allocated by, mapped by, or registered with - * a ::CUcontext which uses unified virtual addressing then + * a ::CUcontext which uses unified virtual addressing then * ::CUDA_ERROR_INVALID_VALUE is returned. * - * If the current ::CUcontext does not support unified virtual + * If the current ::CUcontext does not support unified virtual * addressing then ::CUDA_ERROR_INVALID_CONTEXT is returned. - * + * * - ::CU_POINTER_ATTRIBUTE_DEVICE_POINTER: - * + * * Returns in \p *data the device pointer value through which - * \p ptr may be accessed by kernels running in the current + * \p ptr may be accessed by kernels running in the current * ::CUcontext. * The type of \p data must be CUdeviceptr *. - * + * * If there exists no device pointer value through which * kernels running in the current ::CUcontext may access * \p ptr then ::CUDA_ERROR_INVALID_VALUE is returned. - * - * If there is no current ::CUcontext then + * + * If there is no current ::CUcontext then * ::CUDA_ERROR_INVALID_CONTEXT is returned. - * - * Except in the exceptional disjoint addressing cases discussed - * below, the value returned in \p *data will equal the input + * + * Except in the exceptional disjoint addressing cases discussed + * below, the value returned in \p *data will equal the input * value \p ptr. - * + * * - ::CU_POINTER_ATTRIBUTE_HOST_POINTER: - * - * Returns in \p *data the host pointer value through which + * + * Returns in \p *data the host pointer value through which * \p ptr may be accessed by by the host program. * The type of \p data must be void **. * If there exists no host pointer value through which - * the host program may directly access \p ptr then + * the host program may directly access \p ptr then * ::CUDA_ERROR_INVALID_VALUE is returned. - * - * Except in the exceptional disjoint addressing cases discussed - * below, the value returned in \p *data will equal the input + * + * Except in the exceptional disjoint addressing cases discussed + * below, the value returned in \p *data will equal the input * value \p ptr. * * - ::CU_POINTER_ATTRIBUTE_P2P_TOKENS: @@ -7497,7 +8347,7 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * Querying this attribute has a side effect of setting the attribute * ::CU_POINTER_ATTRIBUTE_SYNC_MEMOPS for the region of memory that * \p ptr points to. - * + * * - ::CU_POINTER_ATTRIBUTE_SYNC_MEMOPS: * * A boolean attribute which when set, ensures that synchronous memory operations @@ -7522,22 +8372,27 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * Returns in \p *data a boolean that indicates whether the pointer points to * managed memory or not. * + * - ::CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL: + * + * Returns in \p *data an integer representing a device ordinal of a device against + * which the memory was allocated or registered. + * * \par * * Note that for most allocations in the unified virtual address space - * the host and device pointer for accessing the allocation will be the + * the host and device pointer for accessing the allocation will be the * same. The exceptions to this are - * - user memory registered using ::cuMemHostRegister - * - host memory allocated using ::cuMemHostAlloc with the + * - user memory registered using ::cuMemHostRegister + * - host memory allocated using ::cuMemHostAlloc with the * ::CU_MEMHOSTALLOC_WRITECOMBINED flag - * For these types of allocation there will exist separate, disjoint host - * and device addresses for accessing the allocation. In particular - * - The host address will correspond to an invalid unmapped device address - * (which will result in an exception if accessed from the device) - * - The device address will correspond to an invalid unmapped host address + * For these types of allocation there will exist separate, disjoint host + * and device addresses for accessing the allocation. In particular + * - The host address will correspond to an invalid unmapped device address + * (which will result in an exception if accessed from the device) + * - The device address will correspond to an invalid unmapped host address * (which will result in an exception if accessed from the host). - * For these types of allocations, querying ::CU_POINTER_ATTRIBUTE_HOST_POINTER - * and ::CU_POINTER_ATTRIBUTE_DEVICE_POINTER may be used to retrieve the host + * For these types of allocations, querying ::CU_POINTER_ATTRIBUTE_HOST_POINTER + * and ::CU_POINTER_ATTRIBUTE_DEVICE_POINTER may be used to retrieve the host * and device addresses from either address. * * \param data - Returned pointer attribute value @@ -7553,14 +8408,16 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * ::CUDA_ERROR_INVALID_DEVICE * \notefnerr * - * \sa cuPointerSetAttribute, + * \sa + * ::cuPointerSetAttribute, * ::cuMemAlloc, * ::cuMemFree, * ::cuMemAllocHost, * ::cuMemFreeHost, * ::cuMemHostAlloc, * ::cuMemHostRegister, - * ::cuMemHostUnregister + * ::cuMemHostUnregister, + * ::cudaPointerGetAttributes */ CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute, CUdeviceptr ptr); #endif /* __CUDA_API_VERSION >= 4000 */ @@ -7569,8 +8426,8 @@ CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute /** * \brief Prefetches memory to the specified destination device * - * Prefetches memory to the specified destination device. \p devPtr is the - * base device pointer of the memory to be prefetched and \p dstDevice is the + * Prefetches memory to the specified destination device. \p devPtr is the + * base device pointer of the memory to be prefetched and \p dstDevice is the * destination device. \p count specifies the number of bytes to copy. \p hStream * is the stream in which the operation is enqueued. The memory range must refer * to managed memory allocated via ::cuMemAllocManaged or declared via __managed__ variables. @@ -7629,7 +8486,8 @@ CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute * \note_null_stream * * \sa ::cuMemcpy, ::cuMemcpyPeer, ::cuMemcpyAsync, - * ::cuMemcpy3DPeerAsync, ::cuMemAdvise + * ::cuMemcpy3DPeerAsync, ::cuMemAdvise, + * ::cudaMemPrefetchAsync */ CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream); @@ -7640,7 +8498,10 @@ CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice d * starting at \p devPtr with a size of \p count bytes. The start address and end address of the memory * range will be rounded down and rounded up respectively to be aligned to CPU page size before the * advice is applied. The memory range must refer to managed memory allocated via ::cuMemAllocManaged - * or declared via __managed__ variables. + * or declared via __managed__ variables. The memory range could also refer to system-allocated pageable + * memory provided it represents a valid, host-accessible region of memory and all additional constraints + * imposed by \p advice as outlined below are also satisfied. Specifying an invalid system-allocated pageable + * memory range results in an error being returned. * * The \p advice parameter can take the following values: * - ::CU_MEM_ADVISE_SET_READ_MOSTLY: This implies that the data is mostly going to be read @@ -7654,11 +8515,18 @@ CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice d * Also, if a context is created on a device that does not have the device attribute * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS set, then read-duplication will not occur until * all such contexts are destroyed. + * If the memory region refers to valid system-allocated pageable memory, then the accessing device must + * have a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS for a read-only + * copy to be created on that device. Note however that if the accessing device also has a non-zero value for the + * device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, then setting this advice + * will not create a read-only copy when that device accesses this memory region. + * * - ::CU_MEM_ADVISE_UNSET_READ_MOSTLY: Undoes the effect of ::CU_MEM_ADVISE_SET_READ_MOSTLY and also prevents the * Unified Memory driver from attempting heuristic read-duplication on the memory range. Any read-duplicated * copies of the data will be collapsed into a single copy. The location for the collapsed * copy will be the preferred location if the page has a preferred location and one of the read-duplicated * copies was resident at that location. Otherwise, the location chosen is arbitrary. + * * - ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION: This advice sets the preferred location for the * data to be the memory belonging to \p device. Passing in CU_DEVICE_CPU for \p device sets the * preferred location as host memory. If \p device is a GPU, then it must have a non-zero value for the @@ -7675,9 +8543,17 @@ CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice d * memory, the page may eventually be pinned to host memory by the Unified Memory driver. But * if the preferred location is set as device memory, then the page will continue to thrash indefinitely. * If ::CU_MEM_ADVISE_SET_READ_MOSTLY is also set on this memory region or any subset of it, then the - * policies associated with that advice will override the policies of this advice. + * policies associated with that advice will override the policies of this advice, unless read accesses from + * \p device will not result in a read-only copy being created on that device as outlined in description for + * the advice ::CU_MEM_ADVISE_SET_READ_MOSTLY. + * If the memory region refers to valid system-allocated pageable memory, then \p device must have a non-zero + * value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has + * a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, + * then this call has no effect. Note however that this behavior may change in the future. + * * - ::CU_MEM_ADVISE_UNSET_PREFERRED_LOCATION: Undoes the effect of ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION * and changes the preferred location to none. + * * - ::CU_MEM_ADVISE_SET_ACCESSED_BY: This advice implies that the data will be accessed by \p device. * Passing in ::CU_DEVICE_CPU for \p device will set the advice for the CPU. If \p device is a GPU, then * the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS must be non-zero. @@ -7698,8 +8574,17 @@ CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice d * policies associated with that advice will override the policies of this advice. Additionally, if the * preferred location of this memory region or any subset of it is also \p device, then the policies * associated with ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION will override the policies of this advice. + * If the memory region refers to valid system-allocated pageable memory, then \p device must have a non-zero + * value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has + * a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, + * then this call has no effect. + * * - ::CU_MEM_ADVISE_UNSET_ACCESSED_BY: Undoes the effect of ::CU_MEM_ADVISE_SET_ACCESSED_BY. Any mappings to * the data from \p device may be removed at any time causing accesses to result in non-fatal page faults. + * If the memory region refers to valid system-allocated pageable memory, then \p device must have a non-zero + * value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has + * a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, + * then this call has no effect. * * \param devPtr - Pointer to memory to set the advice for * \param count - Size in bytes of the memory range @@ -7715,13 +8600,14 @@ CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice d * \note_null_stream * * \sa ::cuMemcpy, ::cuMemcpyPeer, ::cuMemcpyAsync, - * ::cuMemcpy3DPeerAsync, ::cuMemPrefetchAsync + * ::cuMemcpy3DPeerAsync, ::cuMemPrefetchAsync, + * ::cudaMemAdvise */ CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, CUmem_advise advice, CUdevice device); /** * \brief Query an attribute of a given memory range - * + * * Query an attribute about the memory range starting at \p devPtr with a size of \p count bytes. The * memory range must refer to managed memory allocated via ::cuMemAllocManaged or declared via * __managed__ variables. @@ -7736,7 +8622,7 @@ CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, CUmem_advise advi * if all pages in the memory range have the CPU as their preferred location, or it will be CU_DEVICE_INVALID * if either all the pages don't have the same preferred location or some of the pages don't have a * preferred location at all. Note that the actual location of the pages in the memory range at the time of - * the query may be different from the preferred location. + * the query may be different from the preferred location. * - ::CU_MEM_RANGE_ATTRIBUTE_ACCESSED_BY: If this attribute is specified, \p data will be interpreted * as an array of 32-bit integers, and \p dataSize must be a non-zero multiple of 4. The result returned * will be a list of device ids that had ::CU_MEM_ADVISE_SET_ACCESSED_BY set for that entire memory range. @@ -7772,7 +8658,8 @@ CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, CUmem_advise advi * \note_null_stream * * \sa ::cuMemRangeGetAttributes, ::cuMemPrefetchAsync, - * ::cuMemAdvise + * ::cuMemAdvise, + * ::cudaMemRangeGetAttribute */ CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, CUmem_range_attribute attribute, CUdeviceptr devPtr, size_t count); @@ -7811,7 +8698,8 @@ CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, CUmem_range * \notefnerr * * \sa ::cuMemRangeGetAttribute, ::cuMemAdvise - * ::cuMemPrefetchAsync + * ::cuMemPrefetchAsync, + * ::cudaMemRangeGetAttributes */ CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes, CUmem_range_attribute *attributes, size_t numAttributes, CUdeviceptr devPtr, size_t count); #endif /* __CUDA_API_VERSION >= 8000 */ @@ -7873,6 +8761,7 @@ CUresult CUDAAPI cuPointerSetAttribute(const void *value, CUpointer_attribute at * - ::CU_POINTER_ATTRIBUTE_SYNC_MEMOPS * - ::CU_POINTER_ATTRIBUTE_BUFFER_ID * - ::CU_POINTER_ATTRIBUTE_IS_MANAGED + * - ::CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL * * \param numAttributes - Number of attributes to query * \param attributes - An array of attributes to query @@ -7896,8 +8785,10 @@ CUresult CUDAAPI cuPointerSetAttribute(const void *value, CUpointer_attribute at * ::CUDA_ERROR_INVALID_DEVICE * \notefnerr * - * \sa ::cuPointerGetAttribute, - * ::cuPointerSetAttribute + * \sa + * ::cuPointerGetAttribute, + * ::cuPointerSetAttribute, + * ::cudaPointerGetAttributes */ CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, CUpointer_attribute *attributes, void **data, CUdeviceptr ptr); #endif /* __CUDA_API_VERSION >= 7000 */ @@ -7922,7 +8813,7 @@ CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, CUpointer_at * Creates a stream and returns a handle in \p phStream. The \p Flags argument * determines behaviors of the stream. Valid values for \p Flags are: * - ::CU_STREAM_DEFAULT: Default stream creation flag. - * - ::CU_STREAM_NON_BLOCKING: Specifies that work running in the created + * - ::CU_STREAM_NON_BLOCKING: Specifies that work running in the created * stream may run concurrently with work in stream 0 (the NULL stream), and that * the created stream should perform no implicit synchronization with stream 0. * @@ -7945,7 +8836,9 @@ CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, CUpointer_at * ::cuStreamWaitEvent, * ::cuStreamQuery, * ::cuStreamSynchronize, - * ::cuStreamAddCallback + * ::cuStreamAddCallback, + * ::cudaStreamCreate, + * ::cudaStreamCreateWithFlags */ CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags); @@ -7978,7 +8871,7 @@ CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags); * ::CUDA_ERROR_OUT_OF_MEMORY * \notefnerr * - * \note Stream priorities are supported only on Quadro and Tesla GPUs + * \note Stream priorities are supported only on GPUs * with compute capability 3.5 or higher. * * \note In the current implementation, only compute kernels launched in @@ -7993,7 +8886,8 @@ CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags); * ::cuStreamWaitEvent, * ::cuStreamQuery, * ::cuStreamSynchronize, - * ::cuStreamAddCallback + * ::cuStreamAddCallback, + * ::cudaStreamCreateWithPriority */ CUresult CUDAAPI cuStreamCreateWithPriority(CUstream *phStream, unsigned int flags, int priority); @@ -8023,7 +8917,8 @@ CUresult CUDAAPI cuStreamCreateWithPriority(CUstream *phStream, unsigned int fla * ::cuStreamCreate, * ::cuStreamCreateWithPriority, * ::cuCtxGetStreamPriorityRange, - * ::cuStreamGetFlags + * ::cuStreamGetFlags, + * ::cudaStreamGetPriority */ CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority); @@ -8050,28 +8945,66 @@ CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority); * * \sa ::cuStreamDestroy, * ::cuStreamCreate, - * ::cuStreamGetPriority + * ::cuStreamGetPriority, + * ::cudaStreamGetFlags */ CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags); +#if __CUDA_API_VERSION >= 9020 /** - * \brief Make a compute stream wait on an event + * \brief Query the context associated with a stream + * + * Returns the CUDA context that the stream is associated with. * - * Makes all future work submitted to \p hStream wait until \p hEvent - * reports completion before beginning execution. This synchronization - * will be performed efficiently on the device. The event \p hEvent may - * be from a different context than \p hStream, in which case this function - * will perform cross-device synchronization. + * The stream handle \p hStream can refer to any of the following: + * <ul> + * <li>a stream created via any of the CUDA driver APIs such as ::cuStreamCreate + * and ::cuStreamCreateWithPriority, or their runtime API equivalents such as + * ::cudaStreamCreate, ::cudaStreamCreateWithFlags and ::cudaStreamCreateWithPriority. + * The returned context is the context that was active in the calling thread when the + * stream was created. Passing an invalid handle will result in undefined behavior.</li> + * <li>any of the special streams such as the NULL stream, ::CU_STREAM_LEGACY and + * ::CU_STREAM_PER_THREAD. The runtime API equivalents of these are also accepted, + * which are NULL, ::cudaStreamLegacy and ::cudaStreamPerThread respectively. + * Specifying any of the special handles will return the context current to the + * calling thread. If no context is current to the calling thread, + * ::CUDA_ERROR_INVALID_CONTEXT is returned.</li> + * </ul> * - * The stream \p hStream will wait only for the completion of the most recent - * host call to ::cuEventRecord() on \p hEvent. Once this call has returned, - * any functions (including ::cuEventRecord() and ::cuEventDestroy()) may be - * called on \p hEvent again, and subsequent calls will not have any - * effect on \p hStream. + * \param hStream - Handle to the stream to be queried + * \param pctx - Returned context associated with the stream * - * If ::cuEventRecord() has not been called on \p hEvent, this call acts as if - * the record has already completed, and so is a functional no-op. + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_CONTEXT, + * ::CUDA_ERROR_INVALID_HANDLE, + * \notefnerr + * + * \sa ::cuStreamDestroy, + * ::cuStreamCreateWithPriority, + * ::cuStreamGetPriority, + * ::cuStreamGetFlags, + * ::cuStreamWaitEvent, + * ::cuStreamQuery, + * ::cuStreamSynchronize, + * ::cuStreamAddCallback, + * ::cudaStreamCreate, + * ::cudaStreamCreateWithFlags + */ +CUresult CUDAAPI cuStreamGetCtx(CUstream hStream, CUcontext *pctx); + +#endif /* __CUDA_API_VERSION >= 9020 */ + +/** + * \brief Make a compute stream wait on an event + * + * Makes all future work submitted to \p hStream wait for all work captured in + * \p hEvent. See ::cuEventRecord() for details on what is captured by an event. + * The synchronization will be performed efficiently on the device when applicable. + * \p hEvent may be from a different context or device than \p hStream. * * \param hStream - Stream to wait * \param hEvent - Event to wait on (may not be NULL) @@ -8091,15 +9024,22 @@ CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags); * ::cuStreamQuery, * ::cuStreamSynchronize, * ::cuStreamAddCallback, - * ::cuStreamDestroy + * ::cuStreamDestroy, + * ::cudaStreamWaitEvent */ CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags); /** * \brief Add a callback to a compute stream * + * \note This function is slated for eventual deprecation and removal. If + * you do not require the callback to execute in case of a device error, + * consider using ::cuLaunchHostFunc. Additionally, this function is not + * supported with ::cuStreamBeginCapture and ::cuStreamEndCapture, unlike + * ::cuLaunchHostFunc. + * * Adds a callback to be called on the host after all currently enqueued - * items in the stream have completed. For each + * items in the stream have completed. For each * cuStreamAddCallback call, the callback will be executed exactly once. * The callback will block later work in the stream until it is finished. * @@ -8113,11 +9053,6 @@ CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned in * that are not mandated to run earlier. Callbacks without a mandated order * (in independent streams) execute in undefined order and may be serialized. * - * This API requires compute capability 1.1 or greater. See - * ::cuDeviceGetAttribute or ::cuDeviceGetProperties to query compute - * capability. Attempting to use this API with earlier compute versions will - * return ::CUDA_ERROR_NOT_SUPPORTED. - * * For the purposes of Unified Memory, callback execution makes a number of * guarantees: * <ul> @@ -8129,10 +9064,11 @@ CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned in * the callback. It thus synchronizes streams which have been "joined" * prior to the callback.</li> * <li>Adding device work to any stream does not have the effect of making - * the stream active until all preceding callbacks have executed. Thus, for + * the stream active until all preceding host functions and stream callbacks + * have executed. Thus, for * example, a callback might use global attached memory even if work has - * been added to another stream, if it has been properly ordered with an - * event.</li> + * been added to another stream, if the work has been ordered behind the + * callback with an event.</li> * <li>Completion of a callback does not cause a stream to become * active except as described above. The callback stream will remain idle * if no device work follows the callback, and will remain idle across @@ -8162,10 +9098,211 @@ CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned in * ::cuStreamWaitEvent, * ::cuStreamDestroy, * ::cuMemAllocManaged, - * ::cuStreamAttachMemAsync + * ::cuStreamAttachMemAsync, + * ::cuStreamLaunchHostFunc, + * ::cudaStreamAddCallback */ CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags); +#if __CUDA_API_VERSION >= 10000 + +/** + * \brief Begins graph capture on a stream + * + * Begin graph capture on \p hStream. When a stream is in capture mode, all operations + * pushed into the stream will not be executed, but will instead be captured into + * a graph, which will be returned via ::cuStreamEndCapture. Capture may not be initiated + * if \p stream is CU_STREAM_LEGACY. Capture must be ended on the same stream in which + * it was initiated, and it may only be initiated if the stream is not already in capture + * mode. The capture mode may be queried via ::cuStreamIsCapturing. A unique id + * representing the capture sequence may be queried via ::cuStreamGetCaptureInfo. + * + * If \p mode is not ::CU_STREAM_CAPTURE_MODE_RELAXED, ::cuStreamEndCapture must be + * called on this stream from the same thread. + * + * \param hStream - Stream in which to initiate capture + * \param mode - Controls the interaction of this capture sequence with other API + * calls that are potentially unsafe. For more details see + * ::cuThreadExchangeStreamCaptureMode. + * + * \note Kernels captured using this API must not use texture and surface references. + * Reading or writing through any texture or surface reference is undefined + * behavior. This restriction does not apply to texture and surface objects. + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \notefnerr + * + * \sa + * ::cuStreamCreate, + * ::cuStreamIsCapturing, + * ::cuStreamEndCapture, + * ::cuThreadExchangeStreamCaptureMode + */ +CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream, CUstreamCaptureMode mode); + +#endif /* __CUDA_API_VERSION >= 10000 */ +#if __CUDA_API_VERSION >= 10010 + +/** + * \brief Swaps the stream capture interaction mode for a thread + * + * Sets the calling thread's stream capture interaction mode to the value contained + * in \p *mode, and overwrites \p *mode with the previous mode for the thread. To + * facilitate deterministic behavior across function or module boundaries, callers + * are encouraged to use this API in a push-pop fashion: \code + CUstreamCaptureMode mode = desiredMode; + cuThreadExchangeStreamCaptureMode(&mode); + ... + cuThreadExchangeStreamCaptureMode(&mode); // restore previous mode + * \endcode + * + * During stream capture (see ::cuStreamBeginCapture), some actions, such as a call + * to ::cudaMalloc, may be unsafe. In the case of ::cudaMalloc, the operation is + * not enqueued asynchronously to a stream, and is not observed by stream capture. + * Therefore, if the sequence of operations captured via ::cuStreamBeginCapture + * depended on the allocation being replayed whenever the graph is launched, the + * captured graph would be invalid. + * + * Therefore, stream capture places restrictions on API calls that can be made within + * or concurrently to a ::cuStreamBeginCapture-::cuStreamEndCapture sequence. This + * behavior can be controlled via this API and flags to ::cuStreamBeginCapture. + * + * A thread's mode is one of the following: + * - \p CU_STREAM_CAPTURE_MODE_GLOBAL: This is the default mode. If the local thread has + * an ongoing capture sequence that was not initiated with + * \p CU_STREAM_CAPTURE_MODE_RELAXED at \p cuStreamBeginCapture, or if any other thread + * has a concurrent capture sequence initiated with \p CU_STREAM_CAPTURE_MODE_GLOBAL, + * this thread is prohibited from potentially unsafe API calls. + * - \p CU_STREAM_CAPTURE_MODE_THREAD_LOCAL: If the local thread has an ongoing capture + * sequence not initiated with \p CU_STREAM_CAPTURE_MODE_RELAXED, it is prohibited + * from potentially unsafe API calls. Concurrent capture sequences in other threads + * are ignored. + * - \p CU_STREAM_CAPTURE_MODE_RELAXED: The local thread is not prohibited from potentially + * unsafe API calls. Note that the thread is still prohibited from API calls which + * necessarily conflict with stream capture, for example, attempting ::cuEventQuery + * on an event that was last recorded inside a capture sequence. + * + * \param mode - Pointer to mode value to swap with the current mode + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \notefnerr + * + * \sa + * ::cuStreamBeginCapture + */ +CUresult CUDAAPI cuThreadExchangeStreamCaptureMode(CUstreamCaptureMode *mode); + +#endif /* __CUDA_API_VERSION >= 10010 */ +#if __CUDA_API_VERSION >= 10000 + +/** + * \brief Ends capture on a stream, returning the captured graph + * + * End capture on \p hStream, returning the captured graph via \p phGraph. + * Capture must have been initiated on \p hStream via a call to ::cuStreamBeginCapture. + * If capture was invalidated, due to a violation of the rules of stream capture, then + * a NULL graph will be returned. + * + * If the \p mode argument to ::cuStreamBeginCapture was not + * ::CU_STREAM_CAPTURE_MODE_RELAXED, this call must be from the same thread as + * ::cuStreamBeginCapture. + * + * \param hStream - Stream to query + * \param phGraph - The captured graph + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_STREAM_CAPTURE_WRONG_THREAD + * \notefnerr + * + * \sa + * ::cuStreamCreate, + * ::cuStreamBeginCapture, + * ::cuStreamIsCapturing + */ +CUresult CUDAAPI cuStreamEndCapture(CUstream hStream, CUgraph *phGraph); + +/** + * \brief Returns a stream's capture status + * + * Return the capture status of \p hStream via \p captureStatus. After a successful + * call, \p *captureStatus will contain one of the following: + * - ::CU_STREAM_CAPTURE_STATUS_NONE: The stream is not capturing. + * - ::CU_STREAM_CAPTURE_STATUS_ACTIVE: The stream is capturing. + * - ::CU_STREAM_CAPTURE_STATUS_INVALIDATED: The stream was capturing but an error + * has invalidated the capture sequence. The capture sequence must be terminated + * with ::cuStreamEndCapture on the stream where it was initiated in order to + * continue using \p hStream. + * + * Note that, if this is called on ::CU_STREAM_LEGACY (the "null stream") while + * a blocking stream in the same context is capturing, it will return + * ::CUDA_ERROR_STREAM_CAPTURE_IMPLICIT and \p *captureStatus is unspecified + * after the call. The blocking stream capture is not invalidated. + * + * When a blocking stream is capturing, the legacy stream is in an + * unusable state until the blocking stream capture is terminated. The legacy + * stream is not supported for stream capture, but attempted use would have an + * implicit dependency on the capturing stream(s). + * + * \param hStream - Stream to query + * \param captureStatus - Returns the stream's capture status + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_STREAM_CAPTURE_IMPLICIT + * \notefnerr + * + * \sa + * ::cuStreamCreate, + * ::cuStreamBeginCapture, + * ::cuStreamEndCapture + */ +CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *captureStatus); + +#endif /* __CUDA_API_VERSION >= 10000 */ + +#if __CUDA_API_VERSION >= 10010 + +/** + * \brief Query capture status of a stream + * + * Query the capture status of a stream and and get an id for + * the capture sequence, which is unique over the lifetime of the process. + * + * If called on ::CU_STREAM_LEGACY (the "null stream") while a stream not created + * with ::CU_STREAM_NON_BLOCKING is capturing, returns ::CUDA_ERROR_STREAM_CAPTURE_IMPLICIT. + * + * A valid id is returned only if both of the following are true: + * - the call returns CUDA_SUCCESS + * - captureStatus is set to ::CU_STREAM_CAPTURE_STATUS_ACTIVE + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_STREAM_CAPTURE_IMPLICIT + * \notefnerr + * + * \sa + * ::cuStreamBeginCapture, + * ::cuStreamIsCapturing + */ + CUresult CUDAAPI cuStreamGetCaptureInfo(CUstream hStream, CUstreamCaptureStatus *captureStatus, cuuint64_t *id); + +#endif /* __CUDA_API_VERSION >= 10010 */ + #if __CUDA_API_VERSION >= 6000 /** @@ -8177,12 +9314,20 @@ CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback * only take effect when, previous work in stream has completed. Any * previous association is automatically replaced. * - * \p dptr must point to an address within managed memory space declared - * using the __managed__ keyword or allocated with ::cuMemAllocManaged. + * \p dptr must point to one of the following types of memories: + * - managed memory declared using the __managed__ keyword or allocated with + * ::cuMemAllocManaged. + * - a valid host-accessible region of system-allocated pageable memory. This + * type of memory may only be specified if the device associated with the + * stream reports a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. * - * \p length must be zero, to indicate that the entire allocation's - * stream association is being changed. Currently, it's not possible - * to change stream association for a portion of an allocation. + * For managed allocations, \p length must be either zero or the entire + * allocation's size. Both indicate that the entire allocation's stream + * association is being changed. Currently, it is not possible to change stream + * association for a portion of a managed allocation. + * + * For pageable host allocations, \p length must be non-zero. * * The stream association is specified using \p flags which must be * one of ::CUmemAttach_flags. @@ -8207,7 +9352,7 @@ CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback * Accessing memory on the device from streams that are not associated with * it will produce undefined results. No error checking is performed by the * Unified Memory system to ensure that kernels launched into other streams - * do not access this region. + * do not access this region. * * It is a program's responsibility to order calls to ::cuStreamAttachMemAsync * via events, synchronization or other means to ensure legal access to memory @@ -8222,8 +9367,10 @@ CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback * happen until all work in the stream has completed. * * \param hStream - Stream in which to enqueue the attach operation - * \param dptr - Pointer to memory (must be a pointer to managed memory) - * \param length - Length of memory (must be zero) + * \param dptr - Pointer to memory (must be a pointer to managed memory or + * to a valid host-accessible region of system-allocated + * pageable memory) + * \param length - Length of memory * \param flags - Must be one of ::CUmemAttach_flags * * \return @@ -8241,7 +9388,8 @@ CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback * ::cuStreamSynchronize, * ::cuStreamWaitEvent, * ::cuStreamDestroy, - * ::cuMemAllocManaged + * ::cuMemAllocManaged, + * ::cudaStreamAttachMemAsync */ CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags); @@ -8272,7 +9420,8 @@ CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size * ::cuStreamWaitEvent, * ::cuStreamDestroy, * ::cuStreamSynchronize, - * ::cuStreamAddCallback + * ::cuStreamAddCallback, + * ::cudaStreamQuery */ CUresult CUDAAPI cuStreamQuery(CUstream hStream); @@ -8280,7 +9429,7 @@ CUresult CUDAAPI cuStreamQuery(CUstream hStream); * \brief Wait until a stream's tasks are completed * * Waits until the device has completed all operations in the stream specified - * by \p hStream. If the context was created with the + * by \p hStream. If the context was created with the * ::CU_CTX_SCHED_BLOCKING_SYNC flag, the CPU thread will block until the * stream is finished with all of its tasks. * @@ -8292,6 +9441,7 @@ CUresult CUDAAPI cuStreamQuery(CUstream hStream); * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_HANDLE + * \note_null_stream * \notefnerr * @@ -8299,7 +9449,8 @@ CUresult CUDAAPI cuStreamQuery(CUstream hStream); * ::cuStreamDestroy, * ::cuStreamWaitEvent, * ::cuStreamQuery, - * ::cuStreamAddCallback + * ::cuStreamAddCallback, + * ::cudaStreamSynchronize */ CUresult CUDAAPI cuStreamSynchronize(CUstream hStream); @@ -8307,11 +9458,11 @@ CUresult CUDAAPI cuStreamSynchronize(CUstream hStream); /** * \brief Destroys a stream * - * Destroys the stream specified by \p hStream. + * Destroys the stream specified by \p hStream. * * In case the device is still doing work in the stream \p hStream - * when ::cuStreamDestroy() is called, the function will return immediately - * and the resources associated with \p hStream will be released automatically + * when ::cuStreamDestroy() is called, the function will return immediately + * and the resources associated with \p hStream will be released automatically * once the device has completed all work in \p hStream. * * \param hStream - Stream to destroy @@ -8321,14 +9472,16 @@ CUresult CUDAAPI cuStreamSynchronize(CUstream hStream); * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, * ::CUDA_ERROR_INVALID_CONTEXT, - * ::CUDA_ERROR_INVALID_VALUE + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE * \notefnerr * * \sa ::cuStreamCreate, * ::cuStreamWaitEvent, * ::cuStreamQuery, * ::cuStreamSynchronize, - * ::cuStreamAddCallback + * ::cuStreamAddCallback, + * ::cudaStreamDestroy */ CUresult CUDAAPI cuStreamDestroy(CUstream hStream); #endif /* __CUDA_API_VERSION >= 4000 */ @@ -8351,8 +9504,8 @@ CUresult CUDAAPI cuStreamDestroy(CUstream hStream); /** * \brief Creates an event * - * Creates an event *phEvent with the flags specified via \p Flags. Valid flags - * include: + * Creates an event *phEvent for the current context with the flags specified via + * \p Flags. Valid flags include: * - ::CU_EVENT_DEFAULT: Default event creation flag. * - ::CU_EVENT_BLOCKING_SYNC: Specifies that the created event should use blocking * synchronization. A CPU thread that uses ::cuEventSynchronize() to wait on @@ -8383,23 +9536,29 @@ CUresult CUDAAPI cuStreamDestroy(CUstream hStream); * ::cuEventQuery, * ::cuEventSynchronize, * ::cuEventDestroy, - * ::cuEventElapsedTime + * ::cuEventElapsedTime, + * ::cudaEventCreate, + * ::cudaEventCreateWithFlags */ CUresult CUDAAPI cuEventCreate(CUevent *phEvent, unsigned int Flags); /** * \brief Records an event * - * Records an event. See note on NULL stream behavior. Since operation is - * asynchronous, ::cuEventQuery or ::cuEventSynchronize() must be used - * to determine when the event has actually been recorded. - * - * If ::cuEventRecord() has previously been called on \p hEvent, then this - * call will overwrite any existing state in \p hEvent. Any subsequent calls - * which examine the status of \p hEvent will only examine the completion of - * this most recent call to ::cuEventRecord(). + * Captures in \p hEvent the contents of \p hStream at the time of this call. + * \p hEvent and \p hStream must be from the same context. + * Calls such as ::cuEventQuery() or ::cuStreamWaitEvent() will then + * examine or wait for completion of the work that was captured. Uses of + * \p hStream after this call do not modify \p hEvent. See note on default + * stream behavior for what is captured in the default case. * - * It is necessary that \p hEvent and \p hStream be created on the same context. + * ::cuEventRecord() can be called multiple times on the same event and + * will overwrite the previously captured state. Other APIs such as + * ::cuStreamWaitEvent() use the most recently captured state at the time + * of the API call, and are not affected by later calls to + * ::cuEventRecord(). Before the first call to ::cuEventRecord(), an + * event represents an empty set of work, so for example ::cuEventQuery() + * would return ::CUDA_SUCCESS. * * \param hEvent - Event to record * \param hStream - Stream to record event for @@ -8419,21 +9578,19 @@ CUresult CUDAAPI cuEventCreate(CUevent *phEvent, unsigned int Flags); * ::cuEventSynchronize, * ::cuStreamWaitEvent, * ::cuEventDestroy, - * ::cuEventElapsedTime + * ::cuEventElapsedTime, + * ::cudaEventRecord */ CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream); /** * \brief Queries an event's status * - * Query the status of all device work preceding the most recent - * call to ::cuEventRecord() (in the appropriate compute streams, - * as specified by the arguments to ::cuEventRecord()). + * Queries the status of all work currently captured by \p hEvent. See + * ::cuEventRecord() for details on what is captured by an event. * - * If this work has successfully been completed by the device, or if - * ::cuEventRecord() has not been called on \p hEvent, then ::CUDA_SUCCESS is - * returned. If this work has not yet been completed by the device then - * ::CUDA_ERROR_NOT_READY is returned. + * Returns ::CUDA_SUCCESS if all captured work has been completed, or + * ::CUDA_ERROR_NOT_READY if any captured work is incomplete. * * For the purposes of Unified Memory, a return value of ::CUDA_SUCCESS * is equivalent to having called ::cuEventSynchronize(). @@ -8453,19 +9610,16 @@ CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream); * ::cuEventRecord, * ::cuEventSynchronize, * ::cuEventDestroy, - * ::cuEventElapsedTime + * ::cuEventElapsedTime, + * ::cudaEventQuery */ CUresult CUDAAPI cuEventQuery(CUevent hEvent); /** * \brief Waits for an event to complete * - * Wait until the completion of all device work preceding the most recent - * call to ::cuEventRecord() (in the appropriate compute streams, as specified - * by the arguments to ::cuEventRecord()). - * - * If ::cuEventRecord() has not been called on \p hEvent, ::CUDA_SUCCESS is - * returned immediately. + * Waits until the completion of all work currently captured in \p hEvent. + * See ::cuEventRecord() for details on what is captured by an event. * * Waiting for an event that was created with the ::CU_EVENT_BLOCKING_SYNC * flag will cause the calling CPU thread to block until the event has @@ -8487,7 +9641,8 @@ CUresult CUDAAPI cuEventQuery(CUevent hEvent); * ::cuEventRecord, * ::cuEventQuery, * ::cuEventDestroy, - * ::cuEventElapsedTime + * ::cuEventElapsedTime, + * ::cudaEventSynchronize */ CUresult CUDAAPI cuEventSynchronize(CUevent hEvent); @@ -8497,10 +9652,10 @@ CUresult CUDAAPI cuEventSynchronize(CUevent hEvent); * * Destroys the event specified by \p hEvent. * - * In case \p hEvent has been recorded but has not yet been completed - * when ::cuEventDestroy() is called, the function will return immediately and - * the resources associated with \p hEvent will be released automatically once - * the device has completed \p hEvent. + * An event may be destroyed before it is complete (i.e., while + * ::cuEventQuery() would return ::CUDA_ERROR_NOT_READY). In this case, the + * call does not block on completion of the event, and any associated + * resources will automatically be released asynchronously at completion. * * \param hEvent - Event to destroy * @@ -8516,7 +9671,8 @@ CUresult CUDAAPI cuEventSynchronize(CUevent hEvent); * ::cuEventRecord, * ::cuEventQuery, * ::cuEventSynchronize, - * ::cuEventElapsedTime + * ::cuEventElapsedTime, + * ::cudaEventDestroy */ CUresult CUDAAPI cuEventDestroy(CUevent hEvent); #endif /* __CUDA_API_VERSION >= 4000 */ @@ -8560,10 +9716,520 @@ CUresult CUDAAPI cuEventDestroy(CUevent hEvent); * ::cuEventRecord, * ::cuEventQuery, * ::cuEventSynchronize, - * ::cuEventDestroy + * ::cuEventDestroy, + * ::cudaEventElapsedTime */ CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, CUevent hEnd); +/** @} */ /* END CUDA_EVENT */ + +/** + * \defgroup CUDA_EXTRES_INTEROP External Resource Interoperability + * + * ___MANBRIEF___ External resource interoperability functions of the low-level CUDA driver API + * (___CURRENT_FILE___) ___ENDMANBRIEF___ + * + * This section describes the external resource interoperability functions of the low-level CUDA + * driver application programming interface. + * + * @{ + */ + +#if __CUDA_API_VERSION >= 10000 + + /** + * \brief Imports an external memory object + * + * Imports an externally allocated memory object and returns + * a handle to that in \p extMem_out. + * + * The properties of the handle being imported must be described in + * \p memHandleDesc. The ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC structure + * is defined as follows: + * + * \code + typedef struct CUDA_EXTERNAL_MEMORY_HANDLE_DESC_st { + CUexternalMemoryHandleType type; + union { + int fd; + struct { + void *handle; + const void *name; + } win32; + } handle; + unsigned long long size; + unsigned int flags; + } CUDA_EXTERNAL_MEMORY_HANDLE_DESC; + * \endcode + * + * where ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type specifies the type + * of handle being imported. ::CUexternalMemoryHandleType is + * defined as: + * + * \code + typedef enum CUexternalMemoryHandleType_enum { + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD = 1, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 = 2, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP = 4, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE = 5 + } CUexternalMemoryHandleType; + * \endcode + * + * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is + * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD, then + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::fd must be a valid + * file descriptor referencing a memory object. Ownership of + * the file descriptor is transferred to the CUDA driver when the + * handle is imported successfully. Performing any operations on the + * file descriptor after it is imported results in undefined behavior. + * + * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is + * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32, then exactly one + * of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be + * NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle + * is not NULL, then it must represent a valid shared NT handle that + * references a memory object. Ownership of this handle is + * not transferred to CUDA after the import operation, so the + * application must release the handle using the appropriate system + * call. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name + * is not NULL, then it must point to a NULL-terminated array of + * UTF-16 characters that refers to a memory object. + * + * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is + * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT, then + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle must + * be non-NULL and + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name + * must be NULL. The handle specified must be a globally shared KMT + * handle. This handle does not hold a reference to the underlying + * object, and thus will be invalid when all references to the + * memory object are destroyed. + * + * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is + * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP, then exactly one + * of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be + * NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle + * is not NULL, then it must represent a valid shared NT handle that + * is returned by ID3DDevice::CreateSharedHandle when referring to a + * ID3D12Heap object. This handle holds a reference to the underlying + * object. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name + * is not NULL, then it must point to a NULL-terminated array of + * UTF-16 characters that refers to a ID3D12Heap object. + * + * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is + * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE, then exactly one + * of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be + * NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle + * is not NULL, then it must represent a valid shared NT handle that + * is returned by ID3DDevice::CreateSharedHandle when referring to a + * ID3D12Resource object. This handle holds a reference to the + * underlying object. If + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name + * is not NULL, then it must point to a NULL-terminated array of + * UTF-16 characters that refers to a ID3D12Resource object. + * + * The size of the memory object must be specified in + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::size. + * + * Specifying the flag ::CUDA_EXTERNAL_MEMORY_DEDICATED in + * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::flags indicates that the + * resource is a dedicated resource. The definition of what a + * dedicated resource is outside the scope of this extension. + * + * \param extMem_out - Returned handle to an external memory object + * \param memHandleDesc - Memory import handle descriptor + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \note If the Vulkan memory imported into CUDA is mapped on the CPU then the + * application must use vkInvalidateMappedMemoryRanges/vkFlushMappedMemoryRanges + * as well as appropriate Vulkan pipeline barriers to maintain coherence between + * CPU and GPU. For more information on these APIs, please refer to "Synchronization + * and Cache Control" chapter from Vulkan specification. + * + * \sa ::cuDestroyExternalMemory, + * ::cuExternalMemoryGetMappedBuffer, + * ::cuExternalMemoryGetMappedMipmappedArray + */ +CUresult CUDAAPI cuImportExternalMemory(CUexternalMemory *extMem_out, const CUDA_EXTERNAL_MEMORY_HANDLE_DESC *memHandleDesc); + +/** + * \brief Maps a buffer onto an imported memory object + * + * Maps a buffer onto an imported memory object and returns a device + * pointer in \p devPtr. + * + * The properties of the buffer being mapped must be described in + * \p bufferDesc. The ::CUDA_EXTERNAL_MEMORY_BUFFER_DESC structure is + * defined as follows: + * + * \code + typedef struct CUDA_EXTERNAL_MEMORY_BUFFER_DESC_st { + unsigned long long offset; + unsigned long long size; + unsigned int flags; + } CUDA_EXTERNAL_MEMORY_BUFFER_DESC; + * \endcode + * + * where ::CUDA_EXTERNAL_MEMORY_BUFFER_DESC::offset is the offset in + * the memory object where the buffer's base address is. + * ::CUDA_EXTERNAL_MEMORY_BUFFER_DESC::size is the size of the buffer. + * ::CUDA_EXTERNAL_MEMORY_BUFFER_DESC::flags must be zero. + * + * The offset and size have to be suitably aligned to match the + * requirements of the external API. Mapping two buffers whose ranges + * overlap may or may not result in the same virtual address being + * returned for the overlapped portion. In such cases, the application + * must ensure that all accesses to that region from the GPU are + * volatile. Otherwise writes made via one address are not guaranteed + * to be visible via the other address, even if they're issued by the + * same thread. It is recommended that applications map the combined + * range instead of mapping separate buffers and then apply the + * appropriate offsets to the returned pointer to derive the + * individual buffers. + * + * The returned pointer \p devPtr must be freed using ::cuMemFree. + * + * \param devPtr - Returned device pointer to buffer + * \param extMem - Handle to external memory object + * \param bufferDesc - Buffer descriptor + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \sa ::cuImportExternalMemory + * ::cuDestroyExternalMemory, + * ::cuExternalMemoryGetMappedMipmappedArray + */ +CUresult CUDAAPI cuExternalMemoryGetMappedBuffer(CUdeviceptr *devPtr, CUexternalMemory extMem, const CUDA_EXTERNAL_MEMORY_BUFFER_DESC *bufferDesc); + +/** + * \brief Maps a CUDA mipmapped array onto an external memory object + * + * Maps a CUDA mipmapped array onto an external object and returns a + * handle to it in \p mipmap. + * + * The properties of the CUDA mipmapped array being mapped must be + * described in \p mipmapDesc. The structure + * ::CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC is defined as follows: + * + * \code + typedef struct CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC_st { + unsigned long long offset; + CUDA_ARRAY3D_DESCRIPTOR arrayDesc; + unsigned int numLevels; + } CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC; + * \endcode + * + * where ::CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC::offset is the + * offset in the memory object where the base level of the mipmap + * chain is. + * ::CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC::arrayDesc describes + * the format, dimensions and type of the base level of the mipmap + * chain. For further details on these parameters, please refer to the + * documentation for ::cuMipmappedArrayCreate. Note that if the mipmapped + * array is bound as a color target in the graphics API, then the flag + * ::CUDA_ARRAY3D_COLOR_ATTACHMENT must be specified in + * ::CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC::arrayDesc::Flags. + * ::CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC::numLevels specifies + * the total number of levels in the mipmap chain. + * + * The returned CUDA mipmapped array must be freed using ::cuMipmappedArrayDestroy. + * + * \param mipmap - Returned CUDA mipmapped array + * \param extMem - Handle to external memory object + * \param mipmapDesc - CUDA array descriptor + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \sa ::cuImportExternalMemory + * ::cuDestroyExternalMemory, + * ::cuExternalMemoryGetMappedBuffer + */ +CUresult CUDAAPI cuExternalMemoryGetMappedMipmappedArray(CUmipmappedArray *mipmap, CUexternalMemory extMem, const CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC *mipmapDesc); + +/** + * \brief Destroys an external memory object. + * + * Destroys the specified external memory object. Any existing buffers + * and CUDA mipmapped arrays mapped onto this object must no longer be + * used and must be explicitly freed using ::cuMemFree and + * ::cuMipmappedArrayDestroy respectively. + * + * \param extMem - External memory object to be destroyed + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \sa ::cuImportExternalMemory + * ::cuExternalMemoryGetMappedBuffer, + * ::cuExternalMemoryGetMappedMipmappedArray + */ +CUresult CUDAAPI cuDestroyExternalMemory(CUexternalMemory extMem); + +/** + * \brief Imports an external semaphore + * + * Imports an externally allocated synchronization object and returns + * a handle to that in \p extSem_out. + * + * The properties of the handle being imported must be described in + * \p semHandleDesc. The ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC is + * defined as follows: + * + * \code + typedef struct CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC_st { + CUexternalSemaphoreHandleType type; + union { + int fd; + struct { + void *handle; + const void *name; + } win32; + } handle; + unsigned int flags; + } CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC; + * \endcode + * + * where ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::type specifies the type of + * handle being imported. ::CUexternalSemaphoreHandleType is defined + * as: + * + * \code + typedef enum CUexternalSemaphoreHandleType_enum { + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD = 1, + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 = 2, + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE = 4 + } CUexternalSemaphoreHandleType; + * \endcode + * + * If ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD, then + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::fd must be a valid + * file descriptor referencing a synchronization object. Ownership of + * the file descriptor is transferred to the CUDA driver when the + * handle is imported successfully. Performing any operations on the + * file descriptor after it is imported results in undefined behavior. + * + * If ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32, then exactly one + * of ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::handle and + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::name must not be + * NULL. If + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::handle + * is not NULL, then it must represent a valid shared NT handle that + * references a synchronization object. Ownership of this handle is + * not transferred to CUDA after the import operation, so the + * application must release the handle using the appropriate system + * call. If ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::name + * is not NULL, then it must name a valid synchronization object. + * + * If ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT, then + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::handle must + * be non-NULL and + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::name + * must be NULL. The handle specified must be a globally shared KMT + * handle. This handle does not hold a reference to the underlying + * object, and thus will be invalid when all references to the + * synchronization object are destroyed. + * + * If ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE, then exactly one + * of ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::handle and + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::name must not be + * NULL. If + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::handle + * is not NULL, then it must represent a valid shared NT handle that + * is returned by ID3DDevice::CreateSharedHandle when referring to a + * ID3D12Fence object. This handle holds a reference to the underlying + * object. If + * ::CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC::handle::win32::name + * is not NULL, then it must name a valid synchronization object that + * refers to a valid ID3D12Fence object. + * + * \param extSem_out - Returned handle to an external semaphore + * \param semHandleDesc - Semaphore import handle descriptor + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \sa ::cuDestroyExternalSemaphore, + * ::cuSignalExternalSemaphoresAsync, + * ::cuWaitExternalSemaphoresAsync + */ +CUresult CUDAAPI cuImportExternalSemaphore(CUexternalSemaphore *extSem_out, const CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC *semHandleDesc); + +/** + * \brief Signals a set of external semaphore objects + * + * Enqueues a signal operation on a set of externally allocated + * semaphore object in the specified stream. The operations will be + * executed when all prior operations in the stream complete. + * + * The exact semantics of signaling a semaphore depends on the type of + * the object. + * + * If the semaphore object is any one of the following types: + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD, + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32, + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * then signaling the semaphore will set it to the signaled state. + * + * If the semaphore object is of the type + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE, then the + * semaphore will be set to the value specified in + * ::CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS::params::fence::value. + * + * \param extSemArray - Set of external semaphores to be signaled + * \param paramsArray - Array of semaphore parameters + * \param numExtSems - Number of semaphores to signal + * \param stream - Stream to enqueue the signal operations in + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \sa ::cuImportExternalSemaphore, + * ::cuDestroyExternalSemaphore, + * ::cuWaitExternalSemaphoresAsync + */ +CUresult CUDAAPI cuSignalExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); + +/** + * \brief Waits on a set of external semaphore objects + * + * Enqueues a wait operation on a set of externally allocated + * semaphore object in the specified stream. The operations will be + * executed when all prior operations in the stream complete. + * + * The exact semantics of waiting on a semaphore depends on the type + * of the object. + * + * If the semaphore object is any one of the following types: + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD, + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32, + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * then waiting on the semaphore will wait until the semaphore reaches + * the signaled state. The semaphore will then be reset to the + * unsignaled state. Therefore for every signal operation, there can + * only be one wait operation. + * + * If the semaphore object is of the type + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE, then waiting on + * the semaphore will wait until the value of the semaphore is + * greater than or equal to + * ::CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS::params::fence::value. + * + * \param extSemArray - External semaphores to be waited on + * \param paramsArray - Array of semaphore parameters + * \param numExtSems - Number of semaphores to wait on + * \param stream - Stream to enqueue the wait operations in + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \sa ::cuImportExternalSemaphore, + * ::cuDestroyExternalSemaphore, + * ::cuSignalExternalSemaphoresAsync + */ +CUresult CUDAAPI cuWaitExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); + +/** + * \brief Destroys an external semaphore + * + * Destroys an external semaphore object and releases any references + * to the underlying resource. Any outstanding signals or waits must + * have completed before the semaphore is destroyed. + * + * \param extSem - External semaphore to be destroyed + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_HANDLE + * \notefnerr + * + * \sa ::cuImportExternalSemaphore, + * ::cuSignalExternalSemaphoresAsync, + * ::cuWaitExternalSemaphoresAsync + */ +CUresult CUDAAPI cuDestroyExternalSemaphore(CUexternalSemaphore extSem); + +#endif /* __CUDA_API_VERSION >= 10000 */ + +/** @} */ /* END CUDA_EXTRES_INTEROP */ + +/** + * \defgroup CUDA_MEMOP Stream memory operations + * + * ___MANBRIEF___ Stream memory operations of the low-level CUDA driver API + * (___CURRENT_FILE___) ___ENDMANBRIEF___ + * + * This section describes the stream memory operations of the low-level CUDA + * driver application programming interface. + * + * The whole set of operations is disabled by default. Users are required + * to explicitly enable them, e.g. on Linux by passing the kernel module + * parameter shown below: + * modprobe nvidia NVreg_EnableStreamMemOPs=1 + * There is currently no way to enable these operations on other operating + * systems. + * + * Users can programmatically query whether the device supports these + * operations with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS. + * + * Support for the ::CU_STREAM_WAIT_VALUE_NOR flag can be queried with + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR. + * + * Support for the ::cuStreamWriteValue64() and ::cuStreamWaitValue64() + * functions, as well as for the ::CU_STREAM_MEM_OP_WAIT_VALUE_64 and + * ::CU_STREAM_MEM_OP_WRITE_VALUE_64 flags, can be queried with + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_64_BIT_STREAM_MEM_OPS. + * + * Support for both ::CU_STREAM_WAIT_VALUE_FLUSH and + * ::CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES requires dedicated platform + * hardware features and can be queried with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_FLUSH_REMOTE_WRITES. + * + * Note that all memory pointers passed as parameters to these operations + * are device pointers. Where necessary a device pointer should be + * obtained, for example with ::cuMemHostGetDevicePointer(). + * + * None of the operations accepts pointers to managed memory buffers + * (::cuMemAllocManaged). + * + * @{ + */ + #if __CUDA_API_VERSION >= 8000 /** * \brief Wait on a memory location @@ -8578,8 +10244,11 @@ CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, CUeven * should be obtained with ::cuMemHostGetDevicePointer(). This function cannot * be used with managed memory (::cuMemAllocManaged). * - * On Windows, the device must be using TCC, or the operation is not supported. - * See ::cuDeviceGetAttributes(). + * Support for this can be queried with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS. + * + * Support for CU_STREAM_WAIT_VALUE_NOR can be queried with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR. * * \param stream The stream to synchronize on the memory location. * \param addr The memory location to wait on. @@ -8592,7 +10261,9 @@ CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, CUeven * ::CUDA_ERROR_NOT_SUPPORTED * \notefnerr * - * \sa ::cuStreamWriteValue32, + * \sa ::cuStreamWaitValue64, + * ::cuStreamWriteValue32, + * ::cuStreamWriteValue64 * ::cuStreamBatchMemOp, * ::cuMemHostRegister, * ::cuStreamWaitEvent @@ -8600,6 +10271,41 @@ CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, CUeven CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); /** + * \brief Wait on a memory location + * + * Enqueues a synchronization of the stream on the given memory location. Work + * ordered after the operation will block until the given condition on the + * memory is satisfied. By default, the condition is to wait for + * (int64_t)(*addr - value) >= 0, a cyclic greater-or-equal. + * Other condition types can be specified via \p flags. + * + * If the memory was registered via ::cuMemHostRegister(), the device pointer + * should be obtained with ::cuMemHostGetDevicePointer(). + * + * Support for this can be queried with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_64_BIT_STREAM_MEM_OPS. + * + * \param stream The stream to synchronize on the memory location. + * \param addr The memory location to wait on. + * \param value The value to compare with the memory location. + * \param flags See ::CUstreamWaitValue_flags. + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_NOT_SUPPORTED + * \notefnerr + * + * \sa ::cuStreamWaitValue32, + * ::cuStreamWriteValue32, + * ::cuStreamWriteValue64, + * ::cuStreamBatchMemOp, + * ::cuMemHostRegister, + * ::cuStreamWaitEvent + */ +CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); + +/** * \brief Write a value to memory * * Write a value to memory. Unless the ::CU_STREAM_WRITE_VALUE_NO_MEMORY_BARRIER @@ -8611,8 +10317,8 @@ CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32 * should be obtained with ::cuMemHostGetDevicePointer(). This function cannot * be used with managed memory (::cuMemAllocManaged). * - * On Windows, the device must be using TCC, or the operation is not supported. - * See ::cuDeviceGetAttribute(). + * Support for this can be queried with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS. * * \param stream The stream to do the write in. * \param addr The device address to write to. @@ -8625,7 +10331,9 @@ CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32 * ::CUDA_ERROR_NOT_SUPPORTED * \notefnerr * - * \sa ::cuStreamWaitValue32, + * \sa ::cuStreamWriteValue64, + * ::cuStreamWaitValue32, + * ::cuStreamWaitValue64, * ::cuStreamBatchMemOp, * ::cuMemHostRegister, * ::cuEventRecord @@ -8633,6 +10341,40 @@ CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32 CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); /** + * \brief Write a value to memory + * + * Write a value to memory. Unless the ::CU_STREAM_WRITE_VALUE_NO_MEMORY_BARRIER + * flag is passed, the write is preceded by a system-wide memory fence, + * equivalent to a __threadfence_system() but scoped to the stream + * rather than a CUDA thread. + * + * If the memory was registered via ::cuMemHostRegister(), the device pointer + * should be obtained with ::cuMemHostGetDevicePointer(). + * + * Support for this can be queried with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_64_BIT_STREAM_MEM_OPS. + * + * \param stream The stream to do the write in. + * \param addr The device address to write to. + * \param value The value to write. + * \param flags See ::CUstreamWriteValue_flags. + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_NOT_SUPPORTED + * \notefnerr + * + * \sa ::cuStreamWriteValue32, + * ::cuStreamWaitValue32, + * ::cuStreamWaitValue64, + * ::cuStreamBatchMemOp, + * ::cuMemHostRegister, + * ::cuEventRecord + */ +CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); + +/** * \brief Batch operations to synchronize the stream via memory operations * * This is a batch version of ::cuStreamWaitValue32() and ::cuStreamWriteValue32(). @@ -8641,11 +10383,12 @@ CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint3 * calls. The operations are enqueued in the order they appear in the array. * * See ::CUstreamBatchMemOpType for the full set of supported operations, and - * ::cuStreamWaitValue32() and ::cuStreamWriteValue32() for details of specific - * operations. + * ::cuStreamWaitValue32(), ::cuStreamWaitValue64(), ::cuStreamWriteValue32(), + * and ::cuStreamWriteValue64() for details of specific operations. * - * On Windows, the device must be using TCC, or this call is not supported. See - * ::cuDeviceGetAttribute(). + * Basic support for this can be queried with ::cuDeviceGetAttribute() and + * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS. See related APIs for details + * on querying support for specific operations. * * \param stream The stream to enqueue the operations in. * \param count The number of operations in the array. Must be less than 256. @@ -8659,13 +10402,15 @@ CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint3 * \notefnerr * * \sa ::cuStreamWaitValue32, + * ::cuStreamWaitValue64, * ::cuStreamWriteValue32, + * ::cuStreamWriteValue64, * ::cuMemHostRegister */ CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, unsigned int flags); #endif /* __CUDA_API_VERSION >= 8000 */ -/** @} */ /* END CUDA_EVENT */ +/** @} */ /* END CUDA_MEMOP */ /** * \defgroup CUDA_EXEC Execution Control @@ -8709,8 +10454,12 @@ CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstrea * would return the value 13. Note that this will return a value of 10 for * legacy cubins that do not have a properly-encoded binary architecture * version. - * - ::CU_FUNC_CACHE_MODE_CA: The attribute to indicate whether the function has + * - ::CU_FUNC_CACHE_MODE_CA: The attribute to indicate whether the function has * been compiled with user specified option "-Xptxas --dlcm=ca" set . + * - ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: The maximum size in bytes of + * dynamically-allocated shared memory. + * - ::CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: Preferred shared memory-L1 + * cache split ratio in percent of total shared memory. * * \param pi - Returned attribute value * \param attrib - Attribute requested @@ -8728,10 +10477,61 @@ CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstrea * \sa ::cuCtxGetCacheConfig, * ::cuCtxSetCacheConfig, * ::cuFuncSetCacheConfig, - * ::cuLaunchKernel + * ::cuLaunchKernel, + * ::cudaFuncGetAttributes + * ::cudaFuncSetAttribute */ CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib, CUfunction hfunc); +#if __CUDA_API_VERSION >= 9000 + +/** + * \brief Sets information about a function + * + * This call sets the value of a specified attribute \p attrib on the kernel given + * by \p hfunc to an integer value specified by \p val + * This function returns CUDA_SUCCESS if the new value of the attribute could be + * successfully set. If the set fails, this call will return an error. + * Not all attributes can have values set. Attempting to set a value on a read-only + * attribute will result in an error (CUDA_ERROR_INVALID_VALUE) + * + * Supported attributes for the cuFuncSetAttribute call are: + * - ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: This maximum size in bytes of + * dynamically-allocated shared memory. The value should contain the requested + * maximum size of dynamically-allocated shared memory. The sum of this value and + * the function attribute ::CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES cannot exceed the + * device attribute ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN. + * The maximal size of requestable dynamic shared memory may differ by GPU + * architecture. + * - ::CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: On devices where the L1 + * cache and shared memory use the same hardware resources, this sets the shared memory + * carveout preference, in percent of the total shared memory. + * See ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR + * This is only a hint, and the driver can choose a different ratio if required to execute the function. + * + * \param hfunc - Function to query attribute of + * \param attrib - Attribute requested + * \param value - The value to set + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_CONTEXT, + * ::CUDA_ERROR_INVALID_HANDLE, + * ::CUDA_ERROR_INVALID_VALUE + * \notefnerr + * + * \sa ::cuCtxGetCacheConfig, + * ::cuCtxSetCacheConfig, + * ::cuFuncSetCacheConfig, + * ::cuLaunchKernel, + * ::cudaFuncGetAttributes + * ::cudaFuncSetAttribute + */ +CUresult CUDAAPI cuFuncSetAttribute(CUfunction hfunc, CUfunction_attribute attrib, int value); +#endif // __CUDA_API_VERSION >= 9000 + /** * \brief Sets the preferred cache configuration for a device function * @@ -8771,7 +10571,8 @@ CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib, CUfunc * \sa ::cuCtxGetCacheConfig, * ::cuCtxSetCacheConfig, * ::cuFuncGetAttribute, - * ::cuLaunchKernel + * ::cuLaunchKernel, + * ::cudaFuncSetCacheConfig */ CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config); @@ -8779,28 +10580,28 @@ CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config); /** * \brief Sets the shared memory configuration for a device function. * - * On devices with configurable shared memory banks, this function will + * On devices with configurable shared memory banks, this function will * force all subsequent launches of the specified device function to have * the given shared memory bank size configuration. On any given launch of the * function, the shared memory configuration of the device will be temporarily * changed if needed to suit the function's preferred configuration. Changes in - * shared memory configuration between subsequent launches of functions, + * shared memory configuration between subsequent launches of functions, * may introduce a device side synchronization point. * - * Any per-function setting of shared memory bank size set via + * Any per-function setting of shared memory bank size set via * ::cuFuncSetSharedMemConfig will override the context wide setting set with * ::cuCtxSetSharedMemConfig. * * Changing the shared memory bank size will not increase shared memory usage - * or affect occupancy of kernels, but may have major effects on performance. + * or affect occupancy of kernels, but may have major effects on performance. * Larger bank sizes will allow for greater potential bandwidth to shared memory, - * but will change what kinds of accesses to shared memory will result in bank + * but will change what kinds of accesses to shared memory will result in bank * conflicts. * * This function will do nothing on devices with fixed shared memory bank size. * * The supported bank configurations are: - * - ::CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE: use the context's shared memory + * - ::CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE: use the context's shared memory * configuration when launching this function. * - ::CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE: set shared memory bank width to * be natively four bytes when launching this function. @@ -8823,7 +10624,8 @@ CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config); * ::cuCtxGetSharedMemConfig, * ::cuCtxSetSharedMemConfig, * ::cuFuncGetAttribute, - * ::cuLaunchKernel + * ::cuLaunchKernel, + * ::cudaFuncSetSharedMemConfig */ CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig config); #endif @@ -8937,7 +10739,8 @@ CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig confi * \sa ::cuCtxGetCacheConfig, * ::cuCtxSetCacheConfig, * ::cuFuncSetCacheConfig, - * ::cuFuncGetAttribute + * ::cuFuncGetAttribute, + * ::cudaLaunchKernel */ CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, @@ -8951,6 +10754,299 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, void **kernelParams, void **extra); #endif /* __CUDA_API_VERSION >= 4000 */ +#if __CUDA_API_VERSION >= 9000 +/** + * \brief Launches a CUDA function where thread blocks can cooperate and synchronize as they execute + * + * Invokes the kernel \p f on a \p gridDimX x \p gridDimY x \p gridDimZ + * grid of blocks. Each block contains \p blockDimX x \p blockDimY x + * \p blockDimZ threads. + * + * \p sharedMemBytes sets the amount of dynamic shared memory that will be + * available to each thread block. + * + * The device on which this kernel is invoked must have a non-zero value for + * the device attribute ::CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH. + * + * The total number of blocks launched cannot exceed the maximum number of blocks per + * multiprocessor as returned by ::cuOccupancyMaxActiveBlocksPerMultiprocessor (or + * ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of multiprocessors + * as specified by the device attribute ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. + * + * The kernel cannot make use of CUDA dynamic parallelism. + * + * Kernel parameters must be specified via \p kernelParams. If \p f + * has N parameters, then \p kernelParams needs to be an array of N + * pointers. Each of \p kernelParams[0] through \p kernelParams[N-1] + * must point to a region of memory from which the actual kernel + * parameter will be copied. The number of kernel parameters and their + * offsets and sizes do not need to be specified as that information is + * retrieved directly from the kernel's image. + * + * Calling ::cuLaunchCooperativeKernel() sets persistent function state that is + * the same as function state set through ::cuLaunchKernel API + * + * When the kernel \p f is launched via ::cuLaunchCooperativeKernel(), the previous + * block shape, shared size and parameter info associated with \p f + * is overwritten. + * + * Note that to use ::cuLaunchCooperativeKernel(), the kernel \p f must either have + * been compiled with toolchain version 3.2 or later so that it will + * contain kernel parameter information, or have no kernel parameters. + * If either of these conditions is not met, then ::cuLaunchCooperativeKernel() will + * return ::CUDA_ERROR_INVALID_IMAGE. + * + * \param f - Kernel to launch + * \param gridDimX - Width of grid in blocks + * \param gridDimY - Height of grid in blocks + * \param gridDimZ - Depth of grid in blocks + * \param blockDimX - X dimension of each thread block + * \param blockDimY - Y dimension of each thread block + * \param blockDimZ - Z dimension of each thread block + * \param sharedMemBytes - Dynamic shared-memory size per thread block in bytes + * \param hStream - Stream identifier + * \param kernelParams - Array of pointers to kernel parameters + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_CONTEXT, + * ::CUDA_ERROR_INVALID_HANDLE, + * ::CUDA_ERROR_INVALID_IMAGE, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_LAUNCH_FAILED, + * ::CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES, + * ::CUDA_ERROR_LAUNCH_TIMEOUT, + * ::CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING, + * ::CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE, + * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED + * \note_null_stream + * \notefnerr + * + * \sa ::cuCtxGetCacheConfig, + * ::cuCtxSetCacheConfig, + * ::cuFuncSetCacheConfig, + * ::cuFuncGetAttribute, + * ::cuLaunchCooperativeKernelMultiDevice, + * ::cudaLaunchCooperativeKernel + */ +CUresult CUDAAPI cuLaunchCooperativeKernel(CUfunction f, + unsigned int gridDimX, + unsigned int gridDimY, + unsigned int gridDimZ, + unsigned int blockDimX, + unsigned int blockDimY, + unsigned int blockDimZ, + unsigned int sharedMemBytes, + CUstream hStream, + void **kernelParams); + +/** + * \brief Launches CUDA functions on multiple devices where thread blocks can cooperate and synchronize as they execute + * + * Invokes kernels as specified in the \p launchParamsList array where each element + * of the array specifies all the parameters required to perform a single kernel launch. + * These kernels can cooperate and synchronize as they execute. The size of the array is + * specified by \p numDevices. + * + * No two kernels can be launched on the same device. All the devices targeted by this + * multi-device launch must be identical. All devices must have a non-zero value for the + * device attribute ::CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH. + * + * All kernels launched must be identical with respect to the compiled code. Note that + * any __device__, __constant__ or __managed__ variables present in the module that owns + * the kernel launched on each device, are independently instantiated on every device. + * It is the application's responsiblity to ensure these variables are initialized and + * used appropriately. + * + * The size of the grids as specified in blocks, the size of the blocks themselves + * and the amount of shared memory used by each thread block must also match across + * all launched kernels. + * + * The streams used to launch these kernels must have been created via either ::cuStreamCreate + * or ::cuStreamCreateWithPriority. The NULL stream or ::CU_STREAM_LEGACY or ::CU_STREAM_PER_THREAD + * cannot be used. + * + * The total number of blocks launched per kernel cannot exceed the maximum number of blocks + * per multiprocessor as returned by ::cuOccupancyMaxActiveBlocksPerMultiprocessor (or + * ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of multiprocessors + * as specified by the device attribute ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. Since the + * total number of blocks launched per device has to match across all devices, the maximum + * number of blocks that can be launched per device will be limited by the device with the + * least number of multiprocessors. + * + * The kernels cannot make use of CUDA dynamic parallelism. + * + * The ::CUDA_LAUNCH_PARAMS structure is defined as: + * \code + typedef struct CUDA_LAUNCH_PARAMS_st + { + CUfunction function; + unsigned int gridDimX; + unsigned int gridDimY; + unsigned int gridDimZ; + unsigned int blockDimX; + unsigned int blockDimY; + unsigned int blockDimZ; + unsigned int sharedMemBytes; + CUstream hStream; + void **kernelParams; + } CUDA_LAUNCH_PARAMS; + * \endcode + * where: + * - ::CUDA_LAUNCH_PARAMS::function specifies the kernel to be launched. All functions must + * be identical with respect to the compiled code. + * - ::CUDA_LAUNCH_PARAMS::gridDimX is the width of the grid in blocks. This must match across + * all kernels launched. + * - ::CUDA_LAUNCH_PARAMS::gridDimY is the height of the grid in blocks. This must match across + * all kernels launched. + * - ::CUDA_LAUNCH_PARAMS::gridDimZ is the depth of the grid in blocks. This must match across + * all kernels launched. + * - ::CUDA_LAUNCH_PARAMS::blockDimX is the X dimension of each thread block. This must match across + * all kernels launched. + * - ::CUDA_LAUNCH_PARAMS::blockDimX is the Y dimension of each thread block. This must match across + * all kernels launched. + * - ::CUDA_LAUNCH_PARAMS::blockDimZ is the Z dimension of each thread block. This must match across + * all kernels launched. + * - ::CUDA_LAUNCH_PARAMS::sharedMemBytes is the dynamic shared-memory size per thread block in bytes. + * This must match across all kernels launched. + * - ::CUDA_LAUNCH_PARAMS::hStream is the handle to the stream to perform the launch in. This cannot + * be the NULL stream or ::CU_STREAM_LEGACY or ::CU_STREAM_PER_THREAD. The CUDA context associated + * with this stream must match that associated with ::CUDA_LAUNCH_PARAMS::function. + * - ::CUDA_LAUNCH_PARAMS::kernelParams is an array of pointers to kernel parameters. If + * ::CUDA_LAUNCH_PARAMS::function has N parameters, then ::CUDA_LAUNCH_PARAMS::kernelParams + * needs to be an array of N pointers. Each of ::CUDA_LAUNCH_PARAMS::kernelParams[0] through + * ::CUDA_LAUNCH_PARAMS::kernelParams[N-1] must point to a region of memory from which the actual + * kernel parameter will be copied. The number of kernel parameters and their offsets and sizes + * do not need to be specified as that information is retrieved directly from the kernel's image. + * + * By default, the kernel won't begin execution on any GPU until all prior work in all the specified + * streams has completed. This behavior can be overridden by specifying the flag + * ::CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC. When this flag is specified, each kernel + * will only wait for prior work in the stream corresponding to that GPU to complete before it begins + * execution. + * + * Similarly, by default, any subsequent work pushed in any of the specified streams will not begin + * execution until the kernels on all GPUs have completed. This behavior can be overridden by specifying + * the flag ::CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC. When this flag is specified, + * any subsequent work pushed in any of the specified streams will only wait for the kernel launched + * on the GPU corresponding to that stream to complete before it begins execution. + * + * Calling ::cuLaunchCooperativeKernelMultiDevice() sets persistent function state that is + * the same as function state set through ::cuLaunchKernel API when called individually for each + * element in \p launchParamsList. + * + * When kernels are launched via ::cuLaunchCooperativeKernelMultiDevice(), the previous + * block shape, shared size and parameter info associated with each ::CUDA_LAUNCH_PARAMS::function + * in \p launchParamsList is overwritten. + * + * Note that to use ::cuLaunchCooperativeKernelMultiDevice(), the kernels must either have + * been compiled with toolchain version 3.2 or later so that it will + * contain kernel parameter information, or have no kernel parameters. + * If either of these conditions is not met, then ::cuLaunchCooperativeKernelMultiDevice() will + * return ::CUDA_ERROR_INVALID_IMAGE. + * + * \param launchParamsList - List of launch parameters, one per device + * \param numDevices - Size of the \p launchParamsList array + * \param flags - Flags to control launch behavior + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_CONTEXT, + * ::CUDA_ERROR_INVALID_HANDLE, + * ::CUDA_ERROR_INVALID_IMAGE, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_LAUNCH_FAILED, + * ::CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES, + * ::CUDA_ERROR_LAUNCH_TIMEOUT, + * ::CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING, + * ::CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE, + * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED + * \note_null_stream + * \notefnerr + * + * \sa ::cuCtxGetCacheConfig, + * ::cuCtxSetCacheConfig, + * ::cuFuncSetCacheConfig, + * ::cuFuncGetAttribute, + * ::cuLaunchCooperativeKernel, + * ::cudaLaunchCooperativeKernelMultiDevice + */ +CUresult CUDAAPI cuLaunchCooperativeKernelMultiDevice(CUDA_LAUNCH_PARAMS *launchParamsList, unsigned int numDevices, unsigned int flags); + +#endif /* __CUDA_API_VERSION >= 9000 */ + +#if __CUDA_API_VERSION >= 10000 + +/** + * \brief Enqueues a host function call in a stream + * + * Enqueues a host function to run in a stream. The function will be called + * after currently enqueued work and will block work added after it. + * + * The host function must not make any CUDA API calls. Attempting to use a + * CUDA API may result in ::CUDA_ERROR_NOT_PERMITTED, but this is not required. + * The host function must not perform any synchronization that may depend on + * outstanding CUDA work not mandated to run earlier. Host functions without a + * mandated order (such as in independent streams) execute in undefined order + * and may be serialized. + * + * For the purposes of Unified Memory, execution makes a number of guarantees: + * <ul> + * <li>The stream is considered idle for the duration of the function's + * execution. Thus, for example, the function may always use memory attached + * to the stream it was enqueued in.</li> + * <li>The start of execution of the function has the same effect as + * synchronizing an event recorded in the same stream immediately prior to + * the function. It thus synchronizes streams which have been "joined" + * prior to the function.</li> + * <li>Adding device work to any stream does not have the effect of making + * the stream active until all preceding host functions and stream callbacks + * have executed. Thus, for + * example, a function might use global attached memory even if work has + * been added to another stream, if the work has been ordered behind the + * function call with an event.</li> + * <li>Completion of the function does not cause a stream to become + * active except as described above. The stream will remain idle + * if no device work follows the function, and will remain idle across + * consecutive host functions or stream callbacks without device work in + * between. Thus, for example, + * stream synchronization can be done by signaling from a host function at the + * end of the stream.</li> + * </ul> + * + * Note that, in contrast to ::cuStreamAddCallback, the function will not be + * called in the event of an error in the CUDA context. + * + * \param hStream - Stream to enqueue function call in + * \param fn - The function to call once preceding stream operations are complete + * \param userData - User-specified data to be passed to the function + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_CONTEXT, + * ::CUDA_ERROR_INVALID_HANDLE, + * ::CUDA_ERROR_NOT_SUPPORTED + * \note_null_stream + * \notefnerr + * + * \sa ::cuStreamCreate, + * ::cuStreamQuery, + * ::cuStreamSynchronize, + * ::cuStreamWaitEvent, + * ::cuStreamDestroy, + * ::cuMemAllocManaged, + * ::cuStreamAttachMemAsync, + * ::cuStreamAddCallback + */ +CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn, void *userData); + +#endif /* __CUDA_API_VERSION >= 10000 */ /** @} */ /* END CUDA_EXEC */ @@ -9000,7 +11096,7 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, int y, int z); +__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, int y, int z); /** * \brief Sets the dynamic shared-memory size for the function @@ -9034,7 +11130,7 @@ CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, int y, int z); * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, unsigned int bytes); +__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, unsigned int bytes); /** * \brief Sets the parameter size for the function @@ -9066,7 +11162,7 @@ CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, unsigned int bytes); * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, unsigned int numbytes); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, unsigned int numbytes); /** * \brief Adds an integer parameter to the function's argument list @@ -9099,7 +11195,7 @@ CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, unsigned int numbytes); * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, unsigned int value); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, unsigned int value); /** * \brief Adds a floating-point parameter to the function's argument list @@ -9132,7 +11228,7 @@ CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, unsigned int value); * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, float value); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, float value); /** * \brief Adds arbitrary data to the function's argument list @@ -9167,7 +11263,7 @@ CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, float value); * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset, void *ptr, unsigned int numbytes); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset, void *ptr, unsigned int numbytes); /** * \brief Launches a CUDA function @@ -9204,7 +11300,7 @@ CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset, void *ptr, unsigned i * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuLaunch(CUfunction f); +__CUDA_DEPRECATED CUresult CUDAAPI cuLaunch(CUfunction f); /** * \brief Launches a CUDA function @@ -9243,7 +11339,7 @@ CUresult CUDAAPI cuLaunch(CUfunction f); * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, int grid_height); +__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, int grid_height); /** * \brief Launches a CUDA function @@ -9272,10 +11368,10 @@ CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, int grid_height); * ::CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING, * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED * - * \note In certain cases where cubins are created with no ABI (i.e., using \p ptxas \p --abi-compile \p no), - * this function may serialize kernel launches. In order to force the CUDA driver to retain - * asynchronous behavior, set the ::CU_CTX_LMEM_RESIZE_TO_MAX flag during context creation (see ::cuCtxCreate). - * + * \note In certain cases where cubins are created with no ABI (i.e., using \p ptxas \p --abi-compile \p no), + * this function may serialize kernel launches. In order to force the CUDA driver to retain + * asynchronous behavior, set the ::CU_CTX_LMEM_RESIZE_TO_MAX flag during context creation (see ::cuCtxCreate). + * * \note_null_stream * \notefnerr * @@ -9290,7 +11386,7 @@ CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, int grid_height); * ::cuLaunchGrid, * ::cuLaunchKernel */ -CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, int grid_width, int grid_height, CUstream hStream); +__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, int grid_width, int grid_height, CUstream hStream); /** @@ -9315,9 +11411,1043 @@ CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, int grid_width, int grid_height * ::CUDA_ERROR_INVALID_VALUE * \notefnerr */ -CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc, int texunit, CUtexref hTexRef); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc, int texunit, CUtexref hTexRef); /** @} */ /* END CUDA_EXEC_DEPRECATED */ +#if __CUDA_API_VERSION >= 10000 +/** + * \defgroup CUDA_GRAPH Graph Management + * + * ___MANBRIEF___ graph management functions of the low-level CUDA driver API + * (___CURRENT_FILE___) ___ENDMANBRIEF___ + * + * This section describes the graph management functions of the low-level CUDA + * driver application programming interface. + * + * @{ + */ + +/** + * \brief Creates a graph + * + * Creates an empty graph, which is returned via \p phGraph. + * + * \param phGraph - Returns newly created graph + * \param flags - Graph creation flags, must be 0 + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_OUT_OF_MEMORY + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphAddChildGraphNode, + * ::cuGraphAddEmptyNode, + * ::cuGraphAddKernelNode, + * ::cuGraphAddHostNode, + * ::cuGraphAddMemcpyNode, + * ::cuGraphAddMemsetNode, + * ::cuGraphInstantiate, + * ::cuGraphDestroy, + * ::cuGraphGetNodes, + * ::cuGraphGetRootNodes, + * ::cuGraphGetEdges, + * ::cuGraphClone + */ +CUresult CUDAAPI cuGraphCreate(CUgraph *phGraph, unsigned int flags); + +/** + * \brief Creates a kernel execution node and adds it to a graph + * + * Creates a new kernel execution node and adds it to \p hGraph with \p numDependencies + * dependencies specified via \p dependencies and arguments specified in \p nodeParams. + * It is possible for \p numDependencies to be 0, in which case the node will be placed + * at the root of the graph. \p dependencies may not have any duplicate entries. + * A handle to the new node will be returned in \p phGraphNode. + * + * The CUDA_KERNEL_NODE_PARAMS structure is defined as: + * + * \code + * typedef struct CUDA_KERNEL_NODE_PARAMS_st { + * CUfunction func; + * unsigned int gridDimX; + * unsigned int gridDimY; + * unsigned int gridDimZ; + * unsigned int blockDimX; + * unsigned int blockDimY; + * unsigned int blockDimZ; + * unsigned int sharedMemBytes; + * void **kernelParams; + * void **extra; + * } CUDA_KERNEL_NODE_PARAMS; + * \endcode + * + * When the graph is launched, the node will invoke kernel \p func on a (\p gridDimX x + * \p gridDimY x \p gridDimZ) grid of blocks. Each block contains + * (\p blockDimX x \p blockDimY x \p blockDimZ) threads. + * + * \p sharedMemBytes sets the amount of dynamic shared memory that will be + * available to each thread block. + * + * Kernel parameters to \p func can be specified in one of two ways: + * + * 1) Kernel parameters can be specified via \p kernelParams. If the kernel has N + * parameters, then \p kernelParams needs to be an array of N pointers. Each pointer, + * from \p kernelParams[0] to \p kernelParams[N-1], points to the region of memory from which the actual + * parameter will be copied. The number of kernel parameters and their offsets and sizes do not need + * to be specified as that information is retrieved directly from the kernel's image. + * + * 2) Kernel parameters can also be packaged by the application into a single buffer that is passed in + * via \p extra. This places the burden on the application of knowing each kernel + * parameter's size and alignment/padding within the buffer. The \p extra parameter exists + * to allow this function to take additional less commonly used arguments. \p extra specifies + * a list of names of extra settings and their corresponding values. Each extra setting name is + * immediately followed by the corresponding value. The list must be terminated with either NULL or + * CU_LAUNCH_PARAM_END. + * + * - ::CU_LAUNCH_PARAM_END, which indicates the end of the \p extra + * array; + * - ::CU_LAUNCH_PARAM_BUFFER_POINTER, which specifies that the next + * value in \p extra will be a pointer to a buffer + * containing all the kernel parameters for launching kernel + * \p func; + * - ::CU_LAUNCH_PARAM_BUFFER_SIZE, which specifies that the next + * value in \p extra will be a pointer to a size_t + * containing the size of the buffer specified with + * ::CU_LAUNCH_PARAM_BUFFER_POINTER; + * + * The error ::CUDA_ERROR_INVALID_VALUE will be returned if kernel parameters are specified with both + * \p kernelParams and \p extra (i.e. both \p kernelParams and + * \p extra are non-NULL). + * + * The \p kernelParams or \p extra array, as well as the argument values it points to, + * are copied during this call. + * + * \note Kernels launched using graphs must not use texture and surface references. Reading or + * writing through any texture or surface reference is undefined behavior. + * This restriction does not apply to texture and surface objects. + * + * \param phGraphNode - Returns newly created node + * \param hGraph - Graph to which to add the node + * \param dependencies - Dependencies of the node + * \param numDependencies - Number of dependencies + * \param nodeParams - Parameters for the GPU execution node + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuLaunchKernel, + * ::cuGraphKernelNodeGetParams, + * ::cuGraphKernelNodeSetParams, + * ::cuGraphCreate, + * ::cuGraphDestroyNode, + * ::cuGraphAddChildGraphNode, + * ::cuGraphAddEmptyNode, + * ::cuGraphAddHostNode, + * ::cuGraphAddMemcpyNode, + * ::cuGraphAddMemsetNode + */ +CUresult CUDAAPI cuGraphAddKernelNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_KERNEL_NODE_PARAMS *nodeParams); + +/** + * \brief Returns a kernel node's parameters + * + * Returns the parameters of kernel node \p hNode in \p nodeParams. + * The \p kernelParams or \p extra array returned in \p nodeParams, + * as well as the argument values it points to, are owned by the node. + * This memory remains valid until the node is destroyed or its + * parameters are modified, and should not be modified + * directly. Use ::cuGraphKernelNodeSetParams to update the + * parameters of this node. + * + * The params will contain either \p kernelParams or \p extra, + * according to which of these was most recently set on the node. + * + * \param hNode - Node to get the parameters for + * \param nodeParams - Pointer to return the parameters + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuLaunchKernel, + * ::cuGraphAddKernelNode, + * ::cuGraphKernelNodeSetParams + */ +CUresult CUDAAPI cuGraphKernelNodeGetParams(CUgraphNode hNode, CUDA_KERNEL_NODE_PARAMS *nodeParams); + +/** + * \brief Sets a kernel node's parameters + * + * Sets the parameters of kernel node \p hNode to \p nodeParams. + * + * \param hNode - Node to set the parameters for + * \param nodeParams - Parameters to copy + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_HANDLE, + * ::CUDA_ERROR_OUT_OF_MEMORY + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuLaunchKernel, + * ::cuGraphAddKernelNode, + * ::cuGraphKernelNodeGetParams + */ +CUresult CUDAAPI cuGraphKernelNodeSetParams(CUgraphNode hNode, const CUDA_KERNEL_NODE_PARAMS *nodeParams); + +/** + * \brief Creates a memcpy node and adds it to a graph + * + * Creates a new memcpy node and adds it to \p hGraph with \p numDependencies + * dependencies specified via \p dependencies. + * It is possible for \p numDependencies to be 0, in which case the node will be placed + * at the root of the graph. \p dependencies may not have any duplicate entries. + * A handle to the new node will be returned in \p phGraphNode. + * + * When the graph is launched, the node will perform the memcpy described by \p copyParams. + * See ::cuMemcpy3D() for a description of the structure and its restrictions. + * + * Memcpy nodes have some additional restrictions with regards to managed memory, if the + * system contains at least one device which has a zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. If one or more of the operands refer + * to managed memory, then using the memory type ::CU_MEMORYTYPE_UNIFIED is disallowed + * for those operand(s). The managed memory will be treated as residing on either the + * host or the device, depending on which memory type is specified. + * + * \param phGraphNode - Returns newly created node + * \param hGraph - Graph to which to add the node + * \param dependencies - Dependencies of the node + * \param numDependencies - Number of dependencies + * \param copyParams - Parameters for the memory copy + * \param ctx - Context on which to run the node + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuMemcpy3D, + * ::cuGraphMemcpyNodeGetParams, + * ::cuGraphMemcpyNodeSetParams, + * ::cuGraphCreate, + * ::cuGraphDestroyNode, + * ::cuGraphAddChildGraphNode, + * ::cuGraphAddEmptyNode, + * ::cuGraphAddKernelNode, + * ::cuGraphAddHostNode, + * ::cuGraphAddMemsetNode + */ +CUresult CUDAAPI cuGraphAddMemcpyNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_MEMCPY3D *copyParams, CUcontext ctx); + +/** + * \brief Returns a memcpy node's parameters + * + * Returns the parameters of memcpy node \p hNode in \p nodeParams. + * + * \param hNode - Node to get the parameters for + * \param nodeParams - Pointer to return the parameters + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuMemcpy3D, + * ::cuGraphAddMemcpyNode, + * ::cuGraphMemcpyNodeSetParams + */ +CUresult CUDAAPI cuGraphMemcpyNodeGetParams(CUgraphNode hNode, CUDA_MEMCPY3D *nodeParams); + +/** + * \brief Sets a memcpy node's parameters + * + * Sets the parameters of memcpy node \p hNode to \p nodeParams. + * + * \param hNode - Node to set the parameters for + * \param nodeParams - Parameters to copy + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuMemcpy3D, + * ::cuGraphAddMemcpyNode, + * ::cuGraphMemcpyNodeGetParams + */ +CUresult CUDAAPI cuGraphMemcpyNodeSetParams(CUgraphNode hNode, const CUDA_MEMCPY3D *nodeParams); + +/** + * \brief Creates a memset node and adds it to a graph + * + * Creates a new memset node and adds it to \p hGraph with \p numDependencies + * dependencies specified via \p dependencies. + * It is possible for \p numDependencies to be 0, in which case the node will be placed + * at the root of the graph. \p dependencies may not have any duplicate entries. + * A handle to the new node will be returned in \p phGraphNode. + * + * The element size must be 1, 2, or 4 bytes. + * When the graph is launched, the node will perform the memset described by \p memsetParams. + * + * \param phGraphNode - Returns newly created node + * \param hGraph - Graph to which to add the node + * \param dependencies - Dependencies of the node + * \param numDependencies - Number of dependencies + * \param memsetParams - Parameters for the memory set + * \param ctx - Context on which to run the node + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_INVALID_CONTEXT + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuMemsetD2D32, + * ::cuGraphMemsetNodeGetParams, + * ::cuGraphMemsetNodeSetParams, + * ::cuGraphCreate, + * ::cuGraphDestroyNode, + * ::cuGraphAddChildGraphNode, + * ::cuGraphAddEmptyNode, + * ::cuGraphAddKernelNode, + * ::cuGraphAddHostNode, + * ::cuGraphAddMemcpyNode + */ +CUresult CUDAAPI cuGraphAddMemsetNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_MEMSET_NODE_PARAMS *memsetParams, CUcontext ctx); + +/** + * \brief Returns a memset node's parameters + * + * Returns the parameters of memset node \p hNode in \p nodeParams. + * + * \param hNode - Node to get the parameters for + * \param nodeParams - Pointer to return the parameters + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuMemsetD2D32, + * ::cuGraphAddMemsetNode, + * ::cuGraphMemsetNodeSetParams + */ +CUresult CUDAAPI cuGraphMemsetNodeGetParams(CUgraphNode hNode, CUDA_MEMSET_NODE_PARAMS *nodeParams); + +/** + * \brief Sets a memset node's parameters + * + * Sets the parameters of memset node \p hNode to \p nodeParams. + * + * \param hNode - Node to set the parameters for + * \param nodeParams - Parameters to copy + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuMemsetD2D32, + * ::cuGraphAddMemsetNode, + * ::cuGraphMemsetNodeGetParams + */ +CUresult CUDAAPI cuGraphMemsetNodeSetParams(CUgraphNode hNode, const CUDA_MEMSET_NODE_PARAMS *nodeParams); + +/** + * \brief Creates a host execution node and adds it to a graph + * + * Creates a new CPU execution node and adds it to \p hGraph with \p numDependencies + * dependencies specified via \p dependencies and arguments specified in \p nodeParams. + * It is possible for \p numDependencies to be 0, in which case the node will be placed + * at the root of the graph. \p dependencies may not have any duplicate entries. + * A handle to the new node will be returned in \p phGraphNode. + * + * When the graph is launched, the node will invoke the specified CPU function. + * Host nodes are not supported under MPS with pre-Volta GPUs. + * + * \param phGraphNode - Returns newly created node + * \param hGraph - Graph to which to add the node + * \param dependencies - Dependencies of the node + * \param numDependencies - Number of dependencies + * \param nodeParams - Parameters for the host node + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_NOT_SUPPORTED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuLaunchHostFunc, + * ::cuGraphHostNodeGetParams, + * ::cuGraphHostNodeSetParams, + * ::cuGraphCreate, + * ::cuGraphDestroyNode, + * ::cuGraphAddChildGraphNode, + * ::cuGraphAddEmptyNode, + * ::cuGraphAddKernelNode, + * ::cuGraphAddMemcpyNode, + * ::cuGraphAddMemsetNode + */ +CUresult CUDAAPI cuGraphAddHostNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_HOST_NODE_PARAMS *nodeParams); + +/** + * \brief Returns a host node's parameters + * + * Returns the parameters of host node \p hNode in \p nodeParams. + * + * \param hNode - Node to get the parameters for + * \param nodeParams - Pointer to return the parameters + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuLaunchHostFunc, + * ::cuGraphAddHostNode, + * ::cuGraphHostNodeSetParams + */ +CUresult CUDAAPI cuGraphHostNodeGetParams(CUgraphNode hNode, CUDA_HOST_NODE_PARAMS *nodeParams); + +/** + * \brief Sets a host node's parameters + * + * Sets the parameters of host node \p hNode to \p nodeParams. + * + * \param hNode - Node to set the parameters for + * \param nodeParams - Parameters to copy + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuLaunchHostFunc, + * ::cuGraphAddHostNode, + * ::cuGraphHostNodeGetParams + */ +CUresult CUDAAPI cuGraphHostNodeSetParams(CUgraphNode hNode, const CUDA_HOST_NODE_PARAMS *nodeParams); + +/** + * \brief Creates a child graph node and adds it to a graph + * + * Creates a new node which executes an embedded graph, and adds it to \p hGraph with + * \p numDependencies dependencies specified via \p dependencies. + * It is possible for \p numDependencies to be 0, in which case the node will be placed + * at the root of the graph. \p dependencies may not have any duplicate entries. + * A handle to the new node will be returned in \p phGraphNode. + * + * The node executes an embedded child graph. The child graph is cloned in this call. + * + * \param phGraphNode - Returns newly created node + * \param hGraph - Graph to which to add the node + * \param dependencies - Dependencies of the node + * \param numDependencies - Number of dependencies + * \param childGraph - The graph to clone into this node + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphChildGraphNodeGetGraph, + * ::cuGraphCreate, + * ::cuGraphDestroyNode, + * ::cuGraphAddEmptyNode, + * ::cuGraphAddKernelNode, + * ::cuGraphAddHostNode, + * ::cuGraphAddMemcpyNode, + * ::cuGraphAddMemsetNode, + * ::cuGraphClone + */ +CUresult CUDAAPI cuGraphAddChildGraphNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, CUgraph childGraph); + +/** + * \brief Gets a handle to the embedded graph of a child graph node + * + * Gets a handle to the embedded graph in a child graph node. This call + * does not clone the graph. Changes to the graph will be reflected in + * the node, and the node retains ownership of the graph. + * + * \param hNode - Node to get the embedded graph for + * \param phGraph - Location to store a handle to the graph + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphAddChildGraphNode, + * ::cuGraphNodeFindInClone + */ +CUresult CUDAAPI cuGraphChildGraphNodeGetGraph(CUgraphNode hNode, CUgraph *phGraph); + +/** + * \brief Creates an empty node and adds it to a graph + * + * Creates a new node which performs no operation, and adds it to \p hGraph with + * \p numDependencies dependencies specified via \p dependencies. + * It is possible for \p numDependencies to be 0, in which case the node will be placed + * at the root of the graph. \p dependencies may not have any duplicate entries. + * A handle to the new node will be returned in \p phGraphNode. + * + * An empty node performs no operation during execution, but can be used for + * transitive ordering. For example, a phased execution graph with 2 groups of n + * nodes with a barrier between them can be represented using an empty node and + * 2*n dependency edges, rather than no empty node and n^2 dependency edges. + * + * \param phGraphNode - Returns newly created node + * \param hGraph - Graph to which to add the node + * \param dependencies - Dependencies of the node + * \param numDependencies - Number of dependencies + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE, + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphCreate, + * ::cuGraphDestroyNode, + * ::cuGraphAddChildGraphNode, + * ::cuGraphAddKernelNode, + * ::cuGraphAddHostNode, + * ::cuGraphAddMemcpyNode, + * ::cuGraphAddMemsetNode + */ +CUresult CUDAAPI cuGraphAddEmptyNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies); + +/** + * \brief Clones a graph + * + * This function creates a copy of \p originalGraph and returns it in \p * phGraphClone. + * All parameters are copied into the cloned graph. The original graph may be modified + * after this call without affecting the clone. + * + * Child graph nodes in the original graph are recursively copied into the clone. + * + * \param phGraphClone - Returns newly created cloned graph + * \param originalGraph - Graph to clone + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE, + * ::CUDA_ERROR_OUT_OF_MEMORY + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphCreate, + * ::cuGraphNodeFindInClone + */ +CUresult CUDAAPI cuGraphClone(CUgraph *phGraphClone, CUgraph originalGraph); + +/** + * \brief Finds a cloned version of a node + * + * This function returns the node in \p hClonedGraph corresponding to \p hOriginalNode + * in the original graph. + * + * \p hClonedGraph must have been cloned from \p hOriginalGraph via ::cuGraphClone. + * \p hOriginalNode must have been in \p hOriginalGraph at the time of the call to + * ::cuGraphClone, and the corresponding cloned node in \p hClonedGraph must not have + * been removed. The cloned node is then returned via \p phClonedNode. + * + * \param phNode - Returns handle to the cloned node + * \param hOriginalNode - Handle to the original node + * \param hClonedGraph - Cloned graph to query + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE, + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphClone + */ +CUresult CUDAAPI cuGraphNodeFindInClone(CUgraphNode *phNode, CUgraphNode hOriginalNode, CUgraph hClonedGraph); + +/** + * \brief Returns a node's type + * + * Returns the node type of \p hNode in \p type. + * + * \param hNode - Node to query + * \param type - Pointer to return the node type + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphGetNodes, + * ::cuGraphGetRootNodes, + * ::cuGraphChildGraphNodeGetGraph, + * ::cuGraphKernelNodeGetParams, + * ::cuGraphKernelNodeSetParams, + * ::cuGraphHostNodeGetParams, + * ::cuGraphHostNodeSetParams, + * ::cuGraphMemcpyNodeGetParams, + * ::cuGraphMemcpyNodeSetParams, + * ::cuGraphMemsetNodeGetParams, + * ::cuGraphMemsetNodeSetParams + */ +CUresult CUDAAPI cuGraphNodeGetType(CUgraphNode hNode, CUgraphNodeType *type); + +/** + * \brief Returns a graph's nodes + * + * Returns a list of \p hGraph's nodes. \p nodes may be NULL, in which case this + * function will return the number of nodes in \p numNodes. Otherwise, + * \p numNodes entries will be filled in. If \p numNodes is higher than the actual + * number of nodes, the remaining entries in \p nodes will be set to NULL, and the + * number of nodes actually obtained will be returned in \p numNodes. + * + * \param hGraph - Graph to query + * \param nodes - Pointer to return the nodes + * \param numNodes - See description + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphCreate, + * ::cuGraphGetRootNodes, + * ::cuGraphGetEdges, + * ::cuGraphNodeGetType, + * ::cuGraphNodeGetDependencies, + * ::cuGraphNodeGetDependentNodes + */ +CUresult CUDAAPI cuGraphGetNodes(CUgraph hGraph, CUgraphNode *nodes, size_t *numNodes); + +/** + * \brief Returns a graph's root nodes + * + * Returns a list of \p hGraph's root nodes. \p rootNodes may be NULL, in which case this + * function will return the number of root nodes in \p numRootNodes. Otherwise, + * \p numRootNodes entries will be filled in. If \p numRootNodes is higher than the actual + * number of root nodes, the remaining entries in \p rootNodes will be set to NULL, and the + * number of nodes actually obtained will be returned in \p numRootNodes. + * + * \param hGraph - Graph to query + * \param rootNodes - Pointer to return the root nodes + * \param numRootNodes - See description + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphCreate, + * ::cuGraphGetNodes, + * ::cuGraphGetEdges, + * ::cuGraphNodeGetType, + * ::cuGraphNodeGetDependencies, + * ::cuGraphNodeGetDependentNodes + */ +CUresult CUDAAPI cuGraphGetRootNodes(CUgraph hGraph, CUgraphNode *rootNodes, size_t *numRootNodes); + +/** + * \brief Returns a graph's dependency edges + * + * Returns a list of \p hGraph's dependency edges. Edges are returned via corresponding + * indices in \p from and \p to; that is, the node in \p to[i] has a dependency on the + * node in \p from[i]. \p from and \p to may both be NULL, in which + * case this function only returns the number of edges in \p numEdges. Otherwise, + * \p numEdges entries will be filled in. If \p numEdges is higher than the actual + * number of edges, the remaining entries in \p from and \p to will be set to NULL, and + * the number of edges actually returned will be written to \p numEdges. + * + * \param hGraph - Graph to get the edges from + * \param from - Location to return edge endpoints + * \param to - Location to return edge endpoints + * \param numEdges - See description + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphGetNodes, + * ::cuGraphGetRootNodes, + * ::cuGraphAddDependencies, + * ::cuGraphRemoveDependencies, + * ::cuGraphNodeGetDependencies, + * ::cuGraphNodeGetDependentNodes + */ +CUresult CUDAAPI cuGraphGetEdges(CUgraph hGraph, CUgraphNode *from, CUgraphNode *to, size_t *numEdges); + +/** + * \brief Returns a node's dependencies + * + * Returns a list of \p node's dependencies. \p dependencies may be NULL, in which case this + * function will return the number of dependencies in \p numDependencies. Otherwise, + * \p numDependencies entries will be filled in. If \p numDependencies is higher than the actual + * number of dependencies, the remaining entries in \p dependencies will be set to NULL, and the + * number of nodes actually obtained will be returned in \p numDependencies. + * + * \param hNode - Node to query + * \param dependencies - Pointer to return the dependencies + * \param numDependencies - See description + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphNodeGetDependentNodes, + * ::cuGraphGetNodes, + * ::cuGraphGetRootNodes, + * ::cuGraphGetEdges, + * ::cuGraphAddDependencies, + * ::cuGraphRemoveDependencies + */ +CUresult CUDAAPI cuGraphNodeGetDependencies(CUgraphNode hNode, CUgraphNode *dependencies, size_t *numDependencies); + +/** + * \brief Returns a node's dependent nodes + * + * Returns a list of \p node's dependent nodes. \p dependentNodes may be NULL, in which + * case this function will return the number of dependent nodes in \p numDependentNodes. + * Otherwise, \p numDependentNodes entries will be filled in. If \p numDependentNodes is + * higher than the actual number of dependent nodes, the remaining entries in + * \p dependentNodes will be set to NULL, and the number of nodes actually obtained will + * be returned in \p numDependentNodes. + * + * \param hNode - Node to query + * \param dependentNodes - Pointer to return the dependent nodes + * \param numDependentNodes - See description + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphNodeGetDependencies, + * ::cuGraphGetNodes, + * ::cuGraphGetRootNodes, + * ::cuGraphGetEdges, + * ::cuGraphAddDependencies, + * ::cuGraphRemoveDependencies + */ +CUresult CUDAAPI cuGraphNodeGetDependentNodes(CUgraphNode hNode, CUgraphNode *dependentNodes, size_t *numDependentNodes); + +/** + * \brief Adds dependency edges to a graph + * + * The number of dependencies to be added is defined by \p numDependencies + * Elements in \p from and \p to at corresponding indices define a dependency. + * Each node in \p from and \p to must belong to \p hGraph. + * + * If \p numDependencies is 0, elements in \p from and \p to will be ignored. + * Specifying an existing dependency will return an error. + * + * \param hGraph - Graph to which dependencies are added + * \param from - Array of nodes that provide the dependencies + * \param to - Array of dependent nodes + * \param numDependencies - Number of dependencies to be added + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphRemoveDependencies, + * ::cuGraphGetEdges, + * ::cuGraphNodeGetDependencies, + * ::cuGraphNodeGetDependentNodes + */ +CUresult CUDAAPI cuGraphAddDependencies(CUgraph hGraph, const CUgraphNode *from, const CUgraphNode *to, size_t numDependencies); + +/** + * \brief Removes dependency edges from a graph + * + * The number of \p dependencies to be removed is defined by \p numDependencies. + * Elements in \p from and \p to at corresponding indices define a dependency. + * Each node in \p from and \p to must belong to \p hGraph. + * + * If \p numDependencies is 0, elements in \p from and \p to will be ignored. + * Specifying a non-existing dependency will return an error. + * + * \param hGraph - Graph from which to remove dependencies + * \param from - Array of nodes that provide the dependencies + * \param to - Array of dependent nodes + * \param numDependencies - Number of dependencies to be removed + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphAddDependencies, + * ::cuGraphGetEdges, + * ::cuGraphNodeGetDependencies, + * ::cuGraphNodeGetDependentNodes + */ +CUresult CUDAAPI cuGraphRemoveDependencies(CUgraph hGraph, const CUgraphNode *from, const CUgraphNode *to, size_t numDependencies); + +/** + * \brief Remove a node from the graph + * + * Removes \p hNode from its graph. This operation also severs any dependencies of other nodes + * on \p hNode and vice versa. + * + * \param hNode - Node to remove + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphAddChildGraphNode, + * ::cuGraphAddEmptyNode, + * ::cuGraphAddKernelNode, + * ::cuGraphAddHostNode, + * ::cuGraphAddMemcpyNode, + * ::cuGraphAddMemsetNode + */ +CUresult CUDAAPI cuGraphDestroyNode(CUgraphNode hNode); + +/** + * \brief Creates an executable graph from a graph + * + * Instantiates \p hGraph as an executable graph. The graph is validated for any + * structural constraints or intra-node constraints which were not previously + * validated. If instantiation is successful, a handle to the instantiated graph + * is returned in \p graphExec. + * + * If there are any errors, diagnostic information may be returned in \p errorNode and + * \p logBuffer. This is the primary way to inspect instantiation errors. The output + * will be null terminated unless the diagnostics overflow + * the buffer. In this case, they will be truncated, and the last byte can be + * inspected to determine if truncation occurred. + * + * \param phGraphExec - Returns instantiated graph + * \param hGraph - Graph to instantiate + * \param phErrorNode - In case of an instantiation error, this may be modified to + * indicate a node contributing to the error + * \param logBuffer - A character buffer to store diagnostic messages + * \param bufferSize - Size of the log buffer in bytes + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphCreate, + * ::cuGraphLaunch, + * ::cuGraphExecDestroy + */ +CUresult CUDAAPI cuGraphInstantiate(CUgraphExec *phGraphExec, CUgraph hGraph, CUgraphNode *phErrorNode, char *logBuffer, size_t bufferSize); + + +#if __CUDA_API_VERSION >= 10010 +/** + * \brief Sets the parameters for a kernel node in the given graphExec + * + * Sets the parameters of a kernel node in an executable graph \p hGraphExec. + * The node is identified by the corresponding node \p hNode in the + * non-executable graph, from which the executable graph was instantiated. + * + * \p hNode must not have been removed from the original graph. The \p func field + * of \p nodeParams cannot be modified and must match the original value. + * All other values can be modified. + * + * The modifications take effect at the next launch of \p hGraphExec. Already + * enqueued or running launches of \p hGraphExec are not affected by this call. + * \p hNode is also not modified by this call. + * + * \param hGraphExec - The executable graph in which to set the specified node + * \param hNode - kernel node from the graph from which graphExec was instantiated + * \param nodeParams - Updated Parameters to set + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_INVALID_VALUE, + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphAddKernelNode, + * ::cuGraphKernelNodeSetParams, + * ::cuGraphInstantiate + */ + CUresult CUDAAPI cuGraphExecKernelNodeSetParams(CUgraphExec hGraphExec, CUgraphNode hNode, const CUDA_KERNEL_NODE_PARAMS *nodeParams); + +#endif /* __CUDA_API_VERSION >= 10010 */ + +/** + * \brief Launches an executable graph in a stream + * + * Executes \p hGraphExec in \p hStream. Only one instance of \p hGraphExec may be executing + * at a time. Each launch is ordered behind both any previous work in \p hStream + * and any previous launches of \p hGraphExec. To execute a graph concurrently, it must be + * instantiated multiple times into multiple executable graphs. + * + * \param hGraphExec - Executable graph to launch + * \param hStream - Stream in which to launch the graph + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphInstantiate, + * ::cuGraphExecDestroy + */ +CUresult CUDAAPI cuGraphLaunch(CUgraphExec hGraphExec, CUstream hStream); + +/** + * \brief Destroys an executable graph + * + * Destroys the executable graph specified by \p hGraphExec, as well + * as all of its executable nodes. If the executable graph is + * in-flight, it will not be terminated, but rather freed + * asynchronously on completion. + * + * \param hGraphExec - Executable graph to destroy + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphInstantiate, + * ::cuGraphLaunch + */ +CUresult CUDAAPI cuGraphExecDestroy(CUgraphExec hGraphExec); + +/** + * \brief Destroys a graph + * + * Destroys the graph specified by \p hGraph, as well as all of its nodes. + * + * \param hGraph - Graph to destroy + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_VALUE + * \note_graph_thread_safety + * \notefnerr + * + * \sa + * ::cuGraphCreate + */ +CUresult CUDAAPI cuGraphDestroy(CUgraph hGraph); +/** @} */ /* END CUDA_GRAPH */ +#endif /* __CUDA_API_VERSION >= 10000 */ #if __CUDA_API_VERSION >= 6050 /** @@ -9352,6 +12482,8 @@ CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc, int texunit, CUtexref hTexRe * ::CUDA_ERROR_UNKNOWN * \notefnerr * + * \sa + * ::cudaOccupancyMaxActiveBlocksPerMultiprocessor */ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor(int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize); @@ -9392,9 +12524,11 @@ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor(int *numBlocks, CUf * ::CUDA_ERROR_UNKNOWN * \notefnerr * + * \sa + * ::cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags */ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize, unsigned int flags); - + /** * \brief Suggest a launch configuration with reasonable occupancy * @@ -9442,6 +12576,8 @@ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBl * ::CUDA_ERROR_UNKNOWN * \notefnerr * + * \sa + * ::cudaOccupancyMaxPotentialBlockSize */ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(int *minGridSize, int *blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit); @@ -9486,6 +12622,8 @@ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(int *minGridSize, int *blockSi * ::CUDA_ERROR_UNKNOWN * \notefnerr * + * \sa + * ::cudaOccupancyMaxPotentialBlockSizeWithFlags */ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(int *minGridSize, int *blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit, unsigned int flags); @@ -9493,13 +12631,13 @@ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(int *minGridSize, int #endif /* __CUDA_API_VERSION >= 6050 */ /** - * \defgroup CUDA_TEXREF Texture Reference Management + * \defgroup CUDA_TEXREF_DEPRECATED Texture Reference Management [DEPRECATED] * - * ___MANBRIEF___ texture reference management functions of the low-level CUDA - * driver API (___CURRENT_FILE___) ___ENDMANBRIEF___ + * ___MANBRIEF___ deprecated texture reference management functions of the + * low-level CUDA driver API (___CURRENT_FILE___) ___ENDMANBRIEF___ * - * This section describes the texture reference management functions of the - * low-level CUDA driver application programming interface. + * This section describes the deprecated texture reference management + * functions of the low-level CUDA driver application programming interface. * * @{ */ @@ -9507,6 +12645,8 @@ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(int *minGridSize, int /** * \brief Binds an array as a texture reference * + * \deprecated + * * Binds the CUDA array \p hArray to the texture reference \p hTexRef. Any * previous address or CUDA array state associated with the texture reference * is superseded by this function. \p Flags must be set to @@ -9528,16 +12668,19 @@ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(int *minGridSize, int * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, * ::cuTexRefSetFilterMode, ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTextureToArray */ CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef, CUarray hArray, unsigned int Flags); /** * \brief Binds a mipmapped array to a texture reference * + * \deprecated + * * Binds the CUDA mipmapped array \p hMipmappedArray to the texture reference \p hTexRef. * Any previous address or CUDA array state associated with the texture reference - * is superseded by this function. \p Flags must be set to ::CU_TRSA_OVERRIDE_FORMAT. + * is superseded by this function. \p Flags must be set to ::CU_TRSA_OVERRIDE_FORMAT. * Any CUDA array previously bound to \p hTexRef is unbound. * * \param hTexRef - Texture reference to bind @@ -9555,7 +12698,8 @@ CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef, CUarray hArray, unsigned int * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, * ::cuTexRefSetFilterMode, ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hMipmappedArray, unsigned int Flags); @@ -9563,6 +12707,8 @@ CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hM /** * \brief Binds an address as a texture reference * + * \deprecated + * * Binds a linear address range to the texture reference \p hTexRef. Any * previous address or CUDA array state associated with the texture reference * is superseded by this function. Any memory previously bound to \p hTexRef @@ -9581,7 +12727,7 @@ CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hM * The total number of elements (or texels) in the linear address range * cannot exceed ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH. * The number of elements is computed as (\p bytes / bytesPerElement), - * where bytesPerElement is determined from the data format and number of + * where bytesPerElement is determined from the data format and number of * components set using ::cuTexRefSetFormat(). * * \param ByteOffset - Returned byte offset @@ -9599,13 +12745,16 @@ CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hM * \sa ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFilterMode, ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTexture */ CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, size_t bytes); /** * \brief Binds an address as a 2D texture reference * + * \deprecated + * * Binds a linear address range to the texture reference \p hTexRef. Any * previous address or CUDA array state associated with the texture reference * is superseded by this function. Any memory previously bound to \p hTexRef @@ -9625,14 +12774,14 @@ CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdevi * supplied, ::CUDA_ERROR_INVALID_VALUE is returned. * * \p Pitch has to be aligned to the hardware-specific texture pitch alignment. - * This value can be queried using the device attribute - * ::CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT. If an unaligned \p Pitch is + * This value can be queried using the device attribute + * ::CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT. If an unaligned \p Pitch is * supplied, ::CUDA_ERROR_INVALID_VALUE is returned. * * Width and Height, which are specified in elements (or texels), cannot exceed * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH and * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT respectively. - * \p Pitch, which is specified in bytes, cannot exceed + * \p Pitch, which is specified in bytes, cannot exceed * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH. * * \param hTexRef - Texture reference to bind @@ -9651,7 +12800,8 @@ CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdevi * ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFilterMode, ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTexture2D */ CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, size_t Pitch); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -9659,6 +12809,8 @@ CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIP /** * \brief Sets the format for a texture reference * + * \deprecated + * * Specifies the format of the data to be read by the texture reference * \p hTexRef. \p fmt and \p NumPackedComponents are exactly analogous to the * ::Format and ::NumChannels members of the ::CUDA_ARRAY_DESCRIPTOR structure: @@ -9680,13 +12832,20 @@ CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIP * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFilterMode, ::cuTexRefSetFlags, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaCreateChannelDesc, + * ::cudaBindTexture, + * ::cudaBindTexture2D, + * ::cudaBindTextureToArray, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int NumPackedComponents); /** * \brief Sets the addressing mode for a texture reference * + * \deprecated + * * Specifies the addressing mode \p am for the given dimension \p dim of the * texture reference \p hTexRef. If \p dim is zero, the addressing mode is * applied to the first parameter of the functions used to fetch from the @@ -9702,7 +12861,7 @@ CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int Num * \endcode * * Note that this call has no effect if \p hTexRef is bound to linear memory. - * Also, if the flag, ::CU_TRSF_NORMALIZED_COORDINATES, is not set, the only + * Also, if the flag, ::CU_TRSF_NORMALIZED_COORDINATES, is not set, the only * supported address mode is ::CU_TR_ADDRESS_MODE_CLAMP. * * \param hTexRef - Texture reference @@ -9720,13 +12879,19 @@ CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int Num * ::cuTexRefSetAddress2D, ::cuTexRefSetArray, * ::cuTexRefSetFilterMode, ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTexture, + * ::cudaBindTexture2D, + * ::cudaBindTextureToArray, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetAddressMode(CUtexref hTexRef, int dim, CUaddress_mode am); /** * \brief Sets the filtering mode for a texture reference * + * \deprecated + * * Specifies the filtering mode \p fm to be used when reading memory through * the texture reference \p hTexRef. ::CUfilter_mode_enum is defined as: * @@ -9753,13 +12918,16 @@ CUresult CUDAAPI cuTexRefSetAddressMode(CUtexref hTexRef, int dim, CUaddress_mod * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTextureToArray */ CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode fm); /** * \brief Sets the mipmap filtering mode for a texture reference * + * \deprecated + * * Specifies the mipmap filtering mode \p fm to be used when reading memory through * the texture reference \p hTexRef. ::CUfilter_mode_enum is defined as: * @@ -9786,14 +12954,17 @@ CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode fm); * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetMipmapFilterMode(CUtexref hTexRef, CUfilter_mode fm); /** * \brief Sets the mipmap level bias for a texture reference * - * Specifies the mipmap level bias \p bias to be added to the specified mipmap level when + * \deprecated + * + * Specifies the mipmap level bias \p bias to be added to the specified mipmap level when * reading memory through the texture reference \p hTexRef. * * Note that this call has no effect if \p hTexRef is not bound to a mipmapped array. @@ -9812,15 +12983,18 @@ CUresult CUDAAPI cuTexRefSetMipmapFilterMode(CUtexref hTexRef, CUfilter_mode fm) * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetMipmapLevelBias(CUtexref hTexRef, float bias); /** * \brief Sets the mipmap min/max mipmap level clamps for a texture reference * + * \deprecated + * * Specifies the min/max mipmap level clamps, \p minMipmapLevelClamp and \p maxMipmapLevelClamp - * respectively, to be used when reading memory through the texture reference + * respectively, to be used when reading memory through the texture reference * \p hTexRef. * * Note that this call has no effect if \p hTexRef is not bound to a mipmapped array. @@ -9840,15 +13014,18 @@ CUresult CUDAAPI cuTexRefSetMipmapLevelBias(CUtexref hTexRef, float bias); * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(CUtexref hTexRef, float minMipmapLevelClamp, float maxMipmapLevelClamp); /** * \brief Sets the maximum anisotropy for a texture reference * + * \deprecated + * * Specifies the maximum anisotropy \p maxAniso to be used when reading memory through - * the texture reference \p hTexRef. + * the texture reference \p hTexRef. * * Note that this call has no effect if \p hTexRef is bound to linear memory. * @@ -9866,13 +13043,17 @@ CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(CUtexref hTexRef, float minMipmapLe * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFlags, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTextureToArray, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetMaxAnisotropy(CUtexref hTexRef, unsigned int maxAniso); /** * \brief Sets the border color for a texture reference * + * \deprecated + * * Specifies the value of the RGBA color via the \p pBorderColor to the texture reference * \p hTexRef. The color value supports only float type and holds color components in * the following sequence: @@ -9896,13 +13077,19 @@ CUresult CUDAAPI cuTexRefSetMaxAnisotropy(CUtexref hTexRef, unsigned int maxAnis * ::CUDA_ERROR_INVALID_VALUE * * \sa ::cuTexRefSetAddressMode, - * ::cuTexRefGetAddressMode, ::cuTexRefGetBorderColor + * ::cuTexRefGetAddressMode, ::cuTexRefGetBorderColor, + * ::cudaBindTexture, + * ::cudaBindTexture2D, + * ::cudaBindTextureToArray, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetBorderColor(CUtexref hTexRef, float *pBorderColor); /** * \brief Sets the flags for a texture reference * + * \deprecated + * * Specifies optional flags via \p Flags to specify the behavior of data * returned through the texture reference \p hTexRef. The valid flags are: * @@ -9911,7 +13098,7 @@ CUresult CUDAAPI cuTexRefSetBorderColor(CUtexref hTexRef, float *pBorderColor); * range [0, 1]. Note that texture with 32-bit integer format * would not be promoted, regardless of whether or not this * flag is specified; - * - ::CU_TRSF_NORMALIZED_COORDINATES, which suppresses the + * - ::CU_TRSF_NORMALIZED_COORDINATES, which suppresses the * default behavior of having the texture coordinates range * from [0, Dim) where Dim is the width or height of the CUDA * array. Instead, the texture coordinates [0, 1.0) reference @@ -9931,7 +13118,11 @@ CUresult CUDAAPI cuTexRefSetBorderColor(CUtexref hTexRef, float *pBorderColor); * ::cuTexRefSetAddress2D, ::cuTexRefSetAddressMode, ::cuTexRefSetArray, * ::cuTexRefSetFilterMode, ::cuTexRefSetFormat, * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, - * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat + * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, + * ::cudaBindTexture, + * ::cudaBindTexture2D, + * ::cudaBindTextureToArray, + * ::cudaBindTextureToMipmappedArray */ CUresult CUDAAPI cuTexRefSetFlags(CUtexref hTexRef, unsigned int Flags); @@ -9939,6 +13130,8 @@ CUresult CUDAAPI cuTexRefSetFlags(CUtexref hTexRef, unsigned int Flags); /** * \brief Gets the address associated with a texture reference * + * \deprecated + * * Returns in \p *pdptr the base address bound to the texture reference * \p hTexRef, or returns ::CUDA_ERROR_INVALID_VALUE if the texture reference * is not bound to any device memory range. @@ -9965,6 +13158,8 @@ CUresult CUDAAPI cuTexRefGetAddress(CUdeviceptr *pdptr, CUtexref hTexRef); /** * \brief Gets the array bound to a texture reference * + * \deprecated + * * Returns in \p *phArray the CUDA array bound to the texture reference * \p hTexRef, or returns ::CUDA_ERROR_INVALID_VALUE if the texture reference * is not bound to any CUDA array. @@ -9990,7 +13185,9 @@ CUresult CUDAAPI cuTexRefGetArray(CUarray *phArray, CUtexref hTexRef); /** * \brief Gets the mipmapped array bound to a texture reference * - * Returns in \p *phMipmappedArray the CUDA mipmapped array bound to the texture + * \deprecated + * + * Returns in \p *phMipmappedArray the CUDA mipmapped array bound to the texture * reference \p hTexRef, or returns ::CUDA_ERROR_INVALID_VALUE if the texture reference * is not bound to any CUDA mipmapped array. * @@ -10015,6 +13212,8 @@ CUresult CUDAAPI cuTexRefGetMipmappedArray(CUmipmappedArray *phMipmappedArray, C /** * \brief Gets the addressing mode used by a texture reference * + * \deprecated + * * Returns in \p *pam the addressing mode corresponding to the * dimension \p dim of the texture reference \p hTexRef. Currently, the only * valid value for \p dim are 0 and 1. @@ -10041,6 +13240,8 @@ CUresult CUDAAPI cuTexRefGetAddressMode(CUaddress_mode *pam, CUtexref hTexRef, i /** * \brief Gets the filter-mode used by a texture reference * + * \deprecated + * * Returns in \p *pfm the filtering mode of the texture reference * \p hTexRef. * @@ -10065,6 +13266,8 @@ CUresult CUDAAPI cuTexRefGetFilterMode(CUfilter_mode *pfm, CUtexref hTexRef); /** * \brief Gets the format used by a texture reference * + * \deprecated + * * Returns in \p *pFormat and \p *pNumChannels the format and number * of components of the CUDA array bound to the texture reference \p hTexRef. * If \p pFormat or \p pNumChannels is NULL, it will be ignored. @@ -10091,6 +13294,8 @@ CUresult CUDAAPI cuTexRefGetFormat(CUarray_format *pFormat, int *pNumChannels, C /** * \brief Gets the mipmap filtering mode for a texture reference * + * \deprecated + * * Returns the mipmap filtering mode in \p pfm that's used when reading memory through * the texture reference \p hTexRef. * @@ -10115,6 +13320,8 @@ CUresult CUDAAPI cuTexRefGetMipmapFilterMode(CUfilter_mode *pfm, CUtexref hTexRe /** * \brief Gets the mipmap level bias for a texture reference * + * \deprecated + * * Returns the mipmap level bias in \p pBias that's added to the specified mipmap * level when reading memory through the texture reference \p hTexRef. * @@ -10139,8 +13346,10 @@ CUresult CUDAAPI cuTexRefGetMipmapLevelBias(float *pbias, CUtexref hTexRef); /** * \brief Gets the min/max mipmap level clamps for a texture reference * + * \deprecated + * * Returns the min/max mipmap level clamps in \p pminMipmapLevelClamp and \p pmaxMipmapLevelClamp - * that's used when reading memory through the texture reference \p hTexRef. + * that's used when reading memory through the texture reference \p hTexRef. * * \param pminMipmapLevelClamp - Returned mipmap min level clamp * \param pmaxMipmapLevelClamp - Returned mipmap max level clamp @@ -10164,8 +13373,10 @@ CUresult CUDAAPI cuTexRefGetMipmapLevelClamp(float *pminMipmapLevelClamp, float /** * \brief Gets the maximum anisotropy for a texture reference * + * \deprecated + * * Returns the maximum anisotropy in \p pmaxAniso that's used when reading memory through - * the texture reference \p hTexRef. + * the texture reference \p hTexRef. * * \param pmaxAniso - Returned maximum anisotropy * \param hTexRef - Texture reference @@ -10188,6 +13399,8 @@ CUresult CUDAAPI cuTexRefGetMaxAnisotropy(int *pmaxAniso, CUtexref hTexRef); /** * \brief Gets the border color used by a texture reference * + * \deprecated + * * Returns in \p pBorderColor, values of the RGBA color used by * the texture reference \p hTexRef. * The color value is of type float and holds color components in @@ -10210,11 +13423,13 @@ CUresult CUDAAPI cuTexRefGetMaxAnisotropy(int *pmaxAniso, CUtexref hTexRef); * \sa ::cuTexRefSetAddressMode, * ::cuTexRefSetAddressMode, ::cuTexRefSetBorderColor */ -CUresult CUDAAPI cuTexRefGetBorderColor(float *pBorderColor, CUtexref hTexRef); +CUresult CUDAAPI cuTexRefGetBorderColor(float *pBorderColor, CUtexref hTexRef); /** * \brief Gets the flags used by a texture reference * + * \deprecated + * * Returns in \p *pFlags the flags of the texture reference \p hTexRef. * * \param pFlags - Returned flags @@ -10235,20 +13450,6 @@ CUresult CUDAAPI cuTexRefGetBorderColor(float *pBorderColor, CUtexref hTexRef); */ CUresult CUDAAPI cuTexRefGetFlags(unsigned int *pFlags, CUtexref hTexRef); -/** @} */ /* END CUDA_TEXREF */ - -/** - * \defgroup CUDA_TEXREF_DEPRECATED Texture Reference Management [DEPRECATED] - * - * ___MANBRIEF___ deprecated texture reference management functions of the - * low-level CUDA driver API (___CURRENT_FILE___) ___ENDMANBRIEF___ - * - * This section describes the deprecated texture reference management - * functions of the low-level CUDA driver application programming interface. - * - * @{ - */ - /** * \brief Creates a texture reference * @@ -10298,7 +13499,7 @@ CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef); /** - * \defgroup CUDA_SURFREF Surface Reference Management + * \defgroup CUDA_SURFREF_DEPRECATED Surface Reference Management [DEPRECATED] * * ___MANBRIEF___ surface reference management functions of the low-level CUDA * driver API (___CURRENT_FILE___) ___ENDMANBRIEF___ @@ -10312,6 +13513,8 @@ CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef); /** * \brief Sets the CUDA array for a surface reference. * + * \deprecated + * * Sets the CUDA array \p hArray to be read and written by the surface reference * \p hSurfRef. Any previous CUDA array state associated with the surface * reference is superseded by this function. \p Flags must be set to 0. @@ -10329,13 +13532,18 @@ CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef); * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuModuleGetSurfRef, ::cuSurfRefGetArray + * \sa + * ::cuModuleGetSurfRef, + * ::cuSurfRefGetArray, + * ::cudaBindSurfaceToArray */ CUresult CUDAAPI cuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, unsigned int Flags); /** * \brief Passes back the CUDA array bound to a surface reference. * + * \deprecated + * * Returns in \p *phArray the CUDA array bound to the surface reference * \p hSurfRef, or returns ::CUDA_ERROR_INVALID_VALUE if the surface reference * is not bound to any CUDA array. @@ -10354,7 +13562,7 @@ CUresult CUDAAPI cuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, unsigned */ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); -/** @} */ /* END CUDA_SURFREF */ +/** @} */ /* END CUDA_SURFREF_DEPRECATED */ #if __CUDA_API_VERSION >= 5000 /** @@ -10442,7 +13650,7 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); * must be set to a valid device pointer, that is aligned to ::CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT. * ::CUDA_RESOURCE_DESC::res::linear::format and ::CUDA_RESOURCE_DESC::res::linear::numChannels * describe the format of each component and the number of components per array element. ::CUDA_RESOURCE_DESC::res::linear::sizeInBytes - * specifies the size of the array in bytes. The total number of elements in the linear address range cannot exceed + * specifies the size of the array in bytes. The total number of elements in the linear address range cannot exceed * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH. The number of elements is computed as (sizeInBytes / (sizeof(format) * numChannels)). * * \par @@ -10452,7 +13660,7 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); * describe the format of each component and the number of components per array element. ::CUDA_RESOURCE_DESC::res::pitch2D::width * and ::CUDA_RESOURCE_DESC::res::pitch2D::height specify the width and height of the array in elements, and cannot exceed * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH and ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT respectively. - * ::CUDA_RESOURCE_DESC::res::pitch2D::pitchInBytes specifies the pitch between two rows in bytes and has to be aligned to + * ::CUDA_RESOURCE_DESC::res::pitch2D::pitchInBytes specifies the pitch between two rows in bytes and has to be aligned to * ::CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT. Pitch cannot exceed ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH. * * - ::flags must be set to zero. @@ -10481,7 +13689,7 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); CU_TR_ADDRESS_MODE_BORDER = 3 } CUaddress_mode; * \endcode - * This is ignored if ::CUDA_RESOURCE_DESC::resType is ::CU_RESOURCE_TYPE_LINEAR. Also, if the flag, ::CU_TRSF_NORMALIZED_COORDINATES + * This is ignored if ::CUDA_RESOURCE_DESC::resType is ::CU_RESOURCE_TYPE_LINEAR. Also, if the flag, ::CU_TRSF_NORMALIZED_COORDINATES * is not set, the only supported address mode is ::CU_TR_ADDRESS_MODE_CLAMP. * * - ::CUDA_TEXTURE_DESC::filterMode specifies the filtering mode to be used when fetching from the texture. CUfilter_mode is defined as: @@ -10556,14 +13764,14 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); * - ::CUDA_RESOURCE_VIEW_DESC::firstLayer specifies the first layer index for layered textures. This will be the new layer zero. * For non-layered resources, this value has to be zero. * - * - ::CUDA_RESOURCE_VIEW_DESC::lastLayer specifies the last layer index for layered textures. For non-layered resources, + * - ::CUDA_RESOURCE_VIEW_DESC::lastLayer specifies the last layer index for layered textures. For non-layered resources, * this value has to be zero. * * * \param pTexObject - Texture object to create * \param pResDesc - Resource descriptor * \param pTexDesc - Texture descriptor - * \param pResViewDesc - Resource view descriptor + * \param pResViewDesc - Resource view descriptor * * \return * ::CUDA_SUCCESS, @@ -10572,7 +13780,9 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuTexObjectDestroy + * \sa + * ::cuTexObjectDestroy, + * ::cudaCreateTextureObject */ CUresult CUDAAPI cuTexObjectCreate(CUtexObject *pTexObject, const CUDA_RESOURCE_DESC *pResDesc, const CUDA_TEXTURE_DESC *pTexDesc, const CUDA_RESOURCE_VIEW_DESC *pResViewDesc); @@ -10590,7 +13800,9 @@ CUresult CUDAAPI cuTexObjectCreate(CUtexObject *pTexObject, const CUDA_RESOURCE_ * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuTexObjectCreate + * \sa + * ::cuTexObjectCreate, + * ::cudaDestroyTextureObject */ CUresult CUDAAPI cuTexObjectDestroy(CUtexObject texObject); @@ -10609,7 +13821,9 @@ CUresult CUDAAPI cuTexObjectDestroy(CUtexObject texObject); * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuTexObjectCreate + * \sa + * ::cuTexObjectCreate, + * ::cudaGetTextureObjectResourceDesc, */ CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUtexObject texObject); @@ -10628,7 +13842,9 @@ CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUtexO * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuTexObjectCreate + * \sa + * ::cuTexObjectCreate, + * ::cudaGetTextureObjectTextureDesc */ CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc, CUtexObject texObject); @@ -10648,7 +13864,9 @@ CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc, CUtexObj * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuTexObjectCreate + * \sa + * ::cuTexObjectCreate, + * ::cudaGetTextureObjectResourceViewDesc */ CUresult CUDAAPI cuTexObjectGetResourceViewDesc(CUDA_RESOURCE_VIEW_DESC *pResViewDesc, CUtexObject texObject); @@ -10671,7 +13889,7 @@ CUresult CUDAAPI cuTexObjectGetResourceViewDesc(CUDA_RESOURCE_VIEW_DESC *pResVie * \brief Creates a surface object * * Creates a surface object and returns it in \p pSurfObject. \p pResDesc describes - * the data to perform surface load/stores on. ::CUDA_RESOURCE_DESC::resType must be + * the data to perform surface load/stores on. ::CUDA_RESOURCE_DESC::resType must be * ::CU_RESOURCE_TYPE_ARRAY and ::CUDA_RESOURCE_DESC::res::array::hArray * must be set to a valid CUDA array handle. ::CUDA_RESOURCE_DESC::flags must be set to zero. * @@ -10689,7 +13907,9 @@ CUresult CUDAAPI cuTexObjectGetResourceViewDesc(CUDA_RESOURCE_VIEW_DESC *pResVie * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuSurfObjectDestroy + * \sa + * ::cuSurfObjectDestroy, + * ::cudaCreateSurfaceObject */ CUresult CUDAAPI cuSurfObjectCreate(CUsurfObject *pSurfObject, const CUDA_RESOURCE_DESC *pResDesc); @@ -10707,7 +13927,9 @@ CUresult CUDAAPI cuSurfObjectCreate(CUsurfObject *pSurfObject, const CUDA_RESOUR * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuSurfObjectCreate + * \sa + * ::cuSurfObjectCreate, + * ::cudaDestroySurfaceObject */ CUresult CUDAAPI cuSurfObjectDestroy(CUsurfObject surfObject); @@ -10726,26 +13948,29 @@ CUresult CUDAAPI cuSurfObjectDestroy(CUsurfObject surfObject); * ::CUDA_ERROR_INVALID_CONTEXT, * ::CUDA_ERROR_INVALID_VALUE * - * \sa ::cuSurfObjectCreate + * \sa + * ::cuSurfObjectCreate, + * ::cudaGetSurfaceObjectResourceDesc */ CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUsurfObject surfObject); /** @} */ /* END CUDA_SURFOBJECT */ #endif /* __CUDA_API_VERSION >= 5000 */ -#if __CUDA_API_VERSION >= 4000 /** * \defgroup CUDA_PEER_ACCESS Peer Context Memory Access * * ___MANBRIEF___ direct peer context memory access functions of the low-level * CUDA driver API (___CURRENT_FILE___) ___ENDMANBRIEF___ * - * This section describes the direct peer context memory access functions + * This section describes the direct peer context memory access functions * of the low-level CUDA driver application programming interface. * * @{ */ - + +#if __CUDA_API_VERSION >= 4000 + /** * \brief Queries if a device may directly access a peer device's memory. * @@ -10757,7 +13982,7 @@ CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUsur * \param canAccessPeer - Returned access capability * \param dev - Device from which allocations on \p peerDev are to * be directly accessed. - * \param peerDev - Device on which the allocations to be directly accessed + * \param peerDev - Device on which the allocations to be directly accessed * by \p dev reside. * * \return @@ -10767,47 +13992,12 @@ CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUsur * ::CUDA_ERROR_INVALID_DEVICE * \notefnerr * - * \sa ::cuCtxEnablePeerAccess, - * ::cuCtxDisablePeerAccess - */ -CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, CUdevice peerDev); - - -/** - * \brief Queries attributes of the link between two devices. - * - * Returns in \p *value the value of the requested attribute \p attrib of the - * link between \p srcDevice and \p dstDevice. The supported attributes are: - * - ::CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK: A relative value indicating the - * performance of the link between two devices. - * - ::CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED P2P: 1 if P2P Access is enable. - * - ::CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED: 1 if Atomic operations over - * the link are supported. - * - * Returns ::CUDA_ERROR_INVALID_DEVICE if \p srcDevice or \p dstDevice are not valid - * or if they represent the same device. - * - * Returns ::CUDA_ERROR_INVALID_VALUE if \p attrib is not valid or if \p value is - * a null pointer. - * - * \param value - Returned value of the requested attribute - * \param attrib - The requested attribute of the link between \p srcDevice and \p dstDevice. - * \param srcDevice - The source device of the target link. - * \param dstDevice - The destination device of the target link. - * - * \return - * ::CUDA_SUCCESS, - * ::CUDA_ERROR_DEINITIALIZED, - * ::CUDA_ERROR_NOT_INITIALIZED, - * ::CUDA_ERROR_INVALID_DEVICE, - * ::CUDA_ERROR_INVALID_VALUE - * \notefnerr - * - * \sa ::cuCtxEnablePeerAccess, + * \sa + * ::cuCtxEnablePeerAccess, * ::cuCtxDisablePeerAccess, - * ::cuCtxCanAccessPeer + * ::cudaDeviceCanAccessPeer */ -CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attrib, CUdevice srcDevice, CUdevice dstDevice); +CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, CUdevice peerDev); /** * \brief Enables direct access to memory allocations in a peer context. @@ -10819,7 +14009,7 @@ CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attri * details. * * Note that access granted by this call is unidirectional and that in order to access - * memory from the current context in \p peerContext, a separate symmetric call + * memory from the current context in \p peerContext, a separate symmetric call * to ::cuCtxEnablePeerAccess() is required. * * There is a system-wide maximum of eight peer connections per device. @@ -10831,7 +14021,7 @@ CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attri * Returns ::CUDA_ERROR_PEER_ACCESS_ALREADY_ENABLED if direct access of * \p peerContext from the current context has already been enabled. * - * Returns ::CUDA_ERROR_TOO_MANY_PEERS if direct peer access is not possible + * Returns ::CUDA_ERROR_TOO_MANY_PEERS if direct peer access is not possible * because hardware resources required for peer access have been exhausted. * * Returns ::CUDA_ERROR_INVALID_CONTEXT if there is no current context, \p peerContext @@ -10853,16 +14043,18 @@ CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attri * ::CUDA_ERROR_INVALID_VALUE * \notefnerr * - * \sa ::cuDeviceCanAccessPeer, - * ::cuCtxDisablePeerAccess + * \sa + * ::cuDeviceCanAccessPeer, + * ::cuCtxDisablePeerAccess, + * ::cudaDeviceEnablePeerAccess */ CUresult CUDAAPI cuCtxEnablePeerAccess(CUcontext peerContext, unsigned int Flags); /** - * \brief Disables direct access to memory allocations in a peer context and + * \brief Disables direct access to memory allocations in a peer context and * unregisters any registered allocations. * - Returns ::CUDA_ERROR_PEER_ACCESS_NOT_ENABLED if direct peer access has + Returns ::CUDA_ERROR_PEER_ACCESS_NOT_ENABLED if direct peer access has * not yet been enabled from \p peerContext to the current context. * * Returns ::CUDA_ERROR_INVALID_CONTEXT if there is no current context, or if @@ -10878,14 +14070,61 @@ CUresult CUDAAPI cuCtxEnablePeerAccess(CUcontext peerContext, unsigned int Flags * ::CUDA_ERROR_INVALID_CONTEXT, * \notefnerr * - * \sa ::cuDeviceCanAccessPeer, - * ::cuCtxEnablePeerAccess + * \sa + * ::cuDeviceCanAccessPeer, + * ::cuCtxEnablePeerAccess, + * ::cudaDeviceDisablePeerAccess */ CUresult CUDAAPI cuCtxDisablePeerAccess(CUcontext peerContext); -/** @} */ /* END CUDA_PEER_ACCESS */ #endif /* __CUDA_API_VERSION >= 4000 */ +#if __CUDA_API_VERSION >= 8000 + +/** + * \brief Queries attributes of the link between two devices. + * + * Returns in \p *value the value of the requested attribute \p attrib of the + * link between \p srcDevice and \p dstDevice. The supported attributes are: + * - ::CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK: A relative value indicating the + * performance of the link between two devices. + * - ::CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED P2P: 1 if P2P Access is enable. + * - ::CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED: 1 if Atomic operations over + * the link are supported. + * - ::CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED: 1 if cudaArray can + * be accessed over the link. + * + * Returns ::CUDA_ERROR_INVALID_DEVICE if \p srcDevice or \p dstDevice are not valid + * or if they represent the same device. + * + * Returns ::CUDA_ERROR_INVALID_VALUE if \p attrib is not valid or if \p value is + * a null pointer. + * + * \param value - Returned value of the requested attribute + * \param attrib - The requested attribute of the link between \p srcDevice and \p dstDevice. + * \param srcDevice - The source device of the target link. + * \param dstDevice - The destination device of the target link. + * + * \return + * ::CUDA_SUCCESS, + * ::CUDA_ERROR_DEINITIALIZED, + * ::CUDA_ERROR_NOT_INITIALIZED, + * ::CUDA_ERROR_INVALID_DEVICE, + * ::CUDA_ERROR_INVALID_VALUE + * \notefnerr + * + * \sa + * ::cuCtxEnablePeerAccess, + * ::cuCtxDisablePeerAccess, + * ::cuDeviceCanAccessPeer, + * ::cudaDeviceGetP2PAttribute + */ +CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attrib, CUdevice srcDevice, CUdevice dstDevice); + +#endif /* __CUDA_API_VERSION >= 8000 */ + +/** @} */ /* END CUDA_PEER_ACCESS */ + /** * \defgroup CUDA_GRAPHICS Graphics Interoperability * @@ -10923,7 +14162,8 @@ CUresult CUDAAPI cuCtxDisablePeerAccess(CUcontext peerContext); * ::cuGraphicsD3D10RegisterResource, * ::cuGraphicsD3D11RegisterResource, * ::cuGraphicsGLRegisterBuffer, - * ::cuGraphicsGLRegisterImage + * ::cuGraphicsGLRegisterImage, + * ::cudaGraphicsUnregisterResource */ CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource); @@ -10961,7 +14201,9 @@ CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource); * ::CUDA_ERROR_NOT_MAPPED_AS_ARRAY * \notefnerr * - * \sa ::cuGraphicsResourceGetMappedPointer + * \sa + * ::cuGraphicsResourceGetMappedPointer, + * ::cudaGraphicsSubResourceGetMappedArray */ CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex, unsigned int mipLevel); @@ -10970,8 +14212,8 @@ CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphics /** * \brief Get a mipmapped array through which to access a mapped graphics resource. * - * Returns in \p *pMipmappedArray a mipmapped array through which the mapped graphics - * resource \p resource. The value set in \p *pMipmappedArray may change every time + * Returns in \p *pMipmappedArray a mipmapped array through which the mapped graphics + * resource \p resource. The value set in \p *pMipmappedArray may change every time * that \p resource is mapped. * * If \p resource is not a texture then it cannot be accessed via a mipmapped array and @@ -10992,7 +14234,9 @@ CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphics * ::CUDA_ERROR_NOT_MAPPED_AS_ARRAY * \notefnerr * - * \sa ::cuGraphicsResourceGetMappedPointer + * \sa + * ::cuGraphicsResourceGetMappedPointer, + * ::cudaGraphicsResourceGetMappedMipmappedArray */ CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray(CUmipmappedArray *pMipmappedArray, CUgraphicsResource resource); @@ -11028,7 +14272,8 @@ CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray(CUmipmappedArray *pMi * * \sa * ::cuGraphicsMapResources, - * ::cuGraphicsSubResourceGetMappedArray + * ::cuGraphicsSubResourceGetMappedArray, + * ::cudaGraphicsResourceGetMappedPointer */ CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource); #endif /* __CUDA_API_VERSION >= 3020 */ @@ -11069,7 +14314,8 @@ CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t * \notefnerr * * \sa - * ::cuGraphicsMapResources + * ::cuGraphicsMapResources, + * ::cudaGraphicsResourceSetMapFlags */ CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsigned int flags); @@ -11108,7 +14354,8 @@ CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsi * \sa * ::cuGraphicsResourceGetMappedPointer, * ::cuGraphicsSubResourceGetMappedArray, - * ::cuGraphicsUnmapResources + * ::cuGraphicsUnmapResources, + * ::cudaGraphicsMapResources */ CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); @@ -11144,7 +14391,8 @@ CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource * * \notefnerr * * \sa - * ::cuGraphicsMapResources + * ::cuGraphicsMapResources, + * ::cudaGraphicsUnmapResources */ CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); @@ -11223,6 +14471,7 @@ CUresult CUDAAPI cuGetExportTable(const void **ppExportTable, const CUuuid *pExp #undef cuMemsetD2D32Async #undef cuStreamGetPriority #undef cuStreamGetFlags + #undef cuStreamGetCtx #undef cuStreamWaitEvent #undef cuStreamAddCallback #undef cuStreamAttachMemAsync @@ -11230,12 +14479,23 @@ CUresult CUDAAPI cuGetExportTable(const void **ppExportTable, const CUuuid *pExp #undef cuStreamSynchronize #undef cuEventRecord #undef cuLaunchKernel + #undef cuLaunchHostFunc #undef cuGraphicsMapResources #undef cuGraphicsUnmapResources - #undef cuMemPrefetchAsync #undef cuStreamWriteValue32 #undef cuStreamWaitValue32 + #undef cuStreamWriteValue64 + #undef cuStreamWaitValue64 #undef cuStreamBatchMemOp + #undef cuMemPrefetchAsync + #undef cuLaunchCooperativeKernel + #undef cuSignalExternalSemaphoresAsync + #undef cuWaitExternalSemaphoresAsync + #undef cuStreamBeginCapture + #undef cuStreamEndCapture + #undef cuStreamIsCapturing + #undef cuStreamGetCaptureInfo + #undef cuGraphLaunch #endif /* __CUDA_API_VERSION_INTERNAL */ #if defined(__CUDA_API_VERSION_INTERNAL) || (__CUDA_API_VERSION >= 4000 && __CUDA_API_VERSION < 6050) @@ -11453,6 +14713,7 @@ CUresult CUDAAPI cuEventDestroy(CUevent hEvent); CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority); CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags); + CUresult CUDAAPI cuStreamGetCtx(CUstream hStream, CUcontext *pctx); CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags); CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags); CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags); @@ -11460,12 +14721,25 @@ CUresult CUDAAPI cuEventDestroy(CUevent hEvent); CUresult CUDAAPI cuStreamSynchronize(CUstream hStream); CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream); CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void **kernelParams, void **extra); + CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn, void *userData); CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); - CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream); CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); + CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); + CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, unsigned int flags); + CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream); + CUresult CUDAAPI cuLaunchCooperativeKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void **kernelParams); + CUresult CUDAAPI cuSignalExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); + CUresult CUDAAPI cuWaitExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); + CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream); + CUresult CUDAAPI cuStreamBeginCapture_ptsz(CUstream hStream); + CUresult CUDAAPI cuStreamBeginCapture_v2(CUstream hStream, CUstreamCaptureMode mode); + CUresult CUDAAPI cuStreamEndCapture(CUstream hStream, CUgraph *phGraph); + CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *captureStatus); + CUresult CUDAAPI cuStreamGetCaptureInfo(CUstream hStream, CUstreamCaptureStatus *captureStatus, cuuint64_t *id); + CUresult CUDAAPI cuGraphLaunch(CUgraphExec hGraph, CUstream hStream); #endif #ifdef __cplusplus @@ -11473,6 +14747,6 @@ CUresult CUDAAPI cuEventDestroy(CUevent hEvent); #endif #undef __CUDA_API_VERSION +#undef __CUDA_DEPRECATED #endif /* __cuda_cuda_h__ */ - diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h new file mode 100644 index 0000000..0054697 --- /dev/null +++ b/libcuda/cuda_api_object.h @@ -0,0 +1,209 @@ +#ifndef __cuda_api_object_h__ +#define __cuda_api_object_h__ + +#include <list> +#include <map> +#include <set> +#include <string> + +#include "builtin_types.h" + +#include "../src/gpgpu-sim/gpu-sim.h" +#include "../src/cuda-sim/ptx_ir.h" +#include "../src/abstract_hardware_model.h" +#include "cuobjdump.h" + +typedef std::list<gpgpu_ptx_sim_arg> gpgpu_ptx_sim_arg_list_t; + +#ifndef OPENGL_SUPPORT +typedef unsigned long GLuint; +#endif + +struct glbmap_entry { + GLuint m_bufferObj; + void *m_devPtr; + size_t m_size; + struct glbmap_entry *m_next; +}; + +typedef struct glbmap_entry glbmap_entry_t; + +struct _cuda_device_id { + _cuda_device_id(gpgpu_sim* gpu) {m_id = 0; m_next = NULL; m_gpgpu=gpu;} + struct _cuda_device_id *next() { return m_next; } + unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); } + int num_devices() const { + if( m_next == NULL ) return 1; + else return 1 + m_next->num_devices(); + } + struct _cuda_device_id *get_device( unsigned n ) + { + assert( n < (unsigned)num_devices() ); + struct _cuda_device_id *p=this; + for(unsigned i=0; i<n; i++) + p = p->m_next; + return p; + } + const struct cudaDeviceProp *get_prop() const + { + return m_gpgpu->get_prop(); + } + unsigned get_id() const { return m_id; } + + gpgpu_sim *get_gpgpu() { return m_gpgpu; } +private: + unsigned m_id; + class gpgpu_sim *m_gpgpu; + struct _cuda_device_id *m_next; +}; + +struct CUctx_st { + CUctx_st( _cuda_device_id *gpu ) + { + m_gpu = gpu; + m_binary_info.cmem = 0; + m_binary_info.gmem = 0; + no_of_ptx=0; + } + + _cuda_device_id *get_device() { return m_gpu; } + + void add_binary( symbol_table *symtab, unsigned fat_cubin_handle ) + { + m_code[fat_cubin_handle] = symtab; + m_last_fat_cubin_handle = fat_cubin_handle; + } + + void add_ptxinfo( const char *deviceFun, const struct gpgpu_ptx_sim_info &info ) + { + symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun); + assert( s != NULL ); + function_info *f = s->get_pc(); + assert( f != NULL ); + f->set_kernel_info(info); + } + + void add_ptxinfo( const struct gpgpu_ptx_sim_info &info ) + { + m_binary_info = info; + } + + void register_function( unsigned fat_cubin_handle, const char *hostFun, const char *deviceFun ) + { + if( m_code.find(fat_cubin_handle) != m_code.end() ) { + symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun); + if(s != NULL) { + function_info *f = s->get_pc(); + assert( f != NULL ); + m_kernel_lookup[hostFun] = f; + } + else { + printf("Warning: cannot find deviceFun %s\n", deviceFun); + m_kernel_lookup[hostFun] = NULL; + } + // assert( s != NULL ); + // function_info *f = s->get_pc(); + // assert( f != NULL ); + // m_kernel_lookup[hostFun] = f; + } else { + m_kernel_lookup[hostFun] = NULL; + } + } + + void register_hostFun_function( const char*hostFun, function_info* f){ + m_kernel_lookup[hostFun] = f; + } + + function_info *get_kernel(const char *hostFun) + { + std::map<const void*,function_info*>::iterator i=m_kernel_lookup.find(hostFun); + assert( i != m_kernel_lookup.end() ); + return i->second; + } + + int no_of_ptx; + +private: + _cuda_device_id *m_gpu; // selected gpu + std::map<unsigned,symbol_table*> m_code; // fat binary handle => global symbol table + unsigned m_last_fat_cubin_handle; + std::map<const void*,function_info*> m_kernel_lookup; // unique id (CUDA app function address) => kernel entry point + struct gpgpu_ptx_sim_info m_binary_info; + +}; + +class kernel_config { +public: + kernel_config( dim3 GridDim, dim3 BlockDim, size_t sharedMem, struct CUstream_st *stream ) + { + m_GridDim=GridDim; + m_BlockDim=BlockDim; + m_sharedMem=sharedMem; + m_stream = stream; + } + kernel_config() + { + m_GridDim=dim3(-1,-1,-1); + m_BlockDim=dim3(-1,-1,-1); + m_sharedMem=0; + m_stream =NULL; + } + void set_arg( const void *arg, size_t size, size_t offset ) + { + m_args.push_front( gpgpu_ptx_sim_arg(arg,size,offset) ); + } + dim3 grid_dim() const { return m_GridDim; } + dim3 block_dim() const { return m_BlockDim; } + void set_grid_dim(dim3 *d) { m_GridDim = *d; } + void set_block_dim(dim3 *d) { m_BlockDim = *d; } + gpgpu_ptx_sim_arg_list_t get_args() { return m_args; } + struct CUstream_st *get_stream() { return m_stream; } + +private: + dim3 m_GridDim; + dim3 m_BlockDim; + size_t m_sharedMem; + struct CUstream_st *m_stream; + gpgpu_ptx_sim_arg_list_t m_args; +}; + +class cuda_runtime_api { + public: + cuda_runtime_api() { + g_glbmap = NULL; + g_active_device = 0; //active gpu that runs the code + } + // global list + std::list<cuobjdumpSection*> cuobjdumpSectionList; + std::list<cuobjdumpSection*> libSectionList; + std::list<kernel_config> g_cuda_launch_stack; + std::map<int, bool>fatbin_registered; + std::map<int, std::string> fatbinmap; + std::map<std::string, symbol_table*> name_symtab; + std::map<unsigned long long, size_t> g_mallocPtr_Size; + //maps sm version number to set of filenames + std::map<unsigned, std::set<std::string> > version_filename; + std::map<void *,void **> pinned_memory; //support for pinned memories added + std::map<void *, size_t> pinned_memory_size; + glbmap_entry_t* g_glbmap; + int g_active_device; //active gpu that runs the code + // member function list + void cuobjdumpInit(); + void extract_code_using_cuobjdump(); + void extract_ptx_files_using_cuobjdump(CUctx_st *context); + std::list<cuobjdumpSection*> pruneSectionList(CUctx_st *context); + std::list<cuobjdumpSection*> mergeMatchingSections(std::string identifier); + std::list<cuobjdumpSection*> mergeSections(); + cuobjdumpELFSection* findELFSection(const std::string identifier); + cuobjdumpPTXSection* findPTXSection(const std::string identifier); + void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context); + kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, + gpgpu_ptx_sim_arg_list_t args, + struct dim3 gridDim, + struct dim3 blockDim, + struct CUctx_st* context ); + int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ); + int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ); + +}; +#endif /* __cuda_api_object_h__ */ diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 44f0f4e..61f8415 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -131,6 +131,8 @@ #if (CUDART_VERSION < 8000) #include "__cudaFatFormat.h" #endif +#include "gpgpu_context.h" +#include "cuda_api_object.h" #include "../src/gpgpu-sim/gpu-sim.h" #include "../src/cuda-sim/ptx_loader.h" #include "../src/cuda-sim/cuda-sim.h" @@ -147,25 +149,10 @@ #include <mach-o/dyld.h> #endif -std::map<void *,void **> pinned_memory; //support for pinned memories added -std::map<void *, size_t> pinned_memory_size; -std::map<unsigned long long, size_t> g_mallocPtr_Size; -int no_of_ptx=0; -//maps sm version number to set of filenames -std::map<unsigned, std::set<std::string> > version_filename; extern void synchronize(); extern void exit_simulation(); -static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ); -static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ); - -static kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, - gpgpu_ptx_sim_arg_list_t args, - struct dim3 gridDim, - struct dim3 blockDim, - struct CUctx_st* context ); - /*DEVICE_BUILTIN*/ struct cudaArray { @@ -189,7 +176,7 @@ struct cudaArray cudaError_t g_last_cudaError = cudaSuccess; -extern stream_manager *g_stream_manager; +//extern stream_manager *g_stream_manager(); void register_ptx_function( const char *name, function_info *impl ) { @@ -210,145 +197,10 @@ void register_ptx_function( const char *name, function_info *impl ) # endif #endif -struct _cuda_device_id { - _cuda_device_id(gpgpu_sim* gpu) {m_id = 0; m_next = NULL; m_gpgpu=gpu;} - struct _cuda_device_id *next() { return m_next; } - unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); } - int num_devices() const { - if( m_next == NULL ) return 1; - else return 1 + m_next->num_devices(); - } - struct _cuda_device_id *get_device( unsigned n ) - { - assert( n < (unsigned)num_devices() ); - struct _cuda_device_id *p=this; - for(unsigned i=0; i<n; i++) - p = p->m_next; - return p; - } - const struct cudaDeviceProp *get_prop() const - { - return m_gpgpu->get_prop(); - } - unsigned get_id() const { return m_id; } - - gpgpu_sim *get_gpgpu() { return m_gpgpu; } -private: - unsigned m_id; - class gpgpu_sim *m_gpgpu; - struct _cuda_device_id *m_next; -}; - -struct CUctx_st { - CUctx_st( _cuda_device_id *gpu ) - { - m_gpu = gpu; - m_binary_info.cmem = 0; - m_binary_info.gmem = 0; - } - - _cuda_device_id *get_device() { return m_gpu; } - - void add_binary( symbol_table *symtab, unsigned fat_cubin_handle ) - { - m_code[fat_cubin_handle] = symtab; - m_last_fat_cubin_handle = fat_cubin_handle; - } - - void add_ptxinfo( const char *deviceFun, const struct gpgpu_ptx_sim_info &info ) - { - symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun); - assert( s != NULL ); - function_info *f = s->get_pc(); - assert( f != NULL ); - f->set_kernel_info(info); - } - - void add_ptxinfo( const struct gpgpu_ptx_sim_info &info ) - { - m_binary_info = info; - } - - void register_function( unsigned fat_cubin_handle, const char *hostFun, const char *deviceFun ) - { - if( m_code.find(fat_cubin_handle) != m_code.end() ) { - symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun); - if(s != NULL) { - function_info *f = s->get_pc(); - assert( f != NULL ); - m_kernel_lookup[hostFun] = f; - } - else { - printf("Warning: cannot find deviceFun %s\n", deviceFun); - m_kernel_lookup[hostFun] = NULL; - } - // assert( s != NULL ); - // function_info *f = s->get_pc(); - // assert( f != NULL ); - // m_kernel_lookup[hostFun] = f; - } else { - m_kernel_lookup[hostFun] = NULL; - } - } - - void register_hostFun_function( const char*hostFun, function_info* f){ - m_kernel_lookup[hostFun] = f; - } - - function_info *get_kernel(const char *hostFun) - { - std::map<const void*,function_info*>::iterator i=m_kernel_lookup.find(hostFun); - assert( i != m_kernel_lookup.end() ); - return i->second; - } - -private: - _cuda_device_id *m_gpu; // selected gpu - std::map<unsigned,symbol_table*> m_code; // fat binary handle => global symbol table - unsigned m_last_fat_cubin_handle; - std::map<const void*,function_info*> m_kernel_lookup; // unique id (CUDA app function address) => kernel entry point - struct gpgpu_ptx_sim_info m_binary_info; - -}; - -class kernel_config { -public: - kernel_config( dim3 GridDim, dim3 BlockDim, size_t sharedMem, struct CUstream_st *stream ) - { - m_GridDim=GridDim; - m_BlockDim=BlockDim; - m_sharedMem=sharedMem; - m_stream = stream; - } - kernel_config() - { - m_GridDim=dim3(-1,-1,-1); - m_BlockDim=dim3(-1,-1,-1); - m_sharedMem=0; - m_stream =NULL; - } - void set_arg( const void *arg, size_t size, size_t offset ) - { - m_args.push_front( gpgpu_ptx_sim_arg(arg,size,offset) ); - } - dim3 grid_dim() const { return m_GridDim; } - dim3 block_dim() const { return m_BlockDim; } - void set_grid_dim(dim3 *d) { m_GridDim = *d; } - void set_block_dim(dim3 *d) { m_BlockDim = *d; } - gpgpu_ptx_sim_arg_list_t get_args() { return m_args; } - struct CUstream_st *get_stream() { return m_stream; } - -private: - dim3 m_GridDim; - dim3 m_BlockDim; - size_t m_sharedMem; - struct CUstream_st *m_stream; - gpgpu_ptx_sim_arg_list_t m_args; -}; - struct _cuda_device_id *GPGPUSim_Init() { - static _cuda_device_id *the_device = NULL; + //static _cuda_device_id *the_device = NULL; + _cuda_device_id *the_device = GPGPUsim_ctx_ptr()->the_cude_device; if( !the_device ) { gpgpu_sim *the_gpu = gpgpu_ptx_sim_init_perf(); @@ -393,7 +245,8 @@ struct _cuda_device_id *GPGPUSim_Init() prop->maxThreadsPerMultiProcessor = the_gpu->threads_per_core(); #endif the_gpu->set_prop(prop); - the_device = new _cuda_device_id(the_gpu); + GPGPUsim_ctx_ptr()->the_cude_device = new _cuda_device_id(the_gpu); + the_device = GPGPUsim_ctx_ptr()->the_cude_device; } start_sim_thread(1); return the_device; @@ -401,14 +254,25 @@ struct _cuda_device_id *GPGPUSim_Init() static CUctx_st* GPGPUSim_Context() { - static CUctx_st *the_context = NULL; + //static CUctx_st *the_context = NULL; + CUctx_st *the_context = GPGPUsim_ctx_ptr()->the_context; if( the_context == NULL ) { _cuda_device_id *the_gpu = GPGPUSim_Init(); - the_context = new CUctx_st(the_gpu); + GPGPUsim_ctx_ptr()->the_context = new CUctx_st(the_gpu); + the_context = GPGPUsim_ctx_ptr()->the_context; } return the_context; } +gpgpu_context* GPGPU_Context() +{ + static gpgpu_context *gpgpu_ctx = NULL; + if( gpgpu_ctx == NULL ) { + gpgpu_ctx = new gpgpu_context(); + } + return gpgpu_ctx; +} + void ptxinfo_addinfo() { if(!get_ptxinfo_kname()){ @@ -480,29 +344,515 @@ typedef std::map<unsigned,CUevent_st*> event_tracker_t; int CUevent_st::m_next_event_uid; event_tracker_t g_timer_events; -int g_active_device = 0; //active gpu that runs the code -std::list<kernel_config> g_cuda_launch_stack; -/******************************************************************************* - * * - * * - * * - *******************************************************************************/ +extern int cuobjdump_lex_init(yyscan_t* scanner); +extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); +extern int cuobjdump_parse(yyscan_t scanner, struct cuobjdump_parser* parser, std::list<cuobjdumpSection*> &cuobjdumpSectionList); +extern int cuobjdump_lex_destroy(yyscan_t scanner); + +enum cuobjdumpSectionType { + PTXSECTION=0, + ELFSECTION +}; + + +// sectiontype: 0 for ptx, 1 for elf +void addCuobjdumpSection(int sectiontype, std::list<cuobjdumpSection*> &cuobjdumpSectionList){ + if (sectiontype) + cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); + else + cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); + printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); +} + +void setCuobjdumparch(const char* arch, std::list<cuobjdumpSection*> &cuobjdumpSectionList){ + unsigned archnum; + sscanf(arch, "sm_%u", &archnum); + assert (archnum && "cannot have sm_0"); + printf("Adding arch: %s\n", arch); + cuobjdumpSectionList.front()->setArch(archnum); +} + +void setCuobjdumpidentifier(const char* identifier, std::list<cuobjdumpSection*> &cuobjdumpSectionList){ + printf("Adding identifier: %s\n", identifier); + cuobjdumpSectionList.front()->setIdentifier(identifier); +} + +void setCuobjdumpptxfilename(const char* filename, std::list<cuobjdumpSection*> &cuobjdumpSectionList){ + printf("Adding ptx filename: %s\n", filename); + cuobjdumpSection* x = cuobjdumpSectionList.front(); + if (dynamic_cast<cuobjdumpPTXSection*>(x) == NULL){ + assert (0 && "You shouldn't be trying to add a ptxfilename to an elf section"); + } + (dynamic_cast<cuobjdumpPTXSection*>(x))->setPTXfilename(filename); +} + +void setCuobjdumpelffilename(const char* filename, std::list<cuobjdumpSection*> &cuobjdumpSectionList){ + if (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()) == NULL){ + assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); + } + (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()))->setELFfilename(filename); +} + +void setCuobjdumpsassfilename(const char* filename, std::list<cuobjdumpSection*> &cuobjdumpSectionList){ + if (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()) == NULL){ + assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); + } + (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()))->setSASSfilename(filename); +} + +//! Return the executable file of the process containing the PTX/SASS code +//! +//! This Function returns the executable file ran by the process. This +//! executable is supposed to contain the PTX/SASS code. It provides workaround +//! for processes running on valgrind by dereferencing /proc/<pid>/exe within the +//! GPGPU-Sim process before calling cuobjdump to extract PTX/SASS. This is +//! needed because valgrind uses x86 emulation to detect memory leak. Other +//! processes (e.g. cuobjdump) reading /proc/<pid>/exe will see the emulator +//! executable instead of the application binary. +//! +std::string get_app_binary(){ + char self_exe_path[1025]; +#ifdef __APPLE__ + uint32_t size = sizeof(self_exe_path); + if( _NSGetExecutablePath(self_exe_path,&size) != 0 ) { + printf("GPGPU-Sim ** ERROR: _NSGetExecutablePath input buffer too small\n"); + exit(1); + } +#else + std::stringstream exec_link; + exec_link << "/proc/self/exe"; + + ssize_t path_length = readlink(exec_link.str().c_str(), self_exe_path, 1024); + assert(path_length != -1); + self_exe_path[path_length] = '\0'; +#endif + + printf("self exe links to: %s\n", self_exe_path); + return self_exe_path; +} + +//above func gives abs path whereas this give just the name of application. +char* get_app_binary_name(std::string abs_path){ + char *self_exe_path; +#ifdef __APPLE__ + //TODO: get apple device and check the result. + printf("WARNING: not tested for Apple-mac devices \n"); + abort(); +#else + char* buf = strdup(abs_path.c_str()); + char *token = strtok(buf, "/"); + while(token !=NULL){ + self_exe_path = token; + token = strtok(NULL,"/"); + } +#endif + self_exe_path = strtok(self_exe_path, "."); + printf("self exe links to: %s\n", self_exe_path); + return self_exe_path; +} + +static int get_app_cuda_version() { + int app_cuda_version = 0; + char fname[1024]; + snprintf(fname,1024,"_app_cuda_version_XXXXXX"); + int fd=mkstemp(fname); + close(fd); + std::string app_cuda_version_command = "ldd " + get_app_binary() + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + fname; + system(app_cuda_version_command.c_str()); + FILE * cmd = fopen(fname, "r"); + char buf[256]; + while (fgets(buf, sizeof(buf), cmd) != 0) { + std::cout << buf; + app_cuda_version = atoi(buf); + } + fclose(cmd); + if ( app_cuda_version == 0 ) { + printf( "Error - Cannot detect the app's CUDA version.\n" ); + exit(1); + } + return app_cuda_version; +} + +//! Keep track of the association between filename and cubin handle +void cuda_runtime_api::cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ + fatbinmap[handle] = filename; +} + -extern "C" { /******************************************************************************* - * * - * * - * * + * Add internal cuda runtime API call to accept gpgpu_context * *******************************************************************************/ -cudaError_t cudaPeekAtLastError(void) +cudaError_t cudaSetDeviceInternal(int device, gpgpu_context* gpgpu_ctx = NULL) { - return g_last_cudaError; + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //set the active device to run cuda + if ( device <= GPGPUSim_Init()->num_devices() ) { + ctx->api->g_active_device = device; + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } } -__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) +cudaError_t cudaGetDeviceInternal(int *device, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + *device = ctx->api->g_active_device; + return g_last_cudaError = cudaSuccess; +} + + +void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } +#if (CUDART_VERSION < 2010) + printf("GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or higher\n"); + exit(1); +#endif + CUctx_st *context = GPGPUSim_Context(); + static unsigned next_fat_bin_handle = 1; + if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) { + // The following workaround has only been verified on 64-bit systems. + if (sizeof(void*) == 4) + printf("GPGPU-Sim PTX: FatBin file name extraction has not been tested on 32-bit system.\n"); + + // This code will get the CUDA version the app was compiled with. + // We need this to determine how to handle the parsing of the binary. + // Making this a runtime variable based on the app, enables GPGPU-Sim compiled + // with a newer version of CUDA to run apps compiled with older versions of + // CUDA. This is especially useful for PTXPLUS execution. + //Skip cuda version check for pytorch application + std::string app_binary_path = get_app_binary(); + int pos = app_binary_path.find("python"); + if (pos==std::string::npos){ + // Not pytorch app : checking cuda version + int app_cuda_version = get_app_cuda_version(); + assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); + } + + //int app_cuda_version = get_app_cuda_version(); + //assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); + const char* filename; +#if CUDART_VERSION < 6000 + // FatBin handle from the .fatbin.c file (one of the intermediate files generated by NVCC) + typedef struct {int m; int v; const unsigned long long* d; char* f;} __fatDeviceText __attribute__ ((aligned (8))); + __fatDeviceText * fatDeviceText = (__fatDeviceText *) fatCubin; + + // Extract the source code file name that generate the given FatBin. + // - Obtains the pointer to the actual fatbin structure from the FatBin handle (fatCubin). + // - An integer inside the fatbin structure contains the relative offset to the source code file name. + // - This offset differs among different CUDA and GCC versions. + char * pfatbin = (char*) fatDeviceText->d; + int offset = *((int*)(pfatbin+48)); + filename = (pfatbin+16+offset); +#else + filename = "default"; +#endif + + // The extracted file name is associated with a fat_cubin_handle passed + // into cudaLaunch(). Inside cudaLaunch(), the associated file name is + // used to find the PTX/SASS section from cuobjdump, which contains the + // PTX/SASS code for the launched kernel function. + // This allows us to work around the fact that cuobjdump only outputs the + // file name associated with each section. + unsigned long long fat_cubin_handle = next_fat_bin_handle; + next_fat_bin_handle++; + printf("GPGPU-Sim PTX: __cudaRegisterFatBinary, fat_cubin_handle = %llu, filename=%s\n", fat_cubin_handle, filename); + /*! + * This function extracts all data from all files in first call + * then for next calls, only returns the appropriate number + */ + assert(fat_cubin_handle >= 1); + if (fat_cubin_handle==1) ctx->api->cuobjdumpInit(); + ctx->api->cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); + + return (void**)fat_cubin_handle; + } +#if (CUDART_VERSION < 8000) + else { + static unsigned source_num=1; + unsigned long long fat_cubin_handle = next_fat_bin_handle++; + __cudaFatCudaBinary *info = (__cudaFatCudaBinary *)fatCubin; + assert( info->version >= 3 ); + unsigned num_ptx_versions=0; + unsigned max_capability=0; + unsigned selected_capability=0; + bool found=false; + unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); + if (!info->ptx){ + printf("ERROR: Cannot find ptx code in cubin file\n" + "\tIf you are using CUDA 4.0 or higher, please enable -gpgpu_ptx_use_cuobjdump or downgrade to CUDA 3.1\n"); + exit(1); + } + while( info->ptx[num_ptx_versions].gpuProfileName != NULL ) { + unsigned capability=0; + sscanf(info->ptx[num_ptx_versions].gpuProfileName,"compute_%u",&capability); + printf("GPGPU-Sim PTX: __cudaRegisterFatBinary found PTX versions for '%s', ", info->ident); + printf("capability = %s\n", info->ptx[num_ptx_versions].gpuProfileName ); + if( forced_max_capability ) { + if( capability > max_capability && capability <= forced_max_capability ) { + found = true; + max_capability=capability; + selected_capability = num_ptx_versions; + } + } else { + if( capability > max_capability ) { + found = true; + max_capability=capability; + selected_capability = num_ptx_versions; + } + } + num_ptx_versions++; + } + if( found ) { + printf("GPGPU-Sim PTX: Loading PTX for %s, capability = %s\n", + info->ident, info->ptx[selected_capability].gpuProfileName ); + symbol_table *symtab; + const char *ptx = info->ptx[selected_capability].ptx; + if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { + printf("GPGPU-Sim PTX: ERROR ** PTXPlus is only supported through cuobjdump\n" + "\tEither enable cuobjdump or disable PTXPlus in your configuration file\n"); + exit(1); + } else { + symtab=ctx->gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); + context->add_binary(symtab,fat_cubin_handle); + ctx->gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); + } + source_num++; + ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + } else { + printf("GPGPU-Sim PTX: warning -- did not find an appropriate PTX in cubin\n"); + } + return (void**)fat_cubin_handle; + } +#else + else { + printf("ERROR ** __cudaRegisterFatBinary() needs to be updated\n"); + abort(); + } +#endif +} + +void cudaRegisterFunctionInternal( + void **fatCubinHandle, + const char *hostFun, + char *deviceFun, + const char *deviceName, + int thread_limit, + uint3 *tid, + uint3 *bid, + dim3 *bDim, + dim3 *gDim, + gpgpu_context *gpgpu_ctx = NULL +) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(); + unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle; + printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", + deviceFun, hostFun, fat_cubin_handle); + if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) + ctx->cuobjdumpParseBinary(fat_cubin_handle); + context->register_function( fat_cubin_handle, hostFun, deviceFun ); +} + +void cudaRegisterVarInternal( + void **fatCubinHandle, + char *hostVar, //pointer to...something + char *deviceAddress, //name of variable + const char *deviceName, //name of variable (same as above) + int ext, + int size, + int constant, + int global, + gpgpu_context *gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); + printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); + if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) + ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); + fflush(stdout); + if ( constant && !global && !ext ) { + gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); + } else if ( !constant && !global && !ext ) { + gpgpu_ptx_sim_register_global_variable(hostVar,deviceName,size); + } else cuda_not_implemented(__my_func__,__LINE__); +} + +cudaError_t cudaConfigureCallInternal(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + struct CUstream_st *s = (struct CUstream_st *)stream; + ctx->api->g_cuda_launch_stack.push_back( kernel_config(gridDim,blockDim,sharedMem,s) ); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaSetupArgumentInternal(const void *arg, size_t size, size_t offset, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + gpgpusim_ptx_assert( !ctx->api->g_cuda_launch_stack.empty(), "empty launch stack" ); + kernel_config &config = ctx->api->g_cuda_launch_stack.back(); + config.set_arg(arg,size,offset); + printf("GPGPU-Sim PTX: Setting up arguments for %zu bytes starting at 0x%llx..\n",size, (unsigned long long) arg); + + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = NULL ) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st* context = GPGPUSim_Context(); + char *mode = getenv("PTX_SIM_MODE_FUNC"); + if( mode ) + sscanf(mode,"%u", &g_ptx_sim_mode); + gpgpusim_ptx_assert( !ctx->api->g_cuda_launch_stack.empty(), "empty launch stack" ); + kernel_config config = ctx->api->g_cuda_launch_stack.back(); + { + dim3 gridDim = config.grid_dim(); + dim3 blockDim = config.block_dim(); + if (gridDim.x * gridDim.y * gridDim.z == 0 || blockDim.x * blockDim.y * blockDim.z == 0) + { + //can't launch + printf("can't launch a empty kernel\n"); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaErrorInvalidConfiguration; + } + } + struct CUstream_st *stream = config.get_stream(); + printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, + g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); + kernel_info_t *grid = ctx->api->gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); + //do dynamic PDOM analysis for performance simulation scenario + std::string kname = grid->name(); + function_info *kernel_func_info = grid->entry(); + if (kernel_func_info->is_pdom_set()) { + printf("GPGPU-Sim PTX: PDOM analysis already done for %s \n", kname.c_str() ); + } else { + printf("GPGPU-Sim PTX: finding reconvergence points for \'%s\'...\n", kname.c_str() ); + kernel_func_info->do_pdom(); + kernel_func_info->set_pdom(); + } + dim3 gridDim = config.grid_dim(); + dim3 blockDim = config.block_dim(); + + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + checkpoint *g_checkpoint; + g_checkpoint = new checkpoint(); + class memory_space *global_mem; + global_mem = gpu->get_global_memory(); + + if(gpu->resume_option ==1 && (grid->get_uid()==gpu->resume_kernel)) + { + + char f1name[2048]; + snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); + + g_checkpoint->load_global_mem(global_mem, f1name); + for (int i=0;i<gpu->resume_CTA;i++) + grid->increment_cta_id(); + } + if(gpu->resume_option==1 && (grid->get_uid()<gpu->resume_kernel)) + { + char f1name[2048]; + snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); + + g_checkpoint->load_global_mem(global_mem, f1name); + printf("Skipping kernel %d as resuming from kernel %d\n",grid->get_uid(),gpu->resume_kernel ); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; + + } + if(gpu->checkpoint_option==1 && (grid->get_uid()>gpu->checkpoint_kernel)) + { + printf("Skipping kernel %d as checkpoint from kernel %d\n",grid->get_uid(),gpu->checkpoint_kernel ); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; + + } + printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", + kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); + stream_operation op(grid,g_ptx_sim_mode,stream); + g_stream_manager()->push(op); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaMallocInternal(void **devPtr, size_t size, gpgpu_context* gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } if(g_debug_execution >= 3){ announce_call(__my_func__); } @@ -510,7 +860,7 @@ __host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); if(g_debug_execution >= 3){ printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr); - g_mallocPtr_Size[(unsigned long long)*devPtr] = size; + ctx->api->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; } if ( *devPtr ) { return g_last_cudaError = cudaSuccess; @@ -519,20 +869,221 @@ __host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) } } -__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) +cudaError_t cudaMallocHostInternal(void **ptr, size_t size, gpgpu_context* gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } if(g_debug_execution >= 3){ announce_call(__my_func__); } - GPGPUSim_Context(); *ptr = malloc(size); if ( *ptr ) { //track pinned memory size allocated in the host so that same amount of memory is also allocated in GPU. - pinned_memory_size[*ptr]=size; + ctx->api->pinned_memory_size[*ptr]=size; + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //only cpu memory allocation happens in cudaHostAlloc. Linking with device pointer to pinned memory happens here. + //TODO: once kernel is executed, the contents in global pointer of GPU must be copied back to CPU host pointer! + flags=0; + CUctx_st* context = GPGPUSim_Context(); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + std::map<void *, size_t>::const_iterator i = ctx->api->pinned_memory_size.find(pHost); + assert(i != ctx->api->pinned_memory_size.end()); + size_t size = i->second; + *pDevice = gpu->gpu_malloc(size); + if(g_debug_execution >= 3){ + printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *pDevice); + ctx->api->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; + } + if ( *pDevice ) { + ctx->api->pinned_memory[pHost]=pDevice; + //Copy contents in cpu to gpu + gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } +#ifdef OPENGL_SUPPORT + GLint buffer_size=0; + CUctx_st* context = GPGPUSim_Context(); + + glbmap_entry_t *p = ctx->api->g_glbmap; + while ( p && p->m_bufferObj != bufferObj ) + p = p->m_next; + if ( p == NULL ) { + glBindBuffer(GL_ARRAY_BUFFER,bufferObj); + glGetBufferParameteriv(GL_ARRAY_BUFFER,GL_BUFFER_SIZE,&buffer_size); + assert( buffer_size != 0 ); + *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(buffer_size); + + // create entry and insert to front of list + glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t)); + n->m_next = ctx->api->g_glbmap; + ctx->api->g_glbmap = n; + + // initialize entry + n->m_bufferObj = bufferObj; + n->m_devPtr = *devPtr; + n->m_size = buffer_size; + + p = n; + } else { + buffer_size = p->m_size; + *devPtr = p->m_devPtr; + } + + if ( *devPtr ) { + char *data = (char *) calloc(p->m_size,1); + glGetBufferSubData(GL_ARRAY_BUFFER,0,buffer_size,data); + memcpy_to_gpu( (size_t) *devPtr, data, buffer_size ); + free(data); + printf("GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", (size_t)buffer_size, + (unsigned long long) *devPtr); return g_last_cudaError = cudaSuccess; } else { return g_last_cudaError = cudaErrorMemoryAllocation; } + + return g_last_cudaError = cudaSuccess; +#else + fflush(stdout); + fflush(stderr); + printf("GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- exiting\n"); + fflush(stdout); + exit(50); +#endif +} + +#if CUDART_VERSION >= 6050 +CUresult +cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, + unsigned int numOptions, CUjit_option *options, void **optionValues, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + static bool addedFile = false; + if (addedFile){ + printf("GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple files\n"); + abort(); + } + + //blocking + assert(type==CU_JIT_INPUT_PTX); + CUctx_st *context = GPGPUSim_Context(); + char *file = getenv("PTX_JIT_PATH"); + if(file==NULL){ + printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n"); + abort(); + } + strcat(file,"/"); + strcat(file,path); + symbol_table *symtab = ctx->gpgpu_ptx_sim_load_ptx_from_filename( file ); + std::string fname(path); + ctx->api->name_symtab[fname] = symtab; + context->add_binary(symtab, 1); + ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + addedFile = true; + return CUDA_SUCCESS; +} +#endif + +#if (CUDART_VERSION >= 2010) + +cudaError_t cudaHostAllocInternal(void **pHost, size_t bytes, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + *pHost = malloc(bytes); + //need to track the size allocated so that cudaHostGetDevicePointer() can function properly. + //TODO: vary this function behavior based on flags value (following nvidia documentation) + ctx->api->pinned_memory_size[*pHost]=bytes; + if( *pHost ) + return g_last_cudaError = cudaSuccess; + else + return g_last_cudaError = cudaErrorMemoryAllocation; +} + +#endif + +/******************************************************************************* + * * + * * + * * + *******************************************************************************/ + +extern "C" { + +/******************************************************************************* + * * + * * + * * + *******************************************************************************/ +cudaError_t cudaPeekAtLastError(void) +{ + return g_last_cudaError; +} + +__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) +{ + return cudaMallocInternal(devPtr, size); +} + +__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) +{ + return cudaMallocHostInternal(ptr, size); } __host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, size_t width, size_t height) { @@ -617,21 +1168,21 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t cou if(g_debug_execution >= 3) printf("GPGPU-Sim PTX: cudaMemcpy(): devPtr = %p\n", dst); if( kind == cudaMemcpyHostToDevice ) - g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); + g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) ); else if( kind == cudaMemcpyDeviceToHost ) - g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); + g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) ); else if( kind == cudaMemcpyDeviceToDevice ) - g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); + g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) ); else if ( kind == cudaMemcpyDefault ) { if ((size_t)src >= GLOBAL_HEAP_START) { if ((size_t)dst >= GLOBAL_HEAP_START) - g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device + g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device else - g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); // device to host + g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) ); // device to host } else { if ((size_t)dst >= GLOBAL_HEAP_START) - g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); + g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) ); else { printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported transfer: host to host\n"); abort(); @@ -773,7 +1324,7 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void assert(kind == cudaMemcpyHostToDevice); printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); //stream_operation( const char *symbol, const void *src, size_t count, size_t offset ) - g_stream_manager->push( stream_operation(src,symbol,count,offset,0) ); + g_stream_manager()->push( stream_operation(src,symbol,count,offset,0) ); //gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); return g_last_cudaError = cudaSuccess; } @@ -787,7 +1338,7 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbo //CUctx_st *context = GPGPUSim_Context(); assert(kind == cudaMemcpyDeviceToHost); printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); - g_stream_manager->push( stream_operation(symbol,dst,count,offset,0) ); + g_stream_manager()->push( stream_operation(symbol,dst,count,offset,0) ); //gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); return g_last_cudaError = cudaSuccess; } @@ -816,9 +1367,9 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_ } struct CUstream_st *s = (struct CUstream_st *)stream; switch( kind ) { - case cudaMemcpyHostToDevice: g_stream_manager->push( stream_operation(src,(size_t)dst,count,s) ); break; - case cudaMemcpyDeviceToHost: g_stream_manager->push( stream_operation((size_t)src,dst,count,s) ); break; - case cudaMemcpyDeviceToDevice: g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; + case cudaMemcpyHostToDevice: g_stream_manager()->push( stream_operation(src,(size_t)dst,count,s) ); break; + case cudaMemcpyDeviceToHost: g_stream_manager()->push( stream_operation((size_t)src,dst,count,s) ); break; + case cudaMemcpyDeviceToDevice: g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; default: abort(); } @@ -1154,6 +1705,12 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDevic break; case 88: case 89: + case 90: + case 91: + case 92: + case 93: + case 94: + case 95: *value= 0; break; default: @@ -1179,25 +1736,12 @@ __host__ cudaError_t CUDARTAPI cudaChooseDevice(int *device, const struct cudaDe __host__ cudaError_t CUDARTAPI cudaSetDevice(int device) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //set the active device to run cuda - if ( device <= GPGPUSim_Init()->num_devices() ) { - g_active_device = device; - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } + return cudaSetDeviceInternal(device); } __host__ cudaError_t CUDARTAPI cudaGetDevice(int *device) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *device = g_active_device; - return g_last_cudaError = cudaSuccess; + return cudaGetDeviceInternal(device); } __host__ cudaError_t CUDARTAPI cudaDeviceGetLimit ( size_t* pValue, cudaLimit limit ) @@ -1438,128 +1982,35 @@ __host__ const char* CUDARTAPI cudaGetErrorString(cudaError_t error) return strdup(buf); } -__host__ cudaError_t CUDARTAPI cudaConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - struct CUstream_st *s = (struct CUstream_st *)stream; - g_cuda_launch_stack.push_back( kernel_config(gridDim,blockDim,sharedMem,s) ); - return g_last_cudaError = cudaSuccess; -} - __host__ cudaError_t CUDARTAPI cudaSetupArgument(const void *arg, size_t size, size_t offset) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - gpgpusim_ptx_assert( !g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config &config = g_cuda_launch_stack.back(); - config.set_arg(arg,size,offset); - printf("GPGPU-Sim PTX: Setting up arguments for %zu bytes starting at 0x%llx..\n",size, (unsigned long long) arg); - - return g_last_cudaError = cudaSuccess; + return cudaSetupArgumentInternal(arg, size, offset); } __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st* context = GPGPUSim_Context(); - char *mode = getenv("PTX_SIM_MODE_FUNC"); - if( mode ) - sscanf(mode,"%u", &g_ptx_sim_mode); - gpgpusim_ptx_assert( !g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config config = g_cuda_launch_stack.back(); - { - dim3 gridDim = config.grid_dim(); - dim3 blockDim = config.block_dim(); - if (gridDim.x * gridDim.y * gridDim.z == 0 || blockDim.x * blockDim.y * blockDim.z == 0) - { - //can't launch - printf("can't launch a empty kernel\n"); - g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaErrorInvalidConfiguration; - } - } - struct CUstream_st *stream = config.get_stream(); - printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, - g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); - kernel_info_t *grid = gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); - //do dynamic PDOM analysis for performance simulation scenario - std::string kname = grid->name(); - function_info *kernel_func_info = grid->entry(); - if (kernel_func_info->is_pdom_set()) { - printf("GPGPU-Sim PTX: PDOM analysis already done for %s \n", kname.c_str() ); - } else { - printf("GPGPU-Sim PTX: finding reconvergence points for \'%s\'...\n", kname.c_str() ); - kernel_func_info->do_pdom(); - kernel_func_info->set_pdom(); - } - dim3 gridDim = config.grid_dim(); - dim3 blockDim = config.block_dim(); - - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - checkpoint *g_checkpoint; - g_checkpoint = new checkpoint(); - class memory_space *global_mem; - global_mem = gpu->get_global_memory(); - - if(gpu->resume_option ==1 && (grid->get_uid()==gpu->resume_kernel)) - { - - char f1name[2048]; - snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); - - g_checkpoint->load_global_mem(global_mem, f1name); - for (int i=0;i<gpu->resume_CTA;i++) - grid->increment_cta_id(); - } - if(gpu->resume_option==1 && (grid->get_uid()<gpu->resume_kernel)) - { - char f1name[2048]; - snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); - - g_checkpoint->load_global_mem(global_mem, f1name); - printf("Skipping kernel %d as resuming from kernel %d\n",grid->get_uid(),gpu->resume_kernel ); - g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; - - } - if(gpu->checkpoint_option==1 && (grid->get_uid()>gpu->checkpoint_kernel)) - { - printf("Skipping kernel %d as checkpoint from kernel %d\n",grid->get_uid(),gpu->checkpoint_kernel ); - g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; - - } - printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", - kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); - stream_operation op(grid,g_ptx_sim_mode,stream); - g_stream_manager->push(op); - g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; + return cudaLaunchInternal( hostFun ); } __host__ cudaError_t CUDARTAPI cudaLaunchKernel ( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream ) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - function_info *entry = context->get_kernel(hostFun); - - cudaConfigureCall(gridDim, blockDim, sharedMem, stream); - for(unsigned i = 0; i < entry->num_args(); i++){ - std::pair<size_t, unsigned> p = entry->get_param_config(i); - cudaSetupArgument(args[i], p.first, p.second); - } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(); + function_info *entry = context->get_kernel(hostFun); +#if CUDART_VERSION < 10000 + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); +#endif + for(unsigned i = 0; i < entry->num_args(); i++){ + std::pair<size_t, unsigned> p = entry->get_param_config(i); + cudaSetupArgumentInternal(args[i], p.first, p.second); + } - cudaLaunch(hostFun); - return g_last_cudaError = cudaSuccess; + cudaLaunchInternal(hostFun); + return g_last_cudaError = cudaSuccess; } @@ -1577,7 +2028,7 @@ __host__ cudaError_t CUDARTAPI cudaStreamCreate(cudaStream_t *stream) printf("GPGPU-Sim PTX: cudaStreamCreate\n"); #if (CUDART_VERSION >= 3000) *stream = new struct CUstream_st(); - g_stream_manager->add_stream(*stream); + g_stream_manager()->add_stream(*stream); #else *stream = 0; printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); @@ -1616,7 +2067,7 @@ __host__ cudaError_t CUDARTAPI cudaStreamDestroy(cudaStream_t stream) //per-stream synchronization required for application using external libraries without explicit synchronization in the code to //avoid the stream_manager from spinning forever to destroy non-empty streams without making any forward progress. stream->synchronize(); - g_stream_manager->destroy_stream(stream); + g_stream_manager()->destroy_stream(stream); #endif return g_last_cudaError = cudaSuccess; } @@ -1696,7 +2147,7 @@ __host__ cudaError_t CUDARTAPI cudaEventRecord(cudaEvent_t event, cudaStream_t s if( !e ) return g_last_cudaError = cudaErrorUnknown; struct CUstream_st *s = (struct CUstream_st *)stream; stream_operation op(e,s); - g_stream_manager->push(op); + g_stream_manager()->push(op); return g_last_cudaError = cudaSuccess; } @@ -1712,11 +2163,11 @@ __host__ cudaError_t CUDARTAPI cudaStreamWaitEvent(cudaStream_t stream, cudaEven return g_last_cudaError = cudaSuccess; } if (!stream){ - g_stream_manager->pushCudaStreamWaitEventToAllStreams(e, flags); + g_stream_manager()->pushCudaStreamWaitEventToAllStreams(e, flags); } else { struct CUstream_st *s = (struct CUstream_st *)stream; stream_operation op(s,e,flags); - g_stream_manager->push(op); + g_stream_manager()->push(op); } return g_last_cudaError = cudaSuccess; } @@ -1867,177 +2318,8 @@ __host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, co //#include "../../cuobjdump_to_ptxplus/cuobjdump_parser.h" -enum cuobjdumpSectionType { - PTXSECTION=0, - ELFSECTION -}; - - -class cuobjdumpSection { -public: - //Constructor - cuobjdumpSection() { - arch = 0; - identifier = ""; - } - virtual ~cuobjdumpSection() {} - unsigned getArch() {return arch;} - void setArch(unsigned a) {arch = a;} - std::string getIdentifier() {return identifier;} - void setIdentifier(std::string i) {identifier = i;} - virtual void print(){std::cout << "cuobjdump Section: unknown type" << std::endl;} -private: - unsigned arch; - std::string identifier; -}; - -class cuobjdumpELFSection : public cuobjdumpSection -{ -public: - cuobjdumpELFSection() {} - virtual ~cuobjdumpELFSection() { - elffilename = ""; - sassfilename = ""; - } - std::string getELFfilename() {return elffilename;} - void setELFfilename(std::string f) {elffilename = f;} - std::string getSASSfilename() {return sassfilename;} - void setSASSfilename(std::string f) {sassfilename = f;} - virtual void print() { - std::cout << "ELF Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "elf filename: " << getELFfilename() << std::endl; - std::cout << "sass filename: " << getSASSfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string elffilename; - std::string sassfilename; -}; - -class cuobjdumpPTXSection : public cuobjdumpSection -{ -public: - cuobjdumpPTXSection(){ - ptxfilename = ""; - } - std::string getPTXfilename() {return ptxfilename;} - void setPTXfilename(std::string f) {ptxfilename = f;} - virtual void print() { - std::cout << "PTX Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "ptx filename: " << getPTXfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string ptxfilename; -}; - -std::list<cuobjdumpSection*> cuobjdumpSectionList; -std::list<cuobjdumpSection*> libSectionList; - -// sectiontype: 0 for ptx, 1 for elf -void addCuobjdumpSection(int sectiontype){ - if (sectiontype) - cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); - else - cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); - printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); -} - -void setCuobjdumparch(const char* arch){ - unsigned archnum; - sscanf(arch, "sm_%u", &archnum); - assert (archnum && "cannot have sm_0"); - printf("Adding arch: %s\n", arch); - cuobjdumpSectionList.front()->setArch(archnum); -} - -void setCuobjdumpidentifier(const char* identifier){ - printf("Adding identifier: %s\n", identifier); - cuobjdumpSectionList.front()->setIdentifier(identifier); -} - -void setCuobjdumpptxfilename(const char* filename){ - printf("Adding ptx filename: %s\n", filename); - cuobjdumpSection* x = cuobjdumpSectionList.front(); - if (dynamic_cast<cuobjdumpPTXSection*>(x) == NULL){ - assert (0 && "You shouldn't be trying to add a ptxfilename to an elf section"); - } - (dynamic_cast<cuobjdumpPTXSection*>(x))->setPTXfilename(filename); -} - -void setCuobjdumpelffilename(const char* filename){ - if (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); - } - (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()))->setELFfilename(filename); -} - -void setCuobjdumpsassfilename(const char* filename){ - if (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); - } - (dynamic_cast<cuobjdumpELFSection*>(cuobjdumpSectionList.front()))->setSASSfilename(filename); -} -extern int cuobjdump_parse(); -extern FILE *cuobjdump_in; - -//! Return the executable file of the process containing the PTX/SASS code -//! -//! This Function returns the executable file ran by the process. This -//! executable is supposed to contain the PTX/SASS code. It provides workaround -//! for processes running on valgrind by dereferencing /proc/<pid>/exe within the -//! GPGPU-Sim process before calling cuobjdump to extract PTX/SASS. This is -//! needed because valgrind uses x86 emulation to detect memory leak. Other -//! processes (e.g. cuobjdump) reading /proc/<pid>/exe will see the emulator -//! executable instead of the application binary. -//! -std::string get_app_binary(){ - char self_exe_path[1025]; -#ifdef __APPLE__ - uint32_t size = sizeof(self_exe_path); - if( _NSGetExecutablePath(self_exe_path,&size) != 0 ) { - printf("GPGPU-Sim ** ERROR: _NSGetExecutablePath input buffer too small\n"); - exit(1); - } -#else - std::stringstream exec_link; - exec_link << "/proc/self/exe"; - - ssize_t path_length = readlink(exec_link.str().c_str(), self_exe_path, 1024); - assert(path_length != -1); - self_exe_path[path_length] = '\0'; -#endif - - printf("self exe links to: %s\n", self_exe_path); - return self_exe_path; -} - -//above func gives abs path whereas this give just the name of application. -char* get_app_binary_name(std::string abs_path){ - char *self_exe_path; -#ifdef __APPLE__ - //TODO: get apple device and check the result. - printf("WARNING: not tested for Apple-mac devices \n"); - abort(); -#else - char* buf = strdup(abs_path.c_str()); - char *token = strtok(buf, "/"); - while(token !=NULL){ - self_exe_path = token; - token = strtok(NULL,"/"); - } -#endif - self_exe_path = strtok(self_exe_path, "."); - printf("self exe links to: %s\n", self_exe_path); - return self_exe_path; -} - //extracts all ptx files from binary and dumps into prog_name.unique_no.sm_<>.ptx files -void extract_ptx_files_using_cuobjdump(){ +void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ extern bool g_cdp_enabled; char command[1000]; char *pytorch_bin = getenv("PYTORCH_BIN"); @@ -2074,11 +2356,11 @@ void extract_ptx_files_using_cuobjdump(){ printf("ERROR: command: %s failed \n",command); exit(0); } - no_of_ptx++; + context->no_of_ptx++; } } - if(!no_of_ptx){ + if(!context->no_of_ptx){ printf("WARNING: Number of ptx in the executable file are 0. One of the reasons might be\n"); printf("\t1. CDP is enabled\n"); printf("\t2. When using PyTorch, PYTORCH_BIN is not set correctly\n"); @@ -2106,28 +2388,6 @@ void extract_ptx_files_using_cuobjdump(){ } -static int get_app_cuda_version() { - int app_cuda_version = 0; - char fname[1024]; - snprintf(fname,1024,"_app_cuda_version_XXXXXX"); - int fd=mkstemp(fname); - close(fd); - std::string app_cuda_version_command = "ldd " + get_app_binary() + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + fname; - system(app_cuda_version_command.c_str()); - FILE * cmd = fopen(fname, "r"); - char buf[256]; - while (fgets(buf, sizeof(buf), cmd) != 0) { - std::cout << buf; - app_cuda_version = atoi(buf); - } - fclose(cmd); - if ( app_cuda_version == 0 ) { - printf( "Error - Cannot detect the app's CUDA version.\n" ); - exit(1); - } - return app_cuda_version; -} - //! Call cuobjdump to extract everything (-elf -sass -ptx) /*! @@ -2137,7 +2397,7 @@ static int get_app_cuda_version() { * It is also responsible for extracting the libraries linked to the binary if the option is * enabled * */ -void extract_code_using_cuobjdump(){ +void cuda_runtime_api::extract_code_using_cuobjdump(){ CUctx_st *context = GPGPUSim_Context(); unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); @@ -2159,7 +2419,7 @@ void extract_code_using_cuobjdump(){ //dump ptx for all individial ptx files into sepearte files which is later used by ptxas. int result=0; #if (CUDART_VERSION >= 6000) - extract_ptx_files_using_cuobjdump(); + extract_ptx_files_using_cuobjdump(context); return; #endif //TODO: redundant to dump twice. how can it be prevented? @@ -2192,9 +2452,16 @@ void extract_code_using_cuobjdump(){ if (parse_output) { printf("Parsing file %s\n", fname); + FILE *cuobjdump_in; cuobjdump_in = fopen(fname, "r"); - cuobjdump_parse(); + struct cuobjdump_parser parser; + parser.elfserial = 1; + parser.ptxserial = 1; + cuobjdump_lex_init(&(parser.scanner)); + cuobjdump_set_in(cuobjdump_in, (parser.scanner)); + cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); + cuobjdump_lex_destroy(parser.scanner); fclose(cuobjdump_in); printf("Done parsing!!!\n"); } else { @@ -2242,8 +2509,15 @@ void extract_code_using_cuobjdump(){ std::cout << "Done" << std::endl; std::cout << "Trying to parse " << libcodfn.str() << std::endl; + FILE *cuobjdump_in; cuobjdump_in = fopen(libcodfn.str().c_str(), "r"); - cuobjdump_parse(); + struct cuobjdump_parser parser; + parser.elfserial = 1; + parser.ptxserial = 1; + cuobjdump_lex_init(&(parser.scanner)); + cuobjdump_set_in(cuobjdump_in, (parser.scanner)); + cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); + cuobjdump_lex_destroy(parser.scanner); fclose(cuobjdump_in); std::getline(libsf, line); } @@ -2293,7 +2567,7 @@ void printSectionList(std::list<cuobjdumpSection*> sl) { } //! Remove unecessary sm versions from the section list -std::list<cuobjdumpSection*> pruneSectionList(std::list<cuobjdumpSection*> cuobjdumpSectionList, CUctx_st *context) { +std::list<cuobjdumpSection*> cuda_runtime_api::pruneSectionList(CUctx_st *context) { unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); //For ptxplus, force the max capability to 19 if it's higher or unspecified(0) @@ -2346,7 +2620,7 @@ std::list<cuobjdumpSection*> pruneSectionList(std::list<cuobjdumpSection*> cuobj } //! Merge all PTX sections that have a specific identifier into one file -std::list<cuobjdumpSection*> mergeMatchingSections(std::list<cuobjdumpSection*> cuobjdumpSectionList, std::string identifier){ +std::list<cuobjdumpSection*> cuda_runtime_api::mergeMatchingSections(std::string identifier){ const char *ptxcode = ""; std::list<cuobjdumpSection*>::iterator old_iter; cuobjdumpPTXSection* old_ptxsection = NULL; @@ -2389,7 +2663,7 @@ std::list<cuobjdumpSection*> mergeMatchingSections(std::list<cuobjdumpSection*> } //! Merge any PTX sections with matching identifiers -std::list<cuobjdumpSection*> mergeSections(std::list<cuobjdumpSection*> cuobjdumpSectionList){ +std::list<cuobjdumpSection*> cuda_runtime_api::mergeSections(){ std::vector<std::string> identifier; cuobjdumpPTXSection* ptxsection; @@ -2411,7 +2685,7 @@ std::list<cuobjdumpSection*> mergeSections(std::list<cuobjdumpSection*> cuobjdum for ( std::vector<std::string>::iterator iter = identifier.begin(); iter != identifier.end(); iter++) { - cuobjdumpSectionList = mergeMatchingSections(cuobjdumpSectionList, *iter); + cuobjdumpSectionList = mergeMatchingSections(*iter); } return cuobjdumpSectionList; @@ -2436,7 +2710,7 @@ cuobjdumpELFSection* findELFSectionInList(std::list<cuobjdumpSection*> sectionli } //! Find an ELF section in all the known lists -cuobjdumpELFSection* findELFSection(const std::string identifier){ +cuobjdumpELFSection* cuda_runtime_api::findELFSection(const std::string identifier){ cuobjdumpELFSection* sec = findELFSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findELFSectionInList(libSectionList, identifier); @@ -2447,7 +2721,7 @@ cuobjdumpELFSection* findELFSection(const std::string identifier){ } //! Within the section list, find the PTX section corresponding to a given identifier -cuobjdumpPTXSection* findPTXSectionInList(std::list<cuobjdumpSection*> sectionlist, const std::string identifier){ +cuobjdumpPTXSection* findPTXSectionInList(std::list<cuobjdumpSection*> §ionlist, const std::string identifier){ std::list<cuobjdumpSection*>::iterator iter; for ( iter = sectionlist.begin(); iter != sectionlist.end(); @@ -2471,7 +2745,7 @@ cuobjdumpPTXSection* findPTXSectionInList(std::list<cuobjdumpSection*> sectionli } //! Find an PTX section in all the known lists -cuobjdumpPTXSection* findPTXSection(const std::string identifier){ +cuobjdumpPTXSection* cuda_runtime_api::findPTXSection(const std::string identifier){ cuobjdumpPTXSection* sec = findPTXSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findPTXSectionInList(libSectionList, identifier); @@ -2484,35 +2758,27 @@ cuobjdumpPTXSection* findPTXSection(const std::string identifier){ //! Extract the code using cuobjdump and remove unnecessary sections -void cuobjdumpInit(){ +void cuda_runtime_api::cuobjdumpInit(){ CUctx_st *context = GPGPUSim_Context(); extract_code_using_cuobjdump(); //extract all the output of cuobjdump to _cuobjdump_*.* const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if (pre_load ==NULL || strlen(pre_load)==0){ - cuobjdumpSectionList = pruneSectionList(cuobjdumpSectionList, context); - cuobjdumpSectionList = mergeSections(cuobjdumpSectionList); + cuobjdumpSectionList = pruneSectionList(context); + cuobjdumpSectionList = mergeSections(); } } -std::map<int, std::string> fatbinmap; -std::map<int, bool>fatbin_registered; -std::map<std::string, symbol_table*> name_symtab; - -//! Keep track of the association between filename and cubin handle -void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename){ - fatbinmap[handle] = filename; -} //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it -void cuobjdumpParseBinary(unsigned int handle){ +void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ - if(fatbin_registered[handle]) return; - fatbin_registered[handle] = true; CUctx_st *context = GPGPUSim_Context(); - std::string fname = fatbinmap[handle]; + if(api->fatbin_registered[handle]) return; + api->fatbin_registered[handle] = true; + std::string fname = api->fatbinmap[handle]; - if (name_symtab.find(fname) != name_symtab.end()) { - symbol_table *symtab = name_symtab[fname]; + if (api->name_symtab.find(fname) != api->name_symtab.end()) { + symbol_table *symtab = api->name_symtab[fname]; context->add_binary(symtab, handle); return; } @@ -2521,7 +2787,7 @@ void cuobjdumpParseBinary(unsigned int handle){ #if (CUDART_VERSION >= 6000) //loops through all ptx files from smallest sm version to largest std::map<unsigned,std::set<std::string> >::iterator itr_m; - for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ + for (itr_m = api->version_filename.begin(); itr_m!=api->version_filename.end(); itr_m++){ std::set<std::string>::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2529,11 +2795,11 @@ void cuobjdumpParseBinary(unsigned int handle){ symtab = gpgpu_ptx_sim_load_ptx_from_filename( ptx_filename.c_str() ); } } - name_symtab[fname] = symtab; + api->name_symtab[fname] = symtab; context->add_binary(symtab, handle); - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ + api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + for (itr_m = api->version_filename.begin(); itr_m!=api->version_filename.end(); itr_m++){ std::set<std::string>::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2545,8 +2811,8 @@ void cuobjdumpParseBinary(unsigned int handle){ #endif unsigned max_capability = 0; - for ( std::list<cuobjdumpSection*>::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); + for ( std::list<cuobjdumpSection*>::iterator iter = api->cuobjdumpSectionList.begin(); + iter != api->cuobjdumpSectionList.end(); iter++){ unsigned capability = (*iter)->getArch(); if (capability > max_capability) max_capability = capability; @@ -2557,7 +2823,7 @@ void cuobjdumpParseBinary(unsigned int handle){ cuobjdumpPTXSection* ptx = NULL; const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if(pre_load==NULL || strlen(pre_load)==0) - ptx = findPTXSection(fname); + ptx = api->findPTXSection(fname); char *ptxcode; const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or strlen(getenv("PTX_SIM_USE_PTX_FILE"))==0) { @@ -2567,7 +2833,7 @@ void cuobjdumpParseBinary(unsigned int handle){ ptxcode = readfile(override_ptx_name); } if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier()); + cuobjdumpELFSection* elfsection = api->findELFSection(ptx->getIdentifier()); assert (elfsection!= NULL); char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( ptx->getPTXfilename(), @@ -2576,178 +2842,62 @@ void cuobjdumpParseBinary(unsigned int handle){ symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxplus_str, handle); printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), handle); context->add_binary(symtab, handle); - gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability ); + gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability, context->no_of_ptx ); delete[] ptxplus_str; } else { symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxcode, handle); //if CUOBJDUMP_SIM_FILE is not set, ptx is NULL. So comment below. //printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), handle); context->add_binary(symtab, handle); - gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability ); + gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability, context->no_of_ptx ); } - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - name_symtab[fname] = symtab; + api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + api->name_symtab[fname] = symtab; //TODO: Remove temporarily files as per configurations } +} + +extern "C" { void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) { if(g_debug_execution >= 3){ announce_call(__my_func__); } -#if (CUDART_VERSION < 2010) - printf("GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or higher\n"); - exit(1); -#endif - CUctx_st *context = GPGPUSim_Context(); - static unsigned next_fat_bin_handle = 1; - if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) { - // The following workaround has only been verified on 64-bit systems. - if (sizeof(void*) == 4) - printf("GPGPU-Sim PTX: FatBin file name extraction has not been tested on 32-bit system.\n"); - - // This code will get the CUDA version the app was compiled with. - // We need this to determine how to handle the parsing of the binary. - // Making this a runtime variable based on the app, enables GPGPU-Sim compiled - // with a newer version of CUDA to run apps compiled with older versions of - // CUDA. This is especially useful for PTXPLUS execution. - //Skip cuda version check for pytorch application - std::string app_binary_path = get_app_binary(); - int pos = app_binary_path.find("python"); - if (pos==std::string::npos){ - // Not pytorch app : checking cuda version - int app_cuda_version = get_app_cuda_version(); - assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); - } - - //int app_cuda_version = get_app_cuda_version(); - //assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); - const char* filename; -#if CUDART_VERSION < 6000 - // FatBin handle from the .fatbin.c file (one of the intermediate files generated by NVCC) - typedef struct {int m; int v; const unsigned long long* d; char* f;} __fatDeviceText __attribute__ ((aligned (8))); - __fatDeviceText * fatDeviceText = (__fatDeviceText *) fatCubin; - - // Extract the source code file name that generate the given FatBin. - // - Obtains the pointer to the actual fatbin structure from the FatBin handle (fatCubin). - // - An integer inside the fatbin structure contains the relative offset to the source code file name. - // - This offset differs among different CUDA and GCC versions. - char * pfatbin = (char*) fatDeviceText->d; - int offset = *((int*)(pfatbin+48)); - filename = (pfatbin+16+offset); -#else - filename = "default"; -#endif - - // The extracted file name is associated with a fat_cubin_handle passed - // into cudaLaunch(). Inside cudaLaunch(), the associated file name is - // used to find the PTX/SASS section from cuobjdump, which contains the - // PTX/SASS code for the launched kernel function. - // This allows us to work around the fact that cuobjdump only outputs the - // file name associated with each section. - unsigned long long fat_cubin_handle = next_fat_bin_handle; - next_fat_bin_handle++; - printf("GPGPU-Sim PTX: __cudaRegisterFatBinary, fat_cubin_handle = %llu, filename=%s\n", fat_cubin_handle, filename); - /*! - * This function extracts all data from all files in first call - * then for next calls, only returns the appropriate number - */ - assert(fat_cubin_handle >= 1); - if (fat_cubin_handle==1) cuobjdumpInit(); - cuobjdumpRegisterFatBinary(fat_cubin_handle, filename); - - return (void**)fat_cubin_handle; - } -#if (CUDART_VERSION < 8000) - else { - static unsigned source_num=1; - unsigned long long fat_cubin_handle = next_fat_bin_handle++; - __cudaFatCudaBinary *info = (__cudaFatCudaBinary *)fatCubin; - assert( info->version >= 3 ); - unsigned num_ptx_versions=0; - unsigned max_capability=0; - unsigned selected_capability=0; - bool found=false; - unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); - if (!info->ptx){ - printf("ERROR: Cannot find ptx code in cubin file\n" - "\tIf you are using CUDA 4.0 or higher, please enable -gpgpu_ptx_use_cuobjdump or downgrade to CUDA 3.1\n"); - exit(1); - } - while( info->ptx[num_ptx_versions].gpuProfileName != NULL ) { - unsigned capability=0; - sscanf(info->ptx[num_ptx_versions].gpuProfileName,"compute_%u",&capability); - printf("GPGPU-Sim PTX: __cudaRegisterFatBinary found PTX versions for '%s', ", info->ident); - printf("capability = %s\n", info->ptx[num_ptx_versions].gpuProfileName ); - if( forced_max_capability ) { - if( capability > max_capability && capability <= forced_max_capability ) { - found = true; - max_capability=capability; - selected_capability = num_ptx_versions; - } - } else { - if( capability > max_capability ) { - found = true; - max_capability=capability; - selected_capability = num_ptx_versions; - } - } - num_ptx_versions++; - } - if( found ) { - printf("GPGPU-Sim PTX: Loading PTX for %s, capability = %s\n", - info->ident, info->ptx[selected_capability].gpuProfileName ); - symbol_table *symtab; - const char *ptx = info->ptx[selected_capability].ptx; - if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - printf("GPGPU-Sim PTX: ERROR ** PTXPlus is only supported through cuobjdump\n" - "\tEither enable cuobjdump or disable PTXPlus in your configuration file\n"); - exit(1); - } else { - symtab=gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); - context->add_binary(symtab,fat_cubin_handle); - gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability ); - } - source_num++; - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - } else { - printf("GPGPU-Sim PTX: warning -- did not find an appropriate PTX in cubin\n"); - } - return (void**)fat_cubin_handle; - } -#else - else { - printf("ERROR ** __cudaRegisterFatBinary() needs to be updated\n"); - abort(); - } -#endif + return cudaRegisterFatBinaryInternal(fatCubin); } -void __cudaUnregisterFatBinary(void **fatCubinHandle) +void CUDARTAPI __cudaRegisterFatBinaryEnd( void **fatCubinHandle ) { if(g_debug_execution >= 3){ announce_call(__my_func__); } - ; } -cudaError_t cudaDeviceReset ( void ) { - // Should reset the simulated GPU +unsigned CUDARTAPI __cudaPushCallConfiguration(dim3 gridDim, + dim3 blockDim, + size_t sharedMem = 0, + struct CUstream_st *stream = 0) +{ if(g_debug_execution >= 3){ announce_call(__my_func__); } - return g_last_cudaError = cudaSuccess; + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); } -cudaError_t CUDARTAPI cudaDeviceSynchronize(void){ + +cudaError_t CUDARTAPI __cudaPopCallConfiguration( + dim3 *gridDim, + dim3 *blockDim, + size_t *sharedMem, + void *stream +) +{ if(g_debug_execution >= 3){ announce_call(__my_func__); } - //Blocks until the device has completed all preceding requested tasks - synchronize(); - return g_last_cudaError = cudaSuccess; + return g_last_cudaError = cudaSuccess; } void CUDARTAPI __cudaRegisterFunction( @@ -2760,18 +2910,19 @@ void CUDARTAPI __cudaRegisterFunction( uint3 *bid, dim3 *bDim, dim3 *gDim -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle; - printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", - deviceFun, hostFun, fat_cubin_handle); - if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary(fat_cubin_handle); - context->register_function( fat_cubin_handle, hostFun, deviceFun ); +) { + cudaRegisterFunctionInternal( + fatCubinHandle, + hostFun, + deviceFun, + deviceName, + thread_limit, + tid, + bid, + bDim, + gDim + ); + } extern void __cudaRegisterVar( @@ -2784,19 +2935,43 @@ extern void __cudaRegisterVar( int constant, int global ) { + cudaRegisterVarInternal( + fatCubinHandle, + hostVar, + deviceAddress, + deviceName, + ext, + size, + constant, + global ); +} + +__host__ cudaError_t CUDARTAPI cudaConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream) +{ + return cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); +} + +void __cudaUnregisterFatBinary(void **fatCubinHandle) +{ if(g_debug_execution >= 3){ announce_call(__my_func__); } - printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); - printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); - if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); - fflush(stdout); - if ( constant && !global && !ext ) { - gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); - } else if ( !constant && !global && !ext ) { - gpgpu_ptx_sim_register_global_variable(hostVar,deviceName,size); - } else cuda_not_implemented(__my_func__,__LINE__); +} + +cudaError_t cudaDeviceReset ( void ) { + // Should reset the simulated GPU + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + return g_last_cudaError = cudaSuccess; +} +cudaError_t CUDARTAPI cudaDeviceSynchronize(void){ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //Blocks until the device has completed all preceding requested tasks + synchronize(); + return g_last_cudaError = cudaSuccess; } @@ -2868,10 +3043,6 @@ char __cudaInitModule( } -#ifndef OPENGL_SUPPORT -typedef unsigned long GLuint; -#endif - cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj) { if(g_debug_execution >= 3){ @@ -2881,70 +3052,9 @@ cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj) return g_last_cudaError = cudaSuccess; } -struct glbmap_entry { - GLuint m_bufferObj; - void *m_devPtr; - size_t m_size; - struct glbmap_entry *m_next; -}; -typedef struct glbmap_entry glbmap_entry_t; - -glbmap_entry_t* g_glbmap = NULL; - cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -#ifdef OPENGL_SUPPORT - GLint buffer_size=0; - CUctx_st* ctx = GPGPUSim_Context(); - - glbmap_entry_t *p = g_glbmap; - while ( p && p->m_bufferObj != bufferObj ) - p = p->m_next; - if ( p == NULL ) { - glBindBuffer(GL_ARRAY_BUFFER,bufferObj); - glGetBufferParameteriv(GL_ARRAY_BUFFER,GL_BUFFER_SIZE,&buffer_size); - assert( buffer_size != 0 ); - *devPtr = ctx->get_device()->get_gpgpu()->gpu_malloc(buffer_size); - - // create entry and insert to front of list - glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t)); - n->m_next = g_glbmap; - g_glbmap = n; - - // initialize entry - n->m_bufferObj = bufferObj; - n->m_devPtr = *devPtr; - n->m_size = buffer_size; - - p = n; - } else { - buffer_size = p->m_size; - *devPtr = p->m_devPtr; - } - - if ( *devPtr ) { - char *data = (char *) calloc(p->m_size,1); - glGetBufferSubData(GL_ARRAY_BUFFER,0,buffer_size,data); - memcpy_to_gpu( (size_t) *devPtr, data, buffer_size ); - free(data); - printf("GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", (size_t)buffer_size, - (unsigned long long) *devPtr); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } - - return g_last_cudaError = cudaSuccess; -#else - fflush(stdout); - fflush(stderr); - printf("GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- exiting\n"); - fflush(stdout); - exit(50); -#endif + return cudaGLMapBufferObjectInternal(devPtr, bufferObj); } cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) @@ -2953,7 +3063,8 @@ cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) announce_call(__my_func__); } #ifdef OPENGL_SUPPORT - glbmap_entry_t *p = g_glbmap; + CUctx_st* ctx = GPGPUSim_Context(); + glbmap_entry_t *p = ctx->api->g_glbmap; while ( p && p->m_bufferObj != bufferObj ) p = p->m_next; if ( p == NULL ) @@ -2987,45 +3098,12 @@ cudaError_t cudaGLUnregisterBufferObject(GLuint bufferObj) cudaError_t CUDARTAPI cudaHostAlloc(void **pHost, size_t bytes, unsigned int flags) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *pHost = malloc(bytes); - //need to track the size allocated so that cudaHostGetDevicePointer() can function properly. - //TODO: vary this function behavior based on flags value (following nvidia documentation) - pinned_memory_size[*pHost]=bytes; - if( *pHost ) - return g_last_cudaError = cudaSuccess; - else - return g_last_cudaError = cudaErrorMemoryAllocation; + return cudaHostAllocInternal(pHost, bytes, flags); } cudaError_t CUDARTAPI cudaHostGetDevicePointer(void **pDevice, void *pHost, unsigned int flags) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //only cpu memory allocation happens in cudaHostAlloc. Linking with device pointer to pinned memory happens here. - //TODO: once kernel is executed, the contents in global pointer of GPU must be copied back to CPU host pointer! - flags=0; - CUctx_st* context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - std::map<void *, size_t>::const_iterator i = pinned_memory_size.find(pHost); - assert(i != pinned_memory_size.end()); - size_t size = i->second; - *pDevice = gpu->gpu_malloc(size); - if(g_debug_execution >= 3){ - printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *pDevice); - g_mallocPtr_Size[(unsigned long long)*pDevice] = size; - } - if ( *pDevice ) { - pinned_memory[pHost]=pDevice; - //Copy contents in cpu to gpu - gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } + return cudaHostGetDevicePointerInternal(pDevice, pHost, flags); } __host__ cudaError_t CUDARTAPI cudaPointerGetAttributes( @@ -3185,6 +3263,52 @@ __host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, size_t v #endif + +#if CUDART_VERSION >= 9000 +/** + * \brief Set attributes for a given function + * + * This function sets the attributes of a function specified via \p entry. + * The parameter \p entry must be a pointer to a function that executes + * on the device. The parameter specified by \p entry must be declared as a \p __global__ + * function. The enumeration defined by \p attr is set to the value defined by \p value + * If the specified function does not exist, then ::cudaErrorInvalidDeviceFunction is returned. + * If the specified attribute cannot be written, or if the value is incorrect, + * then ::cudaErrorInvalidValue is returned. + * + * Valid values for \p attr are: + * ::cuFuncAttrMaxDynamicSharedMem - Maximum size of dynamic shared memory per block + * ::cudaFuncAttributePreferredSharedMemoryCarveout - Preferred shared memory-L1 cache split ratio + * + * \param entry - Function to get attributes of + * \param attr - Attribute to set + * \param value - Value to set + * + * \return + * ::cudaSuccess, + * ::cudaErrorInitializationError, + * ::cudaErrorInvalidDeviceFunction, + * ::cudaErrorInvalidValue + * \notefnerr + * + * \ref ::cudaLaunchKernel(const T *func, dim3 gridDim, dim3 blockDim, void **args, size_t sharedMem, cudaStream_t stream) "cudaLaunchKernel (C++ API)", + * \ref ::cudaFuncSetCacheConfig(T*, enum cudaFuncCache) "cudaFuncSetCacheConfig (C++ API)", + * \ref ::cudaFuncGetAttributes(struct cudaFuncAttributes*, const void*) "cudaFuncGetAttributes (C API)", + * ::cudaSetDoubleForDevice, + * ::cudaSetDoubleForHost, + * \ref ::cudaSetupArgument(T, size_t) "cudaSetupArgument (C++ API)" + */ +cudaError_t CUDARTAPI cudaFuncSetAttribute(const void *func, enum cudaFuncAttribute attr, int value) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s ( func=%p, attr=%d, value=%d )\"\n", + __my_func__, func, attr, value ); + return g_last_cudaError = cudaSuccess; +} +#endif + cudaError_t CUDARTAPI cudaGLSetGLDevice(int device) { if(g_debug_execution >= 3){ @@ -3254,17 +3378,9 @@ int CUDARTAPI __cudaSynchronizeThreads(void**, void*) //////// -extern int ptx_parse(); -extern int ptx__scan_string(const char*); -extern FILE *ptx_in; - -extern int ptxinfo_parse(); -extern int ptxinfo_debug; -extern FILE *ptxinfo_in; - /// static functions -static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ) +int cuda_runtime_api::load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ) { if(g_debug_execution >= 3){ announce_call(__my_func__); @@ -3303,7 +3419,7 @@ static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsign return ng_bytes; } -static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ) +int cuda_runtime_api::load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ) { if(g_debug_execution >= 3){ announce_call(__my_func__); @@ -3349,7 +3465,7 @@ static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu return nc_bytes; } -kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, +kernel_info_t * cuda_runtime_api::gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, gpgpu_ptx_sim_arg_list_t args, struct dim3 gridDim, struct dim3 blockDim, @@ -3845,33 +3961,8 @@ CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, unsigned int numOptions, CUjit_option *options, void **optionValues) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - static bool addedFile = false; - if (addedFile){ - printf("GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple files\n"); - abort(); - } - - //blocking - assert(type==CU_JIT_INPUT_PTX); - CUctx_st *context = GPGPUSim_Context(); - char *file = getenv("PTX_JIT_PATH"); - if(file==NULL){ - printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n"); - abort(); - } - strcat(file,"/"); - strcat(file,path); - symbol_table *symtab = gpgpu_ptx_sim_load_ptx_from_filename( file ); - std::string fname(path); - name_symtab[fname] = symtab; - context->add_binary(symtab, 1); - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - addedFile = true; - return CUDA_SUCCESS; + return cuLinkAddFileInternal(state, type, path, + numOptions, options, optionValues); } #endif @@ -4838,12 +4929,12 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, const char *hostFun = (const char*) f; CUctx_st *context = GPGPUSim_Context(); function_info *entry = context->get_kernel(hostFun); - cudaConfigureCall(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream); + cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream); for(unsigned i = 0; i < entry->num_args(); i++){ std::pair<size_t, unsigned> p = entry->get_param_config(i); cudaSetupArgument(kernelParams[i], p.first, p.second); } - cudaLaunch(hostFun); + cudaLaunchInternal(hostFun); return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4000 */ diff --git a/libcuda/cuobjdump.h b/libcuda/cuobjdump.h new file mode 100644 index 0000000..6ab6778 --- /dev/null +++ b/libcuda/cuobjdump.h @@ -0,0 +1,80 @@ +#ifndef __cuobjdump_h__ +#define __cuobjdump_h__ +#include <string> +#include <list> +#include <iostream> + +typedef void * yyscan_t; +struct cuobjdump_parser { + yyscan_t scanner; + int elfserial; + int ptxserial; + FILE *ptxfile; + FILE *elffile; + FILE *sassfile; + char filename [1024]; +}; + +class cuobjdumpSection { +public: + //Constructor + cuobjdumpSection() { + arch = 0; + identifier = ""; + } + virtual ~cuobjdumpSection() {} + unsigned getArch() {return arch;} + void setArch(unsigned a) {arch = a;} + std::string getIdentifier() {return identifier;} + void setIdentifier(std::string i) {identifier = i;} + virtual void print(){std::cout << "cuobjdump Section: unknown type" << std::endl;} +private: + unsigned arch; + std::string identifier; +}; + +class cuobjdumpELFSection : public cuobjdumpSection +{ +public: + cuobjdumpELFSection() {} + virtual ~cuobjdumpELFSection() { + elffilename = ""; + sassfilename = ""; + } + std::string getELFfilename() {return elffilename;} + void setELFfilename(std::string f) {elffilename = f;} + std::string getSASSfilename() {return sassfilename;} + void setSASSfilename(std::string f) {sassfilename = f;} + virtual void print() { + std::cout << "ELF Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "elf filename: " << getELFfilename() << std::endl; + std::cout << "sass filename: " << getSASSfilename() << std::endl; + std::cout << std::endl; + } +private: + std::string elffilename; + std::string sassfilename; +}; + +class cuobjdumpPTXSection : public cuobjdumpSection +{ +public: + cuobjdumpPTXSection(){ + ptxfilename = ""; + } + std::string getPTXfilename() {return ptxfilename;} + void setPTXfilename(std::string f) {ptxfilename = f;} + virtual void print() { + std::cout << "PTX Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "ptx filename: " << getPTXfilename() << std::endl; + std::cout << std::endl; + } +private: + std::string ptxfilename; +}; + +#endif /* __cuobjdump_h__ */ diff --git a/libcuda/cuobjdump.l b/libcuda/cuobjdump.l index 2b0dac8..5a19d65 100644 --- a/libcuda/cuobjdump.l +++ b/libcuda/cuobjdump.l @@ -30,6 +30,7 @@ %{ #include <stdio.h> #include <string.h> +#include "cuobjdump.h" #include "cuobjdump_parser.h" #define YY_NEVER_INTERACTIVE 1 @@ -38,9 +39,9 @@ #define YYDEBUG 1 -#define yylval cuobjdump_lval - -void cuobjdump_error(const char*); +void cuobjdump_error(yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list<cuobjdumpSection*> &cuobjdumpSectionList, const char* msg); +#define YY_DECL int cuobjdump_lex \ + (YYSTYPE * yylval_param , yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list<cuobjdumpSection*> &cuobjdumpSectionList) %} %option stack @@ -48,6 +49,8 @@ void cuobjdump_error(const char*); %option yylineno %option nounput +%option bison-bridge +%option reentrant %s ptxcode %s sasscode @@ -69,54 +72,54 @@ newlines {newline}+ "ptxasOptions"{notnewline}*{newline} -[1-9]{numeric}* yylval.string_value = strdup(yytext); return DECIMAL; +[1-9]{numeric}* yylval->string_value = strdup(yytext); return DECIMAL; "has debug info"{newline} "Fatbin ptx code:"{newline} { - yy_push_state(ptxcode); - yy_push_state(identifier); - yy_push_state(ptxheader); - yylval.string_value = strdup(yytext); + yy_push_state(ptxcode, yyscanner); + yy_push_state(identifier, yyscanner); + yy_push_state(ptxheader, yyscanner); + yylval->string_value = strdup(yytext); return PTXHEADER; } "Fatbin elf code:"{newline} { - yy_push_state(elfcode); - yy_push_state(identifier); - yy_push_state(elfheader); - yylval.string_value = strdup(yytext); + yy_push_state(elfcode, yyscanner); + yy_push_state(identifier, yyscanner); + yy_push_state(elfheader, yyscanner); + yylval->string_value = strdup(yytext); return ELFHEADER; } /*PTX code tokens*/ -<ptxcode>{notnewline}*{newline} yylval.string_value = strdup(yytext); return PTXLINE; +<ptxcode>{notnewline}*{newline} yylval->string_value = strdup(yytext); return PTXLINE; /*ELF code tokens*/ <elfcode>{whitespace}*"code for sm_"{numeric}+{newline} { BEGIN(sasscode); - yylval.string_value = strdup(yytext); + yylval->string_value = strdup(yytext); return SASSLINE; } -<elfcode>{notnewline}*{newline} yylval.string_value = strdup(yytext); return ELFLINE; +<elfcode>{notnewline}*{newline} yylval->string_value = strdup(yytext); return ELFLINE; /*SASS code tokens*/ -<sasscode>{notnewline}*{newline} yylval.string_value = strdup(yytext); return SASSLINE; +<sasscode>{notnewline}*{newline} yylval->string_value = strdup(yytext); return SASSLINE; <identifier>{newline}"compressed"{newline} BEGIN(conidentifier); return H_COMPRESSED; <identifier>{newline}"identifier = " BEGIN(endidentifier); return H_IDENTIFIER; -<identifier>{newline}{newline} yy_pop_state(); +<identifier>{newline}{newline} yy_pop_state(yyscanner); <conidentifier>"identifier = " BEGIN(endidentifier); return H_IDENTIFIER; -<conidentifier>{newline} yy_pop_state(); +<conidentifier>{newline} yy_pop_state(yyscanner); -<endidentifier>{notnewline}+ yylval.string_value = strdup(yytext); yy_pop_state(); return FILENAME; +<endidentifier>{notnewline}+ yylval->string_value = strdup(yytext); yy_pop_state(yyscanner); return FILENAME; /*Header tokens*/ -<elfheader>[[:alnum:]_]+ yylval.string_value = strdup(yytext); return IDENTIFIER; +<elfheader>[[:alnum:]_]+ yylval->string_value = strdup(yytext); return IDENTIFIER; <elfheader>"================" return H_SEPARATOR; <elfheader>"arch = " return H_ARCH; <elfheader>"code version = " return H_CODEVERSION; @@ -128,7 +131,7 @@ newlines {newline}+ /*Header tokens*/ -<ptxheader>[[:alnum:]_]+ yylval.string_value = strdup(yytext); return IDENTIFIER; +<ptxheader>[[:alnum:]_]+ yylval->string_value = strdup(yytext); return IDENTIFIER; <ptxheader>"================" return H_SEPARATOR; <ptxheader>"arch = " return H_ARCH; <ptxheader>"code version = " return H_CODEVERSION; @@ -143,7 +146,7 @@ newlines {newline}+ /* Looking for the identifier (filename) then the header is done */ -<endheader>{notnewline}+ yylval.string_value = strdup(yytext); yy_pop_state(); return IDENTIFIER; +<endheader>{notnewline}+ yylval->string_value = strdup(yytext); yy_pop_state(yyscanner); return IDENTIFIER; @@ -153,11 +156,12 @@ newlines {newline}+ <<EOF>> BEGIN(INITIAL);return 0; /*No other rule matched. Throw an error*/ -. cuobjdump_error("Invalid token"); +. cuobjdump_error(yyscanner, parser, cuobjdumpSectionList, "Invalid token"); %% -void cuobjdump_error(const char* message) +void cuobjdump_error(yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list<cuobjdumpSection*> &cuobjdumpSectionList, const char* msg) { - printf(" %s near \"%s\"",message, yytext); - printf(" on line %i\n",yylineno); + struct yyguts_t * yyg = (struct yyguts_t*)yyscanner; + printf(" %s near \"%s\"",msg, yytext); + printf(" on line %i\n",yylineno); } diff --git a/libcuda/cuobjdump.y b/libcuda/cuobjdump.y index 31760f7..8d1bca6 100644 --- a/libcuda/cuobjdump.y +++ b/libcuda/cuobjdump.y @@ -29,24 +29,31 @@ %{ #include <stdio.h> -int yylex(void); -void yyerror(const char*); -extern void addCuobjdumpSection(int sectiontype); -void setCuobjdumparch(const char* arch); -void setCuobjdumpidentifier(const char* identifier); -void setCuobjdumpptxfilename(const char* filename); -void setCuobjdumpelffilename(const char* filename); -void setCuobjdumpsassfilename(const char* filename); -int elfserial = 1; -int ptxserial = 1; -FILE *ptxfile; -FILE *elffile; -FILE *sassfile; -char filename [1024]; +typedef void * yyscan_t; +#include "cuobjdump.h" + +extern void addCuobjdumpSection(int sectiontype, std::list<cuobjdumpSection*> &cuobjdumpSectionList); +void setCuobjdumparch(const char* arch, std::list<cuobjdumpSection*> &cuobjdumpSectionList); +void setCuobjdumpidentifier(const char* identifier, std::list<cuobjdumpSection*> &cuobjdumpSectionList); +void setCuobjdumpptxfilename(const char* filename, std::list<cuobjdumpSection*> &cuobjdumpSectionList); +void setCuobjdumpelffilename(const char* filename, std::list<cuobjdumpSection*> &cuobjdumpSectionList); +void setCuobjdumpsassfilename(const char* filename, std::list<cuobjdumpSection*> &cuobjdumpSectionList); %} +%define api.pure full +%parse-param {yyscan_t scanner} +%parse-param {struct cuobjdump_parser* parser} +%parse-param {std::list<cuobjdumpSection*> &cuobjdumpSectionList} +%lex-param {yyscan_t scanner} +%lex-param {struct cuobjdump_parser* parser} +%lex-param {std::list<cuobjdumpSection*> &cuobjdumpSectionList} + %union { char* string_value; } +%{ +int yylex(YYSTYPE * yylval_param, yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list<cuobjdumpSection*> &cuobjdumpSectionList); +void yyerror(yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list<cuobjdumpSection*> &cuobjdumpSectionList, const char* msg); +%} %token <string_value> H_SEPARATOR H_ARCH H_CODEVERSION H_PRODUCER H_HOST H_COMPILESIZE H_IDENTIFIER H_UNKNOWN H_COMPRESSED %token <string_value> CODEVERSION %token <string_value> STRING @@ -70,25 +77,25 @@ emptylines : emptylines NEWLINE | ; section : PTXHEADER { - addCuobjdumpSection(0); - snprintf(filename, 1024, "_cuobjdump_%d.ptx", ptxserial++); - ptxfile = fopen(filename, "w"); - setCuobjdumpptxfilename(filename); + addCuobjdumpSection(0, cuobjdumpSectionList); + snprintf(parser->filename, 1024, "_cuobjdump_%d.ptx", parser->ptxserial++); + parser->ptxfile = fopen(parser->filename, "w"); + setCuobjdumpptxfilename(parser->filename, cuobjdumpSectionList); } headerinfo compressedkeyword identifier ptxcode { - fclose(ptxfile); + fclose(parser->ptxfile); } | ELFHEADER { - addCuobjdumpSection(1); - snprintf(filename, 1024, "_cuobjdump_%d.elf", elfserial); - elffile = fopen(filename, "w"); - setCuobjdumpelffilename(filename); + addCuobjdumpSection(1, cuobjdumpSectionList); + snprintf(parser->filename, 1024, "_cuobjdump_%d.elf", parser->elfserial); + parser->elffile = fopen(parser->filename, "w"); + setCuobjdumpelffilename(parser->filename, cuobjdumpSectionList); } headerinfo compressedkeyword identifier elfcode { - fclose(elffile); - snprintf(filename, 1024, "_cuobjdump_%d.sass", elfserial++); - sassfile = fopen(filename, "w"); - setCuobjdumpsassfilename(filename); + fclose(parser->elffile); + snprintf(parser->filename, 1024, "_cuobjdump_%d.sass", parser->elfserial++); + parser->sassfile = fopen(parser->filename, "w"); + setCuobjdumpsassfilename(parser->filename, cuobjdumpSectionList); } sasscode { - fclose(sassfile); + fclose(parser->sassfile); }; headerinfo : H_SEPARATOR NEWLINE @@ -96,27 +103,27 @@ headerinfo : H_SEPARATOR NEWLINE H_CODEVERSION CODEVERSION NEWLINE H_PRODUCER H_UNKNOWN NEWLINE H_HOST IDENTIFIER NEWLINE - H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4);}; + H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4, cuobjdumpSectionList);}; | H_SEPARATOR NEWLINE H_ARCH IDENTIFIER NEWLINE H_CODEVERSION CODEVERSION NEWLINE H_PRODUCER IDENTIFIER NEWLINE H_HOST IDENTIFIER NEWLINE - H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4);}; + H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4, cuobjdumpSectionList);}; -identifier : H_IDENTIFIER FILENAME emptylines {setCuobjdumpidentifier($2);} - | {setCuobjdumpidentifier("default");}; +identifier : H_IDENTIFIER FILENAME emptylines {setCuobjdumpidentifier($2, cuobjdumpSectionList);} + | {setCuobjdumpidentifier("default", cuobjdumpSectionList);}; compressedkeyword : H_COMPRESSED emptylines | ; -ptxcode : ptxcode PTXLINE {fprintf(ptxfile, "%s", $2);} +ptxcode : ptxcode PTXLINE {fprintf(parser->ptxfile, "%s", $2);} | ; -elfcode : elfcode ELFLINE {fprintf(elffile, "%s", $2);} +elfcode : elfcode ELFLINE {fprintf(parser->elffile, "%s", $2);} | ; -sasscode : sasscode SASSLINE {fprintf(sassfile, "%s", $2);} +sasscode : sasscode SASSLINE {fprintf(parser->sassfile, "%s", $2);} | ; diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h new file mode 100644 index 0000000..fa5d02b --- /dev/null +++ b/libcuda/gpgpu_context.h @@ -0,0 +1,32 @@ +#ifndef __gpgpu_context_h__ +#define __gpgpu_context_h__ +#include "cuda_api_object.h" +#include "../src/cuda-sim/ptx_loader.h" +#include "../src/cuda-sim/ptx_parser.h" + +class gpgpu_context { + public: + gpgpu_context() { + g_global_allfiles_symbol_table = NULL; + api = new cuda_runtime_api(); + ptxinfo = new ptxinfo_data(); + ptx_parser = new ptx_recognizer(); + } + // global list + symbol_table *g_global_allfiles_symbol_table; + // objects pointers for each file + cuda_runtime_api* api; + ptxinfo_data* ptxinfo; + ptx_recognizer* ptx_parser; + // member function list + void cuobjdumpParseBinary(unsigned int handle); + class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); + class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); + void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version); + void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20, int no_of_ptx=0 ); + void print_ptx_file( const char *p, unsigned source_num, const char *filename ); + class symbol_table* init_parser(const char*); +}; +gpgpu_context* GPGPU_Context(); + +#endif /* __gpgpu_context_h__ */ diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc index 97a54d8..50a02fa 100644 --- a/libopencl/opencl_runtime_api.cc +++ b/libopencl/opencl_runtime_api.cc @@ -84,6 +84,7 @@ #include "../src/gpgpusim_entrypoint.h" #include "../src/gpgpu-sim/gpu-sim.h" #include "../src/gpgpu-sim/shader.h" +#include "../libcuda/gpgpu_context.h" //# define __my_func__ __PRETTY_FUNCTION__ # if defined __cplusplus ? __GNUC_PREREQ (2, 6) : __GNUC_PREREQ (2, 4) @@ -424,6 +425,8 @@ void register_ptx_function( const char *name, function_info *impl ) void _cl_program::Build(const char *options) { + gpgpu_context *ctx; + ctx = GPGPU_Context(); printf("GPGPU-Sim OpenCL API: compiling OpenCL kernels...\n"); std::map<cl_uint,pgm_info>::iterator i; for( i = m_pgm.begin(); i!= m_pgm.end(); i++ ) { @@ -576,8 +579,8 @@ void _cl_program::Build(const char *options) } } info.m_asm = tmp; - info.m_symtab = gpgpu_ptx_sim_load_ptx_from_string( tmp, source_num ); - gpgpu_ptxinfo_load_from_string( tmp, source_num ); + info.m_symtab = ctx->gpgpu_ptx_sim_load_ptx_from_string( tmp, source_num ); + ctx->gpgpu_ptxinfo_load_from_string( tmp, source_num ); free(tmp); } printf("GPGPU-Sim OpenCL API: finished compiling OpenCL kernels.\n"); diff --git a/linux-so-version.txt b/linux-so-version.txt index a7c2d3c..45c40dd 100644 --- a/linux-so-version.txt +++ b/linux-so-version.txt @@ -4,5 +4,9 @@ libcudart.so.9.1{ }; libcudart.so.9.2{ }; +libcudart.so.10.0{ +}; +libcudart.so.10.1{ +}; libcuda.so.1{ }; diff --git a/nightly.jenkinsfile b/nightly.jenkinsfile index 5221b3b..51e2d05 100644 --- a/nightly.jenkinsfile +++ b/nightly.jenkinsfile @@ -23,9 +23,8 @@ pipeline { stage('nightly-simulations-build'){ steps{ sh 'rm -rf gpgpu-sim_simulations' - sh 'git clone [email protected]:TimRogersGroup/gpgpu-sim_simulations.git && \ + sh 'git clone [email protected]:purdue-aalp/gpgpu-sim_simulations.git && \ cd gpgpu-sim_simulations && \ - git checkout purdue-cluster && \ git pull && \ ln -s /home/tgrogers-raid/a/common/data_dirs benchmarks/' sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ @@ -38,30 +37,63 @@ pipeline { } stage('nightly-2B-insn-run'){ steps { + sh 'rm -rf ./gpgpu-sim_simulations/util/plotting/correl-html && rm -rf gpgpu-sim-results-repo &&\ + rm -rf ./gpgpu-sim_simulations/util/plotting/htmls' + sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ - source `pwd`/setup_environment &&\ - ./gpgpu-sim_simulations/util/job_launching/run_simulations.py -B `cat ./gpgpu-sim_simulations/util/job_launching/apps/all-apps.list` -C TITANX-2B,TITANX-L1ON-2B,P100-2B,TITANV-2B -N nightly-$$ && \ - PLOTDIR="jenkins/${JOB_NAME}/${BUILD_NUMBER}/getstats" && ssh [email protected] mkdir -p /home/dynamo/a/tgrogers/website/gpgpu-sim-plots/$PLOTDIR && \ - ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -I -S 1800 -v -s stats-$$.csv -N nightly-$$ && \ - ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c stats-$$.csv -p [email protected]:~/website/gpgpu-sim-plots/$PLOTDIR -w https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/$PLOTDIR -n $PLOTDIR' + source `pwd`/setup_environment &&\ + ./gpgpu-sim_simulations/util/job_launching/run_simulations.py \ + -B `cat ./gpgpu-sim_simulations/util/job_launching/apps/all-apps.list` -C TITANV-2B -N nightly-$$ && \ + ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -I -S 1800 -v \ + -s stats-per-app-9.1.csv -T 12 -K -N nightly-$$ && \ + ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c stats-per-app-9.1.csv -P cuda-9.1.nightly' } } stage('nightly-correlate'){ steps { + sh 'git clone [email protected]:purdue-aalp/gpgpu-sim-results-repo.git' + + sh 'if [ ! -d ./gpgpu-sim-results-repo/${JOB_NAME} ]; then \ + mkdir -p ./gpgpu-sim-results-repo/${JOB_NAME}/ ; \ + cp ./gpgpu-sim-results-repo/purdue-aalp/gpgpu-sim_distribution/dev/* \ + ./gpgpu-sim-results-repo/${JOB_NAME}/ ; \ + fi && \ + ./gpgpu-sim_simulations/run_hw/get_hw_data.sh' + sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\ + ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k \ + -B `cat ./gpgpu-sim_simulations/util/job_launching/apps/all-apps.list` \ + -C TITANV-2B > stats-per-kernel-9.1.csv &&\ + ./gpgpu-sim_simulations/util/plotting/merge-stats.py \ + -c ./gpgpu-sim-results-repo/${JOB_NAME}/stats-per-app-9.1.csv,./stats-per-app-9.1.csv -R \ + > per-app-merge-9.1.csv &&\ + ./gpgpu-sim_simulations/util/plotting/merge-stats.py \ + -c ./gpgpu-sim-results-repo/${JOB_NAME}/stats-per-kernel-9.1.csv,./stats-per-kernel-9.1.csv -R \ + > per-kernel-merge-9.1.csv &&\ source `pwd`/setup_environment &&\ - PLOTDIR="jenkins/${JOB_NAME}" &&\ - ./gpgpu-sim_simulations/util/job_launching/get_stats.py -R -K -k -B `cat ./gpgpu-sim_simulations/util/job_launching/apps/all-apps.list` -C TITANX-2B,TITANX-L1ON-2B,P100-2B,TITANV-2B > nightly-stats-per-kernel-9.1.csv &&\ - ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh nightly-stats-per-kernel-9.1.csv $PLOTDIR ${BUILD_NUMBER}' + ./gpgpu-sim_simulations/util/plotting/plot-correlation.py -c per-kernel-merge-9.1.csv \ + -p cuda-9.1.nightly -b ./gpgpu-sim_simulations/util/plotting/known.correlation.outliers.list \ + | grep "Correl=" | tee correl.9.1.txt &&\ + cp stats-per-*.csv ./gpgpu-sim-results-repo/${JOB_NAME}/ &&\ + cd ./gpgpu-sim-results-repo &&\ + git diff --quiet && git diff --staged --quiet || git commit -am "Jenkins automated checkin ${JOB_NAME} Build:${BUILD_NUMBER}" &&\ + git push' + + sh 'PLOTDIR="/home/dynamo/a/tgrogers/website/gpgpu-sim-plots/jenkins/${JOB_NAME}" &&\ + ssh [email protected] mkdir -p $PLOTDIR/${BUILD_NUMBER} && \ + scp ./gpgpu-sim_simulations/util/plotting/correl-html/* [email protected]:$PLOTDIR/${BUILD_NUMBER} &&\ + scp ./gpgpu-sim_simulations/util/plotting/htmls/* [email protected]:$PLOTDIR/${BUILD_NUMBER} &&\ + ssh [email protected] "cd $PLOTDIR && rm -rf latest && cp -r ${BUILD_NUMBER} latest"' } } } post { success { - emailext body: "See ${BUILD_URL}", + emailext body:'''${SCRIPT, template="groovy-html.nightly.success.template"}''', recipientProviders: [[$class: 'CulpritsRecipientProvider'], [$class: 'RequesterRecipientProvider']], subject: "[AALP Jenkins] Build #${BUILD_NUMBER} - Success!", + attachmentsPattern: 'correl.*.txt', to: '[email protected]' } failure { diff --git a/setup_environment b/setup_environment index b420584..58f4713 100644 --- a/setup_environment +++ b/setup_environment @@ -51,7 +51,7 @@ CC_VERSION=`gcc --version | head -1 | awk '{for(i=1;i<=NF;i++){ if(match($i,/^[0 CUDA_VERSION_STRING=`$CUDA_INSTALL_PATH/bin/nvcc --version | awk '/release/ {print $5;}' | sed 's/,//'`; export CUDA_VERSION_NUMBER=`echo $CUDA_VERSION_STRING | sed 's/\./ /' | awk '{printf("%02u%02u", 10*int($1), 10*$2);}'` -if [ $CUDA_VERSION_NUMBER -gt 9100 -o $CUDA_VERSION_NUMBER -lt 2030 ]; then +if [ $CUDA_VERSION_NUMBER -gt 10100 -o $CUDA_VERSION_NUMBER -lt 2030 ]; then echo "ERROR ** GPGPU-Sim version $GPGPUSIM_VERSION_STRING not tested with CUDA version $CUDA_VERSION_STRING (please see README)"; return fi @@ -148,19 +148,15 @@ fi echo ""; echo "----------------------------------------------------------------------------"; echo "INFO - If you only care about PTX execution, ignore this message. GPGPU-Sim supports PTX execution in modern CUDA." -if [ $CUDA_VERSION_NUMBER -gt 4200 ]; then - echo "If you want to run PTXPLUS (sm_1x SASS) with a modern card configuration, the apps and simulator must be compiled with CUDA 4.2." - echo "You can still run a PASCAL configuration when compiling with 4.2 by setting the \$PTXAS_CUDA_INSTALL_PATH directory environment variable." -else - echo "If you want to run PTXPLUS (sm_1x SASS) with a modern card configuration - set the envronment variable" - echo "\$PTXAS_CUDA_INSTALL_PATH to point a CUDA version compabible with your card configurations (i.e. 8+ for PASCAL, 9+ for VOLTA etc..)" - echo "For example: \"export \$PTXAS_CUDA_INSTALL_PATH=/usr/local/cuda-9.1\"" -fi +echo "If you want to run PTXPLUS (sm_1x SASS) with a modern card configuration - set the envronment variable" +echo "\$PTXAS_CUDA_INSTALL_PATH to point a CUDA version compabible with your card configurations (i.e. 8+ for PASCAL, 9+ for VOLTA etc..)" +echo "For example: \"export \$PTXAS_CUDA_INSTALL_PATH=/usr/local/cuda-9.1\"" +echo "" echo "The following text describes why:"; echo "If you are using PTXPLUS, only sm_1x is supported and it requires that the app and simulator binaries are compiled in CUDA 4.2 or less."; echo "The simulator requires it since CUDA headers desribe struct sizes in the exec which change from gen to gen."; echo "The apps require 4.2 because new versions of CUDA tools have dropped parsing support for generating sm_1x"; -echo "When running using modern config (i.e. pascal) and PTXPLUS with CUDA 4.2, the \$PTXAS_CUDA_INSTALL_PATH env variable is required to get proper register usage" +echo "When running using modern config (i.e. volta) and PTXPLUS with CUDA 4.2, the \$PTXAS_CUDA_INSTALL_PATH env variable is required to get proper register usage" echo "(and hence occupancy) using a version of CUDA that knows the register usage on the real card." echo ""; echo "----------------------------------------------------------------------------"; diff --git a/src/Makefile b/src/Makefile index 6001669..3ad511e 100644 --- a/src/Makefile +++ b/src/Makefile @@ -51,6 +51,8 @@ else CXXFLAGS += endif +CXXFLAGS += -I$(CUDA_INSTALL_PATH)/include + OPTFLAGS += -g3 -fPIC CPP = g++ $(SNOW) diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc index cebdb25..7755477 100644 --- a/src/abstract_hardware_model.cc +++ b/src/abstract_hardware_model.cc @@ -34,6 +34,7 @@ #include "cuda-sim/cuda-sim.h" #include "gpgpu-sim/gpu-sim.h" #include "option_parser.h" +#include "gpgpusim_entrypoint.h" #include <algorithm> #include <sys/stat.h> #include <sstream> @@ -198,6 +199,9 @@ gpgpu_t::gpgpu_t( const gpgpu_functional_sim_config &config ) if(m_function_model_config.get_ptx_inst_debug_to_file() != 0) ptx_inst_debug_file = fopen(m_function_model_config.get_ptx_inst_debug_file(), "w"); + + gpu_sim_cycle=0; + gpu_tot_sim_cycle=0; } address_type line_size_based_tag_func(new_addr_type address, new_addr_type line_size) @@ -768,14 +772,14 @@ void kernel_info_t::notify_parent_finished() { extern unsigned long long g_total_param_size; g_total_param_size -= ((m_kernel_entry->get_args_aligned_size() + 255)/256*256); m_parent_kernel->remove_child(this); - g_stream_manager->register_finished_kernel(m_parent_kernel->get_uid()); + g_stream_manager()->register_finished_kernel(m_parent_kernel->get_uid()); } } CUstream_st * kernel_info_t::create_stream_cta(dim3 ctaid) { assert(get_default_stream_cta(ctaid)); CUstream_st * stream = new CUstream_st(); - g_stream_manager->add_stream(stream); + g_stream_manager()->add_stream(stream); assert(m_cta_streams.find(ctaid) != m_cta_streams.end()); assert(m_cta_streams[ctaid].size() >= 1); //must have default stream m_cta_streams[ctaid].push_back(stream); @@ -791,7 +795,7 @@ CUstream_st * kernel_info_t::get_default_stream_cta(dim3 ctaid) { else { m_cta_streams[ctaid] = std::list<CUstream_st *>(); CUstream_st * stream = new CUstream_st(); - g_stream_manager->add_stream(stream); + g_stream_manager()->add_stream(stream); m_cta_streams[ctaid].push_back(stream); return stream; } @@ -823,17 +827,18 @@ void kernel_info_t::destroy_cta_streams() { for(auto s = m_cta_streams.begin(); s != m_cta_streams.end(); s++) { stream_size += s->second.size(); for(auto ss = s->second.begin(); ss != s->second.end(); ss++) - g_stream_manager->destroy_stream(*ss); + g_stream_manager()->destroy_stream(*ss); s->second.clear(); } printf("size %lu\n", stream_size); m_cta_streams.clear(); } -simt_stack::simt_stack( unsigned wid, unsigned warpSize) +simt_stack::simt_stack( unsigned wid, unsigned warpSize, class gpgpu_sim * gpu) { m_warp_id=wid; m_warp_size = warpSize; + m_gpu=gpu; reset(); } @@ -1033,7 +1038,7 @@ void simt_stack::update( simt_mask_t &thread_done, addr_vector_t &next_pc, addre simt_stack_entry new_stack_entry; new_stack_entry.m_pc = tmp_next_pc; new_stack_entry.m_active_mask = tmp_active_mask; - new_stack_entry.m_branch_div_cycle = gpu_sim_cycle+gpu_tot_sim_cycle; + new_stack_entry.m_branch_div_cycle = m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle; new_stack_entry.m_type = STACK_ENTRY_TYPE_CALL; m_stack.push_back(new_stack_entry); return; @@ -1065,7 +1070,7 @@ void simt_stack::update( simt_mask_t &thread_done, addr_vector_t &next_pc, addre new_recvg_pc = recvg_pc; if (new_recvg_pc != top_recvg_pc) { m_stack.back().m_pc = new_recvg_pc; - m_stack.back().m_branch_div_cycle = gpu_sim_cycle+gpu_tot_sim_cycle; + m_stack.back().m_branch_div_cycle = m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle; m_stack.push_back(simt_stack_entry()); } @@ -1157,7 +1162,7 @@ void core_t::initilizeSIMTStack(unsigned warp_count, unsigned warp_size) { m_simt_stack = new simt_stack*[warp_count]; for (unsigned i = 0; i < warp_count; ++i) - m_simt_stack[i] = new simt_stack(i,warp_size); + m_simt_stack[i] = new simt_stack(i,warp_size,m_gpu); m_warp_size = warp_size; m_warp_count = warp_count; } diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index 22ef509..68cb693 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -173,9 +173,7 @@ enum _memory_op_t { #include <algorithm> #if !defined(__VECTOR_TYPES_H__) -struct dim3 { - unsigned int x, y, z; -}; +#include "vector_types.h" #endif struct dim3comp { bool operator() (const dim3 & a, const dim3 & b) const @@ -197,7 +195,7 @@ void increment_x_then_y_then_z( dim3 &i, const dim3 &bound); #include "stream_manager.h" class stream_manager; struct CUstream_st; -extern stream_manager * g_stream_manager; +//extern stream_manager * g_stream_manager; //support for pinned memories added extern std::map<void *,void **> pinned_memory; extern std::map<void *, size_t> pinned_memory_size; @@ -399,7 +397,7 @@ typedef std::vector<address_type> addr_vector_t; class simt_stack { public: - simt_stack( unsigned wid, unsigned warpSize); + simt_stack( unsigned wid, unsigned warpSize, class gpgpu_sim * gpu); void reset(); void launch( address_type start_pc, const simt_mask_t &active_mask ); @@ -416,6 +414,7 @@ protected: unsigned m_warp_id; unsigned m_warp_size; + enum stack_entry_type { STACK_ENTRY_TYPE_NORMAL = 0, STACK_ENTRY_TYPE_CALL @@ -433,6 +432,8 @@ protected: }; std::deque<simt_stack_entry> m_stack; + + class gpgpu_sim * m_gpu; }; #define GLOBAL_HEAP_START 0xC0000000 @@ -451,19 +452,7 @@ protected: #if !defined(__CUDA_RUNTIME_API_H__) -enum cudaChannelFormatKind { - cudaChannelFormatKindSigned, - cudaChannelFormatKindUnsigned, - cudaChannelFormatKindFloat -}; - -struct cudaChannelFormatDesc { - int x; - int y; - int z; - int w; - enum cudaChannelFormatKind f; -}; +#include "builtin_types.h" struct cudaArray { void *devPtr; @@ -475,28 +464,6 @@ struct cudaArray { unsigned dimensions; }; -enum cudaTextureAddressMode { - cudaAddressModeWrap, - cudaAddressModeClamp -}; - -enum cudaTextureFilterMode { - cudaFilterModePoint, - cudaFilterModeLinear -}; - -enum cudaTextureReadMode { - cudaReadModeElementType, - cudaReadModeNormalizedFloat -}; - -struct textureReference { - int normalized; - enum cudaTextureFilterMode filterMode; - enum cudaTextureAddressMode addressMode[3]; - struct cudaChannelFormatDesc channelDesc; -}; - #endif // Struct that record other attributes in the textureReference declaration @@ -571,6 +538,12 @@ public: int resume_CTA; int checkpoint_CTA_t; int checkpoint_insn_Y; + + //Move some cycle core stats here instead of being global + unsigned long long gpu_sim_cycle; + unsigned long long gpu_tot_sim_cycle; + + void* gpu_malloc( size_t size ); void* gpu_mallocarray( size_t count ); void gpu_memset( size_t dst_start_addr, int c, size_t count ); @@ -835,8 +808,8 @@ public: class mem_fetch_allocator { public: - virtual mem_fetch *alloc( new_addr_type addr, mem_access_type type, unsigned size, bool wr ) const = 0; - virtual mem_fetch *alloc( const class warp_inst_t &inst, const mem_access_t &access ) const = 0; + virtual mem_fetch *alloc( new_addr_type addr, mem_access_type type, unsigned size, bool wr, unsigned long long cycle ) const = 0; + virtual mem_fetch *alloc( const class warp_inst_t &inst, const mem_access_t &access, unsigned long long cycle ) const = 0; }; // the maximum number of destination, source, or address uarch operands in a instruction diff --git a/src/cuda-sim/Makefile b/src/cuda-sim/Makefile index 999dad7..85d1c8c 100644 --- a/src/cuda-sim/Makefile +++ b/src/cuda-sim/Makefile @@ -79,16 +79,16 @@ libgpgpu_ptx_sim.a: $(OBJS) ar rcs $(OUTPUT_DIR)/libgpgpu_ptx_sim.a $(OUTPUT_DIR)/ptx.tab.o $(OUTPUT_DIR)/lex.ptx_.o $(OUTPUT_DIR)/ptxinfo.tab.o $(OUTPUT_DIR)/lex.ptxinfo_.o $(OBJS) $(OUTPUT_DIR)/ptx.tab.o: $(OUTPUT_DIR)/ptx.tab.c - $(CPP) -c $(OPT) -DYYDEBUG $(OUTPUT_DIR)/ptx.tab.c -o $(OUTPUT_DIR)/ptx.tab.o + $(CPP) -c $(CXX_OPT) -DYYDEBUG $(OUTPUT_DIR)/ptx.tab.c -o $(OUTPUT_DIR)/ptx.tab.o $(OUTPUT_DIR)/lex.ptx_.o: $(OUTPUT_DIR)/lex.ptx_.c - $(CPP) -c $(OPT) $(OUTPUT_DIR)/lex.ptx_.c -o $(OUTPUT_DIR)/lex.ptx_.o + $(CPP) -c $(CXX_OPT) $(OUTPUT_DIR)/lex.ptx_.c -o $(OUTPUT_DIR)/lex.ptx_.o $(OUTPUT_DIR)/ptxinfo.tab.o: $(OUTPUT_DIR)/ptxinfo.tab.c - $(CPP) -c $(OPT) -DYYDEBUG $(OUTPUT_DIR)/ptxinfo.tab.c -o $(OUTPUT_DIR)/ptxinfo.tab.o + $(CPP) -c $(CXX_OPT) -DYYDEBUG $(OUTPUT_DIR)/ptxinfo.tab.c -o $(OUTPUT_DIR)/ptxinfo.tab.o $(OUTPUT_DIR)/lex.ptxinfo_.o: $(OUTPUT_DIR)/lex.ptxinfo_.c $(OUTPUT_DIR)/ptxinfo.tab.c - $(CPP) -c $(OPT) $(OUTPUT_DIR)/lex.ptxinfo_.c -o $(OUTPUT_DIR)/lex.ptxinfo_.o + $(CPP) -c $(CXX_OPT) $(OUTPUT_DIR)/lex.ptxinfo_.c -o $(OUTPUT_DIR)/lex.ptxinfo_.o $(OUTPUT_DIR)/ptx.tab.c: ptx.y bison --name-prefix=ptx_ -v -d ptx.y --file-prefix=$(OUTPUT_DIR)/ptx diff --git a/src/cuda-sim/cuda-math.h b/src/cuda-sim/cuda-math.h index a5db337..9a5468c 100644 --- a/src/cuda-sim/cuda-math.h +++ b/src/cuda-sim/cuda-math.h @@ -277,10 +277,10 @@ int float2int(float a, enum cudaRoundMode mode) { int tmp; switch (mode) { - case cuda_math::cudaRoundZero: tmp = truncf(a); break; - case cuda_math::cudaRoundNearest: tmp = nearbyintf(a); break; - case cuda_math::cudaRoundMinInf: tmp = floorf(a); break; - case cuda_math::cudaRoundPosInf: tmp = ceilf(a); break; + case cudaRoundZero: tmp = truncf(a); break; + case cudaRoundNearest: tmp = nearbyintf(a); break; + case cudaRoundMinInf: tmp = floorf(a); break; + case cudaRoundPosInf: tmp = ceilf(a); break; default: abort(); } return tmp; @@ -296,10 +296,10 @@ unsigned int float2uint(float a, enum cudaRoundMode mode) { unsigned int tmp; switch (mode) { - case cuda_math::cudaRoundZero: tmp = truncf(a); break; - case cuda_math::cudaRoundNearest: tmp = nearbyintf(a); break; - case cuda_math::cudaRoundMinInf: tmp = floorf(a); break; - case cuda_math::cudaRoundPosInf: tmp = ceilf(a); break; + case cudaRoundZero: tmp = truncf(a); break; + case cudaRoundNearest: tmp = nearbyintf(a); break; + case cudaRoundMinInf: tmp = floorf(a); break; + case cudaRoundPosInf: tmp = ceilf(a); break; default: abort(); } return tmp; diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index f7bb9cc..261d605 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -30,6 +30,8 @@ #include "instructions.h" #include "ptx_ir.h" +class ptx_recognizer; +typedef void * yyscan_t; #include "ptx.tab.h" #include "ptx_sim.h" #include <stdio.h> @@ -446,8 +448,8 @@ void gpgpu_t::memcpy_to_gpu( size_t dst_start_addr, const void *src, size_t coun m_global_mem->write(dst_start_addr+n,1, src_data+n,NULL,NULL); // Copy into the performance model. - extern gpgpu_sim* g_the_gpu; - g_the_gpu->perf_memcpy_to_gpu(dst_start_addr, count); + //extern gpgpu_sim* g_the_gpu; + g_the_gpu()->perf_memcpy_to_gpu(dst_start_addr, count); if(g_debug_execution >= 3) { printf( " done.\n"); fflush(stdout); @@ -465,8 +467,8 @@ void gpgpu_t::memcpy_from_gpu( void *dst, size_t src_start_addr, size_t count ) m_global_mem->read(src_start_addr+n,1,dst_data+n); // Copy into the performance model. - extern gpgpu_sim* g_the_gpu; - g_the_gpu->perf_memcpy_to_gpu(src_start_addr, count); + //extern gpgpu_sim* g_the_gpu; + g_the_gpu()->perf_memcpy_to_gpu(src_start_addr, count); if(g_debug_execution >= 3) { printf( " done.\n"); fflush(stdout); @@ -1268,8 +1270,8 @@ void function_info::finalize( memory_space *param_mem ) void function_info::param_to_shared( memory_space *shared_mem, symbol_table *symtab ) { // TODO: call this only for PTXPlus with GT200 models - extern gpgpu_sim* g_the_gpu; - if (not g_the_gpu->get_config().convert_to_ptxplus()) return; + //extern gpgpu_sim* g_the_gpu; + if (not g_the_gpu()->get_config().convert_to_ptxplus()) return; // copies parameters into simulated shared memory for( std::map<unsigned,param_info>::iterator i=m_ptx_kernel_param_info.begin(); i!=m_ptx_kernel_param_info.end(); i++ ) { @@ -2148,7 +2150,7 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) printf("GPGPU-Sim: Performing Functional Simulation, executing kernel %s...\n",kernel.name().c_str()); //using a shader core object for book keeping, it is not needed but as most function built for performance simulation need it we use it here - extern gpgpu_sim *g_the_gpu; + //extern gpgpu_sim *g_the_gpu; //before we execute, we should do PDOM analysis for functional simulation scenario. function_info *kernel_func_info = kernel.entry(); const struct gpgpu_ptx_sim_info *kernel_info = ptx_sim_kernel_info(kernel_func_info); @@ -2163,7 +2165,7 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) kernel_func_info->set_pdom(); } - unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), g_the_gpu->getShaderCoreConfig()->warp_size, g_the_gpu->getShaderCoreConfig()->n_thread_per_shader, g_the_gpu->getShaderCoreConfig()->gpgpu_shmem_size, g_the_gpu->getShaderCoreConfig()->gpgpu_shader_registers, g_the_gpu->getShaderCoreConfig()->max_cta_per_core); + unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), g_the_gpu()->getShaderCoreConfig()->warp_size, g_the_gpu()->getShaderCoreConfig()->n_thread_per_shader, g_the_gpu()->getShaderCoreConfig()->gpgpu_shmem_size, g_the_gpu()->getShaderCoreConfig()->gpgpu_shader_registers, g_the_gpu()->getShaderCoreConfig()->max_cta_per_core); printf("Max CTA : %d\n",max_cta_tot); @@ -2171,11 +2173,11 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) int inst_count=50; - int cp_op= g_the_gpu->checkpoint_option; - int cp_CTA = g_the_gpu->checkpoint_CTA; - int cp_kernel= g_the_gpu->checkpoint_kernel; - cp_count= g_the_gpu->checkpoint_insn_Y; - cp_cta_resume= g_the_gpu->checkpoint_CTA_t; + int cp_op= g_the_gpu()->checkpoint_option; + int cp_CTA = g_the_gpu()->checkpoint_CTA; + int cp_kernel= g_the_gpu()->checkpoint_kernel; + cp_count= g_the_gpu()->checkpoint_insn_Y; + cp_cta_resume= g_the_gpu()->checkpoint_CTA_t; int cta_launched =0; //we excute the kernel one CTA (Block) at the time, as synchronization functions work block wise @@ -2187,8 +2189,8 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) { functionalCoreSim cta( &kernel, - g_the_gpu, - g_the_gpu->getShaderCoreConfig()->warp_size + g_the_gpu(), + g_the_gpu()->getShaderCoreConfig()->warp_size ); cta.execute(cp_count,temp); @@ -2209,7 +2211,7 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) { char f1name[2048]; snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", kernel.get_uid() ); - g_checkpoint->store_global_mem(g_the_gpu->get_global_memory(), f1name , "%08x"); + g_checkpoint->store_global_mem(g_the_gpu()->get_global_memory(), f1name , "%08x"); } @@ -2219,8 +2221,8 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) //openCL kernel simulation calls don't register the kernel so we don't register its exit if(!openCL) { - extern stream_manager *g_stream_manager; - g_stream_manager->register_finished_kernel(kernel.get_uid()); + //extern stream_manager *g_stream_manager; + g_stream_manager()->register_finished_kernel(kernel.get_uid()); } //******PRINTING******* @@ -2235,7 +2237,7 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) //g_simulation_starttime is initilized by gpgpu_ptx_sim_init_perf() in gpgpusim_entrypoint.cc upon starting gpgpu-sim time_t end_time, elapsed_time, days, hrs, minutes, sec; end_time = time((time_t *)NULL); - elapsed_time = MAX(end_time - g_simulation_starttime, 1); + elapsed_time = MAX(end_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1); //calculating and printing simulation time in terms of days, hours, minutes and seconds diff --git a/src/cuda-sim/cuda_device_runtime.cc b/src/cuda-sim/cuda_device_runtime.cc index 917e7a8..be8369f 100644 --- a/src/cuda-sim/cuda_device_runtime.cc +++ b/src/cuda-sim/cuda_device_runtime.cc @@ -18,6 +18,7 @@ unsigned long long g_max_total_param_size = 0; #include "cuda-sim.h" #include "ptx_ir.h" #include "../stream_manager.h" +#include "../gpgpusim_entrypoint.h" #include "cuda_device_runtime.h" #define DEV_RUNTIME_REPORT(a) \ @@ -64,7 +65,7 @@ public: std::map<void *, device_launch_config_t> g_cuda_device_launch_param_map; std::list<device_launch_operation_t> g_cuda_device_launch_op; -extern stream_manager *g_stream_manager; +//extern stream_manager *g_stream_manager(); //Handling device runtime api: //void * cudaGetParameterBufferV2(void *func, dim3 gridDimension, dim3 blockDimension, unsigned int sharedMemSize) @@ -200,7 +201,7 @@ void gpgpusim_cuda_launchDeviceV2(const ptx_instruction * pI, ptx_thread_info * //create child kernel_info_t and index it with parameter_buffer address gpgpu_t* gpu=thread->get_gpu(); device_grid = new kernel_info_t(config.grid_dim, config.block_dim, device_kernel_entry, gpu->getNameArrayMapping(), gpu->getNameInfoMapping()); - device_grid->launch_cycle = gpu_sim_cycle + gpu_tot_sim_cycle; + device_grid->launch_cycle = gpu->gpu_sim_cycle + gpu->gpu_tot_sim_cycle; kernel_info_t & parent_grid = thread->get_kernel(); DEV_RUNTIME_REPORT("child kernel launched by " << parent_grid.name() << ", cta (" << thread->get_ctaid().x << ", " << thread->get_ctaid().y << ", " << thread->get_ctaid().z << @@ -322,7 +323,7 @@ void launch_one_device_kernel() { device_launch_operation_t &op = g_cuda_device_launch_op.front(); stream_operation stream_op = stream_operation(op.grid, g_ptx_sim_mode, op.stream); - g_stream_manager->push(stream_op); + g_stream_manager()->push(stream_op); g_cuda_device_launch_op.pop_front(); } } diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc index 11001d7..186bb1e 100644 --- a/src/cuda-sim/instructions.cc +++ b/src/cuda-sim/instructions.cc @@ -31,6 +31,8 @@ #include "ptx_ir.h" #include "opcodes.h" #include "ptx_sim.h" +typedef void * yyscan_t; +class ptx_recognizer; #include "ptx.tab.h" #include <stdlib.h> #include <math.h> @@ -2300,12 +2302,12 @@ ptx_reg_t f2x( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign, half_float::half tmp_h; //assert( from_width == 32); - enum cuda_math::cudaRoundMode mode = cuda_math::cudaRoundZero; + enum cudaRoundMode mode = cudaRoundZero; switch (rounding_mode) { - case RZI_OPTION: mode = cuda_math::cudaRoundZero; break; - case RNI_OPTION: mode = cuda_math::cudaRoundNearest; break; - case RMI_OPTION: mode = cuda_math::cudaRoundMinInf; break; - case RPI_OPTION: mode = cuda_math::cudaRoundPosInf; break; + case RZI_OPTION: mode = cudaRoundZero; break; + case RNI_OPTION: mode = cudaRoundNearest; break; + case RMI_OPTION: mode = cudaRoundMinInf; break; + case RPI_OPTION: mode = cudaRoundPosInf; break; default: break; } diff --git a/src/cuda-sim/ptx.l b/src/cuda-sim/ptx.l index 3232361..18952a9 100644 --- a/src/cuda-sim/ptx.l +++ b/src/cuda-sim/ptx.l @@ -31,21 +31,26 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. %option noyywrap %option yylineno %option prefix="ptx_" + +%option bison-bridge +%option reentrant + %{ #include "opcodes.h" +#include "ptx_parser.h" #include "ptx.tab.h" #include <string.h> -#define LINEBUF_SIZE (64*1024) -char linebuf[LINEBUF_SIZE]; -unsigned col = 0; -#define TC col+=strlen(ptx_text); +#define LINEBUF_SIZE (4*1024) +#define TC recognizer->col+=strlen(yytext); #define CHECK_UNSIGNED \ if( yytext[strlen(yytext)-1]=='U' ) { \ printf("GPGPU-Sim: ERROR ** U modifier not implemented\n"); \ abort(); \ } -int ptx_error( const char *s ); +#define YY_DECL int ptx_lex \ + (YYSTYPE * yylval_param , yyscan_t yyscanner, ptx_recognizer* recognizer) +int ptx_error( yyscan_t yyscanner, ptx_recognizer* recognizer, const char *s ); %} %s IN_STRING @@ -54,121 +59,121 @@ int ptx_error( const char *s ); %x NOT_OPCODE %% -abs TC; ptx_lval.int_value = ABS_OP; return OPCODE; -add TC; ptx_lval.int_value = ADD_OP; return OPCODE; -addp TC; ptx_lval.int_value = ADDP_OP; return OPCODE; -addc TC; ptx_lval.int_value = ADDC_OP; return OPCODE; -and TC; ptx_lval.int_value = AND_OP; return OPCODE; -andn TC; ptx_lval.int_value = ANDN_OP; return OPCODE; -atom TC; ptx_lval.int_value = ATOM_OP; return OPCODE; -bar.warp TC; ptx_lval.int_value = NOP_OP; return OPCODE; -bar TC; ptx_lval.int_value = BAR_OP; return OPCODE; -bfe TC; ptx_lval.int_value = BFE_OP; return OPCODE; -bfi TC; ptx_lval.int_value = BFI_OP; return OPCODE; -bfind TC; ptx_lval.int_value = BFIND_OP; return OPCODE; -bra TC; ptx_lval.int_value = BRA_OP; return OPCODE; -brx TC; ptx_lval.int_value = BRX_OP; return OPCODE; -brev TC; ptx_lval.int_value = BREV_OP; return OPCODE; -brkpt TC; ptx_lval.int_value = BRKPT_OP; return OPCODE; +abs TC; yylval->int_value = ABS_OP; return OPCODE; +add TC; yylval->int_value = ADD_OP; return OPCODE; +addp TC; yylval->int_value = ADDP_OP; return OPCODE; +addc TC; yylval->int_value = ADDC_OP; return OPCODE; +and TC; yylval->int_value = AND_OP; return OPCODE; +andn TC; yylval->int_value = ANDN_OP; return OPCODE; +atom TC; yylval->int_value = ATOM_OP; return OPCODE; +bar.warp TC; yylval->int_value = NOP_OP; return OPCODE; +bar TC; yylval->int_value = BAR_OP; return OPCODE; +bfe TC; yylval->int_value = BFE_OP; return OPCODE; +bfi TC; yylval->int_value = BFI_OP; return OPCODE; +bfind TC; yylval->int_value = BFIND_OP; return OPCODE; +bra TC; yylval->int_value = BRA_OP; return OPCODE; +brx TC; yylval->int_value = BRX_OP; return OPCODE; +brev TC; yylval->int_value = BREV_OP; return OPCODE; +brkpt TC; yylval->int_value = BRKPT_OP; return OPCODE; -wmma TC; ptx_lval.int_value = MMA_OP; return OPCODE; -wmma\.load TC; ptx_lval.int_value = MMA_LD_OP; return OPCODE; -wmma\.store TC; ptx_lval.int_value = MMA_ST_OP; return OPCODE; +wmma TC; yylval->int_value = MMA_OP; return OPCODE; +wmma\.load TC; yylval->int_value = MMA_LD_OP; return OPCODE; +wmma\.store TC; yylval->int_value = MMA_ST_OP; return OPCODE; -call TC; BEGIN(NOT_OPCODE); ptx_lval.int_value = CALL_OP; return OPCODE; // blocking opcode token in case the callee has the same name as an opcode -callp TC; BEGIN(NOT_OPCODE); ptx_lval.int_value = CALLP_OP; return OPCODE; -clz TC; ptx_lval.int_value = CLZ_OP; return OPCODE; -cnot TC; ptx_lval.int_value = CNOT_OP; return OPCODE; -cos TC; ptx_lval.int_value = COS_OP; return OPCODE; -cvt TC; ptx_lval.int_value = CVT_OP; return OPCODE; -cvta TC; ptx_lval.int_value = CVTA_OP; return OPCODE; -div TC; ptx_lval.int_value = DIV_OP; return OPCODE; -dp4a TC; ptx_lval.int_value = DP4A_OP; return OPCODE; -ex2 TC; ptx_lval.int_value = EX2_OP; return OPCODE; -exit TC; ptx_lval.int_value = EXIT_OP; return OPCODE; -fma TC; ptx_lval.int_value = FMA_OP; return OPCODE; -isspacep TC; ptx_lval.int_value = ISSPACEP_OP; return OPCODE; -ld TC; ptx_lval.int_value = LD_OP; return OPCODE; -ld.volatile TC; ptx_lval.int_value = LD_OP; return OPCODE; -ldu TC; ptx_lval.int_value = LDU_OP; return OPCODE; -lg2 TC; ptx_lval.int_value = LG2_OP; return OPCODE; -mad24 TC; ptx_lval.int_value = MAD24_OP; return OPCODE; -mad TC; ptx_lval.int_value = MAD_OP; return OPCODE; -madc TC; ptx_lval.int_value = MADC_OP; return OPCODE; -madp TC; ptx_lval.int_value = MADP_OP; return OPCODE; -max TC; ptx_lval.int_value = MAX_OP; return OPCODE; -membar TC; ptx_lval.int_value = MEMBAR_OP; return OPCODE; -min TC; ptx_lval.int_value = MIN_OP; return OPCODE; -mov TC; ptx_lval.int_value = MOV_OP; return OPCODE; -mul24 TC; ptx_lval.int_value = MUL24_OP; return OPCODE; -mul TC; ptx_lval.int_value = MUL_OP; return OPCODE; -neg TC; ptx_lval.int_value = NEG_OP; return OPCODE; -nandn TC; ptx_lval.int_value = NANDN_OP; return OPCODE; -norn TC; ptx_lval.int_value = NORN_OP; return OPCODE; -not TC; ptx_lval.int_value = NOT_OP; return OPCODE; -or TC; ptx_lval.int_value = OR_OP; return OPCODE; -orn TC; ptx_lval.int_value = ORN_OP; return OPCODE; -pmevent TC; ptx_lval.int_value = PMEVENT_OP; return OPCODE; -popc TC; ptx_lval.int_value = POPC_OP; return OPCODE; -prefetch TC; ptx_lval.int_value = PREFETCH_OP; return OPCODE; -prefetchu TC; ptx_lval.int_value = PREFETCHU_OP; return OPCODE; -prmt TC; ptx_lval.int_value = PRMT_OP; return OPCODE; -rcp TC; ptx_lval.int_value = RCP_OP; return OPCODE; -red TC; ptx_lval.int_value = RED_OP; return OPCODE; -rem TC; ptx_lval.int_value = REM_OP; return OPCODE; -ret TC; ptx_lval.int_value = RET_OP; return OPCODE; -retp TC; ptx_lval.int_value = RETP_OP; return OPCODE; -rsqrt TC; ptx_lval.int_value = RSQRT_OP; return OPCODE; -sad TC; ptx_lval.int_value = SAD_OP; return OPCODE; -selp TC; ptx_lval.int_value = SELP_OP; return OPCODE; -setp TC; ptx_lval.int_value = SETP_OP; return OPCODE; -set TC; ptx_lval.int_value = SET_OP; return OPCODE; -shfl TC; ptx_lval.int_value = SHFL_OP; return OPCODE; -shl TC; ptx_lval.int_value = SHL_OP; return OPCODE; -shr TC; ptx_lval.int_value = SHR_OP; return OPCODE; -sin TC; ptx_lval.int_value = SIN_OP; return OPCODE; -slct TC; ptx_lval.int_value = SLCT_OP; return OPCODE; -sqrt TC; ptx_lval.int_value = SQRT_OP; return OPCODE; -sst TC; ptx_lval.int_value = SST_OP; return OPCODE; -ssy TC; ptx_lval.int_value = SSY_OP; return OPCODE; -st TC; ptx_lval.int_value = ST_OP; return OPCODE; -st.volatile TC; ptx_lval.int_value = ST_OP; return OPCODE; -sub TC; ptx_lval.int_value = SUB_OP; return OPCODE; -subc TC; ptx_lval.int_value = SUBC_OP; return OPCODE; -suld TC; ptx_lval.int_value = SULD_OP; return OPCODE; -sured TC; ptx_lval.int_value = SURED_OP; return OPCODE; -surst TC; ptx_lval.int_value = SUST_OP; return OPCODE; -suq TC; ptx_lval.int_value = SUQ_OP; return OPCODE; -tex TC; BEGIN(NOT_OPCODE); ptx_lval.int_value = TEX_OP; return OPCODE; -txq TC; ptx_lval.int_value = TEX_OP; return OPCODE; -trap TC; ptx_lval.int_value = TRAP_OP; return OPCODE; -vabsdiff TC; ptx_lval.int_value = VABSDIFF_OP; return OPCODE; -vadd TC; ptx_lval.int_value = VADD_OP; return OPCODE; -vmad TC; ptx_lval.int_value = VMAD_OP; return OPCODE; -vmax TC; ptx_lval.int_value = VMAX_OP; return OPCODE; -vmin TC; ptx_lval.int_value = VMIN_OP; return OPCODE; -vset TC; ptx_lval.int_value = VSET_OP; return OPCODE; -vshl TC; ptx_lval.int_value = VSHL_OP; return OPCODE; -vshr TC; ptx_lval.int_value = VSHR_OP; return OPCODE; -vsub TC; ptx_lval.int_value = VSUB_OP; return OPCODE; -vote TC; ptx_lval.int_value = VOTE_OP; return OPCODE; -xor TC; ptx_lval.int_value = XOR_OP; return OPCODE; -nop TC; ptx_lval.int_value = NOP_OP; return OPCODE; -break TC; ptx_lval.int_value = BREAK_OP; return OPCODE; -breakaddr TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE; +call TC; BEGIN(NOT_OPCODE); yylval->int_value = CALL_OP; return OPCODE; // blocking opcode token in case the callee has the same name as an opcode +callp TC; BEGIN(NOT_OPCODE); yylval->int_value = CALLP_OP; return OPCODE; +clz TC; yylval->int_value = CLZ_OP; return OPCODE; +cnot TC; yylval->int_value = CNOT_OP; return OPCODE; +cos TC; yylval->int_value = COS_OP; return OPCODE; +cvt TC; yylval->int_value = CVT_OP; return OPCODE; +cvta TC; yylval->int_value = CVTA_OP; return OPCODE; +div TC; yylval->int_value = DIV_OP; return OPCODE; +dp4a TC; yylval->int_value = DP4A_OP; return OPCODE; +ex2 TC; yylval->int_value = EX2_OP; return OPCODE; +exit TC; yylval->int_value = EXIT_OP; return OPCODE; +fma TC; yylval->int_value = FMA_OP; return OPCODE; +isspacep TC; yylval->int_value = ISSPACEP_OP; return OPCODE; +ld TC; yylval->int_value = LD_OP; return OPCODE; +ld.volatile TC; yylval->int_value = LD_OP; return OPCODE; +ldu TC; yylval->int_value = LDU_OP; return OPCODE; +lg2 TC; yylval->int_value = LG2_OP; return OPCODE; +mad24 TC; yylval->int_value = MAD24_OP; return OPCODE; +mad TC; yylval->int_value = MAD_OP; return OPCODE; +madc TC; yylval->int_value = MADC_OP; return OPCODE; +madp TC; yylval->int_value = MADP_OP; return OPCODE; +max TC; yylval->int_value = MAX_OP; return OPCODE; +membar TC; yylval->int_value = MEMBAR_OP; return OPCODE; +min TC; yylval->int_value = MIN_OP; return OPCODE; +mov TC; yylval->int_value = MOV_OP; return OPCODE; +mul24 TC; yylval->int_value = MUL24_OP; return OPCODE; +mul TC; yylval->int_value = MUL_OP; return OPCODE; +neg TC; yylval->int_value = NEG_OP; return OPCODE; +nandn TC; yylval->int_value = NANDN_OP; return OPCODE; +norn TC; yylval->int_value = NORN_OP; return OPCODE; +not TC; yylval->int_value = NOT_OP; return OPCODE; +or TC; yylval->int_value = OR_OP; return OPCODE; +orn TC; yylval->int_value = ORN_OP; return OPCODE; +pmevent TC; yylval->int_value = PMEVENT_OP; return OPCODE; +popc TC; yylval->int_value = POPC_OP; return OPCODE; +prefetch TC; yylval->int_value = PREFETCH_OP; return OPCODE; +prefetchu TC; yylval->int_value = PREFETCHU_OP; return OPCODE; +prmt TC; yylval->int_value = PRMT_OP; return OPCODE; +rcp TC; yylval->int_value = RCP_OP; return OPCODE; +red TC; yylval->int_value = RED_OP; return OPCODE; +rem TC; yylval->int_value = REM_OP; return OPCODE; +ret TC; yylval->int_value = RET_OP; return OPCODE; +retp TC; yylval->int_value = RETP_OP; return OPCODE; +rsqrt TC; yylval->int_value = RSQRT_OP; return OPCODE; +sad TC; yylval->int_value = SAD_OP; return OPCODE; +selp TC; yylval->int_value = SELP_OP; return OPCODE; +setp TC; yylval->int_value = SETP_OP; return OPCODE; +set TC; yylval->int_value = SET_OP; return OPCODE; +shfl TC; yylval->int_value = SHFL_OP; return OPCODE; +shl TC; yylval->int_value = SHL_OP; return OPCODE; +shr TC; yylval->int_value = SHR_OP; return OPCODE; +sin TC; yylval->int_value = SIN_OP; return OPCODE; +slct TC; yylval->int_value = SLCT_OP; return OPCODE; +sqrt TC; yylval->int_value = SQRT_OP; return OPCODE; +sst TC; yylval->int_value = SST_OP; return OPCODE; +ssy TC; yylval->int_value = SSY_OP; return OPCODE; +st TC; yylval->int_value = ST_OP; return OPCODE; +st.volatile TC; yylval->int_value = ST_OP; return OPCODE; +sub TC; yylval->int_value = SUB_OP; return OPCODE; +subc TC; yylval->int_value = SUBC_OP; return OPCODE; +suld TC; yylval->int_value = SULD_OP; return OPCODE; +sured TC; yylval->int_value = SURED_OP; return OPCODE; +surst TC; yylval->int_value = SUST_OP; return OPCODE; +suq TC; yylval->int_value = SUQ_OP; return OPCODE; +tex TC; BEGIN(NOT_OPCODE); yylval->int_value = TEX_OP; return OPCODE; +txq TC; yylval->int_value = TEX_OP; return OPCODE; +trap TC; yylval->int_value = TRAP_OP; return OPCODE; +vabsdiff TC; yylval->int_value = VABSDIFF_OP; return OPCODE; +vadd TC; yylval->int_value = VADD_OP; return OPCODE; +vmad TC; yylval->int_value = VMAD_OP; return OPCODE; +vmax TC; yylval->int_value = VMAX_OP; return OPCODE; +vmin TC; yylval->int_value = VMIN_OP; return OPCODE; +vset TC; yylval->int_value = VSET_OP; return OPCODE; +vshl TC; yylval->int_value = VSHL_OP; return OPCODE; +vshr TC; yylval->int_value = VSHR_OP; return OPCODE; +vsub TC; yylval->int_value = VSUB_OP; return OPCODE; +vote TC; yylval->int_value = VOTE_OP; return OPCODE; +xor TC; yylval->int_value = XOR_OP; return OPCODE; +nop TC; yylval->int_value = NOP_OP; return OPCODE; +break TC; yylval->int_value = BREAK_OP; return OPCODE; +breakaddr TC; yylval->int_value = BREAKADDR_OP; return OPCODE; "CPTX_END" printf("ENDING CUSTOM PTX.\n"); BEGIN(IN_COMMENT); <INITIAL,NOT_OPCODE,IN_INST,IN_FUNC_DECL>{ -\.a\.sync TC; ptx_lval.int_value = LOAD_A; return WMMA_DIRECTIVE; -\.b\.sync TC; ptx_lval.int_value = LOAD_B; return WMMA_DIRECTIVE; -\.c\.sync TC; ptx_lval.int_value = LOAD_C; return WMMA_DIRECTIVE; -\.d\.sync TC; ptx_lval.int_value = STORE_D; return WMMA_DIRECTIVE; -\.mma\.sync TC;ptx_lval.int_value=MMA; return WMMA_DIRECTIVE; +\.a\.sync TC; yylval->int_value = LOAD_A; return WMMA_DIRECTIVE; +\.b\.sync TC; yylval->int_value = LOAD_B; return WMMA_DIRECTIVE; +\.c\.sync TC; yylval->int_value = LOAD_C; return WMMA_DIRECTIVE; +\.d\.sync TC; yylval->int_value = STORE_D; return WMMA_DIRECTIVE; +\.mma\.sync TC;yylval->int_value=MMA; return WMMA_DIRECTIVE; -\.row TC; ptx_lval.int_value = ROW; return LAYOUT; -\.col TC; ptx_lval.int_value = COL; return LAYOUT; -\.m16n16k16 TC; ptx_lval.int_value = M16N16K16; return CONFIGURATION; +\.row TC; yylval->int_value = ROW; return LAYOUT; +\.col TC; yylval->int_value = COL; return LAYOUT; +\.m16n16k16 TC; yylval->int_value = M16N16K16; return CONFIGURATION; \.f4e TC; return PRMT_F4E_MODE; \.b4e TC; return PRMT_B4E_MODE; \.rc8 TC; return PRMT_RC8_MODE; @@ -181,8 +186,8 @@ breakaddr TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE; \.byte TC; return BYTE_DIRECTIVE; /* not in PTX 2.1 */ \.callprototype TC; return CALLPROTOTYPE_DIRECTIVE; \.calltargets TC; return CALLTARGETS_DIRECTIVE; -\.const\[[0-9]+\] TC; ptx_lval.int_value = atoi(yytext+7); return CONST_DIRECTIVE; -\.const TC; ptx_lval.int_value = 0; return CONST_DIRECTIVE; +\.const\[[0-9]+\] TC; yylval->int_value = atoi(yytext+7); return CONST_DIRECTIVE; +\.const TC; yylval->int_value = 0; return CONST_DIRECTIVE; \.entry TC; return ENTRY_DIRECTIVE; \.extern TC; return EXTERN_DIRECTIVE; \.file TC; BEGIN(INITIAL); return FILE_DIRECTIVE; @@ -217,40 +222,40 @@ breakaddr TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE; \.constptr TC; return CONSTPTR_DIRECTIVE; /* Ptx plus directive for pointer to constant memory */ \.ptr TC; return PTR_DIRECTIVE; /* Added for new OpenCL genrated code */ -"%clock" TC; ptx_lval.int_value = CLOCK_REG; return SPECIAL_REGISTER; -"%halfclock" TC; ptx_lval.int_value = HALFCLOCK_ID; return SPECIAL_REGISTER; -"%clock64" TC; ptx_lval.int_value = CLOCK64_REG; return SPECIAL_REGISTER; -"%ctaid" TC; ptx_lval.int_value = CTAID_REG; return SPECIAL_REGISTER; -"%envreg"[0-9]+ TC; sscanf(yytext+7,"%u",&ptx_lval.int_value); ptx_lval.int_value<<=16; ptx_lval.int_value += ENVREG_REG; return SPECIAL_REGISTER; -"%gridid" TC; ptx_lval.int_value = GRIDID_REG; return SPECIAL_REGISTER; -"%laneid" TC; ptx_lval.int_value = LANEID_REG; return SPECIAL_REGISTER; -"%lanemask_eq" TC; ptx_lval.int_value = LANEMASK_EQ_REG; return SPECIAL_REGISTER; -"%lanemask_le" TC; ptx_lval.int_value = LANEMASK_LE_REG; return SPECIAL_REGISTER; -"%lanemask_lt" TC; ptx_lval.int_value = LANEMASK_LT_REG; return SPECIAL_REGISTER; -"%lanemask_ge" TC; ptx_lval.int_value = LANEMASK_GE_REG; return SPECIAL_REGISTER; -"%lanemask_gt" TC; ptx_lval.int_value = LANEMASK_GT_REG; return SPECIAL_REGISTER; -"%nctaid" TC; ptx_lval.int_value = NCTAID_REG; return SPECIAL_REGISTER; -"%ntid" TC; ptx_lval.int_value = NTID_REG; return SPECIAL_REGISTER; -"%nsmid" TC; ptx_lval.int_value = NSMID_REG; return SPECIAL_REGISTER; -"%nwarpid" TC; ptx_lval.int_value = NWARPID_REG; return SPECIAL_REGISTER; -"%pm"[0-3] TC; sscanf(yytext+3,"%u",&ptx_lval.int_value); ptx_lval.int_value<<=16; ptx_lval.int_value += PM_REG; return SPECIAL_REGISTER; -"%smid" TC; ptx_lval.int_value = SMID_REG; return SPECIAL_REGISTER; -"%tid" TC; ptx_lval.int_value = TID_REG; return SPECIAL_REGISTER; -"%warpid" TC; ptx_lval.int_value = WARPID_REG; return SPECIAL_REGISTER; -"WARP_SZ" TC; ptx_lval.int_value = WARPSZ_REG; return SPECIAL_REGISTER; +"%clock" TC; yylval->int_value = CLOCK_REG; return SPECIAL_REGISTER; +"%halfclock" TC; yylval->int_value = HALFCLOCK_ID; return SPECIAL_REGISTER; +"%clock64" TC; yylval->int_value = CLOCK64_REG; return SPECIAL_REGISTER; +"%ctaid" TC; yylval->int_value = CTAID_REG; return SPECIAL_REGISTER; +"%envreg"[0-9]+ TC; sscanf(yytext+7,"%u",&yylval->int_value); yylval->int_value<<=16; yylval->int_value += ENVREG_REG; return SPECIAL_REGISTER; +"%gridid" TC; yylval->int_value = GRIDID_REG; return SPECIAL_REGISTER; +"%laneid" TC; yylval->int_value = LANEID_REG; return SPECIAL_REGISTER; +"%lanemask_eq" TC; yylval->int_value = LANEMASK_EQ_REG; return SPECIAL_REGISTER; +"%lanemask_le" TC; yylval->int_value = LANEMASK_LE_REG; return SPECIAL_REGISTER; +"%lanemask_lt" TC; yylval->int_value = LANEMASK_LT_REG; return SPECIAL_REGISTER; +"%lanemask_ge" TC; yylval->int_value = LANEMASK_GE_REG; return SPECIAL_REGISTER; +"%lanemask_gt" TC; yylval->int_value = LANEMASK_GT_REG; return SPECIAL_REGISTER; +"%nctaid" TC; yylval->int_value = NCTAID_REG; return SPECIAL_REGISTER; +"%ntid" TC; yylval->int_value = NTID_REG; return SPECIAL_REGISTER; +"%nsmid" TC; yylval->int_value = NSMID_REG; return SPECIAL_REGISTER; +"%nwarpid" TC; yylval->int_value = NWARPID_REG; return SPECIAL_REGISTER; +"%pm"[0-3] TC; sscanf(yytext+3,"%u",&yylval->int_value); yylval->int_value<<=16; yylval->int_value += PM_REG; return SPECIAL_REGISTER; +"%smid" TC; yylval->int_value = SMID_REG; return SPECIAL_REGISTER; +"%tid" TC; yylval->int_value = TID_REG; return SPECIAL_REGISTER; +"%warpid" TC; yylval->int_value = WARPID_REG; return SPECIAL_REGISTER; +"WARP_SZ" TC; yylval->int_value = WARPSZ_REG; return SPECIAL_REGISTER; -[a-zA-Z_][a-zA-Z0-9_$]* TC; ptx_lval.string_value = strdup(yytext); return IDENTIFIER; -[$%][a-zA-Z0-9_$]+ TC; ptx_lval.string_value = strdup(yytext); return IDENTIFIER; +[a-zA-Z_][a-zA-Z0-9_$]* TC; yylval->string_value = strdup(yytext); return IDENTIFIER; +[$%][a-zA-Z0-9_$]+ TC; yylval->string_value = strdup(yytext); return IDENTIFIER; -[0-9]+\.[0-9]+ TC; sscanf(yytext,"%lf", &ptx_lval.double_value); return DOUBLE_OPERAND; +[0-9]+\.[0-9]+ TC; sscanf(yytext,"%lf", &yylval->double_value); return DOUBLE_OPERAND; -0[xX][0-9a-fA-F]+U? TC; CHECK_UNSIGNED; sscanf(yytext,"%x", &ptx_lval.int_value); return INT_OPERAND; +0[xX][0-9a-fA-F]+U? TC; CHECK_UNSIGNED; sscanf(yytext,"%x", &yylval->int_value); return INT_OPERAND; 0[0-7]+U? TC; printf("GPGPU-Sim: ERROR ** parsing octal not (yet) implemented\n"); abort(); return INT_OPERAND; 0[bB][01]+U? TC; printf("GPGPU-Sim: ERROR ** parsing binary not (yet) implemented\n"); abort(); return INT_OPERAND; -[-]?[0-9]+U? TC; CHECK_UNSIGNED; ptx_lval.int_value = atoi(yytext); return INT_OPERAND; +[-]?[0-9]+U? TC; CHECK_UNSIGNED; yylval->int_value = atoi(yytext); return INT_OPERAND; -0[fF][0-9a-fA-F]{8} TC; sscanf(yytext+2,"%x", (unsigned*)(void*)&ptx_lval.float_value); return FLOAT_OPERAND; -0[dD][0-9a-fA-F]{16} TC; sscanf(yytext+2,"%Lx", (unsigned long long*)(void*)&ptx_lval.double_value); return DOUBLE_OPERAND; +0[fF][0-9a-fA-F]{8} TC; sscanf(yytext+2,"%x", (unsigned*)(void*)&yylval->float_value); return FLOAT_OPERAND; +0[dD][0-9a-fA-F]{16} TC; sscanf(yytext+2,"%Lx", (unsigned long long*)(void*)&yylval->double_value); return DOUBLE_OPERAND; \.s8 TC; return S8_TYPE; \.s16 TC; return S16_TYPE; @@ -383,12 +388,12 @@ breakaddr TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE; \.2d TC; return GEOM_MODIFIER_2D; \.3d TC; return GEOM_MODIFIER_3D; -\.0 TC; ptx_lval.int_value = 0; return DIMENSION_MODIFIER; -\.1 TC; ptx_lval.int_value = 1; return DIMENSION_MODIFIER; -\.2 TC; ptx_lval.int_value = 2; return DIMENSION_MODIFIER; -\.x TC; ptx_lval.int_value = 0; return DIMENSION_MODIFIER; -\.y TC; ptx_lval.int_value = 1; return DIMENSION_MODIFIER; -\.z TC; ptx_lval.int_value = 2; return DIMENSION_MODIFIER; +\.0 TC; yylval->int_value = 0; return DIMENSION_MODIFIER; +\.1 TC; yylval->int_value = 1; return DIMENSION_MODIFIER; +\.2 TC; yylval->int_value = 2; return DIMENSION_MODIFIER; +\.x TC; yylval->int_value = 0; return DIMENSION_MODIFIER; +\.y TC; yylval->int_value = 1; return DIMENSION_MODIFIER; +\.z TC; yylval->int_value = 2; return DIMENSION_MODIFIER; "-" TC; return MINUS; "+" TC; return PLUS; @@ -411,7 +416,7 @@ breakaddr TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE; "//"[^\n]* TC; // eat single -\n.* col=0; strncpy(linebuf, yytext + 1, LINEBUF_SIZE); yyless( 1 ); +\n.* recognizer->col=0; strncpy(recognizer->linebuf, yytext + 1, LINEBUF_SIZE); yyless( 1 ); " " TC; "\t" TC; @@ -439,28 +444,28 @@ breakaddr TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE; } <IN_STRING>{ "\"" TC; BEGIN(INITIAL); return STRING; -[^\"]* TC; ptx_lval.string_value = strdup(yytext); +[^\"]* TC; yylval->string_value = strdup(yytext); } <*>\t@@DWARF.*\n -<INITIAL,NOT_OPCODE,IN_FUNC_DECL>. TC; ptx_error((const char*)NULL); +<INITIAL,NOT_OPCODE,IN_FUNC_DECL>. TC; ptx_error(yyscanner, recognizer, (const char*)NULL); %% -extern int g_error_detected; extern const char *g_filename; -int ptx_error( const char *s ) +int ptx_error( yyscan_t yyscanner, ptx_recognizer* recognizer, const char *s ) { + struct yyguts_t * yyg = (struct yyguts_t*)yyscanner; int i; - g_error_detected = 1; + recognizer->g_error_detected = 1; fflush(stdout); if( s != NULL ) - printf("%s:%u: Syntax error:\n\n", g_filename, ptx_lineno ); - printf(" %s\n", linebuf ); + printf("%s:%u Syntax error:\n\n", g_filename, yylineno ); + printf(" %s\n", recognizer->linebuf ); printf(" "); - for( i=0; i < col-1; i++ ) { - if( linebuf[i] == '\t' ) printf("\t"); + for( i=0; i < recognizer->col-1; i++ ) { + if( recognizer->linebuf[i] == '\t' ) printf("\t"); else printf(" "); } diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y index 45392fb..837fbe9 100644 --- a/src/cuda-sim/ptx.y +++ b/src/cuda-sim/ptx.y @@ -27,6 +27,17 @@ OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ +%{ +typedef void * yyscan_t; +class ptx_recognizer; +%} + +%define api.pure full +%parse-param {yyscan_t scanner} +%parse-param {ptx_recognizer* recognizer} +%lex-param {yyscan_t scanner} +%lex-param {ptx_recognizer* recognizer} + %union { double double_value; float float_value; @@ -217,10 +228,9 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #include <stdlib.h> #include <string.h> #include <math.h> - void syntax_not_implemented(); - extern int g_func_decl; - int ptx_lex(void); - int ptx_error(const char *); + void syntax_not_implemented(yyscan_t yyscanner, ptx_recognizer* recognizer); + int ptx_lex(YYSTYPE * yylval_param, yyscan_t yyscanner, ptx_recognizer* recognizer); + int ptx_error( yyscan_t yyscanner, ptx_recognizer* recognizer, const char *s ); %} %% @@ -231,96 +241,96 @@ input: /* empty */ | input function_decl ; -function_defn: function_decl { set_symtab($1); func_header(".skip"); } statement_block { end_function(); } - | function_decl { set_symtab($1); } block_spec_list { func_header(".skip"); } statement_block { end_function(); } +function_defn: function_decl { recognizer->set_symtab($1); recognizer->func_header(".skip"); } statement_block { recognizer->end_function(); } + | function_decl { recognizer->set_symtab($1); } block_spec_list { recognizer->func_header(".skip"); } statement_block { recognizer->end_function(); } ; -block_spec: MAXNTID_DIRECTIVE INT_OPERAND COMMA INT_OPERAND COMMA INT_OPERAND {func_header_info_int(".maxntid", $2); - func_header_info_int(",", $4); - func_header_info_int(",", $6); - maxnt_id($2, $4, $6);} - | MINNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".minnctapersm", $2); printf("GPGPU-Sim: Warning: .minnctapersm ignored. \n"); } - | MAXNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".maxnctapersm", $2); printf("GPGPU-Sim: Warning: .maxnctapersm ignored. \n"); } +block_spec: MAXNTID_DIRECTIVE INT_OPERAND COMMA INT_OPERAND COMMA INT_OPERAND {recognizer->func_header_info_int(".maxntid", $2); + recognizer->func_header_info_int(",", $4); + recognizer->func_header_info_int(",", $6); + recognizer->maxnt_id($2, $4, $6);} + | MINNCTAPERSM_DIRECTIVE INT_OPERAND { recognizer->func_header_info_int(".minnctapersm", $2); printf("GPGPU-Sim: Warning: .minnctapersm ignored. \n"); } + | MAXNCTAPERSM_DIRECTIVE INT_OPERAND { recognizer->func_header_info_int(".maxnctapersm", $2); printf("GPGPU-Sim: Warning: .maxnctapersm ignored. \n"); } ; block_spec_list: block_spec | block_spec_list block_spec ; -function_decl: function_decl_header LEFT_PAREN { start_function($1); func_header_info("(");} param_entry RIGHT_PAREN {func_header_info(")");} function_ident_param { $$ = reset_symtab(); } - | function_decl_header { start_function($1); } function_ident_param { $$ = reset_symtab(); } - | function_decl_header { start_function($1); add_function_name(""); g_func_decl=0; $$ = reset_symtab(); } +function_decl: function_decl_header LEFT_PAREN { recognizer->start_function($1); recognizer->func_header_info("(");} param_entry RIGHT_PAREN {recognizer->func_header_info(")");} function_ident_param { $$ = recognizer->reset_symtab(); } + | function_decl_header { recognizer->start_function($1); } function_ident_param { $$ = recognizer->reset_symtab(); } + | function_decl_header { recognizer->start_function($1); recognizer->add_function_name(""); recognizer->g_func_decl=0; $$ = recognizer->reset_symtab(); } ; -function_ident_param: IDENTIFIER { add_function_name($1); } LEFT_PAREN {func_header_info("(");} param_list RIGHT_PAREN { g_func_decl=0; func_header_info(")"); } - | IDENTIFIER { add_function_name($1); g_func_decl=0; } +function_ident_param: IDENTIFIER { recognizer->add_function_name($1); } LEFT_PAREN {recognizer->func_header_info("(");} param_list RIGHT_PAREN { recognizer->g_func_decl=0; recognizer->func_header_info(")"); } + | IDENTIFIER { recognizer->add_function_name($1); recognizer->g_func_decl=0; } ; -function_decl_header: ENTRY_DIRECTIVE { $$ = 1; g_func_decl=1; func_header(".entry"); } - | VISIBLE_DIRECTIVE ENTRY_DIRECTIVE { $$ = 1; g_func_decl=1; func_header(".entry"); } - | WEAK_DIRECTIVE ENTRY_DIRECTIVE { $$ = 1; g_func_decl=1; func_header(".entry"); } - | FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } - | VISIBLE_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } - | WEAK_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } - | EXTERN_DIRECTIVE FUNC_DIRECTIVE { $$ = 2; g_func_decl=1; func_header(".func"); } - | WEAK_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; g_func_decl=1; func_header(".func"); } +function_decl_header: ENTRY_DIRECTIVE { $$ = 1; recognizer->g_func_decl=1; recognizer->func_header(".entry"); } + | VISIBLE_DIRECTIVE ENTRY_DIRECTIVE { $$ = 1; recognizer->g_func_decl=1; recognizer->func_header(".entry"); } + | WEAK_DIRECTIVE ENTRY_DIRECTIVE { $$ = 1; recognizer->g_func_decl=1; recognizer->func_header(".entry"); } + | FUNC_DIRECTIVE { $$ = 0; recognizer->g_func_decl=1; recognizer->func_header(".func"); } + | VISIBLE_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; recognizer->g_func_decl=1; recognizer->func_header(".func"); } + | WEAK_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; recognizer->g_func_decl=1; recognizer->func_header(".func"); } + | EXTERN_DIRECTIVE FUNC_DIRECTIVE { $$ = 2; recognizer->g_func_decl=1; recognizer->func_header(".func"); } + | WEAK_DIRECTIVE FUNC_DIRECTIVE { $$ = 0; recognizer->g_func_decl=1; recognizer->func_header(".func"); } ; param_list: /*empty*/ - | param_entry { add_directive(); } - | param_list COMMA {func_header_info(",");} param_entry { add_directive(); } + | param_entry { recognizer->add_directive(); } + | param_list COMMA {recognizer->func_header_info(",");} param_entry { recognizer->add_directive(); } -param_entry: PARAM_DIRECTIVE { add_space_spec(param_space_unclassified,0); } variable_spec ptr_spec identifier_spec { add_function_arg(); } - | REG_DIRECTIVE { add_space_spec(reg_space,0); } variable_spec identifier_spec { add_function_arg(); } +param_entry: PARAM_DIRECTIVE { recognizer->add_space_spec(param_space_unclassified,0); } variable_spec ptr_spec identifier_spec { recognizer->add_function_arg(); } + | REG_DIRECTIVE { recognizer->add_space_spec(reg_space,0); } variable_spec identifier_spec { recognizer->add_function_arg(); } ptr_spec: /*empty*/ | PTR_DIRECTIVE ptr_space_spec ptr_align_spec | PTR_DIRECTIVE ptr_align_spec -ptr_space_spec: GLOBAL_DIRECTIVE { add_ptr_spec(global_space); } - | LOCAL_DIRECTIVE { add_ptr_spec(local_space); } - | SHARED_DIRECTIVE { add_ptr_spec(shared_space); } - | CONST_DIRECTIVE { add_ptr_spec(global_space); } +ptr_space_spec: GLOBAL_DIRECTIVE { recognizer->add_ptr_spec(global_space); } + | LOCAL_DIRECTIVE { recognizer->add_ptr_spec(local_space); } + | SHARED_DIRECTIVE { recognizer->add_ptr_spec(shared_space); } + | CONST_DIRECTIVE { recognizer->add_ptr_spec(global_space); } ptr_align_spec: ALIGN_DIRECTIVE INT_OPERAND statement_block: LEFT_BRACE statement_list RIGHT_BRACE -statement_list: directive_statement { add_directive(); } - | instruction_statement { add_instruction(); } - | statement_list directive_statement { add_directive(); } - | statement_list instruction_statement { add_instruction(); } - | statement_list {start_inst_group();} statement_block {end_inst_group();} - | {start_inst_group();} statement_block {end_inst_group();} +statement_list: directive_statement { recognizer->add_directive(); } + | instruction_statement { recognizer->add_instruction(); } + | statement_list directive_statement { recognizer->add_directive(); } + | statement_list instruction_statement { recognizer->add_instruction(); } + | statement_list {recognizer->start_inst_group();} statement_block {recognizer->end_inst_group();} + | {recognizer->start_inst_group();} statement_block {recognizer->end_inst_group();} ; directive_statement: variable_declaration SEMI_COLON - | VERSION_DIRECTIVE DOUBLE_OPERAND { add_version_info($2, 0); } - | VERSION_DIRECTIVE DOUBLE_OPERAND PLUS { add_version_info($2,1); } + | VERSION_DIRECTIVE DOUBLE_OPERAND { recognizer->add_version_info($2, 0); } + | VERSION_DIRECTIVE DOUBLE_OPERAND PLUS { recognizer->add_version_info($2,1); } | ADDRESS_SIZE_DIRECTIVE INT_OPERAND {/*Do nothing*/} - | TARGET_DIRECTIVE IDENTIFIER COMMA IDENTIFIER { target_header2($2,$4); } - | TARGET_DIRECTIVE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER { target_header3($2,$4,$6); } - | TARGET_DIRECTIVE IDENTIFIER { target_header($2); } - | FILE_DIRECTIVE INT_OPERAND STRING { add_file($2,$3); } - | FILE_DIRECTIVE INT_OPERAND STRING COMMA INT_OPERAND COMMA INT_OPERAND { add_file($2,$3); } + | TARGET_DIRECTIVE IDENTIFIER COMMA IDENTIFIER { recognizer->target_header2($2,$4); } + | TARGET_DIRECTIVE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER { recognizer->target_header3($2,$4,$6); } + | TARGET_DIRECTIVE IDENTIFIER { recognizer->target_header($2); } + | FILE_DIRECTIVE INT_OPERAND STRING { recognizer->add_file($2,$3); } + | FILE_DIRECTIVE INT_OPERAND STRING COMMA INT_OPERAND COMMA INT_OPERAND { recognizer->add_file($2,$3); } | LOC_DIRECTIVE INT_OPERAND INT_OPERAND INT_OPERAND - | PRAGMA_DIRECTIVE STRING SEMI_COLON { add_pragma($2); } + | PRAGMA_DIRECTIVE STRING SEMI_COLON { recognizer->add_pragma($2); } | function_decl SEMI_COLON {/*Do nothing*/} ; -variable_declaration: variable_spec identifier_list { add_variables(); } - | variable_spec identifier_spec EQUALS initializer_list { add_variables(); } - | variable_spec identifier_spec EQUALS literal_operand { add_variables(); } - | CONSTPTR_DIRECTIVE IDENTIFIER COMMA IDENTIFIER COMMA INT_OPERAND { add_constptr($2, $4, $6); } +variable_declaration: variable_spec identifier_list { recognizer->add_variables(); } + | variable_spec identifier_spec EQUALS initializer_list { recognizer->add_variables(); } + | variable_spec identifier_spec EQUALS literal_operand { recognizer->add_variables(); } + | CONSTPTR_DIRECTIVE IDENTIFIER COMMA IDENTIFIER COMMA INT_OPERAND { recognizer->add_constptr($2, $4, $6); } ; -variable_spec: var_spec_list { set_variable_type(); } +variable_spec: var_spec_list { recognizer->set_variable_type(); } identifier_list: identifier_spec | identifier_list COMMA identifier_spec; -identifier_spec: IDENTIFIER { add_identifier($1,0,NON_ARRAY_IDENTIFIER); func_header_info($1);} - | IDENTIFIER LEFT_ANGLE_BRACKET INT_OPERAND RIGHT_ANGLE_BRACKET { func_header_info($1); func_header_info_int("<", $3); func_header_info(">"); +identifier_spec: IDENTIFIER { recognizer->add_identifier($1,0,NON_ARRAY_IDENTIFIER); recognizer->func_header_info($1);} + | IDENTIFIER LEFT_ANGLE_BRACKET INT_OPERAND RIGHT_ANGLE_BRACKET { recognizer->func_header_info($1); recognizer->func_header_info_int("<", $3); recognizer->func_header_info(">"); int i,lbase,l; char *id = NULL; lbase = strlen($1); @@ -328,12 +338,12 @@ identifier_spec: IDENTIFIER { add_identifier($1,0,NON_ARRAY_IDENTIFIER); func_he l = lbase + (int)log10(i+1)+10; id = (char*) malloc(l); snprintf(id,l,"%s%u",$1,i); - add_identifier(id,0,NON_ARRAY_IDENTIFIER); + recognizer->add_identifier(id,0,NON_ARRAY_IDENTIFIER); } free($1); } - | IDENTIFIER LEFT_SQUARE_BRACKET RIGHT_SQUARE_BRACKET { add_identifier($1,0,ARRAY_IDENTIFIER_NO_DIM); func_header_info($1); func_header_info("["); func_header_info("]");} - | IDENTIFIER LEFT_SQUARE_BRACKET INT_OPERAND RIGHT_SQUARE_BRACKET { add_identifier($1,$3,ARRAY_IDENTIFIER); func_header_info($1); func_header_info_int("[",$3); func_header_info("]");} + | IDENTIFIER LEFT_SQUARE_BRACKET RIGHT_SQUARE_BRACKET { recognizer->add_identifier($1,0,ARRAY_IDENTIFIER_NO_DIM); recognizer->func_header_info($1); recognizer->func_header_info("["); recognizer->func_header_info("]");} + | IDENTIFIER LEFT_SQUARE_BRACKET INT_OPERAND RIGHT_SQUARE_BRACKET { recognizer->add_identifier($1,$3,ARRAY_IDENTIFIER); recognizer->func_header_info($1); recognizer->func_header_info_int("[",$3); recognizer->func_header_info("]");} ; var_spec_list: var_spec @@ -343,93 +353,93 @@ var_spec: space_spec | type_spec | align_spec | VISIBLE_DIRECTIVE - | EXTERN_DIRECTIVE { add_extern_spec(); } + | EXTERN_DIRECTIVE { recognizer->add_extern_spec(); } | WEAK_DIRECTIVE ; -align_spec: ALIGN_DIRECTIVE INT_OPERAND { add_alignment_spec($2); } +align_spec: ALIGN_DIRECTIVE INT_OPERAND { recognizer->add_alignment_spec($2); } -space_spec: REG_DIRECTIVE { add_space_spec(reg_space,0); } - | SREG_DIRECTIVE { add_space_spec(reg_space,0); } +space_spec: REG_DIRECTIVE { recognizer->add_space_spec(reg_space,0); } + | SREG_DIRECTIVE { recognizer->add_space_spec(reg_space,0); } | addressable_spec ; -addressable_spec: CONST_DIRECTIVE { add_space_spec(const_space,$1); } - | GLOBAL_DIRECTIVE { add_space_spec(global_space,0); } - | LOCAL_DIRECTIVE { add_space_spec(local_space,0); } - | PARAM_DIRECTIVE { add_space_spec(param_space_unclassified,0); } - | SHARED_DIRECTIVE { add_space_spec(shared_space,0); } - | SSTARR_DIRECTIVE { add_space_spec(sstarr_space,0); } - | SURF_DIRECTIVE { add_space_spec(surf_space,0); } - | TEX_DIRECTIVE { add_space_spec(tex_space,0); } +addressable_spec: CONST_DIRECTIVE { recognizer->add_space_spec(const_space,$1); } + | GLOBAL_DIRECTIVE { recognizer->add_space_spec(global_space,0); } + | LOCAL_DIRECTIVE { recognizer->add_space_spec(local_space,0); } + | PARAM_DIRECTIVE { recognizer->add_space_spec(param_space_unclassified,0); } + | SHARED_DIRECTIVE { recognizer->add_space_spec(shared_space,0); } + | SSTARR_DIRECTIVE { recognizer->add_space_spec(sstarr_space,0); } + | SURF_DIRECTIVE { recognizer->add_space_spec(surf_space,0); } + | TEX_DIRECTIVE { recognizer->add_space_spec(tex_space,0); } ; type_spec: scalar_type | vector_spec scalar_type ; -vector_spec: V2_TYPE { add_option(V2_TYPE); func_header_info(".v2");} - | V3_TYPE { add_option(V3_TYPE); func_header_info(".v3");} - | V4_TYPE { add_option(V4_TYPE); func_header_info(".v4");} +vector_spec: V2_TYPE { recognizer->add_option(V2_TYPE); recognizer->func_header_info(".v2");} + | V3_TYPE { recognizer->add_option(V3_TYPE); recognizer->func_header_info(".v3");} + | V4_TYPE { recognizer->add_option(V4_TYPE); recognizer->func_header_info(".v4");} ; -scalar_type: S8_TYPE { add_scalar_type_spec( S8_TYPE ); } - | S16_TYPE { add_scalar_type_spec( S16_TYPE ); } - | S32_TYPE { add_scalar_type_spec( S32_TYPE ); } - | S64_TYPE { add_scalar_type_spec( S64_TYPE ); } - | U8_TYPE { add_scalar_type_spec( U8_TYPE ); } - | U16_TYPE { add_scalar_type_spec( U16_TYPE ); } - | U32_TYPE { add_scalar_type_spec( U32_TYPE ); } - | U64_TYPE { add_scalar_type_spec( U64_TYPE ); } - | F16_TYPE { add_scalar_type_spec( F16_TYPE ); } - | F32_TYPE { add_scalar_type_spec( F32_TYPE ); } - | F64_TYPE { add_scalar_type_spec( F64_TYPE ); } - | FF64_TYPE { add_scalar_type_spec( FF64_TYPE ); } - | B8_TYPE { add_scalar_type_spec( B8_TYPE ); } - | B16_TYPE { add_scalar_type_spec( B16_TYPE ); } - | B32_TYPE { add_scalar_type_spec( B32_TYPE ); } - | B64_TYPE { add_scalar_type_spec( B64_TYPE ); } - | BB64_TYPE { add_scalar_type_spec( BB64_TYPE ); } - | BB128_TYPE { add_scalar_type_spec( BB128_TYPE ); } - | PRED_TYPE { add_scalar_type_spec( PRED_TYPE ); } - | TEXREF_TYPE { add_scalar_type_spec( TEXREF_TYPE ); } - | SAMPLERREF_TYPE { add_scalar_type_spec( SAMPLERREF_TYPE ); } - | SURFREF_TYPE { add_scalar_type_spec( SURFREF_TYPE ); } +scalar_type: S8_TYPE { recognizer->add_scalar_type_spec( S8_TYPE ); } + | S16_TYPE { recognizer->add_scalar_type_spec( S16_TYPE ); } + | S32_TYPE { recognizer->add_scalar_type_spec( S32_TYPE ); } + | S64_TYPE { recognizer->add_scalar_type_spec( S64_TYPE ); } + | U8_TYPE { recognizer->add_scalar_type_spec( U8_TYPE ); } + | U16_TYPE { recognizer->add_scalar_type_spec( U16_TYPE ); } + | U32_TYPE { recognizer->add_scalar_type_spec( U32_TYPE ); } + | U64_TYPE { recognizer->add_scalar_type_spec( U64_TYPE ); } + | F16_TYPE { recognizer->add_scalar_type_spec( F16_TYPE ); } + | F32_TYPE { recognizer->add_scalar_type_spec( F32_TYPE ); } + | F64_TYPE { recognizer->add_scalar_type_spec( F64_TYPE ); } + | FF64_TYPE { recognizer->add_scalar_type_spec( FF64_TYPE ); } + | B8_TYPE { recognizer->add_scalar_type_spec( B8_TYPE ); } + | B16_TYPE { recognizer->add_scalar_type_spec( B16_TYPE ); } + | B32_TYPE { recognizer->add_scalar_type_spec( B32_TYPE ); } + | B64_TYPE { recognizer->add_scalar_type_spec( B64_TYPE ); } + | BB64_TYPE { recognizer->add_scalar_type_spec( BB64_TYPE ); } + | BB128_TYPE { recognizer->add_scalar_type_spec( BB128_TYPE ); } + | PRED_TYPE { recognizer->add_scalar_type_spec( PRED_TYPE ); } + | TEXREF_TYPE { recognizer->add_scalar_type_spec( TEXREF_TYPE ); } + | SAMPLERREF_TYPE { recognizer->add_scalar_type_spec( SAMPLERREF_TYPE ); } + | SURFREF_TYPE { recognizer->add_scalar_type_spec( SURFREF_TYPE ); } ; -initializer_list: LEFT_BRACE literal_list RIGHT_BRACE { add_array_initializer(); } - | LEFT_BRACE initializer_list RIGHT_BRACE { syntax_not_implemented(); } +initializer_list: LEFT_BRACE literal_list RIGHT_BRACE { recognizer->add_array_initializer(); } + | LEFT_BRACE initializer_list RIGHT_BRACE { syntax_not_implemented(scanner, recognizer); } literal_list: literal_operand | literal_list COMMA literal_operand; instruction_statement: instruction SEMI_COLON - | IDENTIFIER COLON { add_label($1); } + | IDENTIFIER COLON { recognizer->add_label($1); } | pred_spec instruction SEMI_COLON; -instruction: opcode_spec LEFT_PAREN operand RIGHT_PAREN { set_return(); } COMMA operand COMMA LEFT_PAREN operand_list RIGHT_PAREN +instruction: opcode_spec LEFT_PAREN operand RIGHT_PAREN { recognizer->set_return(); } COMMA operand COMMA LEFT_PAREN operand_list RIGHT_PAREN | opcode_spec operand COMMA LEFT_PAREN operand_list RIGHT_PAREN | opcode_spec operand COMMA LEFT_PAREN RIGHT_PAREN | opcode_spec operand_list | opcode_spec ; -opcode_spec: OPCODE { add_opcode($1); } option_list - | OPCODE { add_opcode($1); } +opcode_spec: OPCODE { recognizer->add_opcode($1); } option_list + | OPCODE { recognizer->add_opcode($1); } -pred_spec: PRED IDENTIFIER { add_pred($2,0, -1); } - | PRED EXCLAMATION IDENTIFIER { add_pred($3,1, -1); } - | PRED IDENTIFIER LT_OPTION { add_pred($2,0,1); } - | PRED IDENTIFIER EQ_OPTION { add_pred($2,0,2); } - | PRED IDENTIFIER LE_OPTION { add_pred($2,0,3); } - | PRED IDENTIFIER NE_OPTION { add_pred($2,0,5); } - | PRED IDENTIFIER GE_OPTION { add_pred($2,0,6); } - | PRED IDENTIFIER EQU_OPTION { add_pred($2,0,10); } - | PRED IDENTIFIER GTU_OPTION { add_pred($2,0,12); } - | PRED IDENTIFIER NEU_OPTION { add_pred($2,0,13); } - | PRED IDENTIFIER CF_OPTION { add_pred($2,0,17); } - | PRED IDENTIFIER SF_OPTION { add_pred($2,0,19); } - | PRED IDENTIFIER NSF_OPTION { add_pred($2,0,28); } +pred_spec: PRED IDENTIFIER { recognizer->add_pred($2,0, -1); } + | PRED EXCLAMATION IDENTIFIER { recognizer->add_pred($3,1, -1); } + | PRED IDENTIFIER LT_OPTION { recognizer->add_pred($2,0,1); } + | PRED IDENTIFIER EQ_OPTION { recognizer->add_pred($2,0,2); } + | PRED IDENTIFIER LE_OPTION { recognizer->add_pred($2,0,3); } + | PRED IDENTIFIER NE_OPTION { recognizer->add_pred($2,0,5); } + | PRED IDENTIFIER GE_OPTION { recognizer->add_pred($2,0,6); } + | PRED IDENTIFIER EQU_OPTION { recognizer->add_pred($2,0,10); } + | PRED IDENTIFIER GTU_OPTION { recognizer->add_pred($2,0,12); } + | PRED IDENTIFIER NEU_OPTION { recognizer->add_pred($2,0,13); } + | PRED IDENTIFIER CF_OPTION { recognizer->add_pred($2,0,17); } + | PRED IDENTIFIER SF_OPTION { recognizer->add_pred($2,0,19); } + | PRED IDENTIFIER NSF_OPTION { recognizer->add_pred($2,0,28); } ; option_list: option @@ -441,108 +451,108 @@ option: type_spec | rounding_mode | wmma_spec | prmt_spec - | SYNC_OPTION { add_option(SYNC_OPTION); } - | ARRIVE_OPTION { add_option(ARRIVE_OPTION); } - | RED_OPTION { add_option(RED_OPTION); } - | UNI_OPTION { add_option(UNI_OPTION); } - | WIDE_OPTION { add_option(WIDE_OPTION); } - | ANY_OPTION { add_option(ANY_OPTION); } - | ALL_OPTION { add_option(ALL_OPTION); } - | BALLOT_OPTION { add_option(BALLOT_OPTION); } - | GLOBAL_OPTION { add_option(GLOBAL_OPTION); } - | CTA_OPTION { add_option(CTA_OPTION); } - | SYS_OPTION { add_option(SYS_OPTION); } - | GEOM_MODIFIER_1D { add_option(GEOM_MODIFIER_1D); } - | GEOM_MODIFIER_2D { add_option(GEOM_MODIFIER_2D); } - | GEOM_MODIFIER_3D { add_option(GEOM_MODIFIER_3D); } - | SAT_OPTION { add_option(SAT_OPTION); } - | FTZ_OPTION { add_option(FTZ_OPTION); } - | NEG_OPTION { add_option(NEG_OPTION); } - | APPROX_OPTION { add_option(APPROX_OPTION); } - | FULL_OPTION { add_option(FULL_OPTION); } - | EXIT_OPTION { add_option(EXIT_OPTION); } - | ABS_OPTION { add_option(ABS_OPTION); } + | SYNC_OPTION { recognizer->add_option(SYNC_OPTION); } + | ARRIVE_OPTION { recognizer->add_option(ARRIVE_OPTION); } + | RED_OPTION { recognizer->add_option(RED_OPTION); } + | UNI_OPTION { recognizer->add_option(UNI_OPTION); } + | WIDE_OPTION { recognizer->add_option(WIDE_OPTION); } + | ANY_OPTION { recognizer->add_option(ANY_OPTION); } + | ALL_OPTION { recognizer->add_option(ALL_OPTION); } + | BALLOT_OPTION { recognizer->add_option(BALLOT_OPTION); } + | GLOBAL_OPTION { recognizer->add_option(GLOBAL_OPTION); } + | CTA_OPTION { recognizer->add_option(CTA_OPTION); } + | SYS_OPTION { recognizer->add_option(SYS_OPTION); } + | GEOM_MODIFIER_1D { recognizer->add_option(GEOM_MODIFIER_1D); } + | GEOM_MODIFIER_2D { recognizer->add_option(GEOM_MODIFIER_2D); } + | GEOM_MODIFIER_3D { recognizer->add_option(GEOM_MODIFIER_3D); } + | SAT_OPTION { recognizer->add_option(SAT_OPTION); } + | FTZ_OPTION { recognizer->add_option(FTZ_OPTION); } + | NEG_OPTION { recognizer->add_option(NEG_OPTION); } + | APPROX_OPTION { recognizer->add_option(APPROX_OPTION); } + | FULL_OPTION { recognizer->add_option(FULL_OPTION); } + | EXIT_OPTION { recognizer->add_option(EXIT_OPTION); } + | ABS_OPTION { recognizer->add_option(ABS_OPTION); } | atomic_operation_spec ; - | TO_OPTION { add_option(TO_OPTION); } - | HALF_OPTION { add_option(HALF_OPTION); } - | EXTP_OPTION { add_option(EXTP_OPTION); } - | CA_OPTION { add_option(CA_OPTION); } - | CG_OPTION { add_option(CG_OPTION); } - | CS_OPTION { add_option(CS_OPTION); } - | LU_OPTION { add_option(LU_OPTION); } - | CV_OPTION { add_option(CV_OPTION); } - | WB_OPTION { add_option(WB_OPTION); } - | WT_OPTION { add_option(WT_OPTION); } - | NC_OPTION { add_option(NC_OPTION); } - | UP_OPTION { add_option(UP_OPTION); } - | DOWN_OPTION { add_option(DOWN_OPTION); } - | BFLY_OPTION { add_option(BFLY_OPTION); } - | IDX_OPTION { add_option(IDX_OPTION); } + | TO_OPTION { recognizer->add_option(TO_OPTION); } + | HALF_OPTION { recognizer->add_option(HALF_OPTION); } + | EXTP_OPTION { recognizer->add_option(EXTP_OPTION); } + | CA_OPTION { recognizer->add_option(CA_OPTION); } + | CG_OPTION { recognizer->add_option(CG_OPTION); } + | CS_OPTION { recognizer->add_option(CS_OPTION); } + | LU_OPTION { recognizer->add_option(LU_OPTION); } + | CV_OPTION { recognizer->add_option(CV_OPTION); } + | WB_OPTION { recognizer->add_option(WB_OPTION); } + | WT_OPTION { recognizer->add_option(WT_OPTION); } + | NC_OPTION { recognizer->add_option(NC_OPTION); } + | UP_OPTION { recognizer->add_option(UP_OPTION); } + | DOWN_OPTION { recognizer->add_option(DOWN_OPTION); } + | BFLY_OPTION { recognizer->add_option(BFLY_OPTION); } + | IDX_OPTION { recognizer->add_option(IDX_OPTION); } ; -atomic_operation_spec: ATOMIC_AND { add_option(ATOMIC_AND); } - | ATOMIC_POPC { add_option(ATOMIC_POPC); } - | ATOMIC_OR { add_option(ATOMIC_OR); } - | ATOMIC_XOR { add_option(ATOMIC_XOR); } - | ATOMIC_CAS { add_option(ATOMIC_CAS); } - | ATOMIC_EXCH { add_option(ATOMIC_EXCH); } - | ATOMIC_ADD { add_option(ATOMIC_ADD); } - | ATOMIC_INC { add_option(ATOMIC_INC); } - | ATOMIC_DEC { add_option(ATOMIC_DEC); } - | ATOMIC_MIN { add_option(ATOMIC_MIN); } - | ATOMIC_MAX { add_option(ATOMIC_MAX); } +atomic_operation_spec: ATOMIC_AND { recognizer->add_option(ATOMIC_AND); } + | ATOMIC_POPC { recognizer->add_option(ATOMIC_POPC); } + | ATOMIC_OR { recognizer->add_option(ATOMIC_OR); } + | ATOMIC_XOR { recognizer->add_option(ATOMIC_XOR); } + | ATOMIC_CAS { recognizer->add_option(ATOMIC_CAS); } + | ATOMIC_EXCH { recognizer->add_option(ATOMIC_EXCH); } + | ATOMIC_ADD { recognizer->add_option(ATOMIC_ADD); } + | ATOMIC_INC { recognizer->add_option(ATOMIC_INC); } + | ATOMIC_DEC { recognizer->add_option(ATOMIC_DEC); } + | ATOMIC_MIN { recognizer->add_option(ATOMIC_MIN); } + | ATOMIC_MAX { recognizer->add_option(ATOMIC_MAX); } ; rounding_mode: floating_point_rounding_mode | integer_rounding_mode; -floating_point_rounding_mode: RN_OPTION { add_option(RN_OPTION); } - | RZ_OPTION { add_option(RZ_OPTION); } - | RM_OPTION { add_option(RM_OPTION); } - | RP_OPTION { add_option(RP_OPTION); } +floating_point_rounding_mode: RN_OPTION { recognizer->add_option(RN_OPTION); } + | RZ_OPTION { recognizer->add_option(RZ_OPTION); } + | RM_OPTION { recognizer->add_option(RM_OPTION); } + | RP_OPTION { recognizer->add_option(RP_OPTION); } ; -integer_rounding_mode: RNI_OPTION { add_option(RNI_OPTION); } - | RZI_OPTION { add_option(RZI_OPTION); } - | RMI_OPTION { add_option(RMI_OPTION); } - | RPI_OPTION { add_option(RPI_OPTION); } +integer_rounding_mode: RNI_OPTION { recognizer->add_option(RNI_OPTION); } + | RZI_OPTION { recognizer->add_option(RZI_OPTION); } + | RMI_OPTION { recognizer->add_option(RMI_OPTION); } + | RPI_OPTION { recognizer->add_option(RPI_OPTION); } ; -compare_spec:EQ_OPTION { add_option(EQ_OPTION); } - | NE_OPTION { add_option(NE_OPTION); } - | LT_OPTION { add_option(LT_OPTION); } - | LE_OPTION { add_option(LE_OPTION); } - | GT_OPTION { add_option(GT_OPTION); } - | GE_OPTION { add_option(GE_OPTION); } - | LO_OPTION { add_option(LO_OPTION); } - | LS_OPTION { add_option(LS_OPTION); } - | HI_OPTION { add_option(HI_OPTION); } - | HS_OPTION { add_option(HS_OPTION); } - | EQU_OPTION { add_option(EQU_OPTION); } - | NEU_OPTION { add_option(NEU_OPTION); } - | LTU_OPTION { add_option(LTU_OPTION); } - | LEU_OPTION { add_option(LEU_OPTION); } - | GTU_OPTION { add_option(GTU_OPTION); } - | GEU_OPTION { add_option(GEU_OPTION); } - | NUM_OPTION { add_option(NUM_OPTION); } - | NAN_OPTION { add_option(NAN_OPTION); } +compare_spec:EQ_OPTION { recognizer->add_option(EQ_OPTION); } + | NE_OPTION { recognizer->add_option(NE_OPTION); } + | LT_OPTION { recognizer->add_option(LT_OPTION); } + | LE_OPTION { recognizer->add_option(LE_OPTION); } + | GT_OPTION { recognizer->add_option(GT_OPTION); } + | GE_OPTION { recognizer->add_option(GE_OPTION); } + | LO_OPTION { recognizer->add_option(LO_OPTION); } + | LS_OPTION { recognizer->add_option(LS_OPTION); } + | HI_OPTION { recognizer->add_option(HI_OPTION); } + | HS_OPTION { recognizer->add_option(HS_OPTION); } + | EQU_OPTION { recognizer->add_option(EQU_OPTION); } + | NEU_OPTION { recognizer->add_option(NEU_OPTION); } + | LTU_OPTION { recognizer->add_option(LTU_OPTION); } + | LEU_OPTION { recognizer->add_option(LEU_OPTION); } + | GTU_OPTION { recognizer->add_option(GTU_OPTION); } + | GEU_OPTION { recognizer->add_option(GEU_OPTION); } + | NUM_OPTION { recognizer->add_option(NUM_OPTION); } + | NAN_OPTION { recognizer->add_option(NAN_OPTION); } ; -prmt_spec: PRMT_F4E_MODE { add_option( PRMT_F4E_MODE); } - | PRMT_B4E_MODE { add_option( PRMT_B4E_MODE); } - | PRMT_RC8_MODE { add_option( PRMT_RC8_MODE); } - | PRMT_RC16_MODE{ add_option( PRMT_RC16_MODE);} - | PRMT_ECL_MODE { add_option( PRMT_ECL_MODE); } - | PRMT_ECR_MODE { add_option( PRMT_ECR_MODE); } +prmt_spec: PRMT_F4E_MODE { recognizer->add_option( PRMT_F4E_MODE); } + | PRMT_B4E_MODE { recognizer->add_option( PRMT_B4E_MODE); } + | PRMT_RC8_MODE { recognizer->add_option( PRMT_RC8_MODE); } + | PRMT_RC16_MODE{ recognizer->add_option( PRMT_RC16_MODE);} + | PRMT_ECL_MODE { recognizer->add_option( PRMT_ECL_MODE); } + | PRMT_ECR_MODE { recognizer->add_option( PRMT_ECR_MODE); } ; -wmma_spec: WMMA_DIRECTIVE LAYOUT CONFIGURATION{add_space_spec(global_space,0);add_ptr_spec(global_space); add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);} - | WMMA_DIRECTIVE LAYOUT LAYOUT CONFIGURATION{add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);add_wmma_option($4);} +wmma_spec: WMMA_DIRECTIVE LAYOUT CONFIGURATION{recognizer->add_space_spec(global_space,0);recognizer->add_ptr_spec(global_space); recognizer->add_wmma_option($1);recognizer->add_wmma_option($2);recognizer->add_wmma_option($3);} + | WMMA_DIRECTIVE LAYOUT LAYOUT CONFIGURATION{recognizer->add_wmma_option($1);recognizer->add_wmma_option($2);recognizer->add_wmma_option($3);recognizer->add_wmma_option($4);} ; -vp_spec: WMMA_DIRECTIVE LAYOUT CONFIGURATION{add_space_spec(global_space,0);add_ptr_spec(global_space);add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);} - | WMMA_DIRECTIVE LAYOUT LAYOUT CONFIGURATION{add_wmma_option($1);add_wmma_option($2);add_wmma_option($3);add_wmma_option($4);} +vp_spec: WMMA_DIRECTIVE LAYOUT CONFIGURATION{recognizer->add_space_spec(global_space,0);recognizer->add_ptr_spec(global_space);recognizer->add_wmma_option($1);recognizer->add_wmma_option($2);recognizer->add_wmma_option($3);} + | WMMA_DIRECTIVE LAYOUT LAYOUT CONFIGURATION{recognizer->add_wmma_option($1);recognizer->add_wmma_option($2);recognizer->add_wmma_option($3);recognizer->add_wmma_option($4);} ; @@ -550,80 +560,79 @@ vp_spec: WMMA_DIRECTIVE LAYOUT CONFIGURATION{add_space_spec(global_space,0);add_ operand_list: operand | operand COMMA operand_list; -operand: IDENTIFIER { add_scalar_operand( $1 ); } - | EXCLAMATION IDENTIFIER { add_neg_pred_operand( $2 ); } - | MINUS IDENTIFIER { add_scalar_operand( $2 ); change_operand_neg(); } +operand: IDENTIFIER { recognizer->add_scalar_operand( $1 ); } + | EXCLAMATION IDENTIFIER { recognizer->add_neg_pred_operand( $2 ); } + | MINUS IDENTIFIER { recognizer->add_scalar_operand( $2 ); recognizer->change_operand_neg(); } | memory_operand | literal_operand | builtin_operand | vector_operand - | MINUS vector_operand { change_operand_neg(); } + | MINUS vector_operand { recognizer->change_operand_neg(); } | tex_operand - | IDENTIFIER PLUS INT_OPERAND { add_address_operand($1,$3); } - | IDENTIFIER LO_OPTION { add_scalar_operand( $1 ); change_operand_lohi(1);} - | MINUS IDENTIFIER LO_OPTION { add_scalar_operand( $2 ); change_operand_lohi(1); change_operand_neg();} - | IDENTIFIER HI_OPTION { add_scalar_operand( $1 ); change_operand_lohi(2);} - | MINUS IDENTIFIER HI_OPTION { add_scalar_operand( $2 ); change_operand_lohi(2); change_operand_neg();} - | IDENTIFIER PIPE IDENTIFIER { add_2vector_operand($1,$3); change_double_operand_type(-1);} - | IDENTIFIER PIPE IDENTIFIER LO_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-1); change_operand_lohi(1);} - | IDENTIFIER PIPE IDENTIFIER HI_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-1); change_operand_lohi(2);} - | IDENTIFIER BACKSLASH IDENTIFIER { add_2vector_operand($1,$3); change_double_operand_type(-3);} - | IDENTIFIER BACKSLASH IDENTIFIER LO_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-3); change_operand_lohi(1);} - | IDENTIFIER BACKSLASH IDENTIFIER HI_OPTION { add_2vector_operand($1,$3); change_double_operand_type(-3); change_operand_lohi(2);} + | IDENTIFIER PLUS INT_OPERAND { recognizer->add_address_operand($1,$3); } + | IDENTIFIER LO_OPTION { recognizer->add_scalar_operand( $1 ); recognizer->change_operand_lohi(1);} + | MINUS IDENTIFIER LO_OPTION { recognizer->add_scalar_operand( $2 ); recognizer->change_operand_lohi(1); recognizer->change_operand_neg();} + | IDENTIFIER HI_OPTION { recognizer->add_scalar_operand( $1 ); recognizer->change_operand_lohi(2);} + | MINUS IDENTIFIER HI_OPTION { recognizer->add_scalar_operand( $2 ); recognizer->change_operand_lohi(2); recognizer->change_operand_neg();} + | IDENTIFIER PIPE IDENTIFIER { recognizer->add_2vector_operand($1,$3); recognizer->change_double_operand_type(-1);} + | IDENTIFIER PIPE IDENTIFIER LO_OPTION { recognizer->add_2vector_operand($1,$3); recognizer->change_double_operand_type(-1); recognizer->change_operand_lohi(1);} + | IDENTIFIER PIPE IDENTIFIER HI_OPTION { recognizer->add_2vector_operand($1,$3); recognizer->change_double_operand_type(-1); recognizer->change_operand_lohi(2);} + | IDENTIFIER BACKSLASH IDENTIFIER { recognizer->add_2vector_operand($1,$3); recognizer->change_double_operand_type(-3);} + | IDENTIFIER BACKSLASH IDENTIFIER LO_OPTION { recognizer->add_2vector_operand($1,$3); recognizer->change_double_operand_type(-3); recognizer->change_operand_lohi(1);} + | IDENTIFIER BACKSLASH IDENTIFIER HI_OPTION { recognizer->add_2vector_operand($1,$3); recognizer->change_double_operand_type(-3); recognizer->change_operand_lohi(2);} ; -vector_operand: LEFT_BRACE IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_2vector_operand($2,$4); } - | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_3vector_operand($2,$4,$6); } - | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_4vector_operand($2,$4,$6,$8); } - | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { add_8vector_operand($2,$4,$6,$8,$10,$12,$14,$16); } - | LEFT_BRACE IDENTIFIER RIGHT_BRACE { add_1vector_operand($2); } +vector_operand: LEFT_BRACE IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { recognizer->add_2vector_operand($2,$4); } + | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { recognizer->add_3vector_operand($2,$4,$6); } + | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { recognizer->add_4vector_operand($2,$4,$6,$8); } + | LEFT_BRACE IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER COMMA IDENTIFIER RIGHT_BRACE { recognizer->add_8vector_operand($2,$4,$6,$8,$10,$12,$14,$16); } + | LEFT_BRACE IDENTIFIER RIGHT_BRACE { recognizer->add_1vector_operand($2); } ; -tex_operand: LEFT_SQUARE_BRACKET IDENTIFIER COMMA { add_scalar_operand($2); } +tex_operand: LEFT_SQUARE_BRACKET IDENTIFIER COMMA { recognizer->add_scalar_operand($2); } vector_operand RIGHT_SQUARE_BRACKET ; -builtin_operand: SPECIAL_REGISTER DIMENSION_MODIFIER { add_builtin_operand($1,$2); } - | SPECIAL_REGISTER { add_builtin_operand($1,-1); } +builtin_operand: SPECIAL_REGISTER DIMENSION_MODIFIER { recognizer->add_builtin_operand($1,$2); } + | SPECIAL_REGISTER { recognizer->add_builtin_operand($1,-1); } ; -memory_operand : LEFT_SQUARE_BRACKET address_expression RIGHT_SQUARE_BRACKET { add_memory_operand(); } - | IDENTIFIER LEFT_SQUARE_BRACKET address_expression RIGHT_SQUARE_BRACKET { add_memory_operand(); change_memory_addr_space($1); } - | IDENTIFIER LEFT_SQUARE_BRACKET literal_operand RIGHT_SQUARE_BRACKET { change_memory_addr_space($1); } - | IDENTIFIER LEFT_SQUARE_BRACKET twin_operand RIGHT_SQUARE_BRACKET { change_memory_addr_space($1); add_memory_operand();} - | MINUS memory_operand { change_operand_neg(); } +memory_operand : LEFT_SQUARE_BRACKET address_expression RIGHT_SQUARE_BRACKET { recognizer->add_memory_operand(); } + | IDENTIFIER LEFT_SQUARE_BRACKET address_expression RIGHT_SQUARE_BRACKET { recognizer->add_memory_operand(); recognizer->change_memory_addr_space($1); } + | IDENTIFIER LEFT_SQUARE_BRACKET literal_operand RIGHT_SQUARE_BRACKET { recognizer->change_memory_addr_space($1); } + | IDENTIFIER LEFT_SQUARE_BRACKET twin_operand RIGHT_SQUARE_BRACKET { recognizer->change_memory_addr_space($1); recognizer->add_memory_operand();} + | MINUS memory_operand { recognizer->change_operand_neg(); } ; -twin_operand : IDENTIFIER PLUS IDENTIFIER { add_double_operand($1,$3); change_double_operand_type(1); } - | IDENTIFIER PLUS IDENTIFIER LO_OPTION { add_double_operand($1,$3); change_double_operand_type(1); change_operand_lohi(1); } - | IDENTIFIER PLUS IDENTIFIER HI_OPTION { add_double_operand($1,$3); change_double_operand_type(1); change_operand_lohi(2); } - | IDENTIFIER PLUS EQUALS IDENTIFIER { add_double_operand($1,$4); change_double_operand_type(2); } - | IDENTIFIER PLUS EQUALS IDENTIFIER LO_OPTION { add_double_operand($1,$4); change_double_operand_type(2); change_operand_lohi(1); } - | IDENTIFIER PLUS EQUALS IDENTIFIER HI_OPTION { add_double_operand($1,$4); change_double_operand_type(2); change_operand_lohi(2); } - | IDENTIFIER PLUS EQUALS INT_OPERAND { add_address_operand($1,$4); change_double_operand_type(3); } +twin_operand : IDENTIFIER PLUS IDENTIFIER { recognizer->add_double_operand($1,$3); recognizer->change_double_operand_type(1); } + | IDENTIFIER PLUS IDENTIFIER LO_OPTION { recognizer->add_double_operand($1,$3); recognizer->change_double_operand_type(1); recognizer->change_operand_lohi(1); } + | IDENTIFIER PLUS IDENTIFIER HI_OPTION { recognizer->add_double_operand($1,$3); recognizer->change_double_operand_type(1); recognizer->change_operand_lohi(2); } + | IDENTIFIER PLUS EQUALS IDENTIFIER { recognizer->add_double_operand($1,$4); recognizer->change_double_operand_type(2); } + | IDENTIFIER PLUS EQUALS IDENTIFIER LO_OPTION { recognizer->add_double_operand($1,$4); recognizer->change_double_operand_type(2); recognizer->change_operand_lohi(1); } + | IDENTIFIER PLUS EQUALS IDENTIFIER HI_OPTION { recognizer->add_double_operand($1,$4); recognizer->change_double_operand_type(2); recognizer->change_operand_lohi(2); } + | IDENTIFIER PLUS EQUALS INT_OPERAND { recognizer->add_address_operand($1,$4); recognizer->change_double_operand_type(3); } ; -literal_operand : INT_OPERAND { add_literal_int($1); } - | FLOAT_OPERAND { add_literal_float($1); } - | DOUBLE_OPERAND { add_literal_double($1); } +literal_operand : INT_OPERAND { recognizer->add_literal_int($1); } + | FLOAT_OPERAND { recognizer->add_literal_float($1); } + | DOUBLE_OPERAND { recognizer->add_literal_double($1); } ; -address_expression: IDENTIFIER { add_address_operand($1,0); } - | IDENTIFIER LO_OPTION { add_address_operand($1,0); change_operand_lohi(1);} - | IDENTIFIER HI_OPTION { add_address_operand($1,0); change_operand_lohi(2); } - | IDENTIFIER PLUS INT_OPERAND { add_address_operand($1,$3); } - | INT_OPERAND { add_address_operand2($1); } +address_expression: IDENTIFIER { recognizer->add_address_operand($1,0); } + | IDENTIFIER LO_OPTION { recognizer->add_address_operand($1,0); recognizer->change_operand_lohi(1);} + | IDENTIFIER HI_OPTION { recognizer->add_address_operand($1,0); recognizer->change_operand_lohi(2); } + | IDENTIFIER PLUS INT_OPERAND { recognizer->add_address_operand($1,$3); } + | INT_OPERAND { recognizer->add_address_operand2($1); } ; %% -extern int ptx_lineno; extern const char *g_filename; -void syntax_not_implemented() +void syntax_not_implemented(yyscan_t yyscanner, ptx_recognizer* recognizer) { - printf("Parse error (%s:%u): this syntax is not (yet) implemented:\n",g_filename,ptx_lineno); - ptx_error(NULL); + printf("Parse error (%s): this syntax is not (yet) implemented:\n",g_filename); + ptx_error(yyscanner, recognizer, NULL); abort(); } diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc index 73db24d..8cedf79 100644 --- a/src/cuda-sim/ptx_ir.cc +++ b/src/cuda-sim/ptx_ir.cc @@ -28,6 +28,7 @@ #include "ptx_parser.h" #include "ptx_ir.h" +typedef void * yyscan_t; #include "ptx.tab.h" #include "opcodes.h" #include <stdio.h> diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc index 8c6f361..921d1e6 100644 --- a/src/cuda-sim/ptx_loader.cc +++ b/src/cuda-sim/ptx_loader.cc @@ -33,26 +33,28 @@ #include <dirent.h> #include <fstream> #include <sstream> +#include "../../libcuda/gpgpu_context.h" /// globals -memory_space *g_global_mem; -memory_space *g_tex_mem; -memory_space *g_surf_mem; -memory_space *g_param_mem; bool g_override_embedded_ptx = false; /// extern prototypes -extern int ptx_parse(); -extern int ptx__scan_string(const char*); +extern int ptx_error( yyscan_t yyscanner, const char *s ); +extern int ptx_lex_init(yyscan_t* scanner); +extern void ptx_set_in(FILE * _in_str ,yyscan_t yyscanner ); +extern int ptx_parse(yyscan_t scanner, ptx_recognizer* recognizer); +extern int ptx_lex_destroy(yyscan_t scanner); +extern int ptx__scan_string(const char*, yyscan_t scanner); extern std::map<unsigned,const char*> get_duplicate(); -const char *g_ptxinfo_filename; -extern int ptxinfo_parse(); -extern int ptxinfo_debug; -extern FILE *ptxinfo_in; +typedef void * yyscan_t; +extern int ptxinfo_lex_init(yyscan_t* scanner); +extern void ptxinfo_set_in (FILE * _in_str ,yyscan_t yyscanner ); +extern int ptxinfo_parse(yyscan_t scanner, ptxinfo_data* ptxinfo); +extern int ptxinfo_lex_destroy(yyscan_t scanner); static bool g_save_embedded_ptx; static int g_occupancy_sm_number; @@ -79,7 +81,7 @@ void ptx_reg_options(option_parser_t opp) "0"); } -void print_ptx_file( const char *p, unsigned source_num, const char *filename ) +void gpgpu_context::print_ptx_file( const char *p, unsigned source_num, const char *filename ) { printf("\nGPGPU-Sim PTX: file _%u.ptx contents:\n\n", source_num ); char *s = strdup(p); @@ -158,7 +160,7 @@ char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptxfilenam } -symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ) +symbol_table *gpgpu_context::gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ) { char buf[1024]; snprintf(buf,1024,"_%u.ptx", source_num ); @@ -168,8 +170,9 @@ symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source fclose(fp); } symbol_table *symtab=init_parser(buf); - ptx__scan_string(p); - int errors = ptx_parse (); + ptx_lex_init(&(ptx_parser->scanner)); + ptx__scan_string(p, ptx_parser->scanner); + int errors = ptx_parse (ptx_parser->scanner, ptx_parser); if ( errors ) { char fname[1024]; snprintf(fname,1024,"_ptx_errors_XXXXXX"); @@ -182,6 +185,7 @@ symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source abort(); exit(40); } + ptx_lex_destroy(ptx_parser->scanner); if ( g_debug_execution >= 100 ) print_ptx_file(p,source_num,buf); @@ -190,7 +194,7 @@ symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source return symtab; } -symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ) +symbol_table *gpgpu_context::gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ) { symbol_table *symtab=init_parser(filename); printf("GPGPU-Sim PTX: finished parsing EMBEDDED .ptx file %s\n",filename); @@ -327,7 +331,7 @@ char* get_app_binary_name(){ return self_exe_path; } -void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version) +void gpgpu_context::gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version) { std::string ptxas_filename(std::string(filename) + "as"); char buff[1024], extra_flags[1024]; @@ -346,13 +350,17 @@ void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_versio exit(1); } - g_ptxinfo_filename = strdup(ptxas_filename.c_str()); - ptxinfo_in = fopen(g_ptxinfo_filename,"r"); - ptxinfo_parse(); + FILE *ptxinfo_in; + ptxinfo->g_ptxinfo_filename = strdup(ptxas_filename.c_str()); + ptxinfo_in = fopen(ptxinfo->g_ptxinfo_filename,"r"); + ptxinfo_lex_init(&(ptxinfo->scanner)); + ptxinfo_set_in(ptxinfo_in, ptxinfo->scanner); + ptxinfo_parse(ptxinfo->scanner, ptxinfo); + ptxinfo_lex_destroy(ptxinfo->scanner); fclose(ptxinfo_in); } -void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version ) +void gpgpu_context::gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version, int no_of_ptx ) { //do ptxas for individual files instead of one big embedded ptx. This prevents the duplicate defs and declarations. char ptx_file[1000]; @@ -410,9 +418,14 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num if( result != 0 ) { // 65280 = duplicate errors if (result == 65280) { + FILE *ptxinfo_in; ptxinfo_in = fopen(tempfile_ptxinfo,"r"); - g_ptxinfo_filename = tempfile_ptxinfo; - ptxinfo_parse(); + ptxinfo->g_ptxinfo_filename = tempfile_ptxinfo; + ptxinfo_lex_init(&(ptxinfo->scanner)); + ptxinfo_set_in(ptxinfo_in, ptxinfo->scanner); + ptxinfo_parse(ptxinfo->scanner, ptxinfo); + ptxinfo_lex_destroy(ptxinfo->scanner); + fclose(ptxinfo_in); fix_duplicate_errors(fname2); snprintf(commandline,1024,"$CUDA_INSTALL_PATH/bin/ptxas %s -v %s --output-file /dev/null 2> %s", @@ -492,12 +505,17 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num } if(no_of_ptx>0) - g_ptxinfo_filename = final_tempfile_ptxinfo; + ptxinfo->g_ptxinfo_filename = final_tempfile_ptxinfo; else - g_ptxinfo_filename = tempfile_ptxinfo; - ptxinfo_in = fopen(g_ptxinfo_filename,"r"); + ptxinfo->g_ptxinfo_filename = tempfile_ptxinfo; + FILE *ptxinfo_in; + ptxinfo_in = fopen(ptxinfo->g_ptxinfo_filename,"r"); - ptxinfo_parse(); + ptxinfo_lex_init(&(ptxinfo->scanner)); + ptxinfo_set_in(ptxinfo_in, ptxinfo->scanner); + ptxinfo_parse(ptxinfo->scanner, ptxinfo); + ptxinfo_lex_destroy(ptxinfo->scanner); + fclose(ptxinfo_in); snprintf(commandline,1024,"rm -f *info"); if( system(commandline) != 0 ) { @@ -515,4 +533,4 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num exit(1); } } -}
\ No newline at end of file +} diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h index e5df6a9..ee09e16 100644 --- a/src/cuda-sim/ptx_loader.h +++ b/src/cuda-sim/ptx_loader.h @@ -29,13 +29,21 @@ #define PTX_LOADER_H_INCLUDED #include <string> +#define PTXINFO_LINEBUF_SIZE 1024 +typedef void * yyscan_t; +class ptxinfo_data{ + public: + yyscan_t scanner; + char linebuf[PTXINFO_LINEBUF_SIZE]; + unsigned col; + const char *g_ptxinfo_filename; + +}; + + extern bool g_override_embedded_ptx; extern int no_of_ptx; //counter to track number of ptx files to be extracted in an application. -class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); -class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); -void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20 ); -void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version ); char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptx_str, const std::string sass_str, const std::string elf_str); bool keep_intermediate_files(); diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc index 25758dd..2872b84 100644 --- a/src/cuda-sim/ptx_parser.cc +++ b/src/cuda-sim/ptx_parser.cc @@ -27,13 +27,20 @@ #include "ptx_parser.h" #include "ptx_ir.h" +#include "../../libcuda/gpgpu_context.h" + +typedef void * yyscan_t; #include "ptx.tab.h" #include <stdarg.h> -extern int ptx_error( const char *s ); -extern int ptx_lineno; -extern int ptx_parse(); -extern FILE *ptx_in; +extern int ptx_get_lineno (yyscan_t yyscanner ); +extern YYSTYPE* ptx_get_lval (yyscan_t yyscanner ); +extern int ptx_error( yyscan_t yyscanner, const char *s ); +extern int ptx_lex_init(yyscan_t* scanner); +extern void ptx_set_in(FILE * _in_str ,yyscan_t yyscanner ); +extern FILE *ptx_get_in (yyscan_t yyscanner ); +extern int ptx_parse(yyscan_t scanner, ptx_recognizer* recognizer); +extern int ptx_lex_destroy(yyscan_t scanner); static const struct core_config *g_shader_core_config; void set_ptx_warp_size(const struct core_config * warp_size) @@ -43,54 +50,20 @@ void set_ptx_warp_size(const struct core_config * warp_size) static bool g_debug_ir_generation=false; const char *g_filename; -unsigned g_max_regs_per_thread = 0; // the program intermediate representation... -static symbol_table *g_global_allfiles_symbol_table = NULL; -static symbol_table *g_global_symbol_table = NULL; std::map<std::string,symbol_table*> g_sym_name_to_symbol_table; -static symbol_table *g_current_symbol_table = NULL; -static std::list<ptx_instruction*> g_instructions; -static symbol *g_last_symbol = NULL; - -int g_error_detected = 0; - -// type specifier stuff: -memory_space_t g_space_spec = undefined_space; -memory_space_t g_ptr_spec = undefined_space; -int g_scalar_type_spec = -1; -int g_vector_spec = -1; -int g_alignment_spec = -1; -int g_size = -1; -int g_extern_spec = 0; - -// variable declaration stuff: -type_info *g_var_type = NULL; - -// instruction definition stuff: -const symbol *g_pred; -int g_neg_pred; -int g_pred_mod; -symbol *g_label; -int g_opcode = -1; -std::list<operand_info> g_operands; -std::list<int> g_options; -std::list<int> g_wmma_options; -std::list<int> g_scalar_type; #define PTX_PARSE_DPRINTF(...) \ if( g_debug_ir_generation ) { \ - printf(" %s:%u => ",g_filename,ptx_lineno); \ + printf(" %s:%u => ",g_filename,ptx_get_lineno(scanner)); \ printf(" (%s:%u) ", __FILE__, __LINE__); \ printf(__VA_ARGS__); \ printf("\n"); \ fflush(stdout); \ } -static unsigned g_entry_func_param_index=0; -static function_info *g_func_info = NULL; static std::map<unsigned,std::string> g_ptx_token_decode; -static operand_info g_return_var; const char *decode_token( int type ) { @@ -109,7 +82,7 @@ void read_parser_environment_variables() } } -void init_directive_state() +void ptx_recognizer::init_directive_state() { PTX_PARSE_DPRINTF("init_directive_state"); g_space_spec=undefined_space; @@ -125,7 +98,7 @@ void init_directive_state() g_last_symbol = NULL; } -void init_instruction_state() +void ptx_recognizer::init_instruction_state() { PTX_PARSE_DPRINTF("init_instruction_state"); g_pred = NULL; @@ -139,17 +112,16 @@ void init_instruction_state() init_directive_state(); } -symbol_table *init_parser( const char *ptx_filename ) +symbol_table * gpgpu_context::init_parser( const char *ptx_filename ) { g_filename = strdup(ptx_filename); if (g_global_allfiles_symbol_table == NULL) { g_global_allfiles_symbol_table = new symbol_table("global_allfiles", 0, NULL); - g_global_symbol_table = g_current_symbol_table = g_global_allfiles_symbol_table; + ptx_parser->g_global_symbol_table = ptx_parser->g_current_symbol_table = g_global_allfiles_symbol_table; } /*else { g_global_symbol_table = g_current_symbol_table = new symbol_table("global",0,g_global_allfiles_symbol_table); }*/ - ptx_lineno = 1; #define DEF(X,Y) g_ptx_token_decode[X] = Y; #include "ptx_parser_decode.def" @@ -169,18 +141,23 @@ symbol_table *init_parser( const char *ptx_filename ) g_ptx_token_decode[generic_space] = "generic_space"; g_ptx_token_decode[instruction_space] = "instruction_space"; - init_directive_state(); - init_instruction_state(); + ptx_lex_init(&(ptx_parser->scanner)); + ptx_parser->init_directive_state(); + ptx_parser->init_instruction_state(); + FILE *ptx_in; ptx_in = fopen(ptx_filename, "r"); - ptx_parse(); + ptx_set_in(ptx_in, ptx_parser->scanner); + ptx_parse(ptx_parser->scanner, ptx_parser); + ptx_in = ptx_get_in(ptx_parser->scanner); + ptx_lex_destroy(ptx_parser->scanner); fclose(ptx_in); - return g_global_symbol_table; + return ptx_parser->g_global_symbol_table; } static int g_entry_point; -void start_function( int entry_point ) +void ptx_recognizer::start_function( int entry_point ) { PTX_PARSE_DPRINTF("start_function"); init_directive_state(); @@ -190,11 +167,7 @@ void start_function( int entry_point ) g_entry_func_param_index=0; } -char *g_add_identifier_cached__identifier = NULL; -int g_add_identifier_cached__array_dim; -int g_add_identifier_cached__array_ident; - -void add_function_name( const char *name ) +void ptx_recognizer::add_function_name( const char *name ) { PTX_PARSE_DPRINTF("add_function_name %s %s", name, ((g_entry_point==1)?"(entrypoint)":((g_entry_point==2)?"(extern)":""))); bool prior_decl = g_global_symbol_table->add_function_decl( name, g_entry_point, &g_func_info, &g_current_symbol_table ); @@ -210,21 +183,21 @@ void add_function_name( const char *name ) if( prior_decl ) { g_func_info->remove_args(); } - g_global_symbol_table->add_function( g_func_info, g_filename, ptx_lineno ); + g_global_symbol_table->add_function( g_func_info, g_filename, ptx_get_lineno(scanner) ); } //Jin: handle instruction group for cdp -void start_inst_group() { +void ptx_recognizer::start_inst_group() { PTX_PARSE_DPRINTF("start_instruction_group"); g_current_symbol_table = g_current_symbol_table->start_inst_group(); } -void end_inst_group() { +void ptx_recognizer::end_inst_group() { PTX_PARSE_DPRINTF("end_instruction_group"); g_current_symbol_table = g_current_symbol_table->end_inst_group(); } -void add_directive() +void ptx_recognizer::add_directive() { PTX_PARSE_DPRINTF("add_directive"); init_directive_state(); @@ -232,7 +205,7 @@ void add_directive() #define mymax(a,b) ((a)>(b)?(a):(b)) -void end_function() +void ptx_recognizer::end_function() { PTX_PARSE_DPRINTF("end_function"); @@ -250,7 +223,7 @@ void end_function() #define parse_error(msg, ...) parse_error_impl(__FILE__,__LINE__, msg, ##__VA_ARGS__) #define parse_assert(cond,msg, ...) parse_assert_impl((cond),__FILE__,__LINE__, msg, ##__VA_ARGS__) -void parse_error_impl( const char *file, unsigned line, const char *msg, ... ) +void ptx_recognizer::parse_error_impl( const char *file, unsigned line, const char *msg, ... ) { va_list ap; char buf[1024]; @@ -259,13 +232,13 @@ void parse_error_impl( const char *file, unsigned line, const char *msg, ... ) va_end(ap); g_error_detected = 1; - printf("%s:%u: Parse error: %s (%s:%u)\n\n", g_filename, ptx_lineno, buf, file, line); - ptx_error(NULL); + printf("%s:%u: Parse error: %s (%s:%u)\n\n", g_filename, ptx_get_lineno(scanner), buf, file, line); + ptx_error(scanner, NULL); abort(); exit(1); } -void parse_assert_impl( int test_value, const char *file, unsigned line, const char *msg, ... ) +void ptx_recognizer::parse_assert_impl( int test_value, const char *file, unsigned line, const char *msg, ... ) { va_list ap; char buf[1024]; @@ -277,10 +250,9 @@ void parse_assert_impl( int test_value, const char *file, unsigned line, const c parse_error_impl(file,line, msg); } -extern char linebuf[4096]; -void set_return() +void ptx_recognizer::set_return() { parse_assert( (g_opcode == CALL_OP || g_opcode == CALLP_OP), "only call can have return value"); g_operands.front().set_return(); @@ -300,7 +272,7 @@ const ptx_instruction *ptx_instruction_lookup( const char *filename, unsigned li return l->second; } -void add_instruction() +void ptx_recognizer::add_instruction() { PTX_PARSE_DPRINTF("add_instruction: %s", ((g_opcode>0)?g_opcode_string[g_opcode]:"<label>") ); assert( g_shader_core_config != 0 ); @@ -316,15 +288,15 @@ void add_instruction() g_scalar_type, g_space_spec, g_filename, - ptx_lineno, + ptx_get_lineno(scanner), linebuf, g_shader_core_config ); g_instructions.push_back(i); - g_inst_lookup[g_filename][ptx_lineno] = i; + g_inst_lookup[g_filename][ptx_get_lineno(scanner)] = i; init_instruction_state(); } -void add_variables() +void ptx_recognizer::add_variables() { PTX_PARSE_DPRINTF("add_variables"); if ( !g_operands.empty() ) { @@ -334,7 +306,7 @@ void add_variables() init_directive_state(); } -void set_variable_type() +void ptx_recognizer::set_variable_type() { PTX_PARSE_DPRINTF("set_variable_type space_spec=%s scalar_type_spec=%s", g_ptx_token_decode[g_space_spec.get_type()].c_str(), @@ -348,7 +320,7 @@ void set_variable_type() g_extern_spec ); } -bool check_for_duplicates( const char *identifier ) +bool ptx_recognizer::check_for_duplicates( const char *identifier ) { const symbol *s = g_current_symbol_table->lookup(identifier); return ( s != NULL ); @@ -357,10 +329,6 @@ bool check_for_duplicates( const char *identifier ) extern std::set<std::string> g_globals; extern std::set<std::string> g_constants; -int g_func_decl = 0; -int g_ident_add_uid = 0; -unsigned g_const_alloc = 1; - // Returns padding that needs to be inserted ahead of address to make it aligned to min(size, maxalign) /* * @param address the address in bytes @@ -378,7 +346,7 @@ int pad_address (new_addr_type address, unsigned size, unsigned maxalign) { return alignto ? ((alignto - (address % alignto)) % alignto) : 0; } -void add_identifier( const char *identifier, int array_dim, unsigned array_ident ) +void ptx_recognizer::add_identifier( const char *identifier, int array_dim, unsigned array_ident ) { if(array_ident==ARRAY_IDENTIFIER){ g_size *= array_dim; @@ -426,7 +394,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident default: break; } - g_last_symbol = g_current_symbol_table->add_variable(identifier,type,num_bits/8,g_filename,ptx_lineno); + g_last_symbol = g_current_symbol_table->add_variable(identifier,type,num_bits/8,g_filename,ptx_get_lineno(scanner)); switch ( ti.get_memory_space().get_type() ) { case reg_space: { regnum = g_current_symbol_table->next_reg_num(); @@ -569,7 +537,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident } } -void add_constptr(const char* identifier1, const char* identifier2, int offset) +void ptx_recognizer::add_constptr(const char* identifier1, const char* identifier2, int offset) { symbol *s1 = g_current_symbol_table->lookup(identifier1); const symbol *s2 = g_current_symbol_table->lookup(identifier2); @@ -584,7 +552,7 @@ void add_constptr(const char* identifier1, const char* identifier2, int offset) s1->set_address( addr + offset ); } -void add_function_arg() +void ptx_recognizer::add_function_arg() { assert(g_size>0); if( g_func_info ) { @@ -597,20 +565,20 @@ void add_function_arg() } -void add_extern_spec() +void ptx_recognizer::add_extern_spec() { PTX_PARSE_DPRINTF("add_extern_spec"); g_extern_spec = 1; } -void add_alignment_spec( int spec ) +void ptx_recognizer::add_alignment_spec( int spec ) { PTX_PARSE_DPRINTF("add_alignment_spec"); parse_assert( g_alignment_spec == -1, "multiple .align specifiers per variable declaration not allowed." ); g_alignment_spec = spec; } -void add_ptr_spec( enum _memory_space_t spec ) +void ptx_recognizer::add_ptr_spec( enum _memory_space_t spec ) { PTX_PARSE_DPRINTF("add_ptr_spec \"%s\"", g_ptx_token_decode[spec].c_str() ); parse_assert( g_ptr_spec == undefined_space, "multiple ptr space specifiers not allowed." ); @@ -618,7 +586,7 @@ void add_ptr_spec( enum _memory_space_t spec ) g_ptr_spec = spec; } -void add_space_spec( enum _memory_space_t spec, int value ) +void ptx_recognizer::add_space_spec( enum _memory_space_t spec, int value ) { PTX_PARSE_DPRINTF("add_space_spec \"%s\"", g_ptx_token_decode[spec].c_str() ); parse_assert( g_space_spec == undefined_space, "multiple space specifiers not allowed." ); @@ -637,14 +605,14 @@ void add_space_spec( enum _memory_space_t spec, int value ) } } -void add_vector_spec(int spec ) +void ptx_recognizer::add_vector_spec(int spec ) { PTX_PARSE_DPRINTF("add_vector_spec"); parse_assert( g_vector_spec == -1, "multiple vector specifiers not allowed." ); g_vector_spec = spec; } -void add_scalar_type_spec( int type_spec ) +void ptx_recognizer::add_scalar_type_spec( int type_spec ) { //save size of parameter switch ( type_spec ) { @@ -682,23 +650,23 @@ void add_scalar_type_spec( int type_spec ) g_scalar_type_spec = type_spec; } -void add_label( const char *identifier ) +void ptx_recognizer::add_label( const char *identifier ) { PTX_PARSE_DPRINTF("add_label"); symbol *s = g_current_symbol_table->lookup(identifier); if ( s != NULL ) { g_label = s; } else { - g_label = g_current_symbol_table->add_variable(identifier,NULL,0,g_filename,ptx_lineno); + g_label = g_current_symbol_table->add_variable(identifier,NULL,0,g_filename,ptx_get_lineno(scanner)); } } -void add_opcode( int opcode ) +void ptx_recognizer::add_opcode( int opcode ) { g_opcode = opcode; } -void add_pred( const char *identifier, int neg, int predModifier ) +void ptx_recognizer::add_pred( const char *identifier, int neg, int predModifier ) { PTX_PARSE_DPRINTF("add_pred"); const symbol *s = g_current_symbol_table->lookup(identifier); @@ -711,17 +679,17 @@ void add_pred( const char *identifier, int neg, int predModifier ) g_pred_mod = predModifier; } -void add_option( int option ) +void ptx_recognizer::add_option( int option ) { PTX_PARSE_DPRINTF("add_option"); g_options.push_back( option ); } -void add_wmma_option( int option ) +void ptx_recognizer::add_wmma_option( int option ) { PTX_PARSE_DPRINTF("add_option"); g_wmma_options.push_back( option ); } -void add_double_operand( const char *d1, const char *d2 ) +void ptx_recognizer::add_double_operand( const char *d1, const char *d2 ) { //operands that access two variables. //eg. s[$ofs1+$r0], g[$ofs1+=$r0] @@ -734,7 +702,7 @@ void add_double_operand( const char *d1, const char *d2 ) g_operands.push_back( operand_info(s1,s2) ); } -void add_1vector_operand( const char *d1 ) +void ptx_recognizer::add_1vector_operand( const char *d1 ) { // handles the single element vector operand ({%v1}) found in tex.1d instructions PTX_PARSE_DPRINTF("add_1vector_operand"); @@ -743,7 +711,7 @@ void add_1vector_operand( const char *d1 ) g_operands.push_back( operand_info(s1,NULL,NULL,NULL) ); } -void add_2vector_operand( const char *d1, const char *d2 ) +void ptx_recognizer::add_2vector_operand( const char *d1, const char *d2 ) { PTX_PARSE_DPRINTF("add_2vector_operand"); const symbol *s1 = g_current_symbol_table->lookup(d1); @@ -752,7 +720,7 @@ void add_2vector_operand( const char *d1, const char *d2 ) g_operands.push_back( operand_info(s1,s2,NULL,NULL) ); } -void add_3vector_operand( const char *d1, const char *d2, const char *d3 ) +void ptx_recognizer::add_3vector_operand( const char *d1, const char *d2, const char *d3 ) { PTX_PARSE_DPRINTF("add_3vector_operand"); const symbol *s1 = g_current_symbol_table->lookup(d1); @@ -762,7 +730,7 @@ void add_3vector_operand( const char *d1, const char *d2, const char *d3 ) g_operands.push_back( operand_info(s1,s2,s3,NULL) ); } -void add_4vector_operand( const char *d1, const char *d2, const char *d3, const char *d4 ) +void ptx_recognizer::add_4vector_operand( const char *d1, const char *d2, const char *d3, const char *d4 ) { PTX_PARSE_DPRINTF("add_4vector_operand"); const symbol *s1 = g_current_symbol_table->lookup(d1); @@ -776,7 +744,7 @@ void add_4vector_operand( const char *d1, const char *d2, const char *d3, const if ( s4 == null_op ) s4 = NULL; g_operands.push_back( operand_info(s1,s2,s3,s4) ); } -void add_8vector_operand( const char *d1, const char *d2, const char *d3, const char *d4,const char *d5,const char *d6,const char *d7,const char *d8 ) +void ptx_recognizer::add_8vector_operand( const char *d1, const char *d2, const char *d3, const char *d4,const char *d5,const char *d6,const char *d7,const char *d8 ) { PTX_PARSE_DPRINTF("add_8vector_operand"); const symbol *s1 = g_current_symbol_table->lookup(d1); @@ -799,13 +767,13 @@ void add_8vector_operand( const char *d1, const char *d2, const char *d3, const g_operands.push_back( operand_info(s1,s2,s3,s4,s5,s6,s7,s8) ); } -void add_builtin_operand( int builtin, int dim_modifier ) +void ptx_recognizer::add_builtin_operand( int builtin, int dim_modifier ) { PTX_PARSE_DPRINTF("add_builtin_operand"); g_operands.push_back( operand_info(builtin,dim_modifier) ); } -void add_memory_operand() +void ptx_recognizer::add_memory_operand() { PTX_PARSE_DPRINTF("add_memory_operand"); assert( !g_operands.empty() ); @@ -813,7 +781,7 @@ void add_memory_operand() } /*TODO: add other memory locations*/ -void change_memory_addr_space(const char *identifier) +void ptx_recognizer::change_memory_addr_space(const char *identifier) { /*0 = N/A, not reading from memory *1 = global memory @@ -860,7 +828,7 @@ void change_memory_addr_space(const char *identifier) parse_assert(recognizedType, "Error: unrecognized memory type."); } -void change_operand_lohi( int lohi ) +void ptx_recognizer::change_operand_lohi( int lohi ) { /*0 = N/A, read entire operand *1 = lo, reading from lowest bits @@ -874,14 +842,14 @@ void change_operand_lohi( int lohi ) } -void set_immediate_operand_type () +void ptx_recognizer::set_immediate_operand_type () { PTX_PARSE_DPRINTF("set_immediate_operand_type"); assert( !g_operands.empty() ); g_operands.back().set_immediate_addr(); } -void change_double_operand_type( int operand_type ) +void ptx_recognizer::change_double_operand_type( int operand_type ) { /* *-3 = reg / reg (set instruction, but both get same value) @@ -913,7 +881,7 @@ void change_double_operand_type( int operand_type ) } -void change_operand_neg( ) +void ptx_recognizer::change_operand_neg( ) { PTX_PARSE_DPRINTF("change_operand_neg"); assert( !g_operands.empty() ); @@ -922,32 +890,32 @@ void change_operand_neg( ) } -void add_literal_int( int value ) +void ptx_recognizer::add_literal_int( int value ) { PTX_PARSE_DPRINTF("add_literal_int"); g_operands.push_back( operand_info(value) ); } -void add_literal_float( float value ) +void ptx_recognizer::add_literal_float( float value ) { PTX_PARSE_DPRINTF("add_literal_float"); g_operands.push_back( operand_info(value) ); } -void add_literal_double( double value ) +void ptx_recognizer::add_literal_double( double value ) { PTX_PARSE_DPRINTF("add_literal_double"); g_operands.push_back( operand_info(value) ); } -void add_scalar_operand( const char *identifier ) +void ptx_recognizer::add_scalar_operand( const char *identifier ) { PTX_PARSE_DPRINTF("add_scalar_operand"); const symbol *s = g_current_symbol_table->lookup(identifier); if ( s == NULL ) { if ( g_opcode == BRA_OP || g_opcode == CALLP_OP) { // forward branch target... - s = g_current_symbol_table->add_variable(identifier,NULL,0,g_filename,ptx_lineno); + s = g_current_symbol_table->add_variable(identifier,NULL,0,g_filename,ptx_get_lineno(scanner)); } else { std::string msg = std::string("operand \"") + identifier + "\" has no declaration."; parse_error( msg.c_str() ); @@ -956,19 +924,19 @@ void add_scalar_operand( const char *identifier ) g_operands.push_back( operand_info(s) ); } -void add_neg_pred_operand( const char *identifier ) +void ptx_recognizer::add_neg_pred_operand( const char *identifier ) { PTX_PARSE_DPRINTF("add_neg_pred_operand"); const symbol *s = g_current_symbol_table->lookup(identifier); if ( s == NULL ) { - s = g_current_symbol_table->add_variable(identifier,NULL,1,g_filename,ptx_lineno); + s = g_current_symbol_table->add_variable(identifier,NULL,1,g_filename,ptx_get_lineno(scanner)); } operand_info op(s); op.set_neg_pred(); g_operands.push_back( op ); } -void add_address_operand( const char *identifier, int offset ) +void ptx_recognizer::add_address_operand( const char *identifier, int offset ) { PTX_PARSE_DPRINTF("add_address_operand"); const symbol *s = g_current_symbol_table->lookup(identifier); @@ -979,23 +947,23 @@ void add_address_operand( const char *identifier, int offset ) g_operands.push_back( operand_info(s,offset) ); } -void add_address_operand2( int offset ) +void ptx_recognizer::add_address_operand2( int offset ) { PTX_PARSE_DPRINTF("add_address_operand"); g_operands.push_back( operand_info((unsigned)offset) ); } -void add_array_initializer() +void ptx_recognizer::add_array_initializer() { g_last_symbol->add_initializer(g_operands); } -void add_version_info( float ver, unsigned ext ) +void ptx_recognizer::add_version_info( float ver, unsigned ext ) { g_global_symbol_table->set_ptx_version(ver,ext); } -void add_file( unsigned num, const char *filename ) +void ptx_recognizer::add_file( unsigned num, const char *filename ) { if( g_filename == NULL ) { char *b = strdup(filename); @@ -1022,44 +990,44 @@ void add_file( unsigned num, const char *filename ) g_current_symbol_table = g_global_symbol_table; } -void *reset_symtab() +void * ptx_recognizer::reset_symtab() { void *result = g_current_symbol_table; g_current_symbol_table = g_global_symbol_table; return result; } -void set_symtab(void*symtab) +void ptx_recognizer::set_symtab(void*symtab) { g_current_symbol_table = (symbol_table*)symtab; } -void add_pragma( const char *str ) +void ptx_recognizer::add_pragma( const char *str ) { printf("GPGPU-Sim PTX: Warning -- ignoring pragma '%s'\n", str ); } -void version_header(double a) {} //intentional dummy function +void ptx_recognizer::version_header(double a) {} //intentional dummy function -void target_header(char* a) +void ptx_recognizer::target_header(char* a) { g_global_symbol_table->set_sm_target(a,NULL,NULL); } -void target_header2(char* a, char* b) +void ptx_recognizer::target_header2(char* a, char* b) { g_global_symbol_table->set_sm_target(a,b,NULL); } -void target_header3(char* a, char* b, char* c) +void ptx_recognizer::target_header3(char* a, char* b, char* c) { g_global_symbol_table->set_sm_target(a,b,c); } -void maxnt_id(int x, int y, int z) { +void ptx_recognizer::maxnt_id(int x, int y, int z) { g_func_info->set_maxnt_id(x * y * z); } -void func_header(const char* a) {} //intentional dummy function -void func_header_info(const char* a) {} //intentional dummy function -void func_header_info_int(const char* a, int b) {} //intentional dummy function +void ptx_recognizer::func_header(const char* a) {} //intentional dummy function +void ptx_recognizer::func_header_info(const char* a) {} //intentional dummy function +void ptx_recognizer::func_header_info_int(const char* a, int b) {} //intentional dummy function diff --git a/src/cuda-sim/ptx_parser.h b/src/cuda-sim/ptx_parser.h index 7b6e3a2..25c01fe 100644 --- a/src/cuda-sim/ptx_parser.h +++ b/src/cuda-sim/ptx_parser.h @@ -29,77 +29,151 @@ #define ptx_parser_INCLUDED #include "../abstract_hardware_model.h" +#include "ptx_ir.h" extern const char *g_filename; extern int g_error_detected; #ifdef __cplusplus -class symbol_table* init_parser(const char*); const class ptx_instruction *ptx_instruction_lookup( const char *filename, unsigned linenumber ); #endif +typedef void * yyscan_t; +class ptx_recognizer { + public: + ptx_recognizer() { + scanner = NULL; + g_size = -1; + g_add_identifier_cached__identifier = NULL; + g_alignment_spec = -1; + g_var_type = NULL; + g_opcode = -1; + g_space_spec = undefined_space; + g_ptr_spec = undefined_space; + g_scalar_type_spec = -1; + g_vector_spec = -1; + g_extern_spec = 0; + g_func_decl = 0; + g_ident_add_uid = 0; + g_const_alloc = 1; + g_max_regs_per_thread = 0; + g_global_symbol_table = NULL; + g_current_symbol_table = NULL; + g_last_symbol = NULL; + g_error_detected = 0; + g_entry_func_param_index=0; + g_func_info = NULL; + } + // global list + yyscan_t scanner; +#define PTX_LINEBUF_SIZE (4*1024) + char linebuf[PTX_LINEBUF_SIZE]; + unsigned col; + int g_size; + char *g_add_identifier_cached__identifier; + int g_add_identifier_cached__array_dim; + int g_add_identifier_cached__array_ident; + int g_alignment_spec; + // variable declaration stuff: + type_info *g_var_type; + // instruction definition stuff: + const symbol *g_pred; + int g_neg_pred; + int g_pred_mod; + symbol *g_label; + int g_opcode; + std::list<operand_info> g_operands; + std::list<int> g_options; + std::list<int> g_wmma_options; + std::list<int> g_scalar_type; + // type specifier stuff: + memory_space_t g_space_spec; + memory_space_t g_ptr_spec; + int g_scalar_type_spec; + int g_vector_spec; + int g_extern_spec; + int g_func_decl; + int g_ident_add_uid; + unsigned g_const_alloc; + unsigned g_max_regs_per_thread; + symbol_table *g_global_symbol_table; + symbol_table *g_current_symbol_table; + symbol *g_last_symbol; + std::list<ptx_instruction*> g_instructions; + int g_error_detected; + unsigned g_entry_func_param_index; + function_info *g_func_info; + operand_info g_return_var; + + // member function list + void init_directive_state(); + void init_instruction_state(); + void start_function( int entry_point ); + void add_function_name( const char *fname ); + void add_directive(); + void end_function(); + void add_identifier( const char *s, int array_dim, unsigned array_ident ); + void add_function_arg(); + void add_scalar_type_spec( int type_spec ); + void add_scalar_operand( const char *identifier ); + void add_neg_pred_operand( const char *identifier ); + void add_variables(); + void set_variable_type(); + void add_opcode( int opcode ); + void add_pred( const char *identifier, int negate, int predModifier ); + void add_1vector_operand( const char *d1 ); + void add_2vector_operand( const char *d1, const char *d2 ); + void add_3vector_operand( const char *d1, const char *d2, const char *d3 ); + void add_4vector_operand( const char *d1, const char *d2, const char *d3, const char *d4 ); + void add_8vector_operand( const char *d1, const char *d2, const char *d3, const char *d4 ,const char *d5,const char *d6,const char *d7,const char *d8); + void add_option(int option ); + void add_wmma_option(int option ); + void add_builtin_operand( int builtin, int dim_modifier ); + void add_memory_operand( ); + void add_literal_int( int value ); + void add_literal_float( float value ); + void add_literal_double( double value ); + void add_address_operand( const char *identifier, int offset ); + void add_address_operand2( int offset ); + void add_label( const char *idenfiier ); + void add_vector_spec(int spec ); + void add_space_spec( enum _memory_space_t spec, int value ); + void add_ptr_spec( enum _memory_space_t spec ); + void add_extern_spec(); + void add_instruction(); + void set_return(); + void add_alignment_spec( int spec ); + void add_array_initializer(); + void add_file( unsigned num, const char *filename ); + void add_version_info( float ver, unsigned ext); + void *reset_symtab(); + void set_symtab(void*); + void add_pragma( const char *str ); + void func_header(const char* a); + void func_header_info(const char* a); + void func_header_info_int(const char* a, int b); + void add_constptr(const char* identifier1, const char* identifier2, int offset); + void target_header(char* a); + void target_header2(char* a, char* b); + void target_header3(char* a, char* b, char* c); + void add_double_operand( const char *d1, const char *d2 ); + void change_memory_addr_space( const char *identifier ); + void change_operand_lohi( int lohi ); + void change_double_operand_type( int addr_type ); + void change_operand_neg( ); + void set_immediate_operand_type( ); + void version_header(double a); + void maxnt_id(int x, int y, int z); + void parse_error_impl( const char *file, unsigned line, const char *msg, ... ); + void parse_assert_impl( int test_value, const char *file, unsigned line, const char *msg, ... ); + //Jin: handle instructino group for cdp + void start_inst_group(); + void end_inst_group(); + bool check_for_duplicates( const char *identifier ); + +}; + const char *decode_token( int type ); void read_parser_environment_variables(); -void start_function( int entry_point ); -void add_function_name( const char *fname ); -void init_directive_state(); -void add_directive(); -void end_function(); -void add_identifier( const char *s, int array_dim, unsigned array_ident ); -void add_function_arg(); -void add_scalar_type_spec( int type_spec ); -void add_scalar_operand( const char *identifier ); -void add_neg_pred_operand( const char *identifier ); -void add_variables(); -void set_variable_type(); -void add_opcode( int opcode ); -void add_pred( const char *identifier, int negate, int predModifier ); -void add_1vector_operand( const char *d1 ); -void add_2vector_operand( const char *d1, const char *d2 ); -void add_3vector_operand( const char *d1, const char *d2, const char *d3 ); -void add_4vector_operand( const char *d1, const char *d2, const char *d3, const char *d4 ); -void add_8vector_operand( const char *d1, const char *d2, const char *d3, const char *d4 ,const char *d5,const char *d6,const char *d7,const char *d8); -void add_option(int option ); -void add_wmma_option(int option ); -void add_builtin_operand( int builtin, int dim_modifier ); -void add_memory_operand( ); -void add_literal_int( int value ); -void add_literal_float( float value ); -void add_literal_double( double value ); -void add_address_operand( const char *identifier, int offset ); -void add_address_operand2( int offset ); -void add_label( const char *idenfiier ); -void add_vector_spec(int spec ); -void add_space_spec( enum _memory_space_t spec, int value ); -void add_ptr_spec( enum _memory_space_t spec ); -void add_extern_spec(); -void add_instruction(); -void set_return(); -void add_alignment_spec( int spec ); -void add_array_initializer(); -void add_file( unsigned num, const char *filename ); -void add_version_info( float ver, unsigned ext); -void *reset_symtab(); -void set_symtab(void*); -void add_pragma( const char *str ); -void func_header(const char* a); -void func_header_info(const char* a); -void func_header_info_int(const char* a, int b); -void add_constptr(const char* identifier1, const char* identifier2, int offset); -void target_header(char* a); -void target_header2(char* a, char* b); -void target_header3(char* a, char* b, char* c); -void add_double_operand( const char *d1, const char *d2 ); -void change_memory_addr_space( const char *identifier ); -void change_operand_lohi( int lohi ); -void change_double_operand_type( int addr_type ); -void change_operand_neg( ); -void set_immediate_operand_type( ); -void version_header(double a); -void maxnt_id(int x, int y, int z); - -//Jin: handle instructino group for cdp -void start_inst_group(); -void end_inst_group(); #define NON_ARRAY_IDENTIFIER 1 #define ARRAY_IDENTIFIER_NO_DIM 2 diff --git a/src/cuda-sim/ptx_sim.cc b/src/cuda-sim/ptx_sim.cc index 820287d..1b8831a 100644 --- a/src/cuda-sim/ptx_sim.cc +++ b/src/cuda-sim/ptx_sim.cc @@ -28,6 +28,8 @@ #include "ptx_sim.h" #include <string> #include "ptx_ir.h" +class ptx_recognizer; +typedef void * yyscan_t; #include "ptx.tab.h" #include "../gpgpu-sim/gpu-sim.h" #include "../gpgpu-sim/shader.h" @@ -222,7 +224,7 @@ void ptx_thread_info::set_done() { assert( !m_at_barrier ); m_thread_done = true; - m_cycle_done = gpu_sim_cycle; + m_cycle_done = m_gpu->gpu_sim_cycle; } unsigned ptx_thread_info::get_builtin( int builtin_id, unsigned dim_mod ) @@ -230,15 +232,15 @@ unsigned ptx_thread_info::get_builtin( int builtin_id, unsigned dim_mod ) assert( m_valid ); switch ((builtin_id&0xFFFF)) { case CLOCK_REG: - return (unsigned)(gpu_sim_cycle + gpu_tot_sim_cycle); + return (unsigned)(m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle); case CLOCK64_REG: abort(); // change return value to unsigned long long? // GPGPUSim clock is 4 times slower - multiply by 4 - return (gpu_sim_cycle + gpu_tot_sim_cycle)*4; + return (m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle)*4; case HALFCLOCK_ID: // GPGPUSim clock is 4 times slower - multiply by 4 // Hardware clock counter is incremented at half the shader clock frequency - divide by 2 (Henry '10) - return (gpu_sim_cycle + gpu_tot_sim_cycle)*2; + return (m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle)*2; case CTAID_REG: assert( dim_mod < 3 ); if( dim_mod == 0 ) return m_ctaid.x; diff --git a/src/cuda-sim/ptxinfo.l b/src/cuda-sim/ptxinfo.l index 33c2748..b0ada09 100644 --- a/src/cuda-sim/ptxinfo.l +++ b/src/cuda-sim/ptxinfo.l @@ -31,17 +31,22 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. %option noyywrap %option yylineno %option prefix="ptxinfo_" + +%option bison-bridge +%option reentrant + %{ +#include "ptx_loader.h" #include "ptxinfo.tab.h" #include <string.h> #define LINEBUF_SIZE 1024 -char ptxinfo_linebuf[LINEBUF_SIZE]; -unsigned ptxinfo_col = 0; -#define TC if( (ptxinfo_lineno == 1) && ((ptxinfo_col + strlen(ptxinfo_text)) < LINEBUF_SIZE) ) { \ - strncpy(ptxinfo_linebuf+ptxinfo_col,ptxinfo_text,strlen(ptxinfo_text)); \ +#define TC if( (yylineno == 1) && (ptxinfo->col + strlen(yytext) < LINEBUF_SIZE) ) { \ + strncpy(ptxinfo->linebuf+ptxinfo->col,yytext,strlen(yytext)); \ } \ - ptxinfo_col+=strlen(ptxinfo_text); + ptxinfo->col+=strlen(yytext); +#define YY_DECL int ptxinfo_lex \ + (YYSTYPE * yylval_param , yyscan_t yyscanner, ptxinfo_data* ptxinfo) %} %% @@ -61,12 +66,12 @@ unsigned ptxinfo_col = 0; "for" TC; return FOR; "textures" TC; return TEXTURES; "error : Duplicate definition of" TC; return DUPLICATE; -"function" TC; ptxinfo_lval.string_value = strdup(yytext); return FUNCTION; -"variable" TC; ptxinfo_lval.string_value = strdup(yytext); return VARIABLE; +"function" TC; yylval->string_value = strdup(yytext); return FUNCTION; +"variable" TC; yylval->string_value = strdup(yytext); return VARIABLE; "fatal : Ptx assembly aborted due to errors" TC; return FATAL; -[_A-Za-z$%][_0-9A-Za-z$]* TC; ptxinfo_lval.string_value = strdup(yytext); return IDENTIFIER; -[-]{0,1}[0-9]+ TC; ptxinfo_lval.int_value = atoi(yytext); return INT_OPERAND; +[_A-Za-z$%][_0-9A-Za-z$]* TC; yylval->string_value = strdup(yytext); return IDENTIFIER; +[-]{0,1}[0-9]+ TC; yylval->int_value = atoi(yytext); return INT_OPERAND; "+" TC; return PLUS; "," TC; return COMMA; @@ -78,26 +83,26 @@ unsigned ptxinfo_col = 0; " " TC; "\t" TC; -\n.* ptxinfo_col=0; strncpy(ptxinfo_linebuf, yytext + 1, 1024); yyless( 1 ); +\n.* ptxinfo->col=0; strncpy(ptxinfo->linebuf, yytext + 1, 1024); yyless( 1 ); %% extern int g_ptxinfo_error_detected; extern const char *g_filename; -extern const char *g_ptxinfo_filename; -int ptxinfo_error( const char *s ) +int ptxinfo_error(yyscan_t yyscanner, ptxinfo_data* ptxinfo, const char* msg) { + struct yyguts_t * yyg = (struct yyguts_t*)yyscanner; int i; g_ptxinfo_error_detected = 1; fflush(stdout); printf("GPGPU-Sim: ERROR while parsing output of ptxas (used to capture resource usage information)\n"); - if( s != NULL ) - printf("GPGPU-Sim: %s (%s:%u) Syntax error:\n\n", g_filename, g_ptxinfo_filename, ptxinfo_lineno ); - printf(" %s\n", ptxinfo_linebuf ); + if( msg != NULL ) + printf("GPGPU-Sim: %s (%s:%u) Syntax error:\n\n", g_filename, ptxinfo->g_ptxinfo_filename, yylineno ); + printf(" %s\n", ptxinfo->linebuf ); printf(" "); - for( i=0; i < ptxinfo_col-1; i++ ) { - if( ptxinfo_linebuf[i] == '\t' ) printf("\t"); + for( i=0; i < ptxinfo->col-1; i++ ) { + if( ptxinfo->linebuf[i] == '\t' ) printf("\t"); else printf(" "); } diff --git a/src/cuda-sim/ptxinfo.y b/src/cuda-sim/ptxinfo.y index d241d8c..00c81e0 100644 --- a/src/cuda-sim/ptxinfo.y +++ b/src/cuda-sim/ptxinfo.y @@ -27,6 +27,17 @@ OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ +%{ +typedef void * yyscan_t; +class ptxinfo_data; +%} + +%define api.pure full +%parse-param {yyscan_t scanner} +%parse-param {ptxinfo_data* ptxinfo} +%lex-param {yyscan_t scanner} +%lex-param {ptxinfo_data* ptxinfo} + %union { int int_value; char * string_value; @@ -66,7 +77,8 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. static unsigned g_declared; static unsigned g_system; - int ptxinfo_lex(void); + int ptxinfo_lex(YYSTYPE * yylval_param, yyscan_t yyscanner, ptxinfo_data* ptxinfo); + void yyerror(yyscan_t yyscanner, ptxinfo_data* ptxinfo, const char* msg); void ptxinfo_addinfo(); void ptxinfo_function(const char *fname ); void ptxinfo_regs( unsigned nregs ); @@ -74,7 +86,6 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. void ptxinfo_gmem( unsigned declared, unsigned system ); void ptxinfo_smem( unsigned declared, unsigned system ); void ptxinfo_cmem( unsigned nbytes, unsigned bank ); - int ptxinfo_error(const char*); void ptxinfo_linenum( unsigned ); void ptxinfo_dup_type( const char* ); %} diff --git a/src/gpgpu-sim/Makefile b/src/gpgpu-sim/Makefile index f10a8a4..4994577 100644 --- a/src/gpgpu-sim/Makefile +++ b/src/gpgpu-sim/Makefile @@ -53,6 +53,8 @@ else CXXFLAGS += endif +CXXFLAGS += -I$(CUDA_INSTALL_PATH)/include + POWER_FLAGS= ifneq ($(GPGPUSIM_POWER_MODEL),) POWER_FLAGS = -I$(GPGPUSIM_POWER_MODEL) -DGPGPUSIM_POWER_MODEL diff --git a/src/gpgpu-sim/dram.cc b/src/gpgpu-sim/dram.cc index 192cb65..5e36d4b 100644 --- a/src/gpgpu-sim/dram.cc +++ b/src/gpgpu-sim/dram.cc @@ -42,12 +42,13 @@ template class fifo_pipeline<mem_fetch>; template class fifo_pipeline<dram_req_t>; dram_t::dram_t( unsigned int partition_id, const struct memory_config *config, memory_stats_t *stats, - memory_partition_unit *mp ) + memory_partition_unit *mp, gpgpu_sim* gpu ) { id = partition_id; m_memory_partition_unit = mp; m_stats = stats; m_config = config; + m_gpu = gpu; //rowblp access_num=0; @@ -191,11 +192,12 @@ unsigned int dram_t::queue_limit() const } -dram_req_t::dram_req_t( class mem_fetch *mf, unsigned banks, unsigned dram_bnk_indexing_policy) +dram_req_t::dram_req_t( class mem_fetch *mf, unsigned banks, unsigned dram_bnk_indexing_policy, class gpgpu_sim* gpu) { txbytes = 0; dqbytes = 0; data = mf; + m_gpu = gpu; const addrdec_t &tlx = mf->get_tlx_addr(); @@ -226,9 +228,9 @@ dram_req_t::dram_req_t( class mem_fetch *mf, unsigned banks, unsigned dram_bnk_i col = tlx.col; nbytes = mf->get_data_size(); - timestamp = gpu_tot_sim_cycle + gpu_sim_cycle; + timestamp = m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle; addr = mf->get_addr(); - insertion_time = (unsigned) gpu_sim_cycle; + insertion_time = (unsigned) m_gpu->gpu_sim_cycle; rw = data->get_is_write()?WRITE:READ; } @@ -236,9 +238,9 @@ void dram_t::push( class mem_fetch *data ) { assert(id == data->get_tlx_addr().chip); // Ensure request is in correct memory partition - dram_req_t *mrq = new dram_req_t(data,m_config->nbk,m_config->dram_bnk_indexing_policy); + dram_req_t *mrq = new dram_req_t(data,m_config->nbk,m_config->dram_bnk_indexing_policy,m_memory_partition_unit->get_mgpu()); - data->set_status(IN_PARTITION_MC_INTERFACE_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + data->set_status(IN_PARTITION_MC_INTERFACE_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); mrqq->push(mrq); // stats... @@ -259,7 +261,7 @@ void dram_t::scheduler_fifo() if (!mrqq->empty()) { unsigned int bkn; dram_req_t *head_mrqq = mrqq->top(); - head_mrqq->data->set_status(IN_PARTITION_MC_BANK_ARB_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + head_mrqq->data->set_status(IN_PARTITION_MC_BANK_ARB_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); bkn = head_mrqq->bk; if (!bk[bkn]->mrq) bk[bkn]->mrq = mrqq->pop(); @@ -283,7 +285,7 @@ void dram_t::cycle() if (cmd->dqbytes >= cmd->nbytes) { mem_fetch *data = cmd->data; - data->set_status(IN_PARTITION_MC_RETURNQ,gpu_sim_cycle+gpu_tot_sim_cycle); + data->set_status(IN_PARTITION_MC_RETURNQ,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); if( data->get_access_type() != L1_WRBK_ACC && data->get_access_type() != L2_WRBK_ACC ) { data->set_reply(); returnq->push(data); @@ -566,7 +568,7 @@ bool dram_t::issue_col_command(int j) bool issued = false; unsigned grp = get_bankgrp_number(j); if (bk[j]->mrq) { //if currently servicing a memory request - bk[j]->mrq->data->set_status(IN_PARTITION_DRAM,gpu_sim_cycle+gpu_tot_sim_cycle); + bk[j]->mrq->data->set_status(IN_PARTITION_DRAM,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); // correct row activated for a READ if ( !issued && !CCDc && !bk[j]->RCDc && !(bkgrp[grp]->CCDLc) && @@ -654,7 +656,7 @@ bool dram_t::issue_row_command(int j) bool issued = false; unsigned grp = get_bankgrp_number(j); if (bk[j]->mrq) { //if currently servicing a memory request - bk[j]->mrq->data->set_status(IN_PARTITION_DRAM,gpu_sim_cycle+gpu_tot_sim_cycle); + bk[j]->mrq->data->set_status(IN_PARTITION_DRAM,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); // bank is idle //else if ( !issued && !RRDc && diff --git a/src/gpgpu-sim/dram.h b/src/gpgpu-sim/dram.h index 1ab0153..7a3a2da 100644 --- a/src/gpgpu-sim/dram.h +++ b/src/gpgpu-sim/dram.h @@ -48,7 +48,7 @@ class dram_req_t { public: - dram_req_t( class mem_fetch *data , unsigned banks, unsigned dram_bnk_indexing_policy); + dram_req_t( class mem_fetch *data , unsigned banks, unsigned dram_bnk_indexing_policy, class gpgpu_sim* gpu); unsigned int row; unsigned int col; @@ -62,6 +62,7 @@ public: unsigned long long int addr; unsigned int insertion_time; class mem_fetch * data; + class gpgpu_sim * m_gpu; }; struct bankgrp_t @@ -110,7 +111,7 @@ class dram_t { public: dram_t( unsigned int parition_id, const struct memory_config *config, class memory_stats_t *stats, - class memory_partition_unit *mp ); + class memory_partition_unit *mp, class gpgpu_sim* gpu ); bool full(bool is_write) const; void print( FILE* simFile ) const; @@ -129,6 +130,7 @@ public: void dram_log (int task); class memory_partition_unit *m_memory_partition_unit; + class gpgpu_sim* m_gpu; unsigned int id; // Power Model diff --git a/src/gpgpu-sim/dram_sched.cc b/src/gpgpu-sim/dram_sched.cc index ff50050..6ee6271 100644 --- a/src/gpgpu-sim/dram_sched.cc +++ b/src/gpgpu-sim/dram_sched.cc @@ -84,13 +84,13 @@ void frfcfs_scheduler::add_req( dram_req_t *req ) void frfcfs_scheduler::data_collection(unsigned int bank) { - if (gpu_sim_cycle > row_service_timestamp[bank]) { - curr_row_service_time[bank] = gpu_sim_cycle - row_service_timestamp[bank]; + if (m_dram->m_gpu->gpu_sim_cycle > row_service_timestamp[bank]) { + curr_row_service_time[bank] = m_dram->m_gpu->gpu_sim_cycle - row_service_timestamp[bank]; if (curr_row_service_time[bank] > m_stats->max_servicetime2samerow[m_dram->id][bank]) m_stats->max_servicetime2samerow[m_dram->id][bank] = curr_row_service_time[bank]; } curr_row_service_time[bank] = 0; - row_service_timestamp[bank] = gpu_sim_cycle; + row_service_timestamp[bank] = m_dram->m_gpu->gpu_sim_cycle; if (m_stats->concurrent_row_access[m_dram->id][bank] > m_stats->max_conc_access2samerow[m_dram->id][bank]) { m_stats->max_conc_access2samerow[m_dram->id][bank] = m_stats->concurrent_row_access[m_dram->id][bank]; } @@ -215,7 +215,7 @@ void dram_t::scheduler_frfcfs() m_stats->total_n_reads++; } - req->data->set_status(IN_PARTITION_MC_INPUT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + req->data->set_status(IN_PARTITION_MC_INPUT_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); sched->add_req(req); } @@ -228,14 +228,14 @@ void dram_t::scheduler_frfcfs() req = sched->schedule(b, bk[b]->curr_row); if ( req ) { - req->data->set_status(IN_PARTITION_MC_BANK_ARB_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + req->data->set_status(IN_PARTITION_MC_BANK_ARB_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); prio = (prio+1)%m_config->nbk; bk[b]->mrq = req; if (m_config->gpgpu_memlatency_stat) { - mrq_latency = gpu_sim_cycle + gpu_tot_sim_cycle - bk[b]->mrq->timestamp; + mrq_latency = m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle - bk[b]->mrq->timestamp; m_stats->tot_mrq_latency += mrq_latency; m_stats->tot_mrq_num++; - bk[b]->mrq->timestamp = gpu_tot_sim_cycle + gpu_sim_cycle; + bk[b]->mrq->timestamp =m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle; m_stats->mrq_lat_table[LOGB2(mrq_latency)]++; if (mrq_latency > m_stats->max_mrq_latency) { m_stats->max_mrq_latency = mrq_latency; diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc index 370f6e6..f1f6e19 100644 --- a/src/gpgpu-sim/gpu-cache.cc +++ b/src/gpgpu-sim/gpu-cache.cc @@ -26,6 +26,7 @@ // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #include "gpu-cache.h" +#include "gpu-sim.h" #include "stat-tool.h" #include <assert.h> @@ -1258,7 +1259,8 @@ data_cache::wr_miss_wa_naive( new_addr_type addr, mf->get_wid(), mf->get_sid(), mf->get_tpc(), - mf->get_mem_config()); + mf->get_mem_config(), + m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle); bool do_miss = false; bool wb = false; @@ -1276,7 +1278,7 @@ data_cache::wr_miss_wa_naive( new_addr_type addr, if( wb && (m_config.m_write_policy != WRITE_THROUGH) ) { assert(status == MISS); //SECTOR_MISS and HIT_RESERVED should not send write back mem_fetch *wb = m_memfetch_creator->alloc(evicted.m_block_addr, - m_wrbk_type,evicted.m_modified_size,true); + m_wrbk_type,evicted.m_modified_size,true,m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle); send_write_request(wb, cache_event(WRITE_BACK_REQUEST_SENT, evicted), time, events); } return MISS; @@ -1320,7 +1322,7 @@ data_cache::wr_miss_wa_fetch_on_write( new_addr_type addr, // (already modified lower level) if( wb && (m_config.m_write_policy != WRITE_THROUGH) ) { mem_fetch *wb = m_memfetch_creator->alloc(evicted.m_block_addr, - m_wrbk_type,evicted.m_modified_size,true); + m_wrbk_type,evicted.m_modified_size,true,m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle); send_write_request(wb, cache_event(WRITE_BACK_REQUEST_SENT, evicted), time, events); } return MISS; @@ -1372,6 +1374,7 @@ data_cache::wr_miss_wa_fetch_on_write( new_addr_type addr, mf->get_sid(), mf->get_tpc(), mf->get_mem_config(), + m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle, NULL, mf); @@ -1395,7 +1398,7 @@ data_cache::wr_miss_wa_fetch_on_write( new_addr_type addr, // (already modified lower level) if(wb && (m_config.m_write_policy != WRITE_THROUGH) ){ mem_fetch *wb = m_memfetch_creator->alloc(evicted.m_block_addr, - m_wrbk_type,evicted.m_modified_size,true); + m_wrbk_type,evicted.m_modified_size,true,m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle); send_write_request(wb, cache_event(WRITE_BACK_REQUEST_SENT, evicted), time, events); } return MISS; @@ -1448,7 +1451,7 @@ data_cache::wr_miss_wa_lazy_fetch_on_read( new_addr_type addr, // (already modified lower level) if( wb && (m_config.m_write_policy != WRITE_THROUGH) ) { mem_fetch *wb = m_memfetch_creator->alloc(evicted.m_block_addr, - m_wrbk_type,evicted.m_modified_size,true); + m_wrbk_type,evicted.m_modified_size,true,m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle); send_write_request(wb, cache_event(WRITE_BACK_REQUEST_SENT, evicted), time, events); } return MISS; @@ -1533,7 +1536,7 @@ data_cache::rd_miss_base( new_addr_type addr, // (already modified lower level) if(wb && (m_config.m_write_policy != WRITE_THROUGH) ){ mem_fetch *wb = m_memfetch_creator->alloc(evicted.m_block_addr, - m_wrbk_type,evicted.m_modified_size,true); + m_wrbk_type,evicted.m_modified_size,true,m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle); send_write_request(wb, WRITE_BACK_REQUEST_SENT, time, events); } return MISS; diff --git a/src/gpgpu-sim/gpu-cache.h b/src/gpgpu-sim/gpu-cache.h index 337f710..be33d96 100644 --- a/src/gpgpu-sim/gpu-cache.h +++ b/src/gpgpu-sim/gpu-cache.h @@ -1290,12 +1290,13 @@ public: data_cache( const char *name, cache_config &config, int core_id, int type_id, mem_fetch_interface *memport, mem_fetch_allocator *mfcreator, enum mem_fetch_status status, - mem_access_type wr_alloc_type, mem_access_type wrbk_type ) + mem_access_type wr_alloc_type, mem_access_type wrbk_type, class gpgpu_sim* gpu ) : baseline_cache(name,config,core_id,type_id,memport,status) { init( mfcreator ); m_wr_alloc_type = wr_alloc_type; m_wrbk_type = wrbk_type; + m_gpu=gpu; } virtual ~data_cache() {} @@ -1353,16 +1354,19 @@ protected: enum mem_fetch_status status, tag_array* new_tag_array, mem_access_type wr_alloc_type, - mem_access_type wrbk_type) + mem_access_type wrbk_type, + class gpgpu_sim* gpu ) : baseline_cache(name, config, core_id, type_id, memport,status, new_tag_array) { init( mfcreator ); m_wr_alloc_type = wr_alloc_type; m_wrbk_type = wrbk_type; + m_gpu=gpu; } mem_access_type m_wr_alloc_type; // Specifies type of write allocate request (e.g., L1 or L2) mem_access_type m_wrbk_type; // Specifies type of writeback request (e.g., L1 or L2) + class gpgpu_sim* m_gpu; //! A general function that takes the result of a tag_array probe // and performs the correspding functions based on the cache configuration @@ -1519,8 +1523,8 @@ class l1_cache : public data_cache { public: l1_cache(const char *name, cache_config &config, int core_id, int type_id, mem_fetch_interface *memport, - mem_fetch_allocator *mfcreator, enum mem_fetch_status status ) - : data_cache(name,config,core_id,type_id,memport,mfcreator,status, L1_WR_ALLOC_R, L1_WRBK_ACC){} + mem_fetch_allocator *mfcreator, enum mem_fetch_status status, class gpgpu_sim* gpu ) + : data_cache(name,config,core_id,type_id,memport,mfcreator,status, L1_WR_ALLOC_R, L1_WRBK_ACC, gpu){} virtual ~l1_cache(){} @@ -1538,10 +1542,11 @@ protected: mem_fetch_interface *memport, mem_fetch_allocator *mfcreator, enum mem_fetch_status status, - tag_array* new_tag_array ) + tag_array* new_tag_array, + class gpgpu_sim* gpu) : data_cache( name, config, - core_id,type_id,memport,mfcreator,status, new_tag_array, L1_WR_ALLOC_R, L1_WRBK_ACC ){} + core_id,type_id,memport,mfcreator,status, new_tag_array, L1_WR_ALLOC_R, L1_WRBK_ACC, gpu ){} }; @@ -1551,8 +1556,8 @@ class l2_cache : public data_cache { public: l2_cache(const char *name, cache_config &config, int core_id, int type_id, mem_fetch_interface *memport, - mem_fetch_allocator *mfcreator, enum mem_fetch_status status ) - : data_cache(name,config,core_id,type_id,memport,mfcreator,status, L2_WR_ALLOC_R, L2_WRBK_ACC){} + mem_fetch_allocator *mfcreator, enum mem_fetch_status status, class gpgpu_sim* gpu ) + : data_cache(name,config,core_id,type_id,memport,mfcreator,status, L2_WR_ALLOC_R, L2_WRBK_ACC, gpu){} virtual ~l2_cache() {} diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index c1ba934..a557d6f 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -82,23 +82,6 @@ class gpgpu_sim_wrapper {}; bool g_interactive_debugger_enabled=false; - -unsigned long long gpu_sim_cycle = 0; -unsigned long long gpu_tot_sim_cycle = 0; - - -// performance counter for stalls due to congestion. -unsigned int gpu_stall_dramfull = 0; -unsigned int gpu_stall_icnt2sh = 0; -unsigned long long partiton_reqs_in_parallel = 0; -unsigned long long partiton_reqs_in_parallel_total = 0; -unsigned long long partiton_reqs_in_parallel_util = 0; -unsigned long long partiton_reqs_in_parallel_util_total = 0; -unsigned long long gpu_sim_cycle_parition_util = 0; -unsigned long long gpu_tot_sim_cycle_parition_util = 0; -unsigned long long partiton_replys_in_parallel = 0; -unsigned long long partiton_replys_in_parallel_total = 0; - tr1_hash_map<new_addr_type,unsigned> address_random_interleaving; /* Clock Domains */ @@ -112,8 +95,6 @@ tr1_hash_map<new_addr_type,unsigned> address_random_interleaving; #define MEM_LATENCY_STAT_IMPL - - #include "mem_latency_stat.h" void power_config::reg_options(class OptionParser * opp) @@ -731,7 +712,7 @@ gpgpu_sim::gpgpu_sim( const gpgpu_sim_config &config ) #endif m_shader_stats = new shader_core_stats(m_shader_config); - m_memory_stats = new memory_stats_t(m_config.num_shader(),m_shader_config,m_memory_config); + m_memory_stats = new memory_stats_t(m_config.num_shader(),m_shader_config,m_memory_config,this); average_pipeline_duty_cycle = (float *)malloc(sizeof(float)); active_sms=(float *)malloc(sizeof(float)); m_power_stats = new power_stat_t(m_shader_config,average_pipeline_duty_cycle,active_sms,m_shader_stats,m_memory_config,m_memory_stats); @@ -742,6 +723,16 @@ gpgpu_sim::gpgpu_sim( const gpgpu_sim_config &config ) m_total_cta_launched = 0; gpu_deadlock = false; + gpu_stall_dramfull = 0; + gpu_stall_icnt2sh = 0; + partiton_reqs_in_parallel = 0; + partiton_reqs_in_parallel_total = 0; + partiton_reqs_in_parallel_util = 0; + partiton_reqs_in_parallel_util_total = 0; + gpu_sim_cycle_parition_util = 0; + gpu_tot_sim_cycle_parition_util = 0; + partiton_replys_in_parallel = 0; + partiton_replys_in_parallel_total = 0; m_cluster = new simt_core_cluster*[m_shader_config->n_simt_clusters]; for (unsigned i=0;i<m_shader_config->n_simt_clusters;i++) @@ -750,7 +741,7 @@ gpgpu_sim::gpgpu_sim( const gpgpu_sim_config &config ) m_memory_partition_unit = new memory_partition_unit*[m_memory_config->m_n_mem]; m_memory_sub_partition = new memory_sub_partition*[m_memory_config->m_n_mem_sub_partition]; for (unsigned i=0;i<m_memory_config->m_n_mem;i++) { - m_memory_partition_unit[i] = new memory_partition_unit(i, m_memory_config, m_memory_stats); + m_memory_partition_unit[i] = new memory_partition_unit(i, m_memory_config, m_memory_stats, this); for (unsigned p = 0; p < m_memory_config->m_n_sub_partition_per_memory_channel; p++) { unsigned submpid = i * m_memory_config->m_n_sub_partition_per_memory_channel + p; m_memory_sub_partition[submpid] = m_memory_partition_unit[i]->get_sub_partition(p); @@ -1140,7 +1131,7 @@ void gpgpu_sim::gpu_print_stat() time_t curr_time; time(&curr_time); - unsigned long long elapsed_time = MAX( curr_time - g_simulation_starttime, 1 ); + unsigned long long elapsed_time = MAX( curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1 ); printf( "gpu_total_sim_rate=%u\n", (unsigned)( ( gpu_tot_sim_insn + gpu_sim_insn ) / elapsed_time ) ); //shader_print_l1_miss_stat( stdout ); @@ -1504,7 +1495,7 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel ) shader_CTA_count_log(m_sid, 1); SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: cta:%2u, start_tid:%4u, end_tid:%4u, initialized @(%lld,%lld)\n", - free_cta_hw_id, start_thread, end_thread, gpu_sim_cycle, gpu_tot_sim_cycle ); + free_cta_hw_id, start_thread, end_thread, m_gpu->gpu_sim_cycle, m_gpu->gpu_tot_sim_cycle ); } @@ -1710,7 +1701,7 @@ void gpgpu_sim::cycle() time_t days, hrs, minutes, sec; time_t curr_time; time(&curr_time); - unsigned long long elapsed_time = MAX(curr_time - g_simulation_starttime, 1); + unsigned long long elapsed_time = MAX(curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1); if ( (elapsed_time - last_liveness_message_time) >= m_config.liveness_message_freq && DTRACE(LIVENESS) ) { days = elapsed_time/(3600*24); hrs = elapsed_time/3600 - 24*days; @@ -1721,7 +1712,7 @@ void gpgpu_sim::cycle() for (unsigned i=0;i<m_shader_config->n_simt_clusters;i++) { m_cluster[i]->get_current_occupancy(active, total); } - DPRINTF(LIVENESS, "uArch: inst.: %lld (ipc=%4.1f, occ=%0.4f\% [%llu / %llu]) sim_rate=%u (inst/sec) elapsed = %u:%u:%02u:%02u / %s", + DPRINTFG(LIVENESS, "uArch: inst.: %lld (ipc=%4.1f, occ=%0.4f\% [%llu / %llu]) sim_rate=%u (inst/sec) elapsed = %u:%u:%02u:%02u / %s", gpu_tot_sim_insn + gpu_sim_insn, (double)gpu_sim_insn/(double)gpu_sim_cycle, float(active)/float(total) * 100, active, total, diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h index c8dad89..5ea5765 100644 --- a/src/gpgpu-sim/gpu-sim.h +++ b/src/gpgpu-sim/gpu-sim.h @@ -64,7 +64,6 @@ extern tr1_hash_map<new_addr_type,unsigned> address_random_interleaving; - enum dram_ctrl_t { DRAM_FIFO=0, DRAM_FRFCFS=1 @@ -292,9 +291,6 @@ struct memory_config { bool m_perf_sim_memcpy; }; -// global counters and flags (please try not to add to this list!!!) -extern unsigned long long gpu_sim_cycle; -extern unsigned long long gpu_tot_sim_cycle; extern bool g_interactive_debugger_enabled; class gpgpu_sim_config : public power_config, public gpgpu_functional_sim_config { @@ -566,6 +562,18 @@ public: occupancy_stats gpu_occupancy; occupancy_stats gpu_tot_occupancy; + // performance counter for stalls due to congestion. + unsigned int gpu_stall_dramfull; + unsigned int gpu_stall_icnt2sh; + unsigned long long partiton_reqs_in_parallel; + unsigned long long partiton_reqs_in_parallel_total; + unsigned long long partiton_reqs_in_parallel_util; + unsigned long long partiton_reqs_in_parallel_util_total; + unsigned long long gpu_sim_cycle_parition_util; + unsigned long long gpu_tot_sim_cycle_parition_util; + unsigned long long partiton_replys_in_parallel; + unsigned long long partiton_replys_in_parallel_total; + FuncCache get_cache_config(std::string kernel_name); void set_cache_config(std::string kernel_name, FuncCache cacheConfig ); diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc index 0edc3b7..62e70a7 100644 --- a/src/gpgpu-sim/l2cache.cc +++ b/src/gpgpu-sim/l2cache.cc @@ -46,7 +46,7 @@ #include "l2cache_trace.h" -mem_fetch * partition_mf_allocator::alloc(new_addr_type addr, mem_access_type type, unsigned size, bool wr ) const +mem_fetch * partition_mf_allocator::alloc(new_addr_type addr, mem_access_type type, unsigned size, bool wr, unsigned long long cycle ) const { assert( wr ); mem_access_t access( type, addr, size, wr ); @@ -56,22 +56,25 @@ mem_fetch * partition_mf_allocator::alloc(new_addr_type addr, mem_access_type ty -1, -1, -1, - m_memory_config ); + m_memory_config, + cycle); return mf; } memory_partition_unit::memory_partition_unit( unsigned partition_id, const struct memory_config *config, - class memory_stats_t *stats ) -: m_id(partition_id), m_config(config), m_stats(stats), m_arbitration_metadata(config) + class memory_stats_t *stats, + class gpgpu_sim* gpu) +: m_id(partition_id), m_config(config), m_stats(stats), m_arbitration_metadata(config), m_gpu(gpu) { - m_dram = new dram_t(m_id,m_config,m_stats,this); + m_dram = new dram_t(m_id,m_config,m_stats,this,gpu); m_sub_partition = new memory_sub_partition*[m_config->m_n_sub_partition_per_memory_channel]; for (unsigned p = 0; p < m_config->m_n_sub_partition_per_memory_channel; p++) { unsigned sub_partition_id = m_id * m_config->m_n_sub_partition_per_memory_channel + p; - m_sub_partition[p] = new memory_sub_partition(sub_partition_id, m_config, stats); + m_sub_partition[p] = new memory_sub_partition(sub_partition_id, m_config, stats, gpu); } + } void memory_partition_unit::handle_memcpy_to_gpu( size_t addr, unsigned global_subpart_id, mem_access_sector_mask_t mask ) @@ -80,7 +83,7 @@ void memory_partition_unit::handle_memcpy_to_gpu( size_t addr, unsigned global_s std::string mystring = mask.to_string<char,std::string::traits_type,std::string::allocator_type>(); MEMPART_DPRINTF("Copy Engine Request Received For Address=%llx, local_subpart=%u, global_subpart=%u, sector_mask=%s \n", addr, p, global_subpart_id, mystring.c_str()); - m_sub_partition[p]->force_l2_tag_update(addr,gpu_sim_cycle+gpu_tot_sim_cycle, mask); + m_sub_partition[p]->force_l2_tag_update(addr,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle, mask); } memory_partition_unit::~memory_partition_unit() @@ -218,7 +221,7 @@ void memory_partition_unit::dram_cycle() delete mf_return; } else { m_sub_partition[dest_spid]->dram_L2_queue_push(mf_return); - mf_return->set_status(IN_PARTITION_DRAM_TO_L2_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf_return->set_status(IN_PARTITION_DRAM_TO_L2_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_arbitration_metadata.return_credit(dest_spid); MEMPART_DPRINTF("mem_fetch request %p return from dram to sub partition %d\n", mf_return, dest_spid); } @@ -247,9 +250,9 @@ void memory_partition_unit::dram_cycle() MEMPART_DPRINTF("Issue mem_fetch request %p from sub partition %d to dram\n", mf, spid); dram_delay_t d; d.req = mf; - d.ready_cycle = gpu_sim_cycle+gpu_tot_sim_cycle + m_config->dram_latency; + d.ready_cycle = m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle + m_config->dram_latency; m_dram_latency_queue.push_back(d); - mf->set_status(IN_PARTITION_DRAM_LATENCY_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_PARTITION_DRAM_LATENCY_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_arbitration_metadata.borrow_credit(spid); break; // the DRAM should only accept one request per cycle } @@ -258,7 +261,7 @@ void memory_partition_unit::dram_cycle() // DRAM latency queue - if( !m_dram_latency_queue.empty() && ( (gpu_sim_cycle+gpu_tot_sim_cycle) >= m_dram_latency_queue.front().ready_cycle ) && !m_dram->full(m_dram_latency_queue.front().req->is_write()) ) { + if( !m_dram_latency_queue.empty() && ( (m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle) >= m_dram_latency_queue.front().ready_cycle ) && !m_dram->full(m_dram_latency_queue.front().req->is_write()) ) { mem_fetch* mf = m_dram_latency_queue.front().req; m_dram_latency_queue.pop_front(); m_dram->push(mf); @@ -310,11 +313,13 @@ void memory_partition_unit::print( FILE *fp ) const memory_sub_partition::memory_sub_partition( unsigned sub_partition_id, const struct memory_config *config, - class memory_stats_t *stats ) + class memory_stats_t *stats, + class gpgpu_sim* gpu) { m_id = sub_partition_id; m_config=config; m_stats=stats; + m_gpu = gpu; m_memcpy_cycle_offset = 0; assert(m_id < m_config->m_n_mem_sub_partition); @@ -325,7 +330,7 @@ memory_sub_partition::memory_sub_partition( unsigned sub_partition_id, m_mf_allocator = new partition_mf_allocator(config); if(!m_config->m_L2_config.disabled()) - m_L2cache = new l2_cache(L2c_name,m_config->m_L2_config,-1,-1,m_L2interface,m_mf_allocator,IN_PARTITION_L2_MISS_QUEUE); + m_L2cache = new l2_cache(L2c_name,m_config->m_L2_config,-1,-1,m_L2interface,m_mf_allocator,IN_PARTITION_L2_MISS_QUEUE, gpu); unsigned int icnt_L2; unsigned int L2_dram; @@ -357,7 +362,7 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) mem_fetch *mf = m_L2cache->next_access(); if(mf->get_access_type() != L2_WR_ALLOC_R){ // Don't pass write allocate read request back to upper level cache mf->set_reply(); - mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_L2_icnt_queue->push(mf); }else{ if(m_config->m_L2_config.m_write_alloc_policy == FETCH_ON_WRITE) @@ -365,7 +370,7 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) mem_fetch* original_wr_mf = mf->get_original_wr_mf(); assert(original_wr_mf); original_wr_mf->set_reply(); - original_wr_mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + original_wr_mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_L2_icnt_queue->push(original_wr_mf); } m_request_tracker.erase(mf); @@ -379,13 +384,13 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) mem_fetch *mf = m_dram_L2_queue->top(); if ( !m_config->m_L2_config.disabled() && m_L2cache->waiting_for_fill(mf) ) { if (m_L2cache->fill_port_free()) { - mf->set_status(IN_PARTITION_L2_FILL_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); - m_L2cache->fill(mf,gpu_sim_cycle+gpu_tot_sim_cycle+m_memcpy_cycle_offset); + mf->set_status(IN_PARTITION_L2_FILL_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); + m_L2cache->fill(mf,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle+m_memcpy_cycle_offset); m_dram_L2_queue->pop(); } } else if ( !m_L2_icnt_queue->full() ) { if(mf->is_write() && mf->get_type() == WRITE_ACK) - mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_L2_icnt_queue->push(mf); m_dram_L2_queue->pop(); } @@ -406,7 +411,7 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) bool port_free = m_L2cache->data_port_free(); if ( !output_full && port_free ) { std::list<cache_event> events; - enum cache_request_status status = m_L2cache->access(mf->get_addr(),mf,gpu_sim_cycle+gpu_tot_sim_cycle+m_memcpy_cycle_offset,events); + enum cache_request_status status = m_L2cache->access(mf->get_addr(),mf,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle+m_memcpy_cycle_offset,events); bool write_sent = was_write_sent(events); bool read_sent = was_read_sent(events); MEM_SUBPART_DPRINTF("Probing L2 cache Address=%llx, status=%u\n", mf->get_addr(), status); @@ -420,7 +425,7 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) delete mf; } else { mf->set_reply(); - mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_L2_icnt_queue->push(mf); } m_icnt_L2_queue->pop(); @@ -431,7 +436,7 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) } else if ( status != RESERVATION_FAIL ) { if(mf->is_write() && (m_config->m_L2_config.m_write_alloc_policy == FETCH_ON_WRITE || m_config->m_L2_config.m_write_alloc_policy == LAZY_FETCH_ON_READ) && !was_writeallocate_sent(events)) { mf->set_reply(); - mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_PARTITION_L2_TO_ICNT_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_L2_icnt_queue->push(mf); } // L2 cache accepted request @@ -444,7 +449,7 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) } } else { // L2 is disabled or non-texture access to texture-only L2 - mf->set_status(IN_PARTITION_L2_TO_DRAM_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_PARTITION_L2_TO_DRAM_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); m_L2_dram_queue->push(mf); m_icnt_L2_queue->pop(); } @@ -455,7 +460,7 @@ void memory_sub_partition::cache_cycle( unsigned cycle ) mem_fetch* mf = m_rop.front().req; m_rop.pop(); m_icnt_L2_queue->push(mf); - mf->set_status(IN_PARTITION_ICNT_TO_L2_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_PARTITION_ICNT_TO_L2_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); } } @@ -644,6 +649,7 @@ std::vector<mem_fetch*> memory_sub_partition::breakdown_request_to_sector_reques mf->get_sid(), mf->get_tpc(), mf->get_mem_config(), + m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle, mf); result.push_back(n_mf); @@ -673,13 +679,13 @@ void memory_sub_partition::push( mem_fetch* m_req, unsigned long long cycle ) m_request_tracker.insert(req); if( req->istexture() ) { m_icnt_L2_queue->push(req); - req->set_status(IN_PARTITION_ICNT_TO_L2_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + req->set_status(IN_PARTITION_ICNT_TO_L2_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); } else { rop_delay_t r; r.req = req; r.ready_cycle = cycle + m_config->rop_latency; m_rop.push(r); - req->set_status(IN_PARTITION_ROP_DELAY,gpu_sim_cycle+gpu_tot_sim_cycle); + req->set_status(IN_PARTITION_ROP_DELAY,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); } } } diff --git a/src/gpgpu-sim/l2cache.h b/src/gpgpu-sim/l2cache.h index beafdd3..8ff2666 100644 --- a/src/gpgpu-sim/l2cache.h +++ b/src/gpgpu-sim/l2cache.h @@ -42,12 +42,12 @@ public: { m_memory_config = config; } - virtual mem_fetch * alloc(const class warp_inst_t &inst, const mem_access_t &access) const + virtual mem_fetch * alloc(const class warp_inst_t &inst, const mem_access_t &access, unsigned long long cycle) const { abort(); return NULL; } - virtual mem_fetch * alloc(new_addr_type addr, mem_access_type type, unsigned size, bool wr) const; + virtual mem_fetch * alloc(new_addr_type addr, mem_access_type type, unsigned size, bool wr, unsigned long long cycle) const; private: const memory_config *m_memory_config; }; @@ -58,7 +58,7 @@ private: class memory_partition_unit { public: - memory_partition_unit( unsigned partition_id, const struct memory_config *config, class memory_stats_t *stats ); + memory_partition_unit( unsigned partition_id, const struct memory_config *config, class memory_stats_t *stats, class gpgpu_sim* gpu ); ~memory_partition_unit(); bool busy() const; @@ -93,6 +93,8 @@ public: unsigned get_mpid() const { return m_id; } + class gpgpu_sim* get_mgpu() const { return m_gpu; } + private: unsigned m_id; @@ -140,12 +142,14 @@ private: class mem_fetch* req; }; std::list<dram_delay_t> m_dram_latency_queue; + + class gpgpu_sim* m_gpu; }; class memory_sub_partition { public: - memory_sub_partition( unsigned sub_partition_id, const struct memory_config *config, class memory_stats_t *stats ); + memory_sub_partition( unsigned sub_partition_id, const struct memory_config *config, class memory_stats_t *stats, class gpgpu_sim* gpu ); ~memory_sub_partition(); unsigned get_id() const { return m_id; } @@ -196,6 +200,7 @@ private: const struct memory_config *m_config; class l2_cache *m_L2cache; class L2interface *m_L2interface; + class gpgpu_sim* m_gpu; partition_mf_allocator *m_mf_allocator; // model delay of ROP units with a fixed latency diff --git a/src/gpgpu-sim/l2cache_trace.h b/src/gpgpu-sim/l2cache_trace.h index 2235cdc..d2dd948 100644 --- a/src/gpgpu-sim/l2cache_trace.h +++ b/src/gpgpu-sim/l2cache_trace.h @@ -42,7 +42,7 @@ #define MEMPART_DPRINTF(...) do {\ if (MEMPART_DTRACE(MEMORY_PARTITION_UNIT)) {\ printf( MEMPART_PRINT_STR,\ - gpu_sim_cycle + gpu_tot_sim_cycle,\ + m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle,\ Trace::trace_streams_str[Trace::MEMORY_PARTITION_UNIT],\ get_mpid() );\ printf(__VA_ARGS__);\ @@ -52,7 +52,7 @@ #define MEM_SUBPART_DPRINTF(...) do {\ if (MEM_SUBPART_DTRACE(MEMORY_PARTITION_UNIT)) {\ printf( MEM_SUBPART_PRINT_STR,\ - gpu_sim_cycle + gpu_tot_sim_cycle,\ + m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle,\ Trace::trace_streams_str[Trace::MEMORY_SUBPARTITION_UNIT],\ m_id );\ printf(__VA_ARGS__);\ diff --git a/src/gpgpu-sim/mem_fetch.cc b/src/gpgpu-sim/mem_fetch.cc index a260a35..c9b0484 100644 --- a/src/gpgpu-sim/mem_fetch.cc +++ b/src/gpgpu-sim/mem_fetch.cc @@ -40,6 +40,7 @@ mem_fetch::mem_fetch( const mem_access_t &access, unsigned sid, unsigned tpc, const struct memory_config *config, + unsigned long long cycle, mem_fetch *m_original_mf, mem_fetch *m_original_wr_mf) @@ -58,10 +59,10 @@ mem_fetch::mem_fetch( const mem_access_t &access, config->m_address_mapping.addrdec_tlx(access.get_addr(),&m_raw_addr); m_partition_addr = config->m_address_mapping.partition_address(access.get_addr()); m_type = m_access.is_write()?WRITE_REQUEST:READ_REQUEST; - m_timestamp = gpu_sim_cycle + gpu_tot_sim_cycle; + m_timestamp = cycle; m_timestamp2 = 0; m_status = MEM_FETCH_INITIALIZED; - m_status_change = gpu_sim_cycle + gpu_tot_sim_cycle; + m_status_change = cycle; m_mem_config = config; icnt_flit_size = config->icnt_flit_size; original_mf = m_original_mf; diff --git a/src/gpgpu-sim/mem_fetch.h b/src/gpgpu-sim/mem_fetch.h index e5efffd..4eb3a52 100644 --- a/src/gpgpu-sim/mem_fetch.h +++ b/src/gpgpu-sim/mem_fetch.h @@ -56,6 +56,7 @@ public: unsigned sid, unsigned tpc, const struct memory_config *config, + unsigned long long cycle, mem_fetch *original_mf = NULL, mem_fetch *original_wr_mf = NULL); ~mem_fetch(); diff --git a/src/gpgpu-sim/mem_latency_stat.cc b/src/gpgpu-sim/mem_latency_stat.cc index 11624f4..04dc75b 100644 --- a/src/gpgpu-sim/mem_latency_stat.cc +++ b/src/gpgpu-sim/mem_latency_stat.cc @@ -42,7 +42,7 @@ #include <stdlib.h> #include <stdio.h> -memory_stats_t::memory_stats_t( unsigned n_shader, const struct shader_core_config *shader_config, const struct memory_config *mem_config ) +memory_stats_t::memory_stats_t( unsigned n_shader, const struct shader_core_config *shader_config, const struct memory_config *mem_config, const class gpgpu_sim* gpu ) { assert( mem_config->m_valid ); assert( shader_config->m_valid ); @@ -67,6 +67,7 @@ memory_stats_t::memory_stats_t( unsigned n_shader, const struct shader_core_conf m_n_shader=n_shader; m_memory_config=mem_config; + m_gpu=gpu; total_n_access=0; total_n_reads=0; total_n_writes=0; @@ -147,7 +148,7 @@ memory_stats_t::memory_stats_t( unsigned n_shader, const struct shader_core_conf unsigned memory_stats_t::memlatstat_done(mem_fetch *mf ) { unsigned mf_latency; - mf_latency = (gpu_sim_cycle+gpu_tot_sim_cycle) - mf->get_timestamp(); + mf_latency = (m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle) - mf->get_timestamp(); mf_num_lat_pw++; mf_tot_lat_pw += mf_latency; unsigned idx = LOGB2(mf_latency); @@ -167,7 +168,7 @@ void memory_stats_t::memlatstat_read_done(mem_fetch *mf) if (mf_latency > mf_max_lat_table[mf->get_tlx_addr().chip][mf->get_tlx_addr().bk]) mf_max_lat_table[mf->get_tlx_addr().chip][mf->get_tlx_addr().bk] = mf_latency; unsigned icnt2sh_latency; - icnt2sh_latency = (gpu_tot_sim_cycle+gpu_sim_cycle) - mf->get_return_timestamp(); + icnt2sh_latency = (m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle) - mf->get_return_timestamp(); tot_icnt2sh_latency += icnt2sh_latency; icnt2sh_lat_table[LOGB2(icnt2sh_latency)]++; if (icnt2sh_latency > max_icnt2sh_latency) @@ -201,7 +202,7 @@ void memory_stats_t::memlatstat_icnt2mem_pop(mem_fetch *mf) { if (m_memory_config->gpgpu_memlatency_stat) { unsigned icnt2mem_latency; - icnt2mem_latency = (gpu_tot_sim_cycle+gpu_sim_cycle) - mf->get_timestamp(); + icnt2mem_latency = (m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle) - mf->get_timestamp(); tot_icnt2mem_latency += icnt2mem_latency; icnt2mem_lat_table[LOGB2(icnt2mem_latency)]++; if (icnt2mem_latency > max_icnt2mem_latency) diff --git a/src/gpgpu-sim/mem_latency_stat.h b/src/gpgpu-sim/mem_latency_stat.h index 982b9ae..b9285c1 100644 --- a/src/gpgpu-sim/mem_latency_stat.h +++ b/src/gpgpu-sim/mem_latency_stat.h @@ -36,7 +36,8 @@ class memory_stats_t { public: memory_stats_t( unsigned n_shader, const struct shader_core_config *shader_config, - const struct memory_config *mem_config ); + const struct memory_config *mem_config, + const class gpgpu_sim* gpu); unsigned memlatstat_done( class mem_fetch *mf ); void memlatstat_read_done( class mem_fetch *mf ); @@ -54,6 +55,7 @@ public: const struct shader_core_config *m_shader_config; const struct memory_config *m_memory_config; + const class gpgpu_sim* m_gpu; unsigned max_mrq_latency; unsigned max_dq_latency; diff --git a/src/gpgpu-sim/scoreboard.cc b/src/gpgpu-sim/scoreboard.cc index ebec891..80f95c6 100644 --- a/src/gpgpu-sim/scoreboard.cc +++ b/src/gpgpu-sim/scoreboard.cc @@ -32,13 +32,15 @@ //Constructor -Scoreboard::Scoreboard( unsigned sid, unsigned n_warps ) +Scoreboard::Scoreboard( unsigned sid, unsigned n_warps, class gpgpu_t* gpu ) : longopregs() { m_sid = sid; //Initialize size of table reg_table.resize(n_warps); longopregs.resize(n_warps); + + m_gpu = gpu; } // Print scoreboard contents diff --git a/src/gpgpu-sim/scoreboard.h b/src/gpgpu-sim/scoreboard.h index 4a76ea3..a4baa19 100644 --- a/src/gpgpu-sim/scoreboard.h +++ b/src/gpgpu-sim/scoreboard.h @@ -38,7 +38,7 @@ class Scoreboard { public: - Scoreboard( unsigned sid, unsigned n_warps ); + Scoreboard( unsigned sid, unsigned n_warps, class gpgpu_t* gpu ); void reserveRegisters(const warp_inst_t *inst); void releaseRegisters(const warp_inst_t *inst); @@ -59,6 +59,8 @@ private: std::vector< std::set<unsigned> > reg_table; //Register that depend on a long operation (global, local or tex memory) std::vector< std::set<unsigned> > longopregs; + + class gpgpu_t* m_gpu; }; diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index 007ad42..69b619a 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -133,7 +133,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu, m_L1I = new read_only_cache( name,m_config->m_L1I_config,m_sid,get_shader_instruction_cache_id(),m_icnt,IN_L1I_MISS_QUEUE); m_warp.resize(m_config->max_warps_per_shader, shd_warp_t(this, warp_size)); - m_scoreboard = new Scoreboard(m_sid, m_config->max_warps_per_shader); + m_scoreboard = new Scoreboard(m_sid, m_config->max_warps_per_shader, gpu); //scedulers //must currently occur after all inputs have been initialized. @@ -767,7 +767,7 @@ void shader_core_ctx::fetch() m_inst_fetch_buffer = ifetch_buffer_t(m_warp[mf->get_wid()].get_pc(), mf->get_access_size(), mf->get_wid()); assert( m_warp[mf->get_wid()].get_pc() == (mf->get_addr()-PROGRAM_MEM_START)); // Verify that we got the instruction we were expecting. m_inst_fetch_buffer.m_valid = true; - m_warp[mf->get_wid()].set_last_fetch(gpu_sim_cycle); + m_warp[mf->get_wid()].set_last_fetch(m_gpu->gpu_sim_cycle); delete mf; } else { @@ -815,17 +815,19 @@ void shader_core_ctx::fetch() warp_id, m_sid, m_tpc, - m_memory_config ); + m_memory_config, + m_gpu->gpu_tot_sim_cycle+m_gpu->gpu_sim_cycle + ); std::list<cache_event> events; - enum cache_request_status status = m_L1I->access( (new_addr_type)ppc, mf, gpu_sim_cycle+gpu_tot_sim_cycle,events); + enum cache_request_status status = m_L1I->access( (new_addr_type)ppc, mf, m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle,events); if( status == MISS ) { m_last_warp_fetched=warp_id; m_warp[warp_id].set_imiss_pending(); - m_warp[warp_id].set_last_fetch(gpu_sim_cycle); + m_warp[warp_id].set_last_fetch(m_gpu->gpu_sim_cycle); } else if( status == HIT ) { m_last_warp_fetched=warp_id; m_inst_fetch_buffer = ifetch_buffer_t(pc,nbytes,warp_id); - m_warp[warp_id].set_last_fetch(gpu_sim_cycle); + m_warp[warp_id].set_last_fetch(m_gpu->gpu_sim_cycle); delete mf; } else { m_last_warp_fetched=warp_id; @@ -859,7 +861,7 @@ void shader_core_ctx::issue_warp( register_set& pipe_reg_set, const warp_inst_t* m_warp[warp_id].ibuffer_free(); assert(next_inst->valid()); **pipe_reg = *next_inst; // static instruction information - (*pipe_reg)->issue( active_mask, warp_id, gpu_tot_sim_cycle + gpu_sim_cycle, m_warp[warp_id].get_dynamic_warp_id(), sch_id ); // dynamic instruction information + (*pipe_reg)->issue( active_mask, warp_id, m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle, m_warp[warp_id].get_dynamic_warp_id(), sch_id ); // dynamic instruction information m_stats->shader_cycle_distro[2+(*pipe_reg)->active_count()]++; func_exec_inst( **pipe_reg ); if( next_inst->op == BARRIER_OP ){ @@ -1514,7 +1516,7 @@ void shader_core_ctx::warp_inst_complete(const warp_inst_t &inst) m_stats->m_num_sim_winsn[m_sid]++; m_gpu->gpu_sim_insn += inst.active_count(); - inst.completed(gpu_tot_sim_cycle + gpu_sim_cycle); + inst.completed(m_gpu->gpu_tot_sim_cycle + m_gpu->gpu_sim_cycle); } void shader_core_ctx::writeback() @@ -1552,9 +1554,9 @@ void shader_core_ctx::writeback() m_warp[warp_id].dec_inst_in_pipeline(); warp_inst_complete(*pipe_reg); m_gpu->gpu_sim_insn_last_update_sid = m_sid; - m_gpu->gpu_sim_insn_last_update = gpu_sim_cycle; - m_last_inst_gpu_sim_cycle = gpu_sim_cycle; - m_last_inst_gpu_tot_sim_cycle = gpu_tot_sim_cycle; + m_gpu->gpu_sim_insn_last_update = m_gpu->gpu_sim_cycle; + m_last_inst_gpu_sim_cycle = m_gpu->gpu_sim_cycle; + m_last_inst_gpu_tot_sim_cycle = m_gpu->gpu_tot_sim_cycle; pipe_reg->clear(); preg = m_pipeline_reg[EX_WB].get_ready(); pipe_reg = (preg==NULL)? NULL:*preg; @@ -1633,9 +1635,9 @@ mem_stage_stall_type ldst_unit::process_memory_access_queue( cache_t *cache, war return DATA_PORT_STALL; //const mem_access_t &access = inst.accessq_back(); - mem_fetch *mf = m_mf_allocator->alloc(inst,inst.accessq_back()); + mem_fetch *mf = m_mf_allocator->alloc(inst,inst.accessq_back(),m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); std::list<cache_event> events; - enum cache_request_status status = cache->access(mf->get_addr(),mf,gpu_sim_cycle+gpu_tot_sim_cycle,events); + enum cache_request_status status = cache->access(mf->get_addr(),mf,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle,events); return process_cache_access( cache, mf->get_addr(), inst, events, mf, status ); } @@ -1645,7 +1647,7 @@ mem_stage_stall_type ldst_unit::process_memory_access_queue_l1cache( l1_cache *c if( inst.accessq_empty() ) return result; - mem_fetch *mf = m_mf_allocator->alloc(inst,inst.accessq_back()); + mem_fetch *mf = m_mf_allocator->alloc(inst,inst.accessq_back(),m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); if(m_config->m_L1D_config.l1_latency > 0) { @@ -1675,7 +1677,7 @@ mem_stage_stall_type ldst_unit::process_memory_access_queue_l1cache( l1_cache *c else { std::list<cache_event> events; - enum cache_request_status status = cache->access(mf->get_addr(),mf,gpu_sim_cycle+gpu_tot_sim_cycle,events); + enum cache_request_status status = cache->access(mf->get_addr(),mf,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle,events); return process_cache_access( cache, mf->get_addr(), inst, events, mf, status ); } } @@ -1687,7 +1689,7 @@ void ldst_unit::L1_latency_queue_cycle() { mem_fetch* mf_next = l1_latency_queue[0]; std::list<cache_event> events; - enum cache_request_status status = m_L1D->access(mf_next->get_addr(),mf_next,gpu_sim_cycle+gpu_tot_sim_cycle,events); + enum cache_request_status status = m_L1D->access(mf_next->get_addr(),mf_next,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle,events); bool write_sent = was_write_sent(events); bool read_sent = was_read_sent(events); @@ -1804,7 +1806,7 @@ bool ldst_unit::memory_cycle( warp_inst_t &inst, mem_stage_stall_type &stall_rea if( m_icnt->full(size, inst.is_store() || inst.isatomic()) ) { stall_cond = ICNT_RC_FAIL; } else { - mem_fetch *mf = m_mf_allocator->alloc(inst,access); + mem_fetch *mf = m_mf_allocator->alloc(inst,access,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); m_icnt->push(mf); inst.accessq_pop_back(); //inst.clear_active( access.get_warp_mask() ); @@ -1840,7 +1842,7 @@ bool ldst_unit::response_buffer_full() const void ldst_unit::fill( mem_fetch *mf ) { - mf->set_status(IN_SHADER_LDST_RESPONSE_FIFO,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_SHADER_LDST_RESPONSE_FIFO,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); m_response_fifo.push_back(mf); } @@ -2115,7 +2117,8 @@ ldst_unit::ldst_unit( mem_fetch_interface *icnt, get_shader_normal_cache_id(), m_icnt, m_mf_allocator, - IN_L1D_MISS_QUEUE ); + IN_L1D_MISS_QUEUE, + core->get_gpu()); if(m_config->m_L1D_config.l1_latency > 0) { @@ -2202,8 +2205,8 @@ void ldst_unit::writeback() m_core->warp_inst_complete(m_next_wb); } m_next_wb.clear(); - m_last_inst_gpu_sim_cycle = gpu_sim_cycle; - m_last_inst_gpu_tot_sim_cycle = gpu_tot_sim_cycle; + m_last_inst_gpu_sim_cycle = m_core->get_gpu()->gpu_sim_cycle; + m_last_inst_gpu_tot_sim_cycle = m_core->get_gpu()->gpu_tot_sim_cycle; } } @@ -2311,13 +2314,13 @@ void ldst_unit::cycle() mem_fetch *mf = m_response_fifo.front(); if (mf->get_access_type() == TEXTURE_ACC_R) { if (m_L1T->fill_port_free()) { - m_L1T->fill(mf,gpu_sim_cycle+gpu_tot_sim_cycle); + m_L1T->fill(mf,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); m_response_fifo.pop_front(); } } else if (mf->get_access_type() == CONST_ACC_R) { if (m_L1C->fill_port_free()) { - mf->set_status(IN_SHADER_FETCHED,gpu_sim_cycle+gpu_tot_sim_cycle); - m_L1C->fill(mf,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_SHADER_FETCHED,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); + m_L1C->fill(mf,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); m_response_fifo.pop_front(); } } else { @@ -2337,13 +2340,13 @@ void ldst_unit::cycle() } if( bypassL1D ) { if ( m_next_global == NULL ) { - mf->set_status(IN_SHADER_FETCHED,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_SHADER_FETCHED,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); m_response_fifo.pop_front(); m_next_global = mf; } } else { if (m_L1D->fill_port_free()) { - m_L1D->fill(mf,gpu_sim_cycle+gpu_tot_sim_cycle); + m_L1D->fill(mf,m_core->get_gpu()->gpu_sim_cycle+m_core->get_gpu()->gpu_tot_sim_cycle); m_response_fifo.pop_front(); } } @@ -2432,7 +2435,7 @@ void shader_core_ctx::register_cta_thread_exit( unsigned cta_num, kernel_info_t shader_CTA_count_unlog(m_sid, 1); SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Finished CTA #%d (%lld,%lld), %u CTAs running\n", - cta_num, gpu_sim_cycle, gpu_tot_sim_cycle, m_n_active_cta); + cta_num, m_gpu->gpu_sim_cycle, m_gpu->gpu_tot_sim_cycle, m_n_active_cta); if( m_n_active_cta == 0 ) { SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Empty (last released kernel %u \'%s\').\n", @@ -2823,7 +2826,7 @@ void shader_core_ctx::display_pipeline(FILE *fout, int print_mem, int mask ) con { fprintf(fout, "=================================================\n"); fprintf(fout, "shader %u at cycle %Lu+%Lu (%u threads running)\n", m_sid, - gpu_tot_sim_cycle, gpu_sim_cycle, m_not_completed); + m_gpu->gpu_tot_sim_cycle, m_gpu->gpu_sim_cycle, m_not_completed); fprintf(fout, "=================================================\n"); dump_warp_state(fout); @@ -3221,7 +3224,7 @@ void barrier_set_t::warp_reaches_barrier(unsigned cta_id,unsigned warp_id,warp_i cta_to_warp_t::iterator w=m_cta_to_warps.find(cta_id); if( w == m_cta_to_warps.end() ) { // cta is active - printf("ERROR ** cta_id %u not found in barrier set on cycle %llu+%llu...\n", cta_id, gpu_tot_sim_cycle, gpu_sim_cycle ); + printf("ERROR ** cta_id %u not found in barrier set on cycle %llu+%llu...\n", cta_id, m_shader->get_gpu()->gpu_tot_sim_cycle, m_shader->get_gpu()->gpu_sim_cycle ); dump(); abort(); } @@ -3394,8 +3397,8 @@ bool shader_core_ctx::fetch_unit_response_buffer_full() const void shader_core_ctx::accept_fetch_response( mem_fetch *mf ) { - mf->set_status(IN_SHADER_FETCHED,gpu_sim_cycle+gpu_tot_sim_cycle); - m_L1I->fill(mf,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_SHADER_FETCHED,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); + m_L1I->fill(mf,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); } bool shader_core_ctx::ldst_unit_response_buffer_full() const @@ -3957,7 +3960,7 @@ void simt_core_cluster::icnt_inject_request_packet(class mem_fetch *mf) } m_stats->m_outgoing_traffic_stats->record_traffic(mf, packet_size); unsigned destination = mf->get_sub_partition_id(); - mf->set_status(IN_ICNT_TO_MEM,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_ICNT_TO_MEM,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); if (!mf->get_is_write() && !mf->isatomic()) ::icnt_push(m_cluster_id, m_config->mem2device(destination), (void*)mf, mf->get_ctrl_size() ); else @@ -3996,7 +3999,7 @@ void simt_core_cluster::icnt_cycle() // - For write-ack, the packet only has control metadata unsigned int packet_size = (mf->get_is_write())? mf->get_ctrl_size() : mf->size(); m_stats->m_incoming_traffic_stats->record_traffic(mf, packet_size); - mf->set_status(IN_CLUSTER_TO_SHADER_QUEUE,gpu_sim_cycle+gpu_tot_sim_cycle); + mf->set_status(IN_CLUSTER_TO_SHADER_QUEUE,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle); //m_memory_stats->memlatstat_read_done(mf,m_shader_config->max_warps_per_shader); m_response_fifo.push_back(mf); m_stats->n_mem_to_simt[m_cluster_id] += mf->get_num_flits(false); diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index a0c2b63..fde87b6 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -1730,7 +1730,7 @@ public: m_cluster_id = cluster_id; m_memory_config = config; } - mem_fetch *alloc( new_addr_type addr, mem_access_type type, unsigned size, bool wr ) const + mem_fetch *alloc( new_addr_type addr, mem_access_type type, unsigned size, bool wr, unsigned long long cycle ) const { mem_access_t access( type, addr, size, wr ); mem_fetch *mf = new mem_fetch( access, @@ -1739,11 +1739,12 @@ public: -1, m_core_id, m_cluster_id, - m_memory_config ); + m_memory_config, + cycle); return mf; } - mem_fetch *alloc( const warp_inst_t &inst, const mem_access_t &access ) const + mem_fetch *alloc( const warp_inst_t &inst, const mem_access_t &access, unsigned long long cycle ) const { warp_inst_t inst_copy = inst; mem_fetch *mf = new mem_fetch(access, @@ -1752,7 +1753,8 @@ public: inst.warp_id(), m_core_id, m_cluster_id, - m_memory_config); + m_memory_config, + cycle); return mf; } diff --git a/src/gpgpu-sim/shader_trace.h b/src/gpgpu-sim/shader_trace.h index de3e059..ac4e894 100644 --- a/src/gpgpu-sim/shader_trace.h +++ b/src/gpgpu-sim/shader_trace.h @@ -44,7 +44,7 @@ #define SHADER_DPRINTF(x, ...) do {\ if (SHADER_DTRACE(x)) {\ printf( SHADER_PRINT_STR,\ - gpu_sim_cycle + gpu_tot_sim_cycle,\ + m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle,\ Trace::trace_streams_str[Trace::x],\ get_sid() );\ printf(__VA_ARGS__);\ @@ -56,7 +56,7 @@ #define SCHED_DPRINTF(...) do {\ if (SHADER_DTRACE(WARP_SCHEDULER)) {\ printf( SCHED_PRINT_STR,\ - gpu_sim_cycle + gpu_tot_sim_cycle,\ + m_shader->get_gpu()->gpu_sim_cycle + m_shader->get_gpu()->gpu_tot_sim_cycle,\ Trace::trace_streams_str[Trace::WARP_SCHEDULER],\ get_sid(),\ m_id );\ diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc index 9e2bfa2..de937b0 100644 --- a/src/gpgpusim_entrypoint.cc +++ b/src/gpgpusim_entrypoint.cc @@ -36,31 +36,25 @@ #include "gpgpu-sim/icnt_wrapper.h" #include "stream_manager.h" -#include <pthread.h> -#include <semaphore.h> - #define MAX(a,b) (((a)>(b))?(a):(b)) +struct GPGPUsim_ctx* the_gpgpusim = NULL; -struct gpgpu_ptx_sim_arg *grid_params; - -sem_t g_sim_signal_start; -sem_t g_sim_signal_finish; -sem_t g_sim_signal_exit; -time_t g_simulation_starttime; -pthread_t g_simulation_thread; - -gpgpu_sim_config g_the_gpu_config; -gpgpu_sim *g_the_gpu; -stream_manager *g_stream_manager; - - +struct GPGPUsim_ctx* GPGPUsim_ctx_ptr(){ + if(the_gpgpusim == NULL) + the_gpgpusim = new GPGPUsim_ctx(); -static int sg_argc = 3; -static const char *sg_argv[] = {"", "-config","gpgpusim.config"}; + return the_gpgpusim; +} +class gpgpu_sim* g_the_gpu() { + return GPGPUsim_ctx_ptr()->g_the_gpu; +} +class stream_manager* g_stream_manager() { + return GPGPUsim_ctx_ptr()->g_stream_manager; +} static void print_simulation_time(); @@ -69,29 +63,26 @@ void *gpgpu_sim_thread_sequential(void*) // at most one kernel running at a time bool done; do { - sem_wait(&g_sim_signal_start); + sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_start)); done = true; - if( g_the_gpu->get_more_cta_left() ) { + if( GPGPUsim_ctx_ptr()->g_the_gpu->get_more_cta_left() ) { done = false; - g_the_gpu->init(); - while( g_the_gpu->active() ) { - g_the_gpu->cycle(); - g_the_gpu->deadlock_check(); + GPGPUsim_ctx_ptr()->g_the_gpu->init(); + while( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) { + GPGPUsim_ctx_ptr()->g_the_gpu->cycle(); + GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check(); } - g_the_gpu->print_stats(); - g_the_gpu->update_stats(); + GPGPUsim_ctx_ptr()->g_the_gpu->print_stats(); + GPGPUsim_ctx_ptr()->g_the_gpu->update_stats(); print_simulation_time(); } - sem_post(&g_sim_signal_finish); + sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish)); } while(!done); - sem_post(&g_sim_signal_exit); + sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit)); return NULL; } -pthread_mutex_t g_sim_lock = PTHREAD_MUTEX_INITIALIZER; -bool g_sim_active = false; -bool g_sim_done = true; -bool break_limit = false; + static void termination_callback() { @@ -108,19 +99,19 @@ void *gpgpu_sim_thread_concurrent(void*) printf("GPGPU-Sim: *** simulation thread starting and spinning waiting for work ***\n"); fflush(stdout); } - while( g_stream_manager->empty_protected() && !g_sim_done ) + while( GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected() && !GPGPUsim_ctx_ptr()->g_sim_done ) ; if(g_debug_execution >= 3) { printf("GPGPU-Sim: ** START simulation thread (detected work) **\n"); - g_stream_manager->print(stdout); + GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout); fflush(stdout); } - pthread_mutex_lock(&g_sim_lock); - g_sim_active = true; - pthread_mutex_unlock(&g_sim_lock); + pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); + GPGPUsim_ctx_ptr()->g_sim_active = true; + pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); bool active = false; bool sim_cycles = false; - g_the_gpu->init(); + GPGPUsim_ctx_ptr()->g_the_gpu->init(); do { // check if a kernel has completed // launch operation on device if one is pending and can be run @@ -132,70 +123,70 @@ void *gpgpu_sim_thread_concurrent(void*) // another kernel, the gpu is not re-initialized and the inter-kernel // behaviour may be incorrect. Check that a kernel has finished and // no other kernel is currently running. - if(g_stream_manager->operation(&sim_cycles) && !g_the_gpu->active()) + if(GPGPUsim_ctx_ptr()->g_stream_manager->operation(&sim_cycles) && !GPGPUsim_ctx_ptr()->g_the_gpu->active()) break; //functional simulation - if( g_the_gpu->is_functional_sim()) { - kernel_info_t * kernel = g_the_gpu->get_functional_kernel(); + if( GPGPUsim_ctx_ptr()->g_the_gpu->is_functional_sim()) { + kernel_info_t * kernel = GPGPUsim_ctx_ptr()->g_the_gpu->get_functional_kernel(); assert(kernel); gpgpu_cuda_ptx_sim_main_func(*kernel); - g_the_gpu->finish_functional_sim(kernel); + GPGPUsim_ctx_ptr()->g_the_gpu->finish_functional_sim(kernel); } //performance simulation - if( g_the_gpu->active() ) { - g_the_gpu->cycle(); - sim_cycles = true; - g_the_gpu->deadlock_check(); + if( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) { + GPGPUsim_ctx_ptr()->g_the_gpu->cycle(); + sim_cycles = true; + GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check(); }else { - if(g_the_gpu->cycle_insn_cta_max_hit()){ - g_stream_manager->stop_all_running_kernels(); - g_sim_done = true; - break_limit = true; + if(GPGPUsim_ctx_ptr()->g_the_gpu->cycle_insn_cta_max_hit()){ + GPGPUsim_ctx_ptr()->g_stream_manager->stop_all_running_kernels(); + GPGPUsim_ctx_ptr()->g_sim_done = true; + GPGPUsim_ctx_ptr()->break_limit = true; } } - active=g_the_gpu->active() || !g_stream_manager->empty_protected(); + active=GPGPUsim_ctx_ptr()->g_the_gpu->active() || !(GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected()); - } while( active && !g_sim_done); + } while( active && !GPGPUsim_ctx_ptr()->g_sim_done); if(g_debug_execution >= 3) { printf("GPGPU-Sim: ** STOP simulation thread (no work) **\n"); fflush(stdout); } if(sim_cycles) { - g_the_gpu->print_stats(); - g_the_gpu->update_stats(); + GPGPUsim_ctx_ptr()->g_the_gpu->print_stats(); + GPGPUsim_ctx_ptr()->g_the_gpu->update_stats(); print_simulation_time(); } - pthread_mutex_lock(&g_sim_lock); - g_sim_active = false; - pthread_mutex_unlock(&g_sim_lock); - } while( !g_sim_done ); + pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); + GPGPUsim_ctx_ptr()->g_sim_active = false; + pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); + } while( !GPGPUsim_ctx_ptr()->g_sim_done ); printf("GPGPU-Sim: *** simulation thread exiting ***\n"); fflush(stdout); - if(break_limit) { + if(GPGPUsim_ctx_ptr()->break_limit) { printf("GPGPU-Sim: ** break due to reaching the maximum cycles (or instructions) **\n"); exit(1); } - sem_post(&g_sim_signal_exit); + sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit)); return NULL; } void synchronize() { printf("GPGPU-Sim: synchronize waiting for inactive GPU simulation\n"); - g_stream_manager->print(stdout); + GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout); fflush(stdout); // sem_wait(&g_sim_signal_finish); bool done = false; do { - pthread_mutex_lock(&g_sim_lock); - done = ( g_stream_manager->empty() && !g_sim_active ) || g_sim_done; - pthread_mutex_unlock(&g_sim_lock); + pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); + done = ( GPGPUsim_ctx_ptr()->g_stream_manager->empty() && !GPGPUsim_ctx_ptr()->g_sim_active ) || GPGPUsim_ctx_ptr()->g_sim_done; + pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); } while (!done); printf("GPGPU-Sim: detected inactive GPU simulation thread\n"); fflush(stdout); @@ -204,10 +195,10 @@ void synchronize() void exit_simulation() { - g_sim_done=true; + GPGPUsim_ctx_ptr()->g_sim_done=true; printf("GPGPU-Sim: exit_simulation called\n"); fflush(stdout); - sem_wait(&g_sim_signal_exit); + sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit)); printf("GPGPU-Sim: simulation thread signaled exit\n"); fflush(stdout); } @@ -226,36 +217,37 @@ gpgpu_sim *gpgpu_ptx_sim_init_perf() ptx_opcocde_latency_options(opp); icnt_reg_options(opp); - g_the_gpu_config.reg_options(opp); // register GPU microrachitecture options + GPGPUsim_ctx_ptr()->g_the_gpu_config = new gpgpu_sim_config(); + GPGPUsim_ctx_ptr()->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options - option_parser_cmdline(opp, sg_argc, sg_argv); // parse configuration options + option_parser_cmdline(opp, GPGPUsim_ctx_ptr()->sg_argc, GPGPUsim_ctx_ptr()->sg_argv); // parse configuration options fprintf(stdout, "GPGPU-Sim: Configuration options:\n\n"); option_parser_print(opp, stdout); // Set the Numeric locale to a standard locale where a decimal point is a "dot" not a "comma" // so it does the parsing correctly independent of the system environment variables assert(setlocale(LC_NUMERIC,"C")); - g_the_gpu_config.init(); + GPGPUsim_ctx_ptr()->g_the_gpu_config->init(); - g_the_gpu = new gpgpu_sim(g_the_gpu_config); - g_stream_manager = new stream_manager(g_the_gpu,g_cuda_launch_blocking); + GPGPUsim_ctx_ptr()->g_the_gpu = new gpgpu_sim(*(GPGPUsim_ctx_ptr()->g_the_gpu_config)); + GPGPUsim_ctx_ptr()->g_stream_manager = new stream_manager((GPGPUsim_ctx_ptr()->g_the_gpu),g_cuda_launch_blocking); - g_simulation_starttime = time((time_t *)NULL); + GPGPUsim_ctx_ptr()->g_simulation_starttime = time((time_t *)NULL); - sem_init(&g_sim_signal_start,0,0); - sem_init(&g_sim_signal_finish,0,0); - sem_init(&g_sim_signal_exit,0,0); + sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_start),0,0); + sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish),0,0); + sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit),0,0); - return g_the_gpu; + return GPGPUsim_ctx_ptr()->g_the_gpu; } void start_sim_thread(int api) { - if( g_sim_done ) { - g_sim_done = false; + if( GPGPUsim_ctx_ptr()->g_sim_done ) { + GPGPUsim_ctx_ptr()->g_sim_done = false; if( api == 1 ) { - pthread_create(&g_simulation_thread,NULL,gpgpu_sim_thread_concurrent,NULL); + pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_concurrent,NULL); } else { - pthread_create(&g_simulation_thread,NULL,gpgpu_sim_thread_sequential,NULL); + pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_sequential,NULL); } } } @@ -264,7 +256,7 @@ void print_simulation_time() { time_t current_time, difference, d, h, m, s; current_time = time((time_t *)NULL); - difference = MAX(current_time - g_simulation_starttime, 1); + difference = MAX(current_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1); d = difference/(3600*24); h = difference/3600 - 24*d; @@ -274,16 +266,16 @@ void print_simulation_time() fflush(stderr); printf("\n\ngpgpu_simulation_time = %u days, %u hrs, %u min, %u sec (%u sec)\n", (unsigned)d, (unsigned)h, (unsigned)m, (unsigned)s, (unsigned)difference ); - printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(g_the_gpu->gpu_tot_sim_insn / difference) ); - printf("gpgpu_simulation_rate = %u (cycle/sec)\n", (unsigned)(gpu_tot_sim_cycle / difference) ); + printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_insn / difference) ); + printf("gpgpu_simulation_rate = %u (cycle/sec)\n", (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_cycle / difference) ); fflush(stdout); } int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ) { - g_the_gpu->launch(grid); - sem_post(&g_sim_signal_start); - sem_wait(&g_sim_signal_finish); + GPGPUsim_ctx_ptr()->g_the_gpu->launch(grid); + sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_start)); + sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish)); return 0; } diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h index bf03f03..406dd00 100644 --- a/src/gpgpusim_entrypoint.h +++ b/src/gpgpusim_entrypoint.h @@ -29,15 +29,64 @@ #define GPGPUSIM_ENTRYPOINT_H_INCLUDED #include "abstract_hardware_model.h" - +#include <pthread.h> +#include <semaphore.h> #include <time.h> -extern time_t g_simulation_starttime; +//extern time_t g_simulation_starttime; + + +struct GPGPUsim_ctx { + + GPGPUsim_ctx() { + g_sim_active = false; + g_sim_done = true; + break_limit = false; + g_sim_lock = PTHREAD_MUTEX_INITIALIZER; + + sg_argc = 3; + sg_argv = {"", "-config","gpgpusim.config"}; + + g_the_gpu_config=NULL; + g_the_gpu=NULL; + g_stream_manager=NULL; + the_cude_device=NULL; + the_context=NULL; + } + + //struct gpgpu_ptx_sim_arg *grid_params; + sem_t g_sim_signal_start; + sem_t g_sim_signal_finish; + sem_t g_sim_signal_exit; + time_t g_simulation_starttime; + pthread_t g_simulation_thread; + + class gpgpu_sim_config *g_the_gpu_config; + class gpgpu_sim *g_the_gpu; + class stream_manager *g_stream_manager; + + struct _cuda_device_id *the_cude_device; + struct CUctx_st* the_context; + + + int sg_argc; + const char *sg_argv[3]; + + pthread_mutex_t g_sim_lock; + bool g_sim_active; + bool g_sim_done; + bool break_limit; + +}; class gpgpu_sim *gpgpu_ptx_sim_init_perf(); void start_sim_thread(int api); +class gpgpu_sim* g_the_gpu(); +struct GPGPUsim_ctx* GPGPUsim_ctx_ptr(); +class stream_manager* g_stream_manager(); + int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ); int gpgpu_opencl_ptx_sim_main_func( kernel_info_t *grid ); diff --git a/src/intersim2/Makefile b/src/intersim2/Makefile index 7d10b3f..3eeeb70 100644 --- a/src/intersim2/Makefile +++ b/src/intersim2/Makefile @@ -48,6 +48,7 @@ CPPFLAGS += -O3 endif CPPFLAGS += -g CPPFLAGS += -fPIC +CPPFLAGS += -I$(CUDA_INSTALL_PATH)/include LFLAGS += diff --git a/src/stream_manager.cc b/src/stream_manager.cc index 6cd62a2..e07f4e4 100644 --- a/src/stream_manager.cc +++ b/src/stream_manager.cc @@ -190,7 +190,7 @@ bool stream_operation::do_operation( gpgpu_sim *gpu ) case stream_event: { printf("event update\n"); time_t wallclock = time((time_t *)NULL); - m_event->update( gpu_tot_sim_cycle, wallclock ); + m_event->update( gpu->gpu_tot_sim_cycle, wallclock ); m_stream->record_next_done(); } break; diff --git a/src/trace.h b/src/trace.h index a79b4a0..0b96dcf 100644 --- a/src/trace.h +++ b/src/trace.h @@ -31,9 +31,6 @@ #ifndef __TRACE_H__ #define __TRACE_H__ -extern unsigned long long gpu_sim_cycle; -extern unsigned long long gpu_tot_sim_cycle; - namespace Trace { #define TS_TUP_BEGIN(X) enum X { @@ -63,17 +60,26 @@ namespace Trace { #define DPRINTF(x, ...) do {\ if (DTRACE(x)) {\ printf( SIM_PRINT_STR,\ - gpu_sim_cycle + gpu_tot_sim_cycle,\ + m_gpu->gpu_sim_cycle + m_gpu->gpu_tot_sim_cycle,\ Trace::trace_streams_str[Trace::x] );\ printf(__VA_ARGS__);\ }\ } while (0) +#define DPRINTFG(x, ...) do {\ + if (DTRACE(x)) {\ + printf( SIM_PRINT_STR,\ + gpu_sim_cycle + gpu_tot_sim_cycle,\ + Trace::trace_streams_str[Trace::x] );\ + printf(__VA_ARGS__);\ + }\ +} while (0) #else #define DTRACE(x) (false) #define DPRINTF(x, ...) do {} while (0) +#define DPRINTFG(x, ...) do {} while (0) #endif diff --git a/version_detection.mk b/version_detection.mk index 00a86c0..ee71a62 100644 --- a/version_detection.mk +++ b/version_detection.mk @@ -33,8 +33,9 @@ GPGPUSIM_VERSION=$(shell cat $(GPGPUSIM_ROOT)/version | awk '/Version/ {print $$ #Detect Git branch and commit # GIT_COMMIT := $(shell git log -n 1 | head -1 | sed -re 's/commit (.*)/\1/') -GIT_FILES_CHANGED := $(shell git diff --numstat --cached && git diff --numstat | wc | sed -re 's/^\s+([0-9]+).*/\1/') -GPGPUSIM_BUILD := "gpgpu-sim_git-commit-$(GIT_COMMIT)_modified_$(GIT_FILES_CHANGED)" +GIT_FILES_CHANGED_A:=$(shell git diff --numstat | wc | sed -re 's/^\s+([0-9]+).*/\1./') +GIT_FILES_CHANGED:= $(GIT_FILES_CHANGED_A)$(shell git diff --numstat --cached | wc | sed -re 's/^\s+([0-9]+).*/\1/') +GPGPUSIM_BUILD := gpgpu-sim_git-commit-$(GIT_COMMIT)_modified_$(GIT_FILES_CHANGED) endif # Detect CUDA Runtime Version |