Merge pull request #1 from speverel/dev

Dev

10 files changed, 21 insertions, 7 deletions

diff --git a/configs/GeForceGTX750Ti/gpgpusim.config b/configs/GeForceGTX750Ti/gpgpusim.config
index afd3825..be3ae3c 100644
--- a/configs/GeForceGTX750Ti/gpgpusim.config
+++ b/configs/GeForceGTX750Ti/gpgpusim.config
@@ -42,8 +42,8 @@
 -ptx_opcode_latency_dp 8,19,8,8,330
 -ptx_opcode_initiation_dp 8,16,8,8,130
 
--gpgpu_cache:dl1 none
-# 32:128:4,L:L:m:N:H,A:32:8,8
+-gpgpu_cache:dl1 32:128:4,L:L:m:N:H,A:32:8,8
+-gmem_skip_L1D 1
 -gpgpu_shmem_size 65536
 
 # The alternative configuration for fermi in case cudaFuncCachePreferL1 is selected
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index 910bebd..e2626d2 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -109,6 +109,7 @@
 #include <stdarg.h>
 #include <iostream>
 #include <string>
+#include <regex>
 #include <sstream>
 #include <fstream>
 #ifdef OPENGL_SUPPORT
@@ -1816,10 +1817,15 @@ void __cudaRegisterTexture(
 		int ext
 ) //passes in a newly created textureReference
 {
+	std::string devStr (deviceName);
+	#if (CUDART_VERSION > 4020)
+	if (devStr.size() > 2 && devStr.data()[0] == ':' && devStr.data()[1] == ':')
+		devStr = devStr.replace(0, 2, "");
+	#endif
 	CUctx_st *context = GPGPUSim_Context();
 	gpgpu_t *gpu = context->get_device()->get_gpgpu();
 	printf("GPGPU-Sim PTX: in __cudaRegisterTexture:\n");
-	gpu->gpgpu_ptx_sim_bindNameToTexture(deviceName, hostVar, dim, norm, ext);
+	gpu->gpgpu_ptx_sim_bindNameToTexture(devStr.data(), hostVar, dim, norm, ext);
 	printf("GPGPU-Sim PTX:   int dim = %d\n", dim);
 	printf("GPGPU-Sim PTX:   int norm = %d\n", norm);
 	printf("GPGPU-Sim PTX:   int ext = %d\n", ext);
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 715be98..15417d1 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -719,7 +719,7 @@ void ptx_instruction::set_opcode_and_latency()
 		   break;
 	   }
 	   break;
-   case MAD_OP: case MADP_OP:
+   case MAD_OP: case MADC_OP:
 	   //MAD latency
 	   switch(get_type()){
 	   case F32_TYPE:
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index cf7f04a..4dd5ed8 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -2421,7 +2421,7 @@ void mad_impl( const ptx_instruction *pI, ptx_thread_info *thread )
    mad_def(pI, thread, false);
 }
 
-void madp_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
+void madc_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
 {
    mad_def(pI, thread, true);
 }
diff --git a/src/cuda-sim/opcodes.def b/src/cuda-sim/opcodes.def
index 7aaa14f..874acc7 100644
--- a/src/cuda-sim/opcodes.def
+++ b/src/cuda-sim/opcodes.def
@@ -69,7 +69,7 @@ OP_DEF(LDU_OP,ldu_impl,"ldu",1,5)
 OP_DEF(LG2_OP,lg2_impl,"lg2",1,4)
 OP_DEF(MAD24_OP,mad24_impl,"mad24",1,2)
 OP_DEF(MAD_OP,mad_impl,"mad",1,2)
-OP_DEF(MADP_OP,madp_impl,"madp",1,2)
+OP_DEF(MADC_OP,madc_impl,"madc",1,2)
 OP_DEF(MAX_OP,max_impl,"max",1,1)
 OP_DEF(MEMBAR_OP,membar_impl,"membar",1,3)
 OP_DEF(MIN_OP,min_impl,"min",1,1)
diff --git a/src/cuda-sim/ptx.l b/src/cuda-sim/ptx.l
index dfed936..95ab74c 100644
--- a/src/cuda-sim/ptx.l
+++ b/src/cuda-sim/ptx.l
@@ -86,7 +86,7 @@ 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;
-madp    TC; ptx_lval.int_value = MADP_OP; return OPCODE;
+madc	TC; ptx_lval.int_value = MADC_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;
@@ -249,6 +249,7 @@ breakaddr  TC; ptx_lval.int_value = BREAKADDR_OP; return OPCODE;
 \.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;
diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y
index 79faddf..82abcbb 100644
--- a/src/cuda-sim/ptx.y
+++ b/src/cuda-sim/ptx.y
@@ -103,6 +103,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 %token  COMMA
 %token  PRED
 %token  HALF_OPTION
+%token  EXTP_OPTION
 %token  EQ_OPTION
 %token  NE_OPTION
 %token  LT_OPTION
@@ -438,6 +439,7 @@ option: type_spec
 	| 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); }
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 751b3f4..8f9c3d2 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -1167,6 +1167,8 @@ ptx_instruction::ptx_instruction( int opcode,
       case HALF_OPTION:
          m_inst_size = 4; // bytes
          break;
+	  case EXTP_OPTION:
+		 break;
       default:
          assert(0);
          break;
diff --git a/src/cuda-sim/ptxinfo.l b/src/cuda-sim/ptxinfo.l
index 99ee1fc..f9b6846 100644
--- a/src/cuda-sim/ptxinfo.l
+++ b/src/cuda-sim/ptxinfo.l
@@ -59,6 +59,7 @@ unsigned ptxinfo_col = 0;
 "gmem"		TC; return GMEM;
 "line"		TC; return LINE;
 "for"		TC; return FOR;
+"textures"	TC; return TEXTURES;
 
 [_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;
diff --git a/src/cuda-sim/ptxinfo.y b/src/cuda-sim/ptxinfo.y
index 294412d..faa33eb 100644
--- a/src/cuda-sim/ptxinfo.y
+++ b/src/cuda-sim/ptxinfo.y
@@ -54,6 +54,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 %token LINE
 %token <string_value> WARNING
 %token FOR
+%token TEXTURES
 
 %{
 	#include <stdlib.h>
@@ -104,6 +105,7 @@ info: 	  USED INT_OPERAND REGS { ptxinfo_regs($2); }
 	| INT_OPERAND BYTES SMEM { ptxinfo_smem($1,0); }
 	| INT_OPERAND BYTES CMEM { ptxinfo_cmem($1,0); }
 	| INT_OPERAND REGS { ptxinfo_regs($1); }
+	| INT_OPERAND TEXTURES {}
 	;
 
 tuple: INT_OPERAND PLUS INT_OPERAND BYTES { g_declared=$1; g_system=$3; }