Merge branch 'dev-purdue-integration' of https://github.rcac.purdue.edu/abdallm/gpgpu-sim_distribution into dev-purdue-integration-trace

60 files changed, 2464 insertions, 109 deletions

diff --git a/.travis.yml b/.travis.yml
index fabfc5b..6ef6730 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -19,5 +19,7 @@ matrix:
       env: CONFIG=configs.quadro5800.yml
     - services: docker
       env: CONFIG=configs.teslac2050.yml
+    - services: docker
+      env: CONFIG=configs.gtx1080ti.yml
 
 script: docker run -v `pwd`:/home/runner/gpgpu-sim_distribution:rw tgrogers/gpgpu-sim_regress:latest /bin/bash -c "./start_torque.sh; chown -R runner /home/runner/gpgpu-sim_distribution; su - runner -c 'source /home/runner/gpgpu-sim_distribution/setup_environment && make -j -C /home/runner/gpgpu-sim_distribution && cd /home/runner/gpgpu-sim_simulations/ && git pull && /home/runner/gpgpu-sim_simulations/util/job_launching/run_simulations.py -c /home/runner/gpgpu-sim_simulations/util/job_launching/regression_recipies/rodinia_2.0-ft/$CONFIG -N regress && /home/runner/gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -N regress'"
diff --git a/Jenkinsfile b/Jenkinsfile
index 85fc8f8..16e932e 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -3,6 +3,10 @@ pipeline {
         label "purdue-cluster"
     }
 
+    options {
+        disableConcurrentBuilds()
+    }
+
     stages {
         stage('simulator-build') {
             steps {
@@ -10,8 +14,8 @@ pipeline {
                     sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\
                     source `pwd`/setup_environment &&\
                     make -j'
-                }, "8.0" : {
-                    sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/8.0_env_setup.sh &&\
+                }, "9.1" : {
+                    sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\
                     source `pwd`/setup_environment &&\
                     make -j'
                 }
@@ -23,17 +27,20 @@ pipeline {
                 sh 'git clone [email protected]:TimRogersGroup/gpgpu-sim_simulations.git && \
                     cd gpgpu-sim_simulations && \
                     git checkout purdue-cluster && \
-                    git pull'
+                    git pull && \
+                    ln -s /home/tgrogers-raid/a/common/data_dirs benchmarks/'
                 sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\
                     source `pwd`/setup_environment &&\
                     cd gpgpu-sim_simulations && \
                     source ./benchmarks/src/setup_environment && \
-                    make -j -C ./benchmarks/src all'
-                sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/8.0_env_setup.sh &&\
+                    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 &&\
                     source `pwd`/setup_environment &&\
                     cd gpgpu-sim_simulations && \
                     source ./benchmarks/src/setup_environment && \
-                    make -j -C ./benchmarks/src/ all'
+                    make -j -C ./benchmarks/src/ rodinia_2.0-ft sdk-4.2 && \
+                    make -C ./benchmarks/src data'
             }
         }
         stage('regress'){
@@ -41,22 +48,36 @@ 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,PASCALTITANX,PASCALTITANX-PTXPLUS,TITANX-P102 -N regress-$$ && \
-                    ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -N regress-$$'
-                }, "8.0-rodinia": {
-                    sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/8.0_env_setup.sh &&\
-                    source `pwd`/setup_environment &&\
-                    ./gpgpu-sim_simulations/util/job_launching/run_simulations.py -B rodinia_2.0-ft -C GTX480,PASCALTITANX -N regress-$$ && \
-                    ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -N regress-$$'
-                }, "4.2-sdk-4.2": {
-                    sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\
+                    ./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-$$.csv && \
+                    ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c stats-$$.csv -p [email protected]:~/website/gpgpu-sim-plots/$PLOTDIR -w https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/$PLOTDIR -n $PLOTDIR'
+                }, "9.1-functest": {
+                    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 sdk-4.2 -C GTX480,PASCALTITANX -N regress-$$ && \
-                    ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v -N regress-$$'
+                    ./gpgpu-sim_simulations/util/job_launching/run_simulations.py -B rodinia_2.0-ft,sdk-4.2 -C TITANX,TITANX-L1ON,P100,TITANV -N regress-$$ && \
+                    PLOTDIR="jenkins/${JOB_NAME}/${BUILD_NUMBER}/9.1-rodinia" && ssh [email protected] mkdir -p /home/dynamo/a/tgrogers/website/gpgpu-sim-plots/$PLOTDIR && \
+                    ./gpgpu-sim_simulations/util/job_launching/monitor_func_test.py -v  -s stats-$$.csv -N regress-$$ && \
+                    ./gpgpu-sim_simulations/util/plotting/plot-get-stats.py -c stats-$$.csv -p [email protected]:~/website/gpgpu-sim-plots/$PLOTDIR -w https://engineering.purdue.edu/tgrogers/gpgpu-sim-plots/$PLOTDIR -n $PLOTDIR'
                 }
             }
         }
-
+        stage('4.2-correlate'){
+            steps {
+                sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/4.2_env_setup.sh &&\
+                    source `pwd`/setup_environment &&\
+                    PLOTDIR="jenkins/${JOB_NAME}" &&\
+                    ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh  GTX480,GTX480-PTXPLUS $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/plotting/correlate_and_publish.sh TITANX,TITANX-L1ON,P100,TITANV $PLOTDIR ${BUILD_NUMBER}'
+            }
+        }
     }
     post {
         success {
diff --git a/Makefile b/Makefile
index 1f043fa..6c2a6b3 100644
--- a/Makefile
+++ b/Makefile
@@ -144,7 +144,7 @@ no_opencl_support:
 	@echo "Warning: gpgpu-sim is building without opencl support. Make sure NVOPENCL_LIBDIR and NVOPENCL_INCDIR are set"
 
 $(SIM_LIB_DIR)/libcudart.so: makedirs $(LIBS) cudalib
-	g++ -shared -Wl \
+	g++ -shared -Wl,-soname,libcudart_$(GPGPUSIM_BUILD).so -Wl,--version-script=linux-so-version.txt\
 			$(SIM_OBJ_FILES_DIR)/libcuda/*.o \
 			$(SIM_OBJ_FILES_DIR)/cuda-sim/*.o \
 			$(SIM_OBJ_FILES_DIR)/cuda-sim/decuda_pred_table/*.o \
@@ -162,6 +162,8 @@ $(SIM_LIB_DIR)/libcudart.so: makedirs $(LIBS) cudalib
 	if [ ! -f $(SIM_LIB_DIR)/libcudart.so.6.5 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.6.5; fi
 	if [ ! -f $(SIM_LIB_DIR)/libcudart.so.7.5 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.7.5; fi
 	if [ ! -f $(SIM_LIB_DIR)/libcudart.so.8.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.8.0; fi
+	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
 
 $(SIM_LIB_DIR)/gpgpusim.out: makedirs $(LIBS) cudalib $(SIM_LIB_DIR)/libcudart.so
 	g++ -std=c++0x -L$(SIM_LIB_DIR) -lcudart -o $(SIM_LIB_DIR)/gpgpusim.out src/trace-driven/gpgpusim_trace_driven_main.cc 
diff --git a/aerialvision/guiclasses.py b/aerialvision/guiclasses.py
index 0a7013b..45fed26 100644
--- a/aerialvision/guiclasses.py
+++ b/aerialvision/guiclasses.py
@@ -69,7 +69,7 @@ import matplotlib
 matplotlib.use('TkAgg')
 from matplotlib.backends.backend_tkagg import FigureCanvasTkAgg, NavigationToolbar2TkAgg
 from matplotlib.figure import Figure
-from matplotlib import mpl
+import matplotlib as mpl
 from matplotlib.colors import colorConverter
 from matplotlib import pyplot
 import Pmw
diff --git a/configs/3.x-cfgs/SM6_P100/config_fermi_islip.icnt b/configs/3.x-cfgs/SM6_P100/config_fermi_islip.icnt
new file mode 100644
index 0000000..d26c8d9
--- /dev/null
+++ b/configs/3.x-cfgs/SM6_P100/config_fermi_islip.icnt
@@ -0,0 +1,70 @@
+//21*1 fly with 32 flits per packet under gpgpusim injection mode
+use_map = 0;
+flit_size = 32; 
+
+// currently we do not use this, see subnets below
+network_count = 2;
+
+// Topology
+topology = fly;
+k = 60;
+n = 1;
+
+// Routing
+
+routing_function = dest_tag;
+
+// Flow control
+
+num_vcs     = 1;
+vc_buf_size = 8;
+
+wait_for_tail_credit = 0;
+
+// Router architecture
+
+vc_allocator = islip; //separable_input_first;
+sw_allocator = islip; //separable_input_first;
+alloc_iters  = 1;
+
+credit_delay   = 0;
+routing_delay  = 0;
+vc_alloc_delay = 1;
+sw_alloc_delay = 1;
+
+input_speedup     = 2;
+output_speedup    = 1;
+internal_speedup  = 1.0;
+
+// Traffic, GPGPU-Sim does not use this
+
+traffic                = uniform;
+packet_size ={{1,2,3,4},{10,20}};
+packet_size_rate={{1,1,1,1},{2,1}};
+
+// Simulation - Don't change
+
+sim_type       = gpgpusim;
+//sim_type = latency;
+injection_rate = 0.1;
+
+subnets = 2;
+
+// Always use read and write no matter following line
+//use_read_write = 1;
+
+
+read_request_subnet = 0;
+read_reply_subnet = 1;
+write_request_subnet = 0;
+write_reply_subnet = 1;
+
+read_request_begin_vc = 0;
+read_request_end_vc = 0;
+write_request_begin_vc = 0;
+write_request_end_vc = 0;
+read_reply_begin_vc = 0;
+read_reply_end_vc = 0;
+write_reply_begin_vc = 0;
+write_reply_end_vc = 0;
+
diff --git a/configs/3.x-cfgs/SM6_P100/gpgpusim.config b/configs/3.x-cfgs/SM6_P100/gpgpusim.config
new file mode 100644
index 0000000..a5e6736
--- /dev/null
+++ b/configs/3.x-cfgs/SM6_P100/gpgpusim.config
@@ -0,0 +1,156 @@
+# This config models the Pascal GP100
+# For more info about this card, see Nvidia White paper
+# https://images.nvidia.com/content/pdf/tesla/whitepaper/pascal-architecture-whitepaper.pdf
+
+# functional simulator specification
+-gpgpu_ptx_instruction_classification 0
+-gpgpu_ptx_sim_mode 0
+-gpgpu_ptx_force_max_capability 60
+
+# SASS execution (only supported with CUDA >= 4.0)
+-gpgpu_ptx_convert_to_ptxplus 0
+-gpgpu_ptx_save_converted_ptxplus 0
+
+# high level architecture configuration
+-gpgpu_n_clusters 28
+-gpgpu_n_cores_per_cluster 2
+-gpgpu_n_mem 32
+-gpgpu_n_sub_partition_per_mchannel 1
+
+# Pscal clock domains
+#-gpgpu_clock_domains <Core Clock>:<Interconnect Clock>:<L2 Clock>:<DRAM Clock>
+# Pascal NVIDIA GP100  clock domains are adopted from 
+# https://en.wikipedia.org/wiki/Nvidia_Tesla
+-gpgpu_clock_domains 1480.0:1480.0:1480.0:715.0
+
+# shader core pipeline config
+-gpgpu_shader_registers 65536
+
+# This implies a maximum of 64 warps/SM
+-gpgpu_shader_core_pipeline 2048:32 
+-gpgpu_shader_cta 32
+-gpgpu_simd_model 1 
+
+# Pipeline widths and number of FUs
+# ID_OC_SP,ID_OC_DP,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_DP,OC_EX_SFU,OC_EX_MEM,EX_WB
+## Pascal GP102 has 4 SP SIMD units and 1 SFU unit
+## we need to scale the number of pipeline registers to be equal to the number of SP units
+-gpgpu_pipeline_widths 2,2,1,2,2,1,4
+-gpgpu_num_sp_units 2
+-gpgpu_num_sfu_units 1
+
+# Instruction latencies and initiation intervals
+# "ADD,MAX,MUL,MAD,DIV"
+# SFU is 32-width in pascal, then dp units initiation is 1 cycle
+-ptx_opcode_latency_int 4,13,4,5,145
+-ptx_opcode_initiation_int 1,1,1,1,4
+-ptx_opcode_latency_fp 4,13,4,5,39
+-ptx_opcode_initiation_fp 1,2,1,1,4
+-ptx_opcode_latency_dp 8,19,8,8,330
+-ptx_opcode_initiation_dp 2,2,2,2,130
+
+# <nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>:<set_index_fn>,<mshr>:<N>:<merge>,<mq>:**<fifo_entry>
+# ** Optional parameter - Required when mshr_type==Texture Fifo
+# Note: Hashing set index function (H) only applies to a set size of 32 or 64. 
+# Pascal GP100 has 64KB Shared memory
+# Pascal GP100 has 48KB L1 cache
+-gpgpu_cache:dl1  64:128:6,L:L:m:N:H,A:128:8,16
+-gpgpu_cache:dl1PrefL1  64:128:6,L:L:m:N:H,A:128:8,16
+-gpgpu_cache:dl1PrefShared  64:128:6,L:L:m:N:H,A:128:8,16
+-gpgpu_shmem_size 65536
+-gpgpu_shmem_size_PrefL1 65536
+-gpgpu_shmem_size_PrefShared 65536
+-gmem_skip_L1D 0
+
+# 64 sets, each 128 bytes 16-way for each memory sub partition (128 KB per memory sub partition). This gives 3MB L2 cache
+-gpgpu_cache:dl2 64:128:16,L:B:m:W:L,A:128:8,4:0,32
+-gpgpu_cache:dl2_texture_only 0 
+-gpgpu_dram_partition_queues 32:32:32:32
+
+# 4 KB Inst.
+-gpgpu_cache:il1 8:128:4,L:R:f:N:L,A:2:48,4
+# 48 KB Tex 
+-gpgpu_tex_cache:l1 16:128:24,L:R:m:N:L,F:128:4,128:2
+# 12 KB Const
+-gpgpu_const_cache:l1 128:64:2,L:R:f:N:L,A:2:64,4
+
+# enable operand collector
+-gpgpu_operand_collector_num_units_sp 14
+-gpgpu_operand_collector_num_units_sfu 8
+-gpgpu_operand_collector_num_units_mem 10
+-gpgpu_operand_collector_num_in_ports_sp 4
+-gpgpu_operand_collector_num_out_ports_sp 4
+-gpgpu_operand_collector_num_in_ports_sfu 1
+-gpgpu_operand_collector_num_out_ports_sfu 1
+-gpgpu_operand_collector_num_in_ports_mem 1
+-gpgpu_operand_collector_num_out_ports_mem 1
+-gpgpu_num_reg_banks 32
+
+# shared memory bankconflict detection 
+-gpgpu_shmem_num_banks 32
+-gpgpu_shmem_limited_broadcast 0
+-gpgpu_shmem_warp_parts 1
+
+## In Pascal, a warp scheduler can issue 2 insts per cycle
+-gpgpu_max_insn_issue_per_warp 2
+
+# interconnection
+-network_mode 1 
+-inter_config_file config_fermi_islip.icnt
+
+# memory partition latency config 
+-rop_latency 120
+-dram_latency 100
+
+# dram model config
+-gpgpu_dram_scheduler 1
+# The DRAM return queue and the scheduler queue together should provide buffer
+# to sustain the memory level parallelism to tolerate DRAM latency 
+# To allow 100% DRAM utility, there should at least be enough buffer to sustain
+# the minimum DRAM latency (100 core cycles).  I.e. 
+#   Total buffer space required = 100 x 924MHz / 700MHz = 132
+-gpgpu_frfcfs_dram_sched_queue_size 64
+-gpgpu_dram_return_queue_size 116
+
+# for HBM, 32 channles, each (128 bits) 16 bytes width
+-gpgpu_n_mem_per_ctrlr 1
+-gpgpu_dram_buswidth 16
+-gpgpu_dram_burst_length 2
+-dram_data_command_freq_ratio 2  # HBM is DDR
+-gpgpu_mem_address_mask 1
+-gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBBCCC.CCCSSSSS
+
+# HBM timing are adopted from hynix JESD235 standered and nVidia HPCA 2017 paper (http://www.cs.utah.edu/~nil/pubs/hpca17.pdf)
+# Timing for 1 GHZ
+# tRRDl and tWTR are missing, need to be added
+#-gpgpu_dram_timing_opt "nbk=16:CCD=1:RRD=4:RCD=14:RAS=33:RP=14:RC=47:
+#                        CL=14:WL=2:CDLR=3:WR=12:nbkgrp=4:CCDL=2:RTPL=4"
+
+# Timing for 715 MHZ, Tesla Pascal P100 HBM runs at 715 MHZ
+-gpgpu_dram_timing_opt "nbk=16:CCD=1:RRD=3:RCD=10:RAS=24:RP=10:RC=34:
+                        CL=10:WL=2:CDLR=3:WR=9:nbkgrp=4:CCDL=2:RTPL=3"
+
+# Fermi has two schedulers per core
+-gpgpu_num_sched_per_core 2
+# Two Level Scheduler with active and pending pools
+#-gpgpu_scheduler two_level_active:6:0:1
+# Loose round robbin scheduler
+#-gpgpu_scheduler lrr
+# Greedy then oldest scheduler
+-gpgpu_scheduler gto
+
+# stat collection
+-gpgpu_memlatency_stat 14 
+-gpgpu_runtime_stat 500
+-enable_ptx_file_line_stats 1
+-visualizer_enabled 0
+
+# power model configs
+-power_simulation_enabled 0
+-gpuwattch_xml_file gpuwattch_gtx480.xml
+
+# tracing functionality
+#-trace_enabled 1
+#-trace_components WARP_SCHEDULER,SCOREBOARD
+#-trace_sampling_core 0
+
diff --git a/configs/GTX480/gpuwattch_gtx480.xml b/configs/3.x-cfgs/SM6_P100/gpuwattch_gtx480.xml
index 304e0fd..304e0fd 100755
--- a/configs/GTX480/gpuwattch_gtx480.xml
+++ b/configs/3.x-cfgs/SM6_P100/gpuwattch_gtx480.xml
diff --git a/configs/PascalTitanX/gpgpu-sim-3.x/config_fermi_islip.icnt b/configs/3.x-cfgs/SM6_TITANX/config_fermi_islip.icnt
index 602daee..602daee 100644
--- a/configs/PascalTitanX/gpgpu-sim-3.x/config_fermi_islip.icnt
+++ b/configs/3.x-cfgs/SM6_TITANX/config_fermi_islip.icnt
diff --git a/configs/PascalTitanX/gpgpu-sim-3.x/gpgpusim.config b/configs/3.x-cfgs/SM6_TITANX/gpgpusim.config
index 4407870..f78bd02 100644
--- a/configs/PascalTitanX/gpgpu-sim-3.x/gpgpusim.config
+++ b/configs/3.x-cfgs/SM6_TITANX/gpgpusim.config
@@ -33,25 +33,21 @@
 
 # Pipeline widths and number of FUs
 # ID_OC_SP,ID_OC_DP,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_DP,OC_EX_SFU,OC_EX_MEM,EX_WB
-## Pascal GP102 has 4 SP SIMD units and 4 SFU units
+## Pascal GP102 has 4 SP SIMD units and 1 SFU unit
 ## we need to scale the number of pipeline registers to be equal to the number of SP units
--gpgpu_pipeline_widths 4,1,4,1,4,1,4,1,9
+-gpgpu_pipeline_widths 4,1,1,4,1,1,6
 -gpgpu_num_sp_units 4
--gpgpu_num_sfu_units 4
--gpgpu_num_dp_units 1
+-gpgpu_num_sfu_units 1
 
 # Instruction latencies and initiation intervals
 # "ADD,MAX,MUL,MAD,DIV"
-# All Div operations are executed on SFU unit
-# Throughput (initiation latency) are adopted from CUDA SDK document V8, section 5.4.1, Table 2
+# SFU is 32-width in pascal, then dp units initiation is 1 cycle
 -ptx_opcode_latency_int 4,13,4,5,145
 -ptx_opcode_initiation_int 1,1,1,1,4
 -ptx_opcode_latency_fp 4,13,4,5,39
 -ptx_opcode_initiation_fp 1,2,1,1,4
 -ptx_opcode_latency_dp 8,19,8,8,330
--ptx_opcode_initiation_dp 8,8,8,8,130
--ptx_opcode_initiation_sfu 4
--ptx_opcode_latency_sfu 8
+-ptx_opcode_initiation_dp 4,8,4,4,130
 
 # <nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>:<set_index_fn>,<mshr>:<N>:<merge>,<mq>:**<fifo_entry>
 # ** Optional parameter - Required when mshr_type==Texture Fifo
@@ -59,36 +55,37 @@
 # Pascal GP102 has 96KB Shared memory
 # Pascal GP102 has 64KB L1 cache
 # The defulat is to disable the L1 cache, unless cache modifieres is used
--gpgpu_cache:dl1  N:64:128:6,L:L:m:N:H,A:128:8,16
+-gpgpu_cache:dl1  64:128:6,L:L:m:N:H,A:128:8,16
+-gpgpu_cache:dl1PrefL1  64:128:6,L:L:m:N:H,A:128:8,16
+-gpgpu_cache:dl1PrefShared  64:128:6,L:L:m:N:H,A:128:8,16
 -gpgpu_shmem_size 98304
+-gpgpu_shmem_size_PrefL1 98304
+-gpgpu_shmem_size_PrefShared 98304
 -gmem_skip_L1D 1
 
 # 64 sets, each 128 bytes 16-way for each memory sub partition (128 KB per memory sub partition). This gives 3MB L2 cache
--gpgpu_cache:dl2 N:64:128:16,L:B:m:W:L,A:128:8,4:0,32
+-gpgpu_cache:dl2 64:128:16,L:B:m:W:L,A:128:8,4:0,32
 -gpgpu_cache:dl2_texture_only 0 
 -gpgpu_dram_partition_queues 32:32:32:32
 
 # 4 KB Inst.
--gpgpu_cache:il1 N:8:128:4,L:R:f:N:L,A:2:48,4
+-gpgpu_cache:il1 8:128:4,L:R:f:N:L,A:2:48,4
 # 48 KB Tex 
--gpgpu_tex_cache:l1 N:16:128:24,L:R:m:N:L,F:128:4,128:2
+-gpgpu_tex_cache:l1 16:128:24,L:R:m:N:L,F:128:4,128:2
 # 12 KB Const
--gpgpu_const_cache:l1 N:128:64:2,L:R:f:N:L,A:2:64,4
+-gpgpu_const_cache:l1 128:64:2,L:R:f:N:L,A:2:64,4
 
 # enable operand collector 
 ## larger operand collectors and reg_banks are needed for the 4 warp schedulers and 4 SIMD units
--gpgpu_operand_collector_num_units_sp 12
--gpgpu_operand_collector_num_units_sfu 6
+-gpgpu_operand_collector_num_units_sp 20
+-gpgpu_operand_collector_num_units_sfu 4
 -gpgpu_operand_collector_num_units_mem 8
--gpgpu_operand_collector_num_units_dp 6
 -gpgpu_operand_collector_num_in_ports_sp 4
 -gpgpu_operand_collector_num_out_ports_sp 4
--gpgpu_operand_collector_num_in_ports_sfu 4
--gpgpu_operand_collector_num_out_ports_sfu 4
+-gpgpu_operand_collector_num_in_ports_sfu 1
+-gpgpu_operand_collector_num_out_ports_sfu 1
 -gpgpu_operand_collector_num_in_ports_mem 1
 -gpgpu_operand_collector_num_out_ports_mem 1
--gpgpu_operand_collector_num_in_ports_dp 1
--gpgpu_operand_collector_num_out_ports_dp 1
 # gpgpu_num_reg_banks should be increased to 32
 -gpgpu_num_reg_banks 32
 
@@ -105,7 +102,7 @@
 -inter_config_file config_fermi_islip.icnt
 
 # memory partition latency config 
--rop_latency 100
+-rop_latency 120
 -dram_latency 100
 
 # dram model config
@@ -130,10 +127,10 @@
 
 # Use the same GDDR5 timing from hynix H5GQ1H24AFR
 # disable bank groups for now, set nbkgrp to 1 and tCCDL and tRTPL to 0
--gpgpu_dram_timing_opt "nbk=16:CCD=2:RRD=8:RCD=16:RAS=37:RP=16:RC=52:
-                        CL=16:WL=6:CDLR=7:WR=16:nbkgrp=4:CCDL=4:RTPL=3"
+-gpgpu_dram_timing_opt "nbk=16:CCD=2:RRD=6:RCD=12:RAS=28:RP=12:RC=40:
+                        CL=12:WL=4:CDLR=5:WR=12:nbkgrp=1:CCDL=0:RTPL=0"
 
-# Pascal GP102 has four schedulers per core
+# Fermi has two schedulers per core
 -gpgpu_num_sched_per_core 4
 # Two Level Scheduler with active and pending pools
 #-gpgpu_scheduler two_level_active:6:0:1
diff --git a/configs/PascalTitanX/gpgpu-sim-3.x/gpuwattch_gtx480.xml b/configs/3.x-cfgs/SM6_TITANX/gpuwattch_gtx480.xml
index 304e0fd..304e0fd 100755
--- a/configs/PascalTitanX/gpgpu-sim-3.x/gpuwattch_gtx480.xml
+++ b/configs/3.x-cfgs/SM6_TITANX/gpuwattch_gtx480.xml
diff --git a/configs/3.x-cfgs/SM7_TITANV/config_fermi_islip.icnt b/configs/3.x-cfgs/SM7_TITANV/config_fermi_islip.icnt
new file mode 100644
index 0000000..fac792a
--- /dev/null
+++ b/configs/3.x-cfgs/SM7_TITANV/config_fermi_islip.icnt
@@ -0,0 +1,70 @@
+//21*1 fly with 32 flits per packet under gpgpusim injection mode
+use_map = 0;
+flit_size = 32; 
+
+// currently we do not use this, see subnets below
+network_count = 2;
+
+// Topology
+topology = fly;
+k = 64;
+n = 1;
+
+// Routing
+
+routing_function = dest_tag;
+
+// Flow control
+
+num_vcs     = 1;
+vc_buf_size = 8;
+
+wait_for_tail_credit = 0;
+
+// Router architecture
+
+vc_allocator = islip; //separable_input_first;
+sw_allocator = islip; //separable_input_first;
+alloc_iters  = 1;
+
+credit_delay   = 0;
+routing_delay  = 0;
+vc_alloc_delay = 1;
+sw_alloc_delay = 1;
+
+input_speedup     = 2;
+output_speedup    = 1;
+internal_speedup  = 1.0;
+
+// Traffic, GPGPU-Sim does not use this
+
+traffic                = uniform;
+packet_size ={{1,2,3,4},{10,20}};
+packet_size_rate={{1,1,1,1},{2,1}};
+
+// Simulation - Don't change
+
+sim_type       = gpgpusim;
+//sim_type = latency;
+injection_rate = 0.1;
+
+subnets = 2;
+
+// Always use read and write no matter following line
+//use_read_write = 1;
+
+
+read_request_subnet = 0;
+read_reply_subnet = 1;
+write_request_subnet = 0;
+write_reply_subnet = 1;
+
+read_request_begin_vc = 0;
+read_request_end_vc = 0;
+write_request_begin_vc = 0;
+write_request_end_vc = 0;
+read_reply_begin_vc = 0;
+read_reply_end_vc = 0;
+write_reply_begin_vc = 0;
+write_reply_end_vc = 0;
+
diff --git a/configs/3.x-cfgs/SM7_TITANV/gpgpusim.config b/configs/3.x-cfgs/SM7_TITANV/gpgpusim.config
new file mode 100644
index 0000000..f1ae2dc
--- /dev/null
+++ b/configs/3.x-cfgs/SM7_TITANV/gpgpusim.config
@@ -0,0 +1,165 @@
+# This config models the Volta Titan X
+# For more info about this card:
+# http://images.nvidia.com/content/volta-architecture/pdf/volta-architecture-whitepaper.pdf
+# https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8344474&tag=1# 
+# http://on-demand.gputechconf.com/gtc/2018/presentation/s8122-dissecting-the-volta-gpu-architecture-through-microbenchmarking.pdf
+# https://en.wikipedia.org/wiki/Volta_(microarchitecture)
+# https://www.hotchips.org/wp-content/uploads/hc_archives/hc29/HC29.21-Monday-Pub/HC29.21.10-GPU-Gaming-Pub/HC29.21.132-Volta-Choquette-NVIDIA-Final3.pdf
+# https://devblogs.nvidia.com/inside-volta/
+# http://on-demand.gputechconf.com/gtc/2017/presentation/s7798-luke-durant-inside-volta.pdf
+
+# functional simulator specification
+-gpgpu_ptx_instruction_classification 0
+-gpgpu_ptx_sim_mode 0
+-gpgpu_ptx_force_max_capability 70
+
+# SASS execution (only supported with CUDA >= 4.0)
+-gpgpu_ptx_convert_to_ptxplus 0
+-gpgpu_ptx_save_converted_ptxplus 0
+
+# high level architecture configuration
+-gpgpu_n_clusters 40
+-gpgpu_n_cores_per_cluster 2
+-gpgpu_n_mem 24
+-gpgpu_n_sub_partition_per_mchannel 1
+
+# volta clock domains
+#-gpgpu_clock_domains <Core Clock>:<Interconnect Clock>:<L2 Clock>:<DRAM Clock>
+# Pascal NVIDIA GP100  clock domains are adopted from 
+# https://en.wikipedia.org/wiki/Volta_(microarchitecture)
+-gpgpu_clock_domains 1200.0:1200.0:2000.0:850.0
+# boost mode
+# -gpgpu_clock_domains 1455.0:1455.0:2000.0:850.0
+
+# shader core pipeline config
+-gpgpu_shader_registers 65536
+
+# This implies a maximum of 64 warps/SM
+-gpgpu_shader_core_pipeline 2048:32 
+-gpgpu_shader_cta 32
+-gpgpu_simd_model 1 
+
+# Pipeline widths and number of FUs
+# ID_OC_SP,ID_OC_DP,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_DP,OC_EX_SFU,OC_EX_MEM,EX_WB
+## Pascal GP100 has 2 SP SIMD units, 2 SFU units, 2 DP units per core
+## we need to scale the number of pipeline registers to be equal to the number of SP units
+-gpgpu_pipeline_widths 4,4,1,4,4,1,9
+-gpgpu_num_sp_units 4
+-gpgpu_num_sfu_units 4
+
+# Instruction latencies and initiation intervals
+# "ADD,MAX,MUL,MAD,DIV"
+# All Div operations are executed on SFU unit
+# Throughput (initiation latency) are adopted from 
+# http://on-demand.gputechconf.com/gtc/2018/presentation/s8122-dissecting-the-volta-gpu-architecture-through-microbenchmarking.pdf
+-ptx_opcode_latency_int 4,13,4,5,145
+-ptx_opcode_initiation_int 2,2,2,2,8
+-ptx_opcode_latency_fp 4,13,4,5,39
+-ptx_opcode_initiation_fp 2,2,2,2,4
+-ptx_opcode_latency_dp 8,19,8,8,330
+-ptx_opcode_initiation_dp 4,4,4,4,130
+
+# <nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>:<set_index_fn>,<mshr>:<N>:<merge>,<mq>:**<fifo_entry>
+# ** Optional parameter - Required when mshr_type==Texture Fifo
+# Note: Hashing set index function (H) only applies to a set size of 32 or 64. 
+# Pascal GP100 has 64KB Shared memory
+-gpgpu_cache:dl1  64:128:8,L:L:m:N:H,A:256:8,16:0
+-gpgpu_cache:dl1PrefL1  64:128:16,L:L:m:N:H,A:256:8,16:0
+-gpgpu_cache:dl1PrefShared  32:128:6,L:L:m:N:H,A:256:8,16:0
+-gpgpu_shmem_size 65536
+-gpgpu_shmem_size_PrefL1 1
+-gpgpu_shmem_size_PrefShared 98304
+-gmem_skip_L1D 0
+
+# 64 sets, each 128 bytes 24-way for each memory sub partition (192 KB per memory sub partition). This gives  4.5MB L2 cache
+-gpgpu_cache:dl2 64:128:24,L:B:m:W:L,A:256:4,32:0,32
+-gpgpu_cache:dl2_texture_only 0 
+-gpgpu_dram_partition_queues 64:64:64:64
+
+# 128 KB Inst.
+-gpgpu_cache:il1 64:128:16,L:R:f:N:L,A:2:48,4
+# 48 KB Tex 
+-gpgpu_tex_cache:l1 16:128:24,L:R:m:N:L,F:128:4,128:2
+# 64 KB Const
+-gpgpu_const_cache:l1 128:64:8,L:R:f:N:L,A:2:64,4
+
+# enable operand collector
+-gpgpu_operand_collector_num_units_sp 14
+-gpgpu_operand_collector_num_units_sfu 8
+-gpgpu_operand_collector_num_units_mem 10
+-gpgpu_operand_collector_num_in_ports_sp 4
+-gpgpu_operand_collector_num_out_ports_sp 4
+-gpgpu_operand_collector_num_in_ports_sfu 1
+-gpgpu_operand_collector_num_out_ports_sfu 1
+-gpgpu_operand_collector_num_in_ports_mem 1
+-gpgpu_operand_collector_num_out_ports_mem 1
+-gpgpu_num_reg_banks 32
+
+# shared memory bankconflict detection 
+-gpgpu_shmem_num_banks 32
+-gpgpu_shmem_limited_broadcast 0
+-gpgpu_shmem_warp_parts 1
+
+## In Pascal, a warp scheduler can issue 2 insts per cycle
+-gpgpu_max_insn_issue_per_warp 1
+
+# interconnection
+-network_mode 1 
+-inter_config_file config_fermi_islip.icnt
+
+# memory partition latency config 
+-rop_latency 120
+-dram_latency 100
+
+# dram model config
+-gpgpu_dram_scheduler 1
+# The DRAM return queue and the scheduler queue together should provide buffer
+# to sustain the memory level parallelism to tolerate DRAM latency 
+# To allow 100% DRAM utility, there should at least be enough buffer to sustain
+# the minimum DRAM latency (100 core cycles).  I.e. 
+#   Total buffer space required = 100 x 924MHz / 700MHz = 132
+-gpgpu_frfcfs_dram_sched_queue_size 64
+-gpgpu_dram_return_queue_size 116
+
+# for HBM, 32 channles, each (128 bits) 16 bytes width
+-gpgpu_n_mem_per_ctrlr 1
+-gpgpu_dram_buswidth 16
+-gpgpu_dram_burst_length 2
+-dram_data_command_freq_ratio 2  # HBM is DDR
+-gpgpu_mem_address_mask 1
+-gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBBCCC.CCCSSSSS
+
+# HBM timing are adopted from hynix JESD235 standered and nVidia HPCA 2017 paper (http://www.cs.utah.edu/~nil/pubs/hpca17.pdf)
+# Timing for 1 GHZ
+# tRRDl and tWTR are missing, need to be added
+#-gpgpu_dram_timing_opt "nbk=16:CCD=1:RRD=4:RCD=14:RAS=33:RP=14:RC=47:
+#                        CL=14:WL=2:CDLR=3:WR=12:nbkgrp=4:CCDL=2:RTPL=4"
+
+# Timing for 715 MHZ, Tesla Pascal P100 HBM runs at 715 MHZ
+-gpgpu_dram_timing_opt "nbk=16:CCD=1:RRD=3:RCD=10:RAS=24:RP=10:RC=34:
+                        CL=10:WL=2:CDLR=3:WR=9:nbkgrp=4:CCDL=2:RTPL=3"
+
+# Fermi has two schedulers per core
+-gpgpu_num_sched_per_core 2
+# Two Level Scheduler with active and pending pools
+#-gpgpu_scheduler two_level_active:6:0:1
+# Loose round robbin scheduler
+#-gpgpu_scheduler lrr
+# Greedy then oldest scheduler
+-gpgpu_scheduler gto
+
+# stat collection
+-gpgpu_memlatency_stat 14 
+-gpgpu_runtime_stat 500
+-enable_ptx_file_line_stats 1
+-visualizer_enabled 0
+
+# power model configs
+-power_simulation_enabled 0
+-gpuwattch_xml_file gpuwattch_gtx480.xml
+
+# tracing functionality
+#-trace_enabled 1
+#-trace_components WARP_SCHEDULER,SCOREBOARD
+#-trace_sampling_core 0
+
diff --git a/configs/QuadroFX5600/gpgpusim.config b/configs/4.x-cfgs/SM1_QFX5600/gpgpusim.config
index e3cab18..e3cab18 100644
--- a/configs/QuadroFX5600/gpgpusim.config
+++ b/configs/4.x-cfgs/SM1_QFX5600/gpgpusim.config
diff --git a/configs/QuadroFX5600/gpuwattch_quadrofx5600.xml b/configs/4.x-cfgs/SM1_QFX5600/gpuwattch_quadrofx5600.xml
index 2c5a6fc..2c5a6fc 100644
--- a/configs/QuadroFX5600/gpuwattch_quadrofx5600.xml
+++ b/configs/4.x-cfgs/SM1_QFX5600/gpuwattch_quadrofx5600.xml
diff --git a/configs/QuadroFX5600/icnt_config_islip.icnt b/configs/4.x-cfgs/SM1_QFX5600/icnt_config_islip.icnt
index de3bcc8..de3bcc8 100644
--- a/configs/QuadroFX5600/icnt_config_islip.icnt
+++ b/configs/4.x-cfgs/SM1_QFX5600/icnt_config_islip.icnt
diff --git a/configs/QuadroFX5800/config_quadro_islip.icnt b/configs/4.x-cfgs/SM1_QFX5800/config_quadro_islip.icnt
index cfe9cac..cfe9cac 100644
--- a/configs/QuadroFX5800/config_quadro_islip.icnt
+++ b/configs/4.x-cfgs/SM1_QFX5800/config_quadro_islip.icnt
diff --git a/configs/QuadroFX5800/gpgpusim.config b/configs/4.x-cfgs/SM1_QFX5800/gpgpusim.config
index 56dbb17..56dbb17 100644
--- a/configs/QuadroFX5800/gpgpusim.config
+++ b/configs/4.x-cfgs/SM1_QFX5800/gpgpusim.config
diff --git a/configs/TeslaC2050/config_fermi_islip.icnt b/configs/4.x-cfgs/SM2_C2050/config_fermi_islip.icnt
index a11bd8e..a11bd8e 100644
--- a/configs/TeslaC2050/config_fermi_islip.icnt
+++ b/configs/4.x-cfgs/SM2_C2050/config_fermi_islip.icnt
diff --git a/configs/TeslaC2050/gpgpusim.config b/configs/4.x-cfgs/SM2_C2050/gpgpusim.config
index aa5f5f3..aa5f5f3 100644
--- a/configs/TeslaC2050/gpgpusim.config
+++ b/configs/4.x-cfgs/SM2_C2050/gpgpusim.config
diff --git a/configs/GTX480/config_fermi_islip.icnt b/configs/4.x-cfgs/SM2_GTX480/config_fermi_islip.icnt
index 7820e4e..7820e4e 100644
--- a/configs/GTX480/config_fermi_islip.icnt
+++ b/configs/4.x-cfgs/SM2_GTX480/config_fermi_islip.icnt
diff --git a/configs/GTX480/gpgpusim.config b/configs/4.x-cfgs/SM2_GTX480/gpgpusim.config
index 03fcda1..03fcda1 100644
--- a/configs/GTX480/gpgpusim.config
+++ b/configs/4.x-cfgs/SM2_GTX480/gpgpusim.config
diff --git a/configs/4.x-cfgs/SM2_GTX480/gpuwattch_gtx480.xml b/configs/4.x-cfgs/SM2_GTX480/gpuwattch_gtx480.xml
new file mode 100755
index 0000000..304e0fd
--- /dev/null
+++ b/configs/4.x-cfgs/SM2_GTX480/gpuwattch_gtx480.xml
@@ -0,0 +1,538 @@
+<?xml version="1.0" ?>
+<component id="root" name="root">
+	<component id="system" name="system">
+		<!--McPAT will skip the components if number is set to 0 -->
+		<param name="GPU_Architecture" value="1"/><!-- 0-G80; 1-Fermi; others not supported -->
+		<param name="number_of_cores" value="16"/>
+		<param name="architecture" value="1"/> <!-- fermi:1 quadro:2 other: undefined-->
+		<param name="number_of_L1Directories" value="0"/>
+		<param name="number_of_L2Directories" value="0"/>
+		<param name="number_of_L2s" value="1"/> <!-- This number means how many L2 clusters in each cluster there can be multiple banks/ports -->
+		<param name="number_of_L3s" value="0"/> <!-- This number means how many L3 clusters -->
+		<param name="number_of_NoCs" value="1"/> 
+		<param name="homogeneous_cores" value="1"/><!--1 means homo -->
+		<param name="homogeneous_L2s" value="1"/>
+		<param name="homogeneous_L1Directorys" value="1"/>
+		<param name="homogeneous_L2Directorys" value="1"/>
+		<param name="homogeneous_L3s" value="1"/>
+		<param name="homogeneous_ccs" value="1"/><!--cache coherece hardware -->
+		<param name="homogeneous_NoCs" value="1"/>
+		<param name="core_tech_node" value="40"/><!-- nm -->
+		<param name="target_core_clockrate" value="700"/><!--MHz -->
+		<param name="temperature" value="380"/> <!-- Kelvin -->
+		<param name="number_cache_levels" value="2"/>
+		<param name="interconnect_projection_type" value="0"/><!--0: agressive wire technology; 1: conservative wire technology -->
+		<param name="device_type" value="0"/><!--0: HP(High Performance Type); 1: LSTP(Low standby power) 2: LOP (Low Operating Power)  -->
+		<param name="longer_channel_device" value="1"/><!-- 0 no use; 1 use when possible -->
+		<param name="machine_bits" value="32"/>
+		<param name="virtual_address_width" value="32"/>
+		<param name="physical_address_width" value="32"/>
+		<param name="virtual_memory_page_size" value="4096"/>
+		<param name="idle_core_power" value="1.59"/><!-- idle core power for GTX479 -->
+		<!--param name="scaling_coefficients" value="10,0.0884816,10,10,8,10,4.12782,10,2.48832,10,10,10,4.29982,0.387764,0.0714269,0.14302,0.01,0.546811,0.485351,0.806633,0.818073,1.9207,100,100,100,87.9303,100,10,4.3548,10"/-->
+		<param name="TOT_INST" value="2.00" />
+		<param name="FP_INT" value="4.57" />
+		<param name="IC_H" value="2.14" />
+		<param name="IC_M" value="22.47" />
+		<param name="DC_RH" value="22.14" />
+		<param name="DC_RM" value="24.66" />
+		<param name="DC_WH" value="1.53" />
+		<param name="DC_WM" value="39.79" />
+		<param name="TC_H" value="10.21" />
+		<param name="TC_M" value="24.66" />
+		<param name="CC_H" value="11.07" />
+		<param name="CC_M" value="12.33" />
+		<param name="SHRD_ACC" value="7.04" />
+		<param name="REG_RD" value="0.14" />
+		<param name="REG_WR" value="0.21" />
+		<param name="NON_REG_OPs" value="2.11" />
+		<param name="SP_ACC" value="2.38" />
+		<param name="SFU_ACC" value="0.51" />
+		<param name="FPU_ACC" value="0.64" />
+		<param name="MEM_RD" value="0.33" />
+		<param name="MEM_WR" value="0.40" />
+		<param name="MEM_PRE" value="0.11" />
+		<param name="L2_RH" value="13.79" />
+		<param name="L2_RM" value="35.18" />
+		<param name="L2_WH" value="43.07" />
+		<param name="L2_WM" value="28.72" />
+		<param name="NOC_A" value="305.48" />
+		<param name="PIPE_A" value="2.57" />
+		<param name="IDLE_CORE_N" value="1"/>
+		<param name="CONST_DYNAMICN" value="11" />
+		<stat name="num_idle_cores" value="0"/><!-- Average Number of idle cores during this period -->
+		<stat name="total_cycles" value="total_cycles_match_mcpat"/>
+		<stat name="idle_cycles" value="idle_cycles_match_mcpat"/>
+		<stat name="busy_cycles"  value="busy_cycles_match_mcpat"/>
+			<!--This page size(B) is complete different from the page size in Main memo secction. this page size is the size of 
+			virtual memory from OS/Archi perspective; the page size in Main memo secction is the actuall physical line in a DRAM bank  -->
+		<!-- *********************** cores ******************* -->
+		<component id="system.core0" name="core0">
+			<!-- Core property -->
+			<param name="clock_rate" value="700"/>
+			<param name="instruction_length" value="32"/>
+			<param name="opcode_width" value="9"/>
+			<!-- address width determins the tag_width in Cache, LSQ and buffers in cache controller 
+			default value is machine_bits, if not set --> 
+			<param name="machine_type" value="1"/><!-- 1 inorder; 0 OOO-->
+			<!-- inorder/OoO -->
+			<param name="number_hardware_threads" value="32"/>
+			<!-- number_instruction_fetch_ports(icache ports) is always 1 in single-thread processor,
+			it only may be more than one in SMT processors. BTB ports always equals to fetch ports since 
+			branch information in consective branch instructions in the same fetch group can be read out from BTB once.--> 
+			<param name="fetch_width" value="1"/>
+			<!-- fetch_width determins the size of cachelines of L1 cache block -->
+			<param name="number_instruction_fetch_ports" value="1"/>
+			<param name="decode_width" value="1"/>
+			<!-- decode_width determins the number of ports of the 
+			renaming table (both RAM and CAM) scheme -->
+			<param name="issue_width" value="2"/>
+			<!-- issue_width determins the number of ports of Issue window and other logic 
+			as in the complexity effective proccessors paper; issue_width==dispatch_width -->
+			<param name="commit_width" value="2"/>
+			<!-- commit_width determins the number of ports of register files -->
+			<param name="fp_issue_width" value="1"/>
+			<param name="prediction_width" value="0"/> 
+			<!-- number of branch instructions can be predicted simultannouesl-->
+			<!-- Current version of McPAT does not distinguish int and floating point pipelines 
+			Theses parameters are reserved for future use.--> 
+			<param name="pipelines_per_core" value="1,1"/>
+			<!--integer_pipeline and floating_pipelines, if the floating_pipelines is 0, then the pipeline is shared-->
+			<param name="pipeline_depth" value="8,8"/>
+			<!-- pipeline depth of int and fp, if pipeline is shared, the second number is the average cycles of fp ops -->
+			<!-- issue and exe unit-->
+			<param name="ALU_per_core" value="32"/>
+			<!-- contains an adder, a shifter, and a logical unit -->
+			<param name="MUL_per_core" value="4"/>
+			<!-- For MUL and Div -->
+			<param name="FPU_per_core" value="32"/>		
+			<!-- buffer between IF and ID stage -->
+			<param name="instruction_buffer_size" value="1"/>
+			<!-- buffer between ID and sche/exe stage -->
+			<param name="decoded_stream_buffer_size" value="1"/>
+			<param name="instruction_window_scheme" value="0"/><!-- 0 PHYREG based, 1 RSBASED-->
+			<!-- McPAT support 2 types of OoO cores, RS based and physical reg based-->
+			<param name="instruction_window_size" value="1"/>
+			<param name="fp_instruction_window_size" value="1"/>
+			<!-- the instruction issue Q as in Alpha 21264; The RS as in Intel P6 -->
+			<param name="ROB_size" value="0"/>
+			<!-- each in-flight instruction has an entry in ROB -->
+			<!-- registers -->
+			<!-- SM parameters  Added by Syed Gilani -->
+			<param name="rf_banks" value="32"/>
+			<param name="simd_width" value="32"/>
+			<param name="collector_units" value="32"/>
+			<param name="core_clock_ratio" value="2"/>
+			<param name="warp_size" value="32"/>
+
+			<param name="archi_Regs_IRF_size" value="32768"/>			
+			<param name="archi_Regs_FRF_size" value="32"/>
+			<!--  if OoO processor, phy_reg number is needed for renaming logic, 
+			renaming logic is for both integer and floating point insts.  -->
+			<param name="phy_Regs_IRF_size" value="32"/>
+			<param name="phy_Regs_FRF_size" value="32"/>
+			<!-- rename logic -->
+			<param name="rename_scheme" value="0"/>
+			<!-- can be RAM based(0) or CAM based(1) rename scheme 
+			RAM-based scheme will have free list, status table;
+			CAM-based scheme have the valid bit in the data field of the CAM 
+			both RAM and CAM need RAM-based checkpoint table, checkpoint_depth=# of in_flight instructions;
+			Detailed RAT Implementation see TR -->
+			<param name="register_windows_size" value="0"/>
+			<!-- how many windows in the windowed register file, sun processors;
+			no register windowing is used when this number is 0 -->
+			<!-- In OoO cores, loads and stores can be issued whether inorder(Pentium Pro) or (OoO)out-of-order(Alpha),
+			They will always try to exeute out-of-order though. -->
+			<param name="LSU_order" value="inorder"/>
+			<param name="store_buffer_size" value="32"/>
+			<!-- By default, in-order cores do not have load buffers -->
+			<param name="load_buffer_size" value="32"/>	
+			<!-- number of ports refer to sustainable concurrent memory accesses --> 
+			<param name="memory_ports" value="2"/>	
+			<!-- max_allowed_in_flight_memo_instructions determins the # of ports of load and store buffer
+			as well as the ports of Dcache which is connected to LSU -->	
+			<!-- dual-pumped Dcache can be used to save the extra read/write ports -->
+			<param name="RAS_size" value="1"/>						
+			<!-- general stats, defines simulation periods;require total, idle, and busy cycles for senity check  -->
+			<!-- please note: if target architecture is X86, then all the instrucions refer to (fused) micro-ops -->
+			<stat name="total_instructions" value="total_instructions_match_mcpat"/>
+			<stat name="int_instructions" value="int_instruction_match_mcpat"/>
+			<stat name="fp_instructions" value="flt_instruction_match_mcpat"/>
+			<stat name="branch_instructions" value="branch_instruction_match_mcpat"/>
+			<stat name="branch_mispredictions" value="0"/>
+			<stat name="load_instructions" value="load_instruction_match_mcpat"/>
+			<stat name="store_instructions" value="store_instruction_match_mcpat"/>
+			<stat name="committed_instructions" value="total_instructions_match_mcpat"/>
+			<stat name="committed_int_instructions" value="int_instruction_match_mcpat"/>
+			<stat name="committed_fp_instructions" value="flt_instruction_match_mcpat"/>
+			<stat name="pipeline_duty_cycle" value="0.6"/><!--<=1, runtime_ipc/peak_ipc; averaged for all cores if homogenous -->
+			<!-- the following cycle stats are used for heterogeneouse cores only, 
+				please ignore them if homogeneouse cores -->
+			<stat name="total_cycles" value="total_cycles_match_mcpat"/>
+		    <stat name="idle_cycles" value="idle_cycles_match_mcpat"/>
+		    <stat name="busy_cycles"  value="busy_cycles_match_mcpat"/>
+			<!-- instruction buffer stats -->
+			<!-- ROB stats, both RS and Phy based OoOs have ROB
+			performance simulator should capture the difference on accesses,
+			otherwise, McPAT has to guess based on number of commited instructions. -->
+			<stat name="ROB_reads" value="263886"/>
+			<stat name="ROB_writes" value="263886"/>
+			<!-- RAT accesses -->
+			<stat name="rename_accesses" value="263886"/>
+			<stat name="fp_rename_accesses" value="263886"/>
+			<!-- decode and rename stage use this, should be total ic - nop -->
+			<!-- Inst window stats -->
+			<stat name="inst_window_reads" value="263886"/>
+			<stat name="inst_window_writes" value="263886"/>
+			<stat name="inst_window_wakeup_accesses" value="263886"/>
+			<stat name="fp_inst_window_reads" value="263886"/>
+			<stat name="fp_inst_window_writes" value="263886"/>
+			<stat name="fp_inst_window_wakeup_accesses" value="263886"/>
+			<!--  RF accesses -->
+			<stat name="int_regfile_reads" value="int_register_read_access_match_mcpat"/>
+			<stat name="float_regfile_reads" value="int_register_write_access_match_mcpat"/>
+			<stat name="int_regfile_writes" value="float_register_read_access_match_mcpat"/>
+			<stat name="float_regfile_writes" value="float_register_write_access_match_mcpat"/>
+			
+			<!-- The following stat is for operand collector power - Added by Syed -->
+			<stat name="non_rf_operands" value="0"/>
+
+			<!-- accesses to the working reg -->
+			<stat name="function_calls" value="0"/>
+			<stat name="context_switches" value="0"/> <!--not used in the McPAT -->
+			<!-- Number of Windowes switches (number of function calls and returns)-->
+			<!-- Alu stats by default, the processor has one FPU that includes the divider and 
+			 multiplier. The fpu accesses should include accesses to multiplier and divider  -->
+			<stat name="ialu_accesses" value="ialu_accesses_match_mcpat"/>			
+			<stat name="fpu_accesses" value="fpu_accesses_match_mcpat"/>
+			<stat name="mul_accesses" value="mul_accesses_match_mcpat"/>
+			<stat name="cdb_alu_accesses" value="0"/>
+			<stat name="cdb_mul_accesses" value="0"/>
+			<stat name="cdb_fpu_accesses" value="0"/>
+			<!-- multiple cycle accesses should be counted multiple times, 
+			otherwise, McPAT can use internal counter for different floating point instructions 
+			to get final accesses. But that needs detailed info for floating point inst mix -->
+			<!--  currently the performance simulator should 
+			make sure all the numbers are final numbers, 
+			including the explicit read/write accesses, 
+			and the implicite accesses such as replacements and etc.
+			Future versions of McPAT may be able to reason the implicite access
+			based on param and stats of last level cache
+			The same rule applies to all cache access stats too!  -->
+			<!-- following is AF for max power computation. 
+				Do not change them, unless you understand them-->
+			<stat name="IFU_duty_cycle" value="0.25"/>			
+			<stat name="LSU_duty_cycle" value="0.25"/>
+			<stat name="MemManU_I_duty_cycle" value="1"/>
+			<stat name="MemManU_D_duty_cycle" value="0.25"/>
+			<stat name="ALU_duty_cycle" value="0.9"/>
+			<stat name="MUL_duty_cycle" value="0.5"/>
+			<stat name="FPU_duty_cycle" value="1"/><!-- FPU numbers are already average -->
+			<stat name="ALU_cdb_duty_cycle" value="0.9"/>
+			<stat name="MUL_cdb_duty_cycle" value="0.5"/>
+			<stat name="FPU_cdb_duty_cycle" value="15"/>	
+			<component id="system.core0.predictor" name="PBT">
+				<!-- branch predictor; tournament predictor see Alpha implementation -->
+				<param name="local_predictor_size" value="10,3"/>
+				<param name="local_predictor_entries" value="1024"/>
+				<param name="global_predictor_entries" value="4096"/>
+				<param name="global_predictor_bits" value="2"/>
+				<param name="chooser_predictor_entries" value="4096"/>
+				<param name="chooser_predictor_bits" value="2"/>
+				<!-- These parameters can be combined like below in next version
+				<param name="load_predictor" value="10,3,1024"/>
+				<param name="global_predictor" value="4096,2"/>
+				<param name="predictor_chooser" value="4096,2"/>
+				-->
+			</component>
+			<component id="system.core0.itlb" name="itlb">
+				<param name="number_entries" value="1"/>
+				<stat name="total_accesses" value="0"/>
+				<stat name="total_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+				<!-- there is no write requests to itlb although writes happen to itlb after miss, 
+				which is actually a replacement -->
+			</component>
+			<component id="system.core0.icache" name="icache">
+				<!-- there is no write requests to itlb although writes happen to it after miss, 
+				which is actually a replacement -->
+				<param name="icache_config" value="16384,32,4,1,1,3,8,0"/>
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<!-- cache_policy;//0 no write or write-though with non-write allocate;1 write-back with write-allocate -->
+				<param name="buffer_sizes" value="16, 16, 16,0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size--> 
+				<stat name="read_accesses" value="total_instructions_match_mcpat"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="conflicts" value="0"/>				
+			</component>
+			<component id="system.core0.dtlb" name="dtlb">
+				<param name="number_entries" value="1"/>
+				<stat name="total_accesses" value="0"/>
+				<stat name="total_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.ccache" name="ccache">
+			        <!-- all the buffer related are optional -->
+				<param name="ccache_config" value="16384,32,4,1,1,3,8,0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="ccache_read_accesses_match_mcpat"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="ccache_read_misses_match_mcpat"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.tcache" name="tcache">
+			        <!-- all the buffer related are optional -->
+				<param name="tcache_config" value="16384,32,4,1,1,3,8,0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="tcache_read_accesses_match_mcpat"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="tcache_read_misses_match_mcpat"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+      <!--model the shared memory by mimicing dcache-->
+			<component id="system.core0.sharedmemory" name="sharedmemory">
+			        <!-- all the buffer related are optional -->
+				<param name="sharedmemory_config" value="49152,16,1,16,1,3,16,0"/>
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<param name="buffer_sizes" value="16, 16, 16, 16"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="sharedmemory_read_access_match_mcpat"/>
+				<stat name="write_accesses" value="sharedmemory_write_access_match_mcpat"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.dcache" name="dcache">
+			        <!-- all the buffer related are optional -->
+				<param name="dcache_config" value="16384,32,4,1,1,3,8,0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="dcache_read_access_match_mcpat"/>
+				<stat name="write_accesses" value="dcache_write_access_match_mcpat"/>
+				<stat name="read_misses" value="dcache_read_miss_match_mcpat"/>
+				<stat name="write_misses" value="dcache_write_miss_match_mcpat"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.BTB" name="BTB">
+			        <!-- all the buffer related are optional -->
+				<param name="BTB_config" value="8192,4,2,1, 1,3"/>
+				<!-- the parameters are capacity,block_width,associativity,bank, throughput w.r.t. core clock, latency w.r.t. core clock,-->
+			</component>
+	</component>
+		<component id="system.L1Directory0" name="L1Directory0">
+				<param name="Directory_type" value="0"/>
+			    <!--0 cam based shadowed tag. 1 directory cache -->	
+				<param name="Dir_config" value="2048,1,0,1, 4, 4,8"/>
+				<!-- the parameters are capacity,block_width, associativity,bank, throughput w.r.t. core clock, latency w.r.t. core clock,-->
+			    <param name="buffer_sizes" value="8, 8, 8, 8"/>	
+				<!-- all the buffer related are optional -->
+			    <param name="clockrate" value="1400"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw search ports -->
+				<param name="device_type" value="0"/>
+				<!-- altough there are multiple access types, 
+				Performance simulator needs to cast them into reads or writes
+				e.g. the invalidates can be considered as writes -->
+				<stat name="read_accesses" value="800000"/>
+				<stat name="write_accesses" value="27276"/>
+				<stat name="read_misses" value="1632"/>
+				<stat name="write_misses" value="183"/>
+				<stat name="conflicts" value="20"/>	
+				<stat name="duty_cycle" value="0.45"/>	
+		</component>
+		<component id="system.L2Directory0" name="L2Directory0">
+				<param name="Directory_type" value="1"/>
+			    <!--0 cam based shadowed tag. 1 directory cache -->	
+				<param name="Dir_config" value="1048576,16,16,1,2, 100"/>
+				<!-- the parameters are capacity,block_width, associativity,bank, throughput w.r.t. core clock, latency w.r.t. core clock,-->
+			    <param name="buffer_sizes" value="8, 8, 8, 8"/>	
+				<!-- all the buffer related are optional -->
+			    <param name="clockrate" value="1400"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw search ports -->
+				<param name="device_type" value="0"/>
+				<!-- altough there are multiple access types, 
+				Performance simulator needs to cast them into reads or writes
+				e.g. the invalidates can be considered as writes -->
+				<stat name="read_accesses" value="0"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>
+			    <stat name="duty_cycle" value="0.45"/>		
+		</component>
+		<component id="system.L20" name="L20">
+			<!-- all the buffer related are optional -->
+				<param name="L2_config" value="131072,256,8,1, 4,23, 64, 1"/>
+			    <!-- consider 4-way bank interleaving for Niagara 1 -->
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<param name="buffer_sizes" value="16, 16, 16, 16"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<param name="clockrate" value="1400"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw ports -->
+				<param name="device_type" value="0"/>
+				<stat name="read_accesses" value="200000"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			    <stat name="duty_cycle" value="0.5"/>	
+		</component>
+		
+<!--**********************************************************************-->
+<component id="system.L30" name="L30">
+				<param name="L3_config" value="1048576,64,16,1, 2,100, 64,1"/>
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<param name="clockrate" value="3500"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw ports -->
+				<param name="device_type" value="0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 16"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="58824"/>
+				<stat name="write_accesses" value="27276"/>
+				<stat name="read_misses" value="1632"/>
+				<stat name="write_misses" value="183"/>
+				<stat name="conflicts" value="0"/>	
+	            <stat name="duty_cycle" value="0.35"/>	
+		</component>
+
+
+<!--**********************************************************************-->
+		<component id="system.NoC0" name="noc0">
+			<param name="clockrate" value="700"/>
+			<param name="type" value="1"/>
+			<!-- 1 NoC, O bus -->
+			<param name="horizontal_nodes" value="2"/>
+			<param name="vertical_nodes" value="1"/>
+			<param name="has_global_link" value="0"/>
+			<!-- 1 has global link, 0 does not have global link -->
+			<param name="link_throughput" value="1"/><!--w.r.t clock -->
+			<param name="link_latency" value="1"/><!--w.r.t clock -->
+			<!-- througput >= latency -->
+			<!-- Router architecture -->
+			<param name="input_ports" value="6"/>
+			<param name="output_ports" value="6"/>
+			<param name="virtual_channel_per_port" value="1"/>
+			<!-- input buffer; in classic routers only input ports need buffers -->
+			<param name="flit_bits" value="32"/>
+			<param name="input_buffer_entries_per_vc" value="1"/><!--VCs within the same ports share input buffers whose size is propotional to the number of VCs-->
+			<param name="chip_coverage" value="1"/>
+			<!-- When multiple NOC present, one NOC will cover part of the whole chip. chip_coverage <=1 -->
+			<stat name="total_accesses" value="0"/>
+			<!-- This is the number of total accesses within the whole network not for each router -->
+			<stat name="duty_cycle" value="0.6"/>
+		</component>	
+<!--**********************************************************************-->
+<!--**********************************************************************-->
+
+		<component id="system.mem" name="mem">
+			<!-- Main memory property -->
+			<param name="mem_tech_node" value="40"/>
+			<param name="device_clock" value="200"/><!--MHz, this is clock rate of the actual memory device, not the FSB -->
+			<param name="peak_transfer_rate" value="3200"/><!--MB/S-->
+			<param name="internal_prefetch_of_DRAM_chip" value="4"/>
+			<!-- 2 for DDR, 4 for DDR2, 8 for DDR3...-->
+			<!-- the device clock, peak_transfer_rate, and the internal prefetch decide the DIMM property -->
+			<!-- above numbers can be easily found from Wikipedia -->
+			<param name="capacity_per_channel" value="4096"/> <!-- MB -->
+			<!-- capacity_per_Dram_chip=capacity_per_channel/number_of_dimms/number_ranks/Dram_chips_per_rank
+			Current McPAT assumes single DIMMs are used.--> 		
+			<param name="number_ranks" value="2"/>
+			<param name="num_banks_of_DRAM_chip" value="6"/>			
+			<param name="Block_width_of_DRAM_chip" value="64"/> <!-- B -->
+			<param name="output_width_of_DRAM_chip" value="8"/>
+			<!--number of Dram_chips_per_rank=" 72/output_width_of_DRAM_chip-->
+			<!--number of Dram_chips_per_rank=" 72/output_width_of_DRAM_chip-->
+			<param name="page_size_of_DRAM_chip" value="8"/> <!-- 8 or 16 -->
+			<param name="burstlength_of_DRAM_chip" value="8"/>
+			<stat name="memory_accesses" value="1052"/>
+			<stat name="memory_reads" value="1052"/>
+			<stat name="memory_writes" value="1052"/>									
+		</component>
+		<component id="system.mc" name="mc">
+			<!-- Memeory controllers are for DDR(2,3...) DIMMs -->
+			<!-- current version of McPAT uses published values for base parameters of memory controller
+			improvments on MC will be added in later versions. -->
+			<param name="type" value="0"/> <!-- 1: low power; 0 high performance -->
+			<param name="mc_clock" value="1848"/><!--DIMM IO bus clock rate MHz DDR2-400 for Niagara 1--> 
+			<param name="peak_transfer_rate" value="29568"/><!--MB/S  Syed: GTX 470 has 177.4GB/s mem transfer rate with 6 MCs -->
+			<param name="block_size" value="64"/><!--B-->
+			<param name="number_mcs" value="6"/><!-- 6 GDDR5 memory controllers  -->
+			<!-- current McPAT only supports homogeneous memory controllers -->
+			<param name="memory_channels_per_mc" value="2"/>
+			<param name="number_ranks" value="1"/>
+			<param name="withPHY" value="0"/>
+			<!-- # of ranks of each channel-->
+			<param name="req_window_size_per_channel" value="16"/>
+			<param name="IO_buffer_size_per_channel" value="16"/>
+			<param name="databus_width" value="32"/>
+			<param name="addressbus_width" value="32"/>
+			<param name="PRT_entries" value="32"/>
+			<!-- # of empirical DRAM model parameter -->
+			<param name="dram_cmd_coeff" value="0"/>
+			<param name="dram_act_coeff" value="0"/>
+			<param name="dram_nop_coeff" value="0"/>
+			<param name="dram_activity_coeff" value="0"/>
+			<param name="dram_pre_coeff" value="3.8475e-8f"/>
+			<param name="dram_rd_coeff" value="7.74707143e-8f"/>
+			<param name="dram_wr_coeff" value="3.54664286e-8f"/>
+			<param name="dram_req_coeff" value="0"/>
+			<param name="dram_const_coeff" value="0"/>
+
+			<!-- McPAT will add the control bus width to the addressbus width automatically -->
+			<stat name="memory_accesses" value="memory_accesses_match_mcpat"/>
+			<stat name="memory_reads" value="memory_reads_match_mcpat"/>
+			<stat name="memory_writes" value="memory_writes_match_mcpat"/>
+			<!-- McPAT does not track individual mc, instead, it takes the total accesses and calculate 
+			the average power per MC or per channel. This is sufficent for most application. 
+			Further trackdown can be easily added in later versions. -->  			
+		</component>
+<!--**********************************************************************-->
+		<component id="system.niu" name="niu">
+			<!-- On chip 10Gb Ethernet NIC, including XAUI Phy and MAC controller  -->
+			<!-- For a minimum IP packet size of 84B at 10Gb/s, a new packet arrives every 67.2ns. 
+				 the low bound of clock rate of a 10Gb MAC is 150Mhz -->
+			<param name="type" value="0"/> <!-- 1: low power; 0 high performance -->
+			<param name="clockrate" value="350"/>
+			<param name="number_units" value="0"/> <!-- unlike PCIe and memory controllers, each Ethernet controller only have one port -->
+			<stat name="duty_cycle" value="1.0"/> <!-- achievable max load <= 1.0 -->
+			<stat name="total_load_perc" value="0.7"/> <!-- ratio of total achived load to total achivable bandwidth  -->
+			<!-- McPAT does not track individual nic, instead, it takes the total accesses and calculate 
+			the average power per nic or per channel. This is sufficent for most application. -->  			
+		</component>
+<!--**********************************************************************-->
+		<component id="system.pcie" name="pcie">
+			<!-- On chip PCIe controller, including Phy-->
+			<!-- For a minimum PCIe packet size of 84B at 8Gb/s per lane (PCIe 3.0), a new packet arrives every 84ns. 
+				 the low bound of clock rate of a PCIe per lane logic is 120Mhz -->
+			<param name="type" value="0"/> <!-- 1: low power; 0 high performance -->
+			<param name="withPHY" value="1"/>
+			<param name="clockrate" value="350"/>
+			<param name="number_units" value="0"/>
+			<param name="num_channels" value="8"/> <!-- 2 ,4 ,8 ,16 ,32 -->
+			<stat name="duty_cycle" value="1.0"/> <!-- achievable max load <= 1.0 -->
+			<stat name="total_load_perc" value="0.7"/> <!-- Percentage of total achived load to total achivable bandwidth  -->
+			<!-- McPAT does not track individual pcie controllers, instead, it takes the total accesses and calculate 
+			the average power per pcie controller or per channel. This is sufficent for most application. -->  			
+		</component>
+<!--**********************************************************************-->
+		<component id="system.flashc" name="flashc">
+		    <param name="number_flashcs" value="0"/>
+			<param name="type" value="1"/> <!-- 1: low power; 0 high performance -->
+            <param name="withPHY" value="1"/>
+			<param name="peak_transfer_rate" value="200"/><!--Per controller sustainable reak rate MB/S -->
+			<stat name="duty_cycle" value="1.0"/> <!-- achievable max load <= 1.0 -->
+			<stat name="total_load_perc" value="0.7"/> <!-- Percentage of total achived load to total achivable bandwidth  -->
+			<!-- McPAT does not track individual flash controller, instead, it takes the total accesses and calculate 
+			the average power per fc or per channel. This is sufficent for most application -->  			
+		</component>
+<!--**********************************************************************-->
+
+		</component>
+</component>
diff --git a/configs/GeForceGTX750Ti/config_fermi_islip.icnt b/configs/4.x-cfgs/SM5_GTX750/config_fermi_islip.icnt
index 069ca02..069ca02 100644
--- a/configs/GeForceGTX750Ti/config_fermi_islip.icnt
+++ b/configs/4.x-cfgs/SM5_GTX750/config_fermi_islip.icnt
diff --git a/configs/GeForceGTX750Ti/gpgpusim.config b/configs/4.x-cfgs/SM5_GTX750/gpgpusim.config
index 9366f93..9366f93 100644
--- a/configs/GeForceGTX750Ti/gpgpusim.config
+++ b/configs/4.x-cfgs/SM5_GTX750/gpgpusim.config
diff --git a/configs/GeForceGTX750Ti/gpuwattch_gtx750Ti.xml b/configs/4.x-cfgs/SM5_GTX750/gpuwattch_gtx750Ti.xml
index e2b2324..e2b2324 100755
--- a/configs/GeForceGTX750Ti/gpuwattch_gtx750Ti.xml
+++ b/configs/4.x-cfgs/SM5_GTX750/gpuwattch_gtx750Ti.xml
diff --git a/configs/4.x-cfgs/SM6_GTX1080/config_fermi_islip.icnt b/configs/4.x-cfgs/SM6_GTX1080/config_fermi_islip.icnt
new file mode 100644
index 0000000..2a69ddd
--- /dev/null
+++ b/configs/4.x-cfgs/SM6_GTX1080/config_fermi_islip.icnt
@@ -0,0 +1,70 @@
+//21*1 fly with 32 flits per packet under gpgpusim injection mode
+use_map = 0;
+flit_size = 32; 
+
+// currently we do not use this, see subnets below
+network_count = 2;
+
+// Topology
+topology = fly;
+k = 50;
+n = 1;
+
+// Routing
+
+routing_function = dest_tag;
+
+// Flow control
+
+num_vcs     = 1;
+vc_buf_size = 8;
+
+wait_for_tail_credit = 0;
+
+// Router architecture
+
+vc_allocator = islip; //separable_input_first;
+sw_allocator = islip; //separable_input_first;
+alloc_iters  = 1;
+
+credit_delay   = 0;
+routing_delay  = 0;
+vc_alloc_delay = 1;
+sw_alloc_delay = 1;
+
+input_speedup     = 2;
+output_speedup    = 1;
+internal_speedup  = 1.0;
+
+// Traffic, GPGPU-Sim does not use this
+
+traffic                = uniform;
+packet_size ={{1,2,3,4},{10,20}};
+packet_size_rate={{1,1,1,1},{2,1}};
+
+// Simulation - Don't change
+
+sim_type       = gpgpusim;
+//sim_type = latency;
+injection_rate = 0.1;
+
+subnets = 2;
+
+// Always use read and write no matter following line
+//use_read_write = 1;
+
+
+read_request_subnet = 0;
+read_reply_subnet = 1;
+write_request_subnet = 0;
+write_reply_subnet = 1;
+
+read_request_begin_vc = 0;
+read_request_end_vc = 0;
+write_request_begin_vc = 0;
+write_request_end_vc = 0;
+read_reply_begin_vc = 0;
+read_reply_end_vc = 0;
+write_reply_begin_vc = 0;
+write_reply_end_vc = 0;
+
diff --git a/configs/4.x-cfgs/SM6_GTX1080/gpgpusim.config b/configs/4.x-cfgs/SM6_GTX1080/gpgpusim.config
new file mode 100644
index 0000000..47c2b6a
--- /dev/null
+++ b/configs/4.x-cfgs/SM6_GTX1080/gpgpusim.config
@@ -0,0 +1,149 @@
+# This config models the Pascal GP102 (GeForceGTX 1080Ti)
+
+# functional simulator specification
+-gpgpu_ptx_instruction_classification 0
+-gpgpu_ptx_sim_mode 0
+-gpgpu_ptx_force_max_capability 20 
+
+# SASS execution (only supported with CUDA >= 4.0)
+-gpgpu_ptx_convert_to_ptxplus 0
+-gpgpu_ptx_save_converted_ptxplus 0
+
+# high level architecture configuration
+-gpgpu_n_clusters 28
+-gpgpu_n_cores_per_cluster 1
+-gpgpu_n_mem 11
+-gpgpu_n_sub_partition_per_mchannel 2 
+
+# Pascal clock domains
+#-gpgpu_clock_domains <Core Clock>:<Interconnect Clock>:<L2 Clock>:<DRAM Clock>
+# Pascal NVIDIA TITAN X clock domains are adopted from 
+# https://en.wikipedia.org/wiki/GeForce_10_series
+-gpgpu_clock_domains 1481.0:2962.0:1481.0:2750.0
+
+# shader core pipeline config
+-gpgpu_shader_registers 65536
+
+# This implies a maximum of 64 warps/SM
+-gpgpu_shader_core_pipeline 2048:32 
+-gpgpu_shader_cta 32
+-gpgpu_simd_model 1 
+
+# Pipeline widths and number of FUs
+# ID_OC_SP,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_SFU,OC_EX_MEM,EX_WB
+## Pascal GP102 has 4 SP SIMD units and 1 SFU unit
+## we need to scale the number of pipeline registers to be equal to the number of SP units
+-gpgpu_pipeline_widths 4,1,1,4,1,1,6
+-gpgpu_num_sp_units 4
+-gpgpu_num_sfu_units 1
+
+# Instruction latencies and initiation intervals
+# "ADD,MAX,MUL,MAD,DIV"
+# SFU is 32-width in pascal, then dp units initiation is 1 cycle
+-ptx_opcode_latency_int 4,13,4,5,145
+-ptx_opcode_initiation_int 1,2,2,2,8
+-ptx_opcode_latency_fp 4,13,4,5,39
+-ptx_opcode_initiation_fp 1,2,1,1,4
+-ptx_opcode_latency_dp 8,19,8,8,330
+-ptx_opcode_initiation_dp 1,2,1,1,130
+
+# <nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>:<set_index_fn>,<mshr>:<N>:<merge>,<mq>:**<fifo_entry>
+# ** Optional parameter - Required when mshr_type==Texture Fifo
+# Note: Hashing set index function (H) only applies to a set size of 32 or 64. 
+# Pascal GP102 has 96KB Shared memory
+# Pascal GP102 has 64KB L1 cache
+# The default is to disable the L1 cache, unless cache modifieres is used
+-gpgpu_cache:dl1  64:128:6,L:L:m:N:H,A:128:8,8
+-gpgpu_shmem_size 98304
+-gmem_skip_L1D 1
+
+# 64 sets, each 128 bytes 16-way for each memory sub partition (128 KB per memory sub partition). This gives 3MB L2 cache
+-gpgpu_cache:dl2 64:128:16,L:B:m:W:L,A:1024:1024,4:0,32 # used to be 128:4
+-gpgpu_cache:dl2_texture_only 0 
+
+# 4 KB Inst.
+-gpgpu_cache:il1 8:128:4,L:R:f:N:L,A:2:48,4
+# 48 KB Tex 
+-gpgpu_tex_cache:l1 16:128:24,L:R:m:N:L,F:128:4,128:2
+# 12 KB Const
+-gpgpu_const_cache:l1 128:64:2,L:R:f:N:L,A:2:64,4
+
+# enable operand collector 
+## larger operand collectors and reg_banks are needed for the 4 warp schedulers and 4 SIMD units
+-gpgpu_operand_collector_num_units_sp 20
+-gpgpu_operand_collector_num_units_sfu 4
+-gpgpu_operand_collector_num_units_mem 8
+-gpgpu_operand_collector_num_in_ports_sp 4
+-gpgpu_operand_collector_num_out_ports_sp 4
+-gpgpu_operand_collector_num_in_ports_sfu 1
+-gpgpu_operand_collector_num_out_ports_sfu 1
+-gpgpu_operand_collector_num_in_ports_mem 1
+-gpgpu_operand_collector_num_out_ports_mem 1
+# gpgpu_num_reg_banks should be increased to 32, but it gives an error!
+-gpgpu_num_reg_banks 32
+
+# shared memory bankconflict detection 
+-gpgpu_shmem_num_banks 32
+-gpgpu_shmem_limited_broadcast 0
+-gpgpu_shmem_warp_parts 1
+
+## In Pascal, a warp scheduler can issue 2 insts per cycle
+-gpgpu_max_insn_issue_per_warp 2
+
+# interconnection
+-network_mode 1 
+-inter_config_file config_fermi_islip.icnt
+
+# memory partition latency config 
+-rop_latency 120
+-dram_latency 100
+
+# dram model config
+-gpgpu_dram_scheduler 1
+# The DRAM return queue and the scheduler queue together should provide buffer
+# to sustain the memory level parallelism to tolerate DRAM latency 
+# To allow 100% DRAM utility, there should at least be enough buffer to sustain
+# the minimum DRAM latency (100 core cycles).  I.e. 
+#   Total buffer space required = 100 x 924MHz / 700MHz = 132
+-gpgpu_frfcfs_dram_sched_queue_size 64
+-gpgpu_dram_return_queue_size 116
+
+# for NVIDIA GeForceGTX 1080Ti, bus width is 352bits (11 DRAM chips x 32 bits)
+# 11 memory paritions, 4 bytes (1 DRAM chip) per memory partition
+# the atom size of GDDR5X (the smallest read request) is 32 bytes 
+-gpgpu_n_mem_per_ctrlr 1
+-gpgpu_dram_buswidth 4
+-gpgpu_dram_burst_length 8
+-dram_data_command_freq_ratio 4  # GDDR5X is QDR
+-gpgpu_mem_address_mask 1
+-gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBCCCC.BCCSSSSS
+
+# Use the same GDDR5 timing from hynix H5GQ1H24AFR
+# disable bank groups for now, set nbkgrp to 1 and tCCDL and tRTPL to 0
+-gpgpu_dram_timing_opt "nbk=16:CCD=2:RRD=6:RCD=12:RAS=28:RP=12:RC=40:
+                        CL=12:WL=4:CDLR=5:WR=12:nbkgrp=1:CCDL=0:RTPL=0"
+
+# Pascal has four schedulers per core
+-gpgpu_num_sched_per_core 2
+# Two Level Scheduler with active and pending pools
+#-gpgpu_scheduler two_level_active:6:0:1
+# Loose round robbin scheduler
+#-gpgpu_scheduler lrr
+# Greedy then oldest scheduler
+-gpgpu_scheduler gto
+
+# stat collection
+-gpgpu_memlatency_stat 14 
+-gpgpu_runtime_stat 500
+-enable_ptx_file_line_stats 1
+-visualizer_enabled 0
+
+# power model configs
+-power_simulation_enabled 1
+-gpuwattch_xml_file gpuwattch_gtx1080Ti.xml
+
+# tracing functionality
+#-trace_enabled 1
+#-trace_components WARP_SCHEDULER,SCOREBOARD
+#-trace_sampling_core 0
+
diff --git a/configs/4.x-cfgs/SM6_GTX1080/gpuwattch_gtx1080Ti.xml b/configs/4.x-cfgs/SM6_GTX1080/gpuwattch_gtx1080Ti.xml
new file mode 100755
index 0000000..02619ff
--- /dev/null
+++ b/configs/4.x-cfgs/SM6_GTX1080/gpuwattch_gtx1080Ti.xml
@@ -0,0 +1,538 @@
+<?xml version="1.0" ?>
+<component id="root" name="root">
+	<component id="system" name="system">
+		<!--McPAT will skip the components if number is set to 0 -->
+		<param name="GPU_Architecture" value="1"/><!-- 0-G80; 1-Fermi; others not supported -->
+		<param name="number_of_cores" value="28"/>
+		<param name="architecture" value="1"/> <!-- fermi:1 quadro:2 other: undefined-->
+		<param name="number_of_L1Directories" value="0"/>
+		<param name="number_of_L2Directories" value="0"/>
+		<param name="number_of_L2s" value="1"/> <!-- This number means how many L2 clusters in each cluster there can be multiple banks/ports -->
+		<param name="number_of_L3s" value="0"/> <!-- This number means how many L3 clusters -->
+		<param name="number_of_NoCs" value="1"/> 
+		<param name="homogeneous_cores" value="1"/><!--1 means homo -->
+		<param name="homogeneous_L2s" value="1"/>
+		<param name="homogeneous_L1Directorys" value="1"/>
+		<param name="homogeneous_L2Directorys" value="1"/>
+		<param name="homogeneous_L3s" value="1"/>
+		<param name="homogeneous_ccs" value="1"/><!--cache coherece hardware -->
+		<param name="homogeneous_NoCs" value="1"/>
+		<param name="core_tech_node" value="23"/><!-- nm -->
+		<param name="target_core_clockrate" value="1481"/><!--MHz -->
+		<param name="temperature" value="380"/> <!-- Kelvin -->
+		<param name="number_cache_levels" value="2"/>
+		<param name="interconnect_projection_type" value="0"/><!--0: agressive wire technology; 1: conservative wire technology -->
+		<param name="device_type" value="0"/><!--0: HP(High Performance Type); 1: LSTP(Low standby power) 2: LOP (Low Operating Power)  -->
+		<param name="longer_channel_device" value="1"/><!-- 0 no use; 1 use when possible -->
+		<param name="machine_bits" value="32"/>
+		<param name="virtual_address_width" value="32"/>
+		<param name="physical_address_width" value="32"/>
+		<param name="virtual_memory_page_size" value="4096"/>
+		<param name="idle_core_power" value="1.59"/><!-- idle core power for GTX479 -->
+		<!--param name="scaling_coefficients" value="10,0.0884816,10,10,8,10,4.12782,10,2.48832,10,10,10,4.29982,0.387764,0.0714269,0.14302,0.01,0.546811,0.485351,0.806633,0.818073,1.9207,100,100,100,87.9303,100,10,4.3548,10"/-->
+		<param name="TOT_INST" value="10" />
+		<param name="FP_INT" value="10" />
+		<param name="IC_H" value="0.001" />
+		<param name="IC_M" value="10" />
+		<param name="DC_RH" value="1" />
+		<param name="DC_RM" value="1" />
+		<param name="DC_WH" value="1" />
+		<param name="DC_WM" value="1" />
+		<param name="TC_H" value="0.001" />
+		<param name="TC_M" value="10" />
+		<param name="CC_H" value="4.5071" />
+		<param name="CC_M" value="10" />
+		<param name="SHRD_ACC" value="10" />
+		<param name="REG_RD" value="1.6294" />
+		<param name="REG_WR" value="0.5031" />
+		<param name="NON_REG_OPs" value="0.01" />
+		<param name="SP_ACC" value="10" />
+		<param name="SFU_ACC" value="0.0082" />
+		<param name="FPU_ACC" value="0.4126" />
+		<param name="MEM_RD" value="0.1234" />
+		<param name="MEM_WR" value="0.001" />
+		<param name="MEM_PRE" value="0.001" />
+		<param name="L2_RH" value="100" />
+		<param name="L2_RM" value="100" />
+		<param name="L2_WH" value="100" />
+		<param name="L2_WM" value="42.6966" />
+		<param name="NOC_A" value="100" />
+		<param name="PIPE_A" value="44.8085" />
+		<param name="IDLE_CORE_N" value="2.0382"/>
+		<param name="CONST_DYNAMICN" value="5.0005" />
+		<stat name="num_idle_cores" value="0"/><!-- Average Number of idle cores during this period -->
+		<stat name="total_cycles" value="total_cycles_match_mcpat"/>
+		<stat name="idle_cycles" value="idle_cycles_match_mcpat"/>
+		<stat name="busy_cycles"  value="busy_cycles_match_mcpat"/>
+			<!--This page size(B) is complete different from the page size in Main memo secction. this page size is the size of 
+			virtual memory from OS/Archi perspective; the page size in Main memo secction is the actuall physical line in a DRAM bank  -->
+		<!-- *********************** cores ******************* -->
+		<component id="system.core0" name="core0">
+			<!-- Core property -->
+			<param name="clock_rate" value="1481"/>
+			<param name="instruction_length" value="32"/>
+			<param name="opcode_width" value="9"/>
+			<!-- address width determins the tag_width in Cache, LSQ and buffers in cache controller 
+			default value is machine_bits, if not set --> 
+			<param name="machine_type" value="1"/><!-- 1 inorder; 0 OOO-->
+			<!-- inorder/OoO -->
+			<param name="number_hardware_threads" value="32"/>
+			<!-- number_instruction_fetch_ports(icache ports) is always 1 in single-thread processor,
+			it only may be more than one in SMT processors. BTB ports always equals to fetch ports since 
+			branch information in consective branch instructions in the same fetch group can be read out from BTB once.--> 
+			<param name="fetch_width" value="1"/>
+			<!-- fetch_width determins the size of cachelines of L1 cache block -->
+			<param name="number_instruction_fetch_ports" value="1"/>
+			<param name="decode_width" value="1"/>
+			<!-- decode_width determins the number of ports of the 
+			renaming table (both RAM and CAM) scheme -->
+			<param name="issue_width" value="2"/>
+			<!-- issue_width determins the number of ports of Issue window and other logic 
+			as in the complexity effective proccessors paper; issue_width==dispatch_width -->
+			<param name="commit_width" value="2"/>
+			<!-- commit_width determins the number of ports of register files -->
+			<param name="fp_issue_width" value="1"/>
+			<param name="prediction_width" value="0"/> 
+			<!-- number of branch instructions can be predicted simultannouesl-->
+			<!-- Current version of McPAT does not distinguish int and floating point pipelines 
+			Theses parameters are reserved for future use.--> 
+			<param name="pipelines_per_core" value="1,1"/>
+			<!--integer_pipeline and floating_pipelines, if the floating_pipelines is 0, then the pipeline is shared-->
+			<param name="pipeline_depth" value="8,8"/>
+			<!-- pipeline depth of int and fp, if pipeline is shared, the second number is the average cycles of fp ops -->
+			<!-- issue and exe unit-->
+			<param name="ALU_per_core" value="32"/>
+			<!-- contains an adder, a shifter, and a logical unit -->
+			<param name="MUL_per_core" value="4"/>
+			<!-- For MUL and Div -->
+			<param name="FPU_per_core" value="32"/>		
+			<!-- buffer between IF and ID stage -->
+			<param name="instruction_buffer_size" value="1"/>
+			<!-- buffer between ID and sche/exe stage -->
+			<param name="decoded_stream_buffer_size" value="1"/>
+			<param name="instruction_window_scheme" value="0"/><!-- 0 PHYREG based, 1 RSBASED-->
+			<!-- McPAT support 2 types of OoO cores, RS based and physical reg based-->
+			<param name="instruction_window_size" value="1"/>
+			<param name="fp_instruction_window_size" value="1"/>
+			<!-- the instruction issue Q as in Alpha 21264; The RS as in Intel P6 -->
+			<param name="ROB_size" value="0"/>
+			<!-- each in-flight instruction has an entry in ROB -->
+			<!-- registers -->
+			<!-- SM parameters  Added by Syed Gilani -->
+			<param name="rf_banks" value="32"/>
+			<param name="simd_width" value="32"/>
+			<param name="collector_units" value="32"/>
+			<param name="core_clock_ratio" value="2"/>
+			<param name="warp_size" value="32"/>
+
+			<param name="archi_Regs_IRF_size" value="65536"/>			
+			<param name="archi_Regs_FRF_size" value="32"/>
+			<!--  if OoO processor, phy_reg number is needed for renaming logic, 
+			renaming logic is for both integer and floating point insts.  -->
+			<param name="phy_Regs_IRF_size" value="32"/>
+			<param name="phy_Regs_FRF_size" value="32"/>
+			<!-- rename logic -->
+			<param name="rename_scheme" value="0"/>
+			<!-- can be RAM based(0) or CAM based(1) rename scheme 
+			RAM-based scheme will have free list, status table;
+			CAM-based scheme have the valid bit in the data field of the CAM 
+			both RAM and CAM need RAM-based checkpoint table, checkpoint_depth=# of in_flight instructions;
+			Detailed RAT Implementation see TR -->
+			<param name="register_windows_size" value="0"/>
+			<!-- how many windows in the windowed register file, sun processors;
+			no register windowing is used when this number is 0 -->
+			<!-- In OoO cores, loads and stores can be issued whether inorder(Pentium Pro) or (OoO)out-of-order(Alpha),
+			They will always try to exeute out-of-order though. -->
+			<param name="LSU_order" value="inorder"/>
+			<param name="store_buffer_size" value="32"/>
+			<!-- By default, in-order cores do not have load buffers -->
+			<param name="load_buffer_size" value="32"/>	
+			<!-- number of ports refer to sustainable concurrent memory accesses --> 
+			<param name="memory_ports" value="2"/>	
+			<!-- max_allowed_in_flight_memo_instructions determins the # of ports of load and store buffer
+			as well as the ports of Dcache which is connected to LSU -->	
+			<!-- dual-pumped Dcache can be used to save the extra read/write ports -->
+			<param name="RAS_size" value="1"/>						
+			<!-- general stats, defines simulation periods;require total, idle, and busy cycles for senity check  -->
+			<!-- please note: if target architecture is X86, then all the instrucions refer to (fused) micro-ops -->
+			<stat name="total_instructions" value="total_instructions_match_mcpat"/>
+			<stat name="int_instructions" value="int_instruction_match_mcpat"/>
+			<stat name="fp_instructions" value="flt_instruction_match_mcpat"/>
+			<stat name="branch_instructions" value="branch_instruction_match_mcpat"/>
+			<stat name="branch_mispredictions" value="0"/>
+			<stat name="load_instructions" value="load_instruction_match_mcpat"/>
+			<stat name="store_instructions" value="store_instruction_match_mcpat"/>
+			<stat name="committed_instructions" value="total_instructions_match_mcpat"/>
+			<stat name="committed_int_instructions" value="int_instruction_match_mcpat"/>
+			<stat name="committed_fp_instructions" value="flt_instruction_match_mcpat"/>
+			<stat name="pipeline_duty_cycle" value="0.6"/><!--<=1, runtime_ipc/peak_ipc; averaged for all cores if homogenous -->
+			<!-- the following cycle stats are used for heterogeneouse cores only, 
+				please ignore them if homogeneouse cores -->
+			<stat name="total_cycles" value="total_cycles_match_mcpat"/>
+		    <stat name="idle_cycles" value="idle_cycles_match_mcpat"/>
+		    <stat name="busy_cycles"  value="busy_cycles_match_mcpat"/>
+			<!-- instruction buffer stats -->
+			<!-- ROB stats, both RS and Phy based OoOs have ROB
+			performance simulator should capture the difference on accesses,
+			otherwise, McPAT has to guess based on number of commited instructions. -->
+			<stat name="ROB_reads" value="263886"/>
+			<stat name="ROB_writes" value="263886"/>
+			<!-- RAT accesses -->
+			<stat name="rename_accesses" value="263886"/>
+			<stat name="fp_rename_accesses" value="263886"/>
+			<!-- decode and rename stage use this, should be total ic - nop -->
+			<!-- Inst window stats -->
+			<stat name="inst_window_reads" value="263886"/>
+			<stat name="inst_window_writes" value="263886"/>
+			<stat name="inst_window_wakeup_accesses" value="263886"/>
+			<stat name="fp_inst_window_reads" value="263886"/>
+			<stat name="fp_inst_window_writes" value="263886"/>
+			<stat name="fp_inst_window_wakeup_accesses" value="263886"/>
+			<!--  RF accesses -->
+			<stat name="int_regfile_reads" value="int_register_read_access_match_mcpat"/>
+			<stat name="float_regfile_reads" value="int_register_write_access_match_mcpat"/>
+			<stat name="int_regfile_writes" value="float_register_read_access_match_mcpat"/>
+			<stat name="float_regfile_writes" value="float_register_write_access_match_mcpat"/>
+			
+			<!-- The following stat is for operand collector power - Added by Syed -->
+			<stat name="non_rf_operands" value="0"/>
+
+			<!-- accesses to the working reg -->
+			<stat name="function_calls" value="0"/>
+			<stat name="context_switches" value="0"/> <!--not used in the McPAT -->
+			<!-- Number of Windowes switches (number of function calls and returns)-->
+			<!-- Alu stats by default, the processor has one FPU that includes the divider and 
+			 multiplier. The fpu accesses should include accesses to multiplier and divider  -->
+			<stat name="ialu_accesses" value="ialu_accesses_match_mcpat"/>			
+			<stat name="fpu_accesses" value="fpu_accesses_match_mcpat"/>
+			<stat name="mul_accesses" value="mul_accesses_match_mcpat"/>
+			<stat name="cdb_alu_accesses" value="0"/>
+			<stat name="cdb_mul_accesses" value="0"/>
+			<stat name="cdb_fpu_accesses" value="0"/>
+			<!-- multiple cycle accesses should be counted multiple times, 
+			otherwise, McPAT can use internal counter for different floating point instructions 
+			to get final accesses. But that needs detailed info for floating point inst mix -->
+			<!--  currently the performance simulator should 
+			make sure all the numbers are final numbers, 
+			including the explicit read/write accesses, 
+			and the implicite accesses such as replacements and etc.
+			Future versions of McPAT may be able to reason the implicite access
+			based on param and stats of last level cache
+			The same rule applies to all cache access stats too!  -->
+			<!-- following is AF for max power computation. 
+				Do not change them, unless you understand them-->
+			<stat name="IFU_duty_cycle" value="0.25"/>			
+			<stat name="LSU_duty_cycle" value="0.25"/>
+			<stat name="MemManU_I_duty_cycle" value="1"/>
+			<stat name="MemManU_D_duty_cycle" value="0.25"/>
+			<stat name="ALU_duty_cycle" value="0.9"/>
+			<stat name="MUL_duty_cycle" value="0.5"/>
+			<stat name="FPU_duty_cycle" value="1"/><!-- FPU numbers are already average -->
+			<stat name="ALU_cdb_duty_cycle" value="0.9"/>
+			<stat name="MUL_cdb_duty_cycle" value="0.5"/>
+			<stat name="FPU_cdb_duty_cycle" value="15"/>	
+			<component id="system.core0.predictor" name="PBT">
+				<!-- branch predictor; tournament predictor see Alpha implementation -->
+				<param name="local_predictor_size" value="10,3"/>
+				<param name="local_predictor_entries" value="1024"/>
+				<param name="global_predictor_entries" value="4096"/>
+				<param name="global_predictor_bits" value="2"/>
+				<param name="chooser_predictor_entries" value="4096"/>
+				<param name="chooser_predictor_bits" value="2"/>
+				<!-- These parameters can be combined like below in next version
+				<param name="load_predictor" value="10,3,1024"/>
+				<param name="global_predictor" value="4096,2"/>
+				<param name="predictor_chooser" value="4096,2"/>
+				-->
+			</component>
+			<component id="system.core0.itlb" name="itlb">
+				<param name="number_entries" value="1"/>
+				<stat name="total_accesses" value="0"/>
+				<stat name="total_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+				<!-- there is no write requests to itlb although writes happen to itlb after miss, 
+				which is actually a replacement -->
+			</component>
+			<component id="system.core0.icache" name="icache">
+				<!-- there is no write requests to itlb although writes happen to it after miss, 
+				which is actually a replacement -->
+				<param name="icache_config" value="16384,128,4,1,1,3,8,0"/>
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<!-- cache_policy;//0 no write or write-though with non-write allocate;1 write-back with write-allocate -->
+				<param name="buffer_sizes" value="16, 16, 16,0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size--> 
+				<stat name="read_accesses" value="total_instructions_match_mcpat"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="conflicts" value="0"/>				
+			</component>
+			<component id="system.core0.dtlb" name="dtlb">
+				<param name="number_entries" value="1"/>
+				<stat name="total_accesses" value="0"/>
+				<stat name="total_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.ccache" name="ccache">
+			        <!-- all the buffer related are optional -->
+				<param name="ccache_config" value="16384,64,2,1,1,3,8,0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="ccache_read_accesses_match_mcpat"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="ccache_read_misses_match_mcpat"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.tcache" name="tcache">
+			        <!-- all the buffer related are optional -->
+				<param name="tcache_config" value="49152,128,8,1,1,3,8,0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="tcache_read_accesses_match_mcpat"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="tcache_read_misses_match_mcpat"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+      <!--model the shared memory by mimicing dcache-->
+			<component id="system.core0.sharedmemory" name="sharedmemory">
+			        <!-- all the buffer related are optional -->
+				<param name="sharedmemory_config" value="98304,16,1,16,1,3,16,0"/>
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<param name="buffer_sizes" value="16, 16, 16, 16"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="sharedmemory_read_access_match_mcpat"/>
+				<stat name="write_accesses" value="sharedmemory_write_access_match_mcpat"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.dcache" name="dcache">
+			        <!-- all the buffer related are optional -->
+				<param name="dcache_config" value="16384,32,4,1,1,3,8,0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 0"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="dcache_read_access_match_mcpat"/>
+				<stat name="write_accesses" value="dcache_write_access_match_mcpat"/>
+				<stat name="read_misses" value="dcache_read_miss_match_mcpat"/>
+				<stat name="write_misses" value="dcache_write_miss_match_mcpat"/>
+				<stat name="conflicts" value="0"/>	
+			</component>
+			<component id="system.core0.BTB" name="BTB">
+			        <!-- all the buffer related are optional -->
+				<param name="BTB_config" value="8192,4,2,1, 1,3"/>
+				<!-- the parameters are capacity,block_width,associativity,bank, throughput w.r.t. core clock, latency w.r.t. core clock,-->
+			</component>
+	</component>
+		<component id="system.L1Directory0" name="L1Directory0">
+				<param name="Directory_type" value="0"/>
+			    <!--0 cam based shadowed tag. 1 directory cache -->	
+				<param name="Dir_config" value="2048,1,0,1, 4, 4,8"/>
+				<!-- the parameters are capacity,block_width, associativity,bank, throughput w.r.t. core clock, latency w.r.t. core clock,-->
+			    <param name="buffer_sizes" value="8, 8, 8, 8"/>	
+				<!-- all the buffer related are optional -->
+			    <param name="clockrate" value="1400"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw search ports -->
+				<param name="device_type" value="0"/>
+				<!-- altough there are multiple access types, 
+				Performance simulator needs to cast them into reads or writes
+				e.g. the invalidates can be considered as writes -->
+				<stat name="read_accesses" value="800000"/>
+				<stat name="write_accesses" value="27276"/>
+				<stat name="read_misses" value="1632"/>
+				<stat name="write_misses" value="183"/>
+				<stat name="conflicts" value="20"/>	
+				<stat name="duty_cycle" value="0.45"/>	
+		</component>
+		<component id="system.L2Directory0" name="L2Directory0">
+				<param name="Directory_type" value="1"/>
+			    <!--0 cam based shadowed tag. 1 directory cache -->	
+				<param name="Dir_config" value="1048576,16,16,1,2, 100"/>
+				<!-- the parameters are capacity,block_width, associativity,bank, throughput w.r.t. core clock, latency w.r.t. core clock,-->
+			    <param name="buffer_sizes" value="8, 8, 8, 8"/>	
+				<!-- all the buffer related are optional -->
+			    <param name="clockrate" value="1400"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw search ports -->
+				<param name="device_type" value="0"/>
+				<!-- altough there are multiple access types, 
+				Performance simulator needs to cast them into reads or writes
+				e.g. the invalidates can be considered as writes -->
+				<stat name="read_accesses" value="0"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>
+			    <stat name="duty_cycle" value="0.45"/>		
+		</component>
+		<component id="system.L20" name="L20">
+			<!-- all the buffer related are optional -->
+				<param name="L2_config" value="131072,128,16,1, 4,23, 64, 1"/>
+			    <!-- consider 4-way bank interleaving for Niagara 1 -->
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<param name="buffer_sizes" value="16, 16, 16, 16"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<param name="clockrate" value="2962"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw ports -->
+				<param name="device_type" value="0"/>
+				<stat name="read_accesses" value="200000"/>
+				<stat name="write_accesses" value="0"/>
+				<stat name="read_misses" value="0"/>
+				<stat name="write_misses" value="0"/>
+				<stat name="conflicts" value="0"/>	
+			    <stat name="duty_cycle" value="0.5"/>	
+		</component>
+		
+<!--**********************************************************************-->
+<component id="system.L30" name="L30">
+				<param name="L3_config" value="1048576,64,16,1, 2,100, 64,1"/>
+				<!-- the parameters are capacity,block_width, associativity, bank, throughput w.r.t. core clock, latency w.r.t. core clock,output_width, cache policy -->
+				<param name="clockrate" value="3500"/>
+				<param name="ports" value="1,1,1"/>
+				<!-- number of r, w, and rw ports -->
+				<param name="device_type" value="0"/>
+				<param name="buffer_sizes" value="16, 16, 16, 16"/>
+				<!-- cache controller buffer sizes: miss_buffer_size(MSHR),fill_buffer_size,prefetch_buffer_size,wb_buffer_size-->	
+				<stat name="read_accesses" value="58824"/>
+				<stat name="write_accesses" value="27276"/>
+				<stat name="read_misses" value="1632"/>
+				<stat name="write_misses" value="183"/>
+				<stat name="conflicts" value="0"/>	
+	            <stat name="duty_cycle" value="0.35"/>	
+		</component>
+
+
+<!--**********************************************************************-->
+		<component id="system.NoC0" name="noc0">
+			<param name="clockrate" value="700"/>
+			<param name="type" value="1"/>
+			<!-- 1 NoC, O bus -->
+			<param name="horizontal_nodes" value="2"/>
+			<param name="vertical_nodes" value="1"/>
+			<param name="has_global_link" value="0"/>
+			<!-- 1 has global link, 0 does not have global link -->
+			<param name="link_throughput" value="1"/><!--w.r.t clock -->
+			<param name="link_latency" value="1"/><!--w.r.t clock -->
+			<!-- througput >= latency -->
+			<!-- Router architecture -->
+			<param name="input_ports" value="6"/>
+			<param name="output_ports" value="6"/>
+			<param name="virtual_channel_per_port" value="1"/>
+			<!-- input buffer; in classic routers only input ports need buffers -->
+			<param name="flit_bits" value="32"/>
+			<param name="input_buffer_entries_per_vc" value="1"/><!--VCs within the same ports share input buffers whose size is propotional to the number of VCs-->
+			<param name="chip_coverage" value="1"/>
+			<!-- When multiple NOC present, one NOC will cover part of the whole chip. chip_coverage <=1 -->
+			<stat name="total_accesses" value="0"/>
+			<!-- This is the number of total accesses within the whole network not for each router -->
+			<stat name="duty_cycle" value="0.6"/>
+		</component>	
+<!--**********************************************************************-->
+<!--**********************************************************************-->
+
+		<component id="system.mem" name="mem">
+			<!-- Main memory property -->
+			<param name="mem_tech_node" value="23"/>
+			<param name="device_clock" value="200"/><!--MHz, this is clock rate of the actual memory device, not the FSB -->
+			<param name="peak_transfer_rate" value="3200"/><!--MB/S-->
+			<param name="internal_prefetch_of_DRAM_chip" value="4"/>
+			<!-- 2 for DDR, 4 for DDR2, 8 for DDR3...-->
+			<!-- the device clock, peak_transfer_rate, and the internal prefetch decide the DIMM property -->
+			<!-- above numbers can be easily found from Wikipedia -->
+			<param name="capacity_per_channel" value="4096"/> <!-- MB -->
+			<!-- capacity_per_Dram_chip=capacity_per_channel/number_of_dimms/number_ranks/Dram_chips_per_rank
+			Current McPAT assumes single DIMMs are used.--> 		
+			<param name="number_ranks" value="2"/>
+			<param name="num_banks_of_DRAM_chip" value="6"/>			
+			<param name="Block_width_of_DRAM_chip" value="64"/> <!-- B -->
+			<param name="output_width_of_DRAM_chip" value="8"/>
+			<!--number of Dram_chips_per_rank=" 72/output_width_of_DRAM_chip-->
+			<!--number of Dram_chips_per_rank=" 72/output_width_of_DRAM_chip-->
+			<param name="page_size_of_DRAM_chip" value="8"/> <!-- 8 or 16 -->
+			<param name="burstlength_of_DRAM_chip" value="8"/>
+			<stat name="memory_accesses" value="1052"/>
+			<stat name="memory_reads" value="1052"/>
+			<stat name="memory_writes" value="1052"/>									
+		</component>
+		<component id="system.mc" name="mc">
+			<!-- Memeory controllers are for DDR(2,3...) DIMMs -->
+			<!-- current version of McPAT uses published values for base parameters of memory controller
+			improvments on MC will be added in later versions. -->
+			<param name="type" value="0"/> <!-- 1: low power; 0 high performance -->
+			<param name="mc_clock" value="1848"/><!--DIMM IO bus clock rate MHz DDR2-400 for Niagara 1--> 
+			<param name="peak_transfer_rate" value="29568"/><!--MB/S  Syed: GTX 470 has 177.4GB/s mem transfer rate with 6 MCs -->
+			<param name="block_size" value="64"/><!--B-->
+			<param name="number_mcs" value="6"/><!-- 6 GDDR5 memory controllers  -->
+			<!-- current McPAT only supports homogeneous memory controllers -->
+			<param name="memory_channels_per_mc" value="2"/>
+			<param name="number_ranks" value="1"/>
+			<param name="withPHY" value="0"/>
+			<!-- # of ranks of each channel-->
+			<param name="req_window_size_per_channel" value="16"/>
+			<param name="IO_buffer_size_per_channel" value="16"/>
+			<param name="databus_width" value="32"/>
+			<param name="addressbus_width" value="32"/>
+			<param name="PRT_entries" value="32"/>
+			<!-- # of empirical DRAM model parameter -->
+			<param name="dram_cmd_coeff" value="0"/>
+			<param name="dram_act_coeff" value="0"/>
+			<param name="dram_nop_coeff" value="0"/>
+			<param name="dram_activity_coeff" value="0"/>
+			<param name="dram_pre_coeff" value="3.8475e-8f"/>
+			<param name="dram_rd_coeff" value="7.74707143e-8f"/>
+			<param name="dram_wr_coeff" value="3.54664286e-8f"/>
+			<param name="dram_req_coeff" value="0"/>
+			<param name="dram_const_coeff" value="0"/>
+
+			<!-- McPAT will add the control bus width to the addressbus width automatically -->
+			<stat name="memory_accesses" value="memory_accesses_match_mcpat"/>
+			<stat name="memory_reads" value="memory_reads_match_mcpat"/>
+			<stat name="memory_writes" value="memory_writes_match_mcpat"/>
+			<!-- McPAT does not track individual mc, instead, it takes the total accesses and calculate 
+			the average power per MC or per channel. This is sufficent for most application. 
+			Further trackdown can be easily added in later versions. -->  			
+		</component>
+<!--**********************************************************************-->
+		<component id="system.niu" name="niu">
+			<!-- On chip 10Gb Ethernet NIC, including XAUI Phy and MAC controller  -->
+			<!-- For a minimum IP packet size of 84B at 10Gb/s, a new packet arrives every 67.2ns. 
+				 the low bound of clock rate of a 10Gb MAC is 150Mhz -->
+			<param name="type" value="0"/> <!-- 1: low power; 0 high performance -->
+			<param name="clockrate" value="350"/>
+			<param name="number_units" value="0"/> <!-- unlike PCIe and memory controllers, each Ethernet controller only have one port -->
+			<stat name="duty_cycle" value="1.0"/> <!-- achievable max load <= 1.0 -->
+			<stat name="total_load_perc" value="0.7"/> <!-- ratio of total achived load to total achivable bandwidth  -->
+			<!-- McPAT does not track individual nic, instead, it takes the total accesses and calculate 
+			the average power per nic or per channel. This is sufficent for most application. -->  			
+		</component>
+<!--**********************************************************************-->
+		<component id="system.pcie" name="pcie">
+			<!-- On chip PCIe controller, including Phy-->
+			<!-- For a minimum PCIe packet size of 84B at 8Gb/s per lane (PCIe 3.0), a new packet arrives every 84ns. 
+				 the low bound of clock rate of a PCIe per lane logic is 120Mhz -->
+			<param name="type" value="0"/> <!-- 1: low power; 0 high performance -->
+			<param name="withPHY" value="1"/>
+			<param name="clockrate" value="350"/>
+			<param name="number_units" value="0"/>
+			<param name="num_channels" value="8"/> <!-- 2 ,4 ,8 ,16 ,32 -->
+			<stat name="duty_cycle" value="1.0"/> <!-- achievable max load <= 1.0 -->
+			<stat name="total_load_perc" value="0.7"/> <!-- Percentage of total achived load to total achivable bandwidth  -->
+			<!-- McPAT does not track individual pcie controllers, instead, it takes the total accesses and calculate 
+			the average power per pcie controller or per channel. This is sufficent for most application. -->  			
+		</component>
+<!--**********************************************************************-->
+		<component id="system.flashc" name="flashc">
+		    <param name="number_flashcs" value="0"/>
+			<param name="type" value="1"/> <!-- 1: low power; 0 high performance -->
+            <param name="withPHY" value="1"/>
+			<param name="peak_transfer_rate" value="200"/><!--Per controller sustainable reak rate MB/S -->
+			<stat name="duty_cycle" value="1.0"/> <!-- achievable max load <= 1.0 -->
+			<stat name="total_load_perc" value="0.7"/> <!-- Percentage of total achived load to total achivable bandwidth  -->
+			<!-- McPAT does not track individual flash controller, instead, it takes the total accesses and calculate 
+			the average power per fc or per channel. This is sufficent for most application -->  			
+		</component>
+<!--**********************************************************************-->
+
+		</component>
+</component>
diff --git a/configs/Pascal-P100-HBM/config_fermi_islip.icnt b/configs/4.x-cfgs/SM6_P100/config_fermi_islip.icnt
index e7c2c3b..e7c2c3b 100644
--- a/configs/Pascal-P100-HBM/config_fermi_islip.icnt
+++ b/configs/4.x-cfgs/SM6_P100/config_fermi_islip.icnt
diff --git a/configs/Pascal-P100-HBM/gpgpusim.config b/configs/4.x-cfgs/SM6_P100/gpgpusim.config
index 533a865..a43c0bb 100644
--- a/configs/Pascal-P100-HBM/gpgpusim.config
+++ b/configs/4.x-cfgs/SM6_P100/gpgpusim.config
@@ -55,16 +55,21 @@
 # Note: Hashing set index function (H) only applies to a set size of 32 or 64. 
 # Pascal GP100 has 64KB Shared memory
 -gpgpu_cache:dl1  S:64:128:6,L:L:f:N:H,A:256:8,16:0,32
+-gpgpu_cache:dl1PrefL1  S:64:128:6,L:L:f:N:H,A:256:8,16:0,32
+-gpgpu_cache:dl1PrefShared  S:64:128:6,L:L:f:N:H,A:256:8,16:0,32
 -gpgpu_shmem_size 65536
--gmem_skip_L1D 1
+-gpgpu_shmem_size_PrefL1 65536
+-gpgpu_shmem_size_PrefShared 65536
+-gmem_skip_L1D 0
 -icnt_flit_size 40
 -gpgpu_n_cluster_ejection_buffer_size 32
 
 # 32 sets, each 128 bytes 16-way for each memory sub partition (128 KB per memory sub partition). This gives  4MB L2 cache
--gpgpu_cache:dl2 S:64:128:16,L:B:m:W:L,A:256:4,32:0,32
+-gpgpu_cache:dl2 S:64:128:16,L:B:m:F:L,A:256:4,32:0,32
 -gpgpu_cache:dl2_texture_only 0 
 -gpgpu_dram_partition_queues 64:64:64:64
 -perf_sim_memcpy 0
+-memory_partition_indexing 2
 
 # 4 KB Inst.
 -gpgpu_cache:il1 N:8:128:4,L:R:f:N:L,S:2:48,4
@@ -86,7 +91,6 @@
 -gpgpu_operand_collector_num_out_ports_mem 1
 -gpgpu_operand_collector_num_in_ports_dp 1
 -gpgpu_operand_collector_num_out_ports_dp 1
-# gpgpu_num_reg_banks should be increased to 32
 -gpgpu_num_reg_banks 32
 
 # shared memory bankconflict detection 
@@ -105,8 +109,7 @@
 
 # memory partition latency config 
 -rop_latency 120
-# DRAM latency should be lower compared to other configs, due to high-speed interposer connection
--dram_latency 60
+-dram_latency 100
 
 # dram model config
 -gpgpu_dram_scheduler 1
diff --git a/configs/Pascal-P102-GDDR5X/config_fermi_islip.icnt b/configs/4.x-cfgs/SM6_TITANX/config_fermi_islip.icnt
index 714d933..714d933 100644
--- a/configs/Pascal-P102-GDDR5X/config_fermi_islip.icnt
+++ b/configs/4.x-cfgs/SM6_TITANX/config_fermi_islip.icnt
diff --git a/configs/Pascal-P102-GDDR5X/gpgpusim.config b/configs/4.x-cfgs/SM6_TITANX/gpgpusim.config
index 0c6c126..3097d19 100644
--- a/configs/Pascal-P102-GDDR5X/gpgpusim.config
+++ b/configs/4.x-cfgs/SM6_TITANX/gpgpusim.config
@@ -6,6 +6,7 @@
 -gpgpu_ptx_instruction_classification 0
 -gpgpu_ptx_sim_mode 0
 -gpgpu_ptx_force_max_capability 61 
+-gpgpu_ignore_resources_limitation 1
 
 # SASS execution (only supported with CUDA >= 4.0)
 -gpgpu_ptx_convert_to_ptxplus 0
@@ -62,7 +63,11 @@
 # Pascal GP102 has 24KB L1 cache
 # The defulat is to disable the L1 cache, unless cache modifieres is used
 -gpgpu_cache:dl1  S:32:128:6,L:L:f:N:H,A:256:8,16:0,32
+-gpgpu_cache:dl1PrefL1  S:32:128:6,L:L:f:N:H,A:256:8,16:0,32
+-gpgpu_cache:dl1PrefShared  S:32:128:6,L:L:f:N:H,A:256:8,16:0,32
 -gpgpu_shmem_size 49152
+-gpgpu_shmem_size_PrefL1 49152
+-gpgpu_shmem_size_PrefShared 49152
 -gmem_skip_L1D 1
 -icnt_flit_size 40
 -gpgpu_n_cluster_ejection_buffer_size 32
diff --git a/configs/4.x-cfgs/SM7_TITANV/config_fermi_islip.icnt b/configs/4.x-cfgs/SM7_TITANV/config_fermi_islip.icnt
new file mode 100644
index 0000000..616e9f3
--- /dev/null
+++ b/configs/4.x-cfgs/SM7_TITANV/config_fermi_islip.icnt
@@ -0,0 +1,74 @@
+//21*1 fly with 32 flits per packet under gpgpusim injection mode
+use_map = 0;
+flit_size = 40; 
+
+// currently we do not use this, see subnets below
+network_count = 2;
+
+// Topology
+topology = fly;
+k = 64;
+n = 1;
+
+// Routing
+
+routing_function = dest_tag;
+
+
+// Flow control
+
+num_vcs     = 1;
+vc_buf_size = 128;
+input_buffer_size = 256;
+ejection_buffer_size = 128;
+boundary_buffer_size = 128;
+
+wait_for_tail_credit = 0;
+
+// Router architecture
+
+vc_allocator = islip; //separable_input_first;
+sw_allocator = islip; //separable_input_first;
+alloc_iters  = 1;
+
+credit_delay   = 0;
+routing_delay  = 0;
+vc_alloc_delay = 1;
+sw_alloc_delay = 1;
+
+input_speedup     = 2;
+output_speedup    = 1;
+internal_speedup  = 1.0;
+
+// Traffic, GPGPU-Sim does not use this
+
+traffic                = uniform;
+packet_size ={{1,2,3,4},{10,20}};
+packet_size_rate={{1,1,1,1},{2,1}};
+
+// Simulation - Don't change
+
+sim_type       = gpgpusim;
+//sim_type = latency;
+injection_rate = 0.1;
+
+subnets = 2;
+
+// Always use read and write no matter following line
+//use_read_write = 1;
+
+
+read_request_subnet = 0;
+read_reply_subnet = 1;
+write_request_subnet = 0;
+write_reply_subnet = 1;
+
+read_request_begin_vc = 0;
+read_request_end_vc = 0;
+write_request_begin_vc = 0;
+write_request_end_vc = 0;
+read_reply_begin_vc = 0;
+read_reply_end_vc = 0;
+write_reply_begin_vc = 0;
+write_reply_end_vc = 0;
+
diff --git a/configs/4.x-cfgs/SM7_TITANV/gpgpusim.config b/configs/4.x-cfgs/SM7_TITANV/gpgpusim.config
new file mode 100644
index 0000000..8969168
--- /dev/null
+++ b/configs/4.x-cfgs/SM7_TITANV/gpgpusim.config
@@ -0,0 +1,186 @@
+# This config models the Volta Titan X
+# For more info about this card:
+# http://images.nvidia.com/content/volta-architecture/pdf/volta-architecture-whitepaper.pdf
+# https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8344474&tag=1# 
+# http://on-demand.gputechconf.com/gtc/2018/presentation/s8122-dissecting-the-volta-gpu-architecture-through-microbenchmarking.pdf
+# https://en.wikipedia.org/wiki/Volta_(microarchitecture)
+# https://www.hotchips.org/wp-content/uploads/hc_archives/hc29/HC29.21-Monday-Pub/HC29.21.10-GPU-Gaming-Pub/HC29.21.132-Volta-Choquette-NVIDIA-Final3.pdf
+# https://devblogs.nvidia.com/inside-volta/
+# http://on-demand.gputechconf.com/gtc/2017/presentation/s7798-luke-durant-inside-volta.pdf
+
+# functional simulator specification
+-gpgpu_ptx_instruction_classification 0
+-gpgpu_ptx_sim_mode 0
+-gpgpu_ptx_force_max_capability 70 
+
+# SASS execution (only supported with CUDA >= 4.0)
+-gpgpu_ptx_convert_to_ptxplus 0
+-gpgpu_ptx_save_converted_ptxplus 0
+
+# high level architecture configuration
+-gpgpu_n_clusters 40
+-gpgpu_n_cores_per_cluster 2
+-gpgpu_n_mem 24
+-gpgpu_n_sub_partition_per_mchannel 1 
+
+# volta clock domains
+#-gpgpu_clock_domains <Core Clock>:<Interconnect Clock>:<L2 Clock>:<DRAM Clock>
+# Pascal NVIDIA GP100  clock domains are adopted from 
+# https://en.wikipedia.org/wiki/Volta_(microarchitecture)
+-gpgpu_clock_domains 1200.0:1200.0:2000.0:850.0
+# boost mode
+# -gpgpu_clock_domains 1455.0:1455.0:2000.0:850.0
+
+# shader core pipeline config
+-gpgpu_shader_registers 65536
+
+# This implies a maximum of 64 warps/SM
+-gpgpu_shader_core_pipeline 2048:32 
+-gpgpu_shader_cta 32
+-gpgpu_simd_model 1 
+
+# Pipeline widths and number of FUs
+# ID_OC_SP,ID_OC_DP,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_DP,OC_EX_SFU,OC_EX_MEM,EX_WB
+## Pascal GP100 has 2 SP SIMD units, 2 SFU units, 2 DP units per core
+## we need to scale the number of pipeline registers to be equal to the number of SP units
+-gpgpu_pipeline_widths 4,4,4,1,4,4,4,1,12
+-gpgpu_num_sp_units 4
+-gpgpu_num_sfu_units 4
+-gpgpu_num_dp_units 4
+
+# Instruction latencies and initiation intervals
+# "ADD,MAX,MUL,MAD,DIV"
+# All Div operations are executed on SFU unit
+# Throughput (initiation latency) are adopted from 
+# http://on-demand.gputechconf.com/gtc/2018/presentation/s8122-dissecting-the-volta-gpu-architecture-through-microbenchmarking.pdf
+-ptx_opcode_latency_int 4,13,4,5,145
+-ptx_opcode_initiation_int 2,2,2,2,8
+-ptx_opcode_latency_fp 4,13,4,5,39
+-ptx_opcode_initiation_fp 2,2,2,2,4
+-ptx_opcode_latency_dp 8,19,8,8,330
+-ptx_opcode_initiation_dp 4,4,4,4,130
+-ptx_opcode_latency_sfu 100
+-ptx_opcode_initiation_sfu 8
+
+
+# <nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>:<set_index_fn>,<mshr>:<N>:<merge>,<mq>:**<fifo_entry>
+# ** Optional parameter - Required when mshr_type==Texture Fifo
+# Note: Hashing set index function (H) only applies to a set size of 32 or 64. 
+# Pascal GP100 has 64KB Shared memory
+-gpgpu_cache:dl1  S:64:128:8,L:L:f:N:H,A:256:8,16:0,32
+-gpgpu_cache:dl1PrefL1  S:64:128:16,L:L:f:N:H,A:256:8,16:0,32
+-gpgpu_cache:dl1PrefShared  S:32:128:6,L:L:f:N:H,A:256:8,16:0,32
+-gpgpu_shmem_size 65536
+-gpgpu_shmem_size_PrefL1 1
+-gpgpu_shmem_size_PrefShared 98304
+-gmem_skip_L1D 0
+-icnt_flit_size 40
+-gpgpu_n_cluster_ejection_buffer_size 32
+
+# 64 sets, each 128 bytes 24-way for each memory sub partition (192 KB per memory sub partition). This gives  4.5MB L2 cache
+-gpgpu_cache:dl2 S:64:128:24,L:B:m:F:L,A:256:4,32:0,32
+-gpgpu_cache:dl2_texture_only 0 
+-gpgpu_dram_partition_queues 64:64:64:64
+-perf_sim_memcpy 0
+
+# 128 KB Inst.
+-gpgpu_cache:il1 N:64:128:16,L:R:f:N:L,S:2:48,4
+# 48 KB Tex 
+-gpgpu_tex_cache:l1 N:16:128:24,L:R:m:N:L,F:128:4,128:2
+# 64 KB Const
+-gpgpu_const_cache:l1 N:128:64:8,L:R:f:N:L,S:2:64,4
+
+# enable operand collector 
+-gpgpu_operand_collector_num_units_sp 12
+-gpgpu_operand_collector_num_units_sfu 6
+-gpgpu_operand_collector_num_units_mem 8
+-gpgpu_operand_collector_num_units_dp 6
+-gpgpu_operand_collector_num_in_ports_sp 4
+-gpgpu_operand_collector_num_out_ports_sp 4
+-gpgpu_operand_collector_num_in_ports_sfu 1
+-gpgpu_operand_collector_num_out_ports_sfu 1
+-gpgpu_operand_collector_num_in_ports_mem 1
+-gpgpu_operand_collector_num_out_ports_mem 1
+-gpgpu_operand_collector_num_in_ports_dp 1
+-gpgpu_operand_collector_num_out_ports_dp 1
+-gpgpu_num_reg_banks 32
+
+# shared memory bankconflict detection 
+-gpgpu_shmem_num_banks 32
+-gpgpu_shmem_limited_broadcast 0
+-gpgpu_shmem_warp_parts 1
+-gpgpu_coalesce_arch 60
+
+## In Pascal, a warp scheduler can issue 2 insts per cycle using 2 diff execution units
+-gpgpu_max_insn_issue_per_warp 1
+-gpgpu_dual_issue_diff_exec_units 1
+
+# interconnection
+-network_mode 1 
+-inter_config_file config_fermi_islip.icnt
+
+# memory partition latency config 
+-rop_latency 120
+-dram_latency 100
+
+# dram model config
+-gpgpu_dram_scheduler 1
+# The DRAM return queue and the scheduler queue together should provide buffer
+# to sustain the memory level parallelism to tolerate DRAM latency 
+# To allow 100% DRAM utility, there should at least be enough buffer to sustain
+# the minimum DRAM latency (100 core cycles).  I.e. 
+#   Total buffer space required = 100 x 924MHz / 700MHz = 132
+-gpgpu_frfcfs_dram_sched_queue_size 64
+-gpgpu_dram_return_queue_size 192
+
+# for HBM, 32 channles, each (128 bits) 16 bytes width
+-gpgpu_n_mem_per_ctrlr 1
+-gpgpu_dram_buswidth 16
+-gpgpu_dram_burst_length 2
+-dram_data_command_freq_ratio 2  # HBM is DDR
+-gpgpu_mem_address_mask 1
+-gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.RBBBBCCC.CCCSSSSS
+
+# HBM timing are adopted from hynix JESD235 standered and nVidia HPCA 2017 paper (http://www.cs.utah.edu/~nil/pubs/hpca17.pdf)
+# Timing for 1 GHZ
+# tRRDl and tWTR are missing, need to be added
+#-gpgpu_dram_timing_opt "nbk=16:CCD=1:RRD=4:RCD=14:RAS=33:RP=14:RC=47:
+#                        CL=14:WL=2:CDLR=3:WR=12:nbkgrp=4:CCDL=2:RTPL=4"
+
+# Timing for 715 MHZ, Tesla Pascal P100 HBM runs at 715 MHZ
+-gpgpu_dram_timing_opt "nbk=16:CCD=1:RRD=3:RCD=10:RAS=24:RP=10:RC=34:
+                        CL=10:WL=2:CDLR=3:WR=9:nbkgrp=4:CCDL=2:RTPL=3"
+
+# HBM has dual bus interface, in which it can issue two col and row commands at a time
+-dual_bus_interface 1
+# select lower bits for bnkgrp to increase bnkgrp parallelism
+-dram_bnk_indexing_policy 0
+-dram_bnkgrp_indexing_policy 1
+
+#-Seperate_Write_Queue_Enable 1
+#-Write_Queue_Size 64:56:32
+
+# Pascal has two schedulers per core
+-gpgpu_num_sched_per_core 2
+# Two Level Scheduler with active and pending pools
+#-gpgpu_scheduler two_level_active:6:0:1
+# Loose round robbin scheduler
+#-gpgpu_scheduler lrr
+# Greedy then oldest scheduler
+-gpgpu_scheduler gto
+
+# stat collection
+-gpgpu_memlatency_stat 14 
+-gpgpu_runtime_stat 500
+-enable_ptx_file_line_stats 1
+-visualizer_enabled 0
+
+# power model configs, disable it untill we create a real energy model for Pascal 100
+-power_simulation_enabled 0
+-gpuwattch_xml_file gpuwattch_gtx480.xml
+
+# tracing functionality
+#-trace_enabled 1
+#-trace_components WARP_SCHEDULER,SCOREBOARD
+#-trace_sampling_core 0
+
diff --git a/cuobjdump_to_ptxplus/ptx_parser.h b/cuobjdump_to_ptxplus/ptx_parser.h
index 418a733..729eaec 100644
--- a/cuobjdump_to_ptxplus/ptx_parser.h
+++ b/cuobjdump_to_ptxplus/ptx_parser.h
@@ -373,4 +373,8 @@ void func_header_info_int(const char* s, int i)
 		g_headerList->getListEnd().addOperand(buff);
 	}
 }
+
+void maxnt_id(int x, int y, int z) {
+
+}
 #endif //_PTX_PARSER_H_
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index cbe8a11..7ee7dbe 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -336,10 +336,19 @@ class _cuda_device_id *GPGPUSim_Init()
 		prop->minor = 2;
 		prop->totalGlobalMem = 0x80000000 /* 2 GB */;
 		prop->memPitch = 0;
-		prop->maxThreadsPerBlock = 512;
-		prop->maxThreadsDim[0] = 512;
-		prop->maxThreadsDim[1] = 512;
-		prop->maxThreadsDim[2] = 512;
+		if(prop->major >= 2) {
+			prop->maxThreadsPerBlock = 1024;
+			prop->maxThreadsDim[0] = 1024;
+			prop->maxThreadsDim[1] = 1024;
+		}
+		else
+		{
+			prop->maxThreadsPerBlock = 512;
+			prop->maxThreadsDim[0] = 512;
+			prop->maxThreadsDim[1] = 512;
+		}
+
+		prop->maxThreadsDim[2] = 64;
 		prop->maxGridSize[0] = 0x40000000;
 		prop->maxGridSize[1] = 0x40000000;
 		prop->maxGridSize[2] = 0x40000000;
@@ -352,6 +361,9 @@ class _cuda_device_id *GPGPUSim_Init()
 #if (CUDART_VERSION >= 2010)
 		prop->multiProcessorCount = the_gpu->get_config().num_shader();
 #endif
+#if (CUDART_VERSION >= 4000)
+		prop->maxThreadsPerMultiProcessor = the_gpu->threads_per_core();
+#endif
 		the_gpu->set_prop(prop);
 		the_device = new _cuda_device_id(the_gpu);
 	}
@@ -2092,8 +2104,28 @@ cudaError_t CUDARTAPI cudaSetValidDevices(int *device_arr, int len)
 
 cudaError_t CUDARTAPI cudaSetDeviceFlags( int flags )
 {
-	cuda_not_implemented(__my_func__,__LINE__);
-	return g_last_cudaError = cudaErrorUnknown;
+    // This flag is implicitly always on (unless you are using the driver API). It is safe for GPGPU-Sim to
+    // just ignore it.
+    if ( cudaDeviceMapHost == flags ) {
+		return g_last_cudaError = cudaSuccess;
+    } else {
+	    cuda_not_implemented(__my_func__,__LINE__);
+	    return g_last_cudaError = cudaErrorUnknown;
+    }
+}
+
+size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr) {
+  _cuda_device_id *dev = GPGPUSim_Init();
+  struct cudaDeviceProp prop;
+
+  prop = *dev->get_prop();
+
+  size_t max = prop.maxThreadsPerBlock;
+
+  if ((prop.regsPerBlock / attr->numRegs) < max)
+    max = prop.regsPerBlock / attr->numRegs;
+
+  return max;
 }
 
 cudaError_t CUDARTAPI cudaFuncGetAttributes(struct cudaFuncAttributes *attr, const char *hostFun )
@@ -2106,7 +2138,10 @@ cudaError_t CUDARTAPI cudaFuncGetAttributes(struct cudaFuncAttributes *attr, con
 		attr->constSizeBytes  = kinfo->cmem;
 		attr->localSizeBytes  = kinfo->lmem;
 		attr->numRegs         = kinfo->regs;
-		attr->maxThreadsPerBlock = 0; // from pragmas?
+		if(kinfo->maxthreads > 0)
+		  attr->maxThreadsPerBlock = kinfo->maxthreads;
+		else
+		  attr->maxThreadsPerBlock = getMaxThreadsPerBlock(attr);
 #if CUDART_VERSION >= 3000
 		attr->ptxVersion      = kinfo->ptx_version;
 		attr->binaryVersion   = kinfo->sm_target;
diff --git a/linux-so-version.txt b/linux-so-version.txt
new file mode 100644
index 0000000..40f775d
--- /dev/null
+++ b/linux-so-version.txt
@@ -0,0 +1,2 @@
+libcudart.so.9.1{
+};
diff --git a/nightly.jenkinsfile b/nightly.jenkinsfile
new file mode 100644
index 0000000..78f3920
--- /dev/null
+++ b/nightly.jenkinsfile
@@ -0,0 +1,74 @@
+pipeline {
+    agent {
+        label "purdue-cluster"
+    }
+
+    options {
+        disableConcurrentBuilds()
+        overrideIndexTriggers(true)
+    }
+
+    triggers {
+        pollSCM('0 1 * * *')
+    }
+
+    stages {
+        stage('nightly-simulator-build') {
+            steps {
+                sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\
+                source `pwd`/setup_environment &&\
+                make -j'
+            }
+        }
+        stage('nightly-simulations-build'){
+            steps{
+                sh 'rm -rf gpgpu-sim_simulations'
+                sh 'git clone [email protected]:TimRogersGroup/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 &&\
+                    source `pwd`/setup_environment &&\
+                    cd gpgpu-sim_simulations && \
+                    source ./benchmarks/src/setup_environment && \
+                    make -i -j -C ./benchmarks/src/ all && \
+                    make -C ./benchmarks/src data'
+            }
+        }
+        stage('nightly-2B-insn-run'){
+            steps {
+                    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 $PLOTDIR'
+            }
+        }
+        stage('nightly-correlate'){
+            steps {
+                sh 'source /home/tgrogers-raid/a/common/gpgpu-sim-setup/9.1_env_setup.sh &&\
+                    source `pwd`/setup_environment &&\
+                    PLOTDIR="jenkins/${JOB_NAME}" &&\
+                    ./gpgpu-sim_simulations/util/plotting/correlate_and_publish.sh TITANX-2B,TITANX-L1ON-2B,P100-2B,TITANV-2B $PLOTDIR ${BUILD_NUMBER}'
+            }
+        }
+    }
+    post {
+        success {
+            emailext body: "See ${BUILD_URL}",
+                recipientProviders: [[$class: 'CulpritsRecipientProvider'],
+                    [$class: 'RequesterRecipientProvider']],
+                subject: "[AALP Jenkins] Build #${BUILD_NUMBER} - Success!",
+                to: '[email protected]'
+        }
+        failure {
+            emailext body: "See ${BUILD_URL}",
+                recipientProviders: [[$class: 'CulpritsRecipientProvider'],
+                    [$class: 'RequesterRecipientProvider']],
+                subject: "[AALP Jenkins] Build #${BUILD_NUMBER} - ${currentBuild.result}",
+                to: '[email protected]'
+        }
+    }
+}
diff --git a/setup_environment b/setup_environment
index 0895d44..f6a16c5 100644
--- a/setup_environment
+++ b/setup_environment
@@ -6,7 +6,11 @@ export GPGPUSIM_SETUP_ENVIRONMENT_WAS_RUN=
 export GPGPUSIM_ROOT="$( cd "$( dirname "$BASH_SOURCE" )" && pwd )"
 
 GPGPUSIM_VERSION_STRING=`cat $GPGPUSIM_ROOT/version | awk '/Version/ {print $8}'`
-GPGPUSIM_BUILD_STRING=`cat $GPGPUSIM_ROOT/version | awk '/Change/ {print $6}'`
+#Detect Git branch and commit #
+GIT_COMMIT=`git log -n 1 | head -1 | sed -re 's/commit (.*)/\1/'`
+GIT_FILES_CHANGED=`git diff --numstat --cached && git diff --numstat | wc | sed -re 's/^\s+([0-9]+).*/\1/'`
+GPGPUSIM_BUILD_STRING="gpgpu-sim_git-commit-$GIT_COMMIT-modified_$GIT_FILES_CHANGED"
+
 echo -n "GPGPU-Sim version $GPGPUSIM_VERSION_STRING (build $GPGPUSIM_BUILD_STRING) ";
 
 if [ ! -n "$CUDA_INSTALL_PATH" ]; then
@@ -43,11 +47,9 @@ CC_VERSION=`gcc --version | head -1 | awk '{for(i=1;i<=NF;i++){ if(match($i,/^[0
 
 CUDA_VERSION_STRING=`$CUDA_INSTALL_PATH/bin/nvcc --version | awk '/release/ {print $5;}' | sed 's/,//'`;
 CUDA_VERSION_NUMBER=`echo $CUDA_VERSION_STRING | sed 's/\./ /' | awk '{printf("%02u%02u", 10*int($1), 10*$2);}'`
-if [ $CUDA_VERSION_NUMBER -gt 8000 -o $CUDA_VERSION_NUMBER -lt 2030  ]; then
+if [ $CUDA_VERSION_NUMBER -gt 9100 -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
-elif [ $CUDA_VERSION_NUMBER -gt 4020 ]; then
-	echo "WARNING ** GPGPU-Sim version $GPGPUSIM_VERSION_STRING not fully tested with CUDA version $CUDA_VERSION_STRING (please see README)";
 fi
 
 if [ $# = '1' ] ;
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index cec75f9..1b764e2 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -517,7 +517,14 @@ public:
     const struct textureReference* get_texref(const std::string &texname) const
     {
         std::map<std::string, const struct textureReference*>::const_iterator t=m_NameToTextureRef.find(texname);
-        assert( t != m_NameToTextureRef.end() );
+        if( t == m_NameToTextureRef.end() ) {
+	  // search for :: prefixed names
+	  std::string temp("::" + texname);
+	  t=m_NameToTextureRef.find(temp);
+	}
+
+	assert(t != m_NameToTextureRef.end());
+
         return t->second;
     }
     const struct cudaArray* get_texarray( const struct textureReference *texref ) const
@@ -568,6 +575,7 @@ struct gpgpu_ptx_sim_info
    int cmem;
    int gmem;
    int regs;
+   unsigned maxthreads;
    unsigned ptx_version;
    unsigned sm_target;
 };
diff --git a/src/cuda-sim/cuda-math.h b/src/cuda-sim/cuda-math.h
index 4721e8a..f88c526 100644
--- a/src/cuda-sim/cuda-math.h
+++ b/src/cuda-sim/cuda-math.h
@@ -67,6 +67,8 @@
 #ifndef CUDA_MATH
 #define CUDA_MATH
 
+#include <cmath>
+
 // cuda math implementations
 #undef max
 #undef min
@@ -321,7 +323,7 @@ float __internal_accurate_fdividef(float a, float b)
 float __saturatef(float a)
 {
    float b; 
-   if (isnan(a)) b = 0.0f; 
+   if (std::isnan(a)) b = 0.0f; 
    else if (a >= 1.0f) b = 1.0f;
    else if (a <= 0.0f) b = 0.0f; 
    else b = a; 
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index 2f166aa..9246613 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -1159,13 +1159,13 @@ void function_info::finalize( memory_space *param_mem )
       // copy the parameter over word-by-word so that parameter that crosses a memory page can be copied over
       //Jin: copy parameter using aligned rules
       const size_t word_size = 4; 
-      param_address = (param_address + size - 1) / size * size; //aligned with size 
+      //param_address = (param_address + size - 1) / size * size; //aligned with size TODO: align not correct
       for (size_t idx = 0; idx < size; idx += word_size) {
          const char *pdata = reinterpret_cast<const char*>(param_value.pdata) + idx; // cast to char * for ptr arithmetic
          param_mem->write(param_address + idx, word_size, pdata,NULL,NULL); 
       }
       unsigned offset = p.get_offset();
-      assert(offset == param_address);
+      //assert(offset == param_address);
       param->set_address(param_address);
       param_address += size;
    }
@@ -1432,7 +1432,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id)
    if ( (g_ptx_sim_num_insn % 100000) == 0 ) {
       dim3 ctaid = get_ctaid();
       dim3 tid = get_tid();
-      printf("GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
+      DPRINTF(LIVENESS, "GPGPU-Sim PTX: %u instructions simulated : ctaid=(%u,%u,%u) tid=(%u,%u,%u)\n",
              g_ptx_sim_num_insn, ctaid.x,ctaid.y,ctaid.z,tid.x,tid.y,tid.z );
       fflush(stdout);
    }
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 011c285..e3b8970 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -33,6 +33,7 @@
 #include "ptx.tab.h"
 #include <stdlib.h>
 #include <math.h>
+#include <cmath>
 #include <fenv.h>
 #include "cuda-math.h"
 #include "../abstract_hardware_model.h"
@@ -1961,7 +1962,7 @@ ptx_reg_t d2d( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign,
       y.f64 = x.f64;
       break; 
    }
-   if (isnan(y.f64)) {
+   if (std::isnan(y.f64)) {
       y.u64 = 0xfff8000000000000ull;
    } else if (saturation_mode) {
       y.f64 = cuda_math::__saturatef(y.f64); 
@@ -2086,7 +2087,7 @@ void ptx_round(ptx_reg_t& data, int rounding_mode, int type)
       }
    }
    if ((type == F64_TYPE)||(type == FF64_TYPE)) {
-      if (isnan(data.f64)) {
+      if (std::isnan(data.f64)) {
          data.u64 = 0xfff8000000000000ull;
       }
    }
@@ -2648,12 +2649,12 @@ void mad_def( const ptx_instruction *pI, ptx_thread_info *thread, bool use_carry
 
 bool isNaN(float x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 bool isNaN(double x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 void max_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y
index e00aa4b..c0c58a6 100644
--- a/src/cuda-sim/ptx.y
+++ b/src/cuda-sim/ptx.y
@@ -228,7 +228,8 @@ function_defn: function_decl { set_symtab($1); func_header(".skip"); } statement
 
 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); }
+										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"); }
 	;
diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc
index 8ebdcf8..ee36957 100644
--- a/src/cuda-sim/ptx_ir.cc
+++ b/src/cuda-sim/ptx_ir.cc
@@ -222,6 +222,7 @@ bool symbol_table::add_function_decl( const char *name, int entry_point, functio
    } else {
       *func_info = new function_info(entry_point);
       (*func_info)->set_name(name);
+      (*func_info)->set_maxnt_id(0);
       m_function_info_lookup[key] = *func_info;
    }
 
diff --git a/src/cuda-sim/ptx_ir.h b/src/cuda-sim/ptx_ir.h
index 9ad1571..36ef3d5 100644
--- a/src/cuda-sim/ptx_ir.h
+++ b/src/cuda-sim/ptx_ir.h
@@ -1245,6 +1245,7 @@ public:
 
    const struct gpgpu_ptx_sim_info* get_kernel_info () const
    {
+      assert (m_kernel_info.maxthreads == maxnt_id);
       return &m_kernel_info;
    }
 
@@ -1252,6 +1253,8 @@ public:
       m_kernel_info = info;
       m_kernel_info.ptx_version = 10*get_ptx_version().ver();
       m_kernel_info.sm_target = get_ptx_version().target();
+      // THIS DEPENDS ON ptxas being called after the PTX is parsed.
+      m_kernel_info.maxthreads = maxnt_id;
    }
    symbol_table *get_symtab()
    {
@@ -1275,7 +1278,11 @@ public:
    }
    bool is_entry_point() const { return m_entry_point; }
 
+   void set_maxnt_id(unsigned maxthreads) { maxnt_id = maxthreads;}
+   unsigned get_maxnt_id() { return maxnt_id;}
+
 private:
+   unsigned maxnt_id;
    unsigned m_uid;
    unsigned m_local_mem_framesize;
    bool m_entry_point;
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index baa3bcd..e5731a8 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -969,6 +969,10 @@ void 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) {
+  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
diff --git a/src/cuda-sim/ptx_parser.h b/src/cuda-sim/ptx_parser.h
index 32f3903..13042e1 100644
--- a/src/cuda-sim/ptx_parser.h
+++ b/src/cuda-sim/ptx_parser.h
@@ -93,6 +93,7 @@ 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();
diff --git a/src/gpgpu-sim/addrdec.cc b/src/gpgpu-sim/addrdec.cc
index 422576d..cfd90ec 100644
--- a/src/gpgpu-sim/addrdec.cc
+++ b/src/gpgpu-sim/addrdec.cc
@@ -62,6 +62,9 @@ void linear_to_raw_address_translation::addrdec_setoption(option_parser_t opp)
    option_parser_register(opp, "-gpgpu_mem_address_mask", OPT_INT32, &gpgpu_mem_address_mask, 
                "0 = old addressing mask, 1 = new addressing mask, 2 = new add. mask + flipped bank sel and chip sel bits",
                "0");
+   option_parser_register(opp, "-memory_partition_indexing", OPT_UINT32, &memory_partition_indexing,
+                  "0 = no indexing, 1 = bitwise xoring, 2 = IPoly, 3 = custom indexing",
+                  "0");
 }
 
 new_addr_type linear_to_raw_address_translation::partition_address( new_addr_type addr ) const 
@@ -103,6 +106,51 @@ void linear_to_raw_address_translation::addrdec_tlx(new_addr_type addr, addrdec_
       tlx->burst= addrdec_packbits(addrdec_mask[BURST], rest_of_addr, addrdec_mkhigh[BURST], addrdec_mklow[BURST]);
    }
 
+    switch(memory_partition_indexing){
+		case CONSECUTIVE:
+			//Do nothing
+			break;
+		case BITWISE_PERMUTATION:
+			assert(!gap);
+			tlx->chip = (tlx->chip) ^ (tlx->row & (m_n_channel-1));
+			assert(tlx->chip < m_n_channel);
+			break;
+		case IPOLY:
+		    /*
+			* Set Indexing function from "Pseudo-randomly interleaved memory."
+			* Rau, B. R et al.
+			* ISCA 1991
+			*
+			* equations are adopted from:
+			* "Sacat: streaming-aware conflict-avoiding thrashing-resistant gpgpu cache management scheme."
+			* Khairy et al.
+			* IEEE TPDS 2017.
+			*/
+			if(m_n_channel == 32) {
+				std::bitset<64> a(tlx->row);
+				std::bitset<5> chip(tlx->chip);
+				chip[0] = a[13]^a[12]^a[11]^a[10]^a[9]^a[6]^a[5]^a[3]^a[0]^chip[0];
+				chip[1] = a[14]^a[13]^a[12]^a[11]^a[10]^a[7]^a[6]^a[4]^a[1]^chip[1];
+				chip[2] = a[14]^a[10]^a[9]^a[8]^a[7]^a[6]^a[3]^a[2]^a[0]^chip[2];
+				chip[3] = a[11]^a[10]^a[9]^a[8]^a[7]^a[4]^a[3]^a[1]^chip[3]; 
+				chip[4] = a[12]^a[11]^a[10]^a[9]^a[8]^a[5]^a[4]^a[2]^chip[4];
+				tlx->chip = chip.to_ulong();
+				
+			}
+			else{ /* Else incorrect number of channels for the hashing function */
+            assert("\nGPGPU-Sim memory_partition_indexing error: The number of channels should be "
+                    "32 for the hashing IPOLY index function.\n" && 0);
+			}
+			assert(tlx->chip < m_n_channel);
+			break;
+		case CUSTOM:
+			/* No custom set function implemented */
+			break;
+		default:
+			 assert("\nUndefined set index function.\n" && 0);
+			 break;
+	}
+
    // combine the chip address and the lower bits of DRAM bank address to form the subpartition ID
    unsigned sub_partition_addr_mask = m_n_sub_partition_in_channel - 1; 
    tlx->sub_partition = tlx->chip * m_n_sub_partition_in_channel
diff --git a/src/gpgpu-sim/addrdec.h b/src/gpgpu-sim/addrdec.h
index fd9af8d..a18ff63 100644
--- a/src/gpgpu-sim/addrdec.h
+++ b/src/gpgpu-sim/addrdec.h
@@ -35,6 +35,13 @@
 
 #include "../abstract_hardware_model.h"
 
+enum partition_index_function{
+	CONSECUTIVE = 0,
+	BITWISE_PERMUTATION,
+	IPOLY,
+    CUSTOM
+};
+
 struct addrdec_t {
    void print( FILE *fp ) const;
     
@@ -72,6 +79,7 @@ private:
 
    const char *addrdec_option;
    int gpgpu_mem_address_mask;
+   partition_index_function memory_partition_indexing;
    bool run_test; 
 
    int ADDR_CHIP_S;
diff --git a/src/gpgpu-sim/dram.cc b/src/gpgpu-sim/dram.cc
index a57508c..92aa819 100644
--- a/src/gpgpu-sim/dram.cc
+++ b/src/gpgpu-sim/dram.cc
@@ -204,7 +204,7 @@ dram_req_t::dram_req_t( class mem_fetch *mf, unsigned banks, unsigned dram_bnk_i
    }
     else if(dram_bnk_indexing_policy == 1) {
     	 int lbank = log2(banks);
-    	 bk  = tlx.bk ^ (((1<<lbank)-1) & tlx.row);
+    	 bk  = tlx.bk ^ (tlx.row & ((1<<lbank)-1));
     }
     else
     	assert(1);
diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc
index 0602e20..75ec00a 100644
--- a/src/gpgpu-sim/gpu-cache.cc
+++ b/src/gpgpu-sim/gpu-cache.cc
@@ -74,7 +74,7 @@ unsigned l1d_cache_config::set_index(new_addr_type addr) const{
         /*
         * Set Indexing function from "A Detailed GPU Cache Model Based on Reuse Distance Theory"
         * Cedric Nugteren et al.
-        * ISCA 2014
+        * HPCA 2014
         */
         if(m_nset == 32 || m_nset == 64){
             // Lower xor value is bits 7-11
@@ -97,6 +97,36 @@ unsigned l1d_cache_config::set_index(new_addr_type addr) const{
         }
         break;
 
+    case HASH_IPOLY_FUNCTION:
+    	/*
+		* Set Indexing function from "Pseudo-randomly interleaved memory."
+		* Rau, B. R et al.
+		* ISCA 1991
+		*
+		* "Sacat: streaming-aware conflict-avoiding thrashing-resistant gpgpu cache management scheme."
+		* Khairy et al.
+		* IEEE TPDS 2017.
+    	*/
+    	if(m_nset == 32 || m_nset == 64){
+		std::bitset<64> a(addr);
+		std::bitset<6> index;
+		index[0] = a[25]^a[24]^a[23]^a[22]^a[21]^a[18]^a[17]^a[15]^a[12]^a[7]; //10
+		index[1] = a[26]^a[25]^a[24]^a[23]^a[22]^a[19]^a[18]^a[16]^a[13]^a[8]; //10
+		index[2] = a[26]^a[22]^a[21]^a[20]^a[19]^a[18]^a[15]^a[14]^a[12]^a[9]; //10
+		index[3] = a[23]^a[22]^a[21]^a[20]^a[19]^a[16]^a[15]^a[13]^a[10]; //9
+		index[4] = a[24]^a[23]^a[22]^a[21]^a[20]^a[17]^a[16]^a[14]^a[11]; //9
+
+		 if(m_nset == 64)
+			 index[5] = a[12];
+
+		set_index = index.to_ulong();
+
+    	}else{ /* Else incorrect number of sets for the hashing function */
+    	            assert("\nGPGPU-Sim cache configuration error: The number of sets should be "
+    	                    "32 or 64 for the hashing set index function.\n" && 0);
+    	 }
+        break;
+
     case CUSTOM_SET_FUNCTION:
         /* No custom set function implemented */
         break;
@@ -104,6 +134,10 @@ unsigned l1d_cache_config::set_index(new_addr_type addr) const{
     case LINEAR_SET_FUNCTION:
         set_index = (addr >> m_line_sz_log2) & (m_nset-1);
         break;
+
+    default:
+    	 assert("\nUndefined set index function.\n" && 0);
+    	 break;
     }
 
     // Linear function selected or custom set index function not implemented
@@ -658,15 +692,13 @@ void cache_stats::print_stats(FILE *fout, const char *cache_name) const{
     std::string m_cache_name = cache_name;
     for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) {
         for (unsigned status = 0; status < NUM_CACHE_REQUEST_STATUS; ++status) {
-            if(m_stats[type][status] > 0){
-                fprintf(fout, "\t%s[%s][%s] = %u\n",
-                    m_cache_name.c_str(),
-                    mem_access_type_str((enum mem_access_type)type),
-                    cache_request_status_str((enum cache_request_status)status),
-                    m_stats[type][status]);
-                if(status != RESERVATION_FAIL)
-                	 total_access[type]+= m_stats[type][status];
-            }
+            fprintf(fout, "\t%s[%s][%s] = %u\n",
+                m_cache_name.c_str(),
+                mem_access_type_str((enum mem_access_type)type),
+                cache_request_status_str((enum cache_request_status)status),
+                m_stats[type][status]);
+            if(status != RESERVATION_FAIL)
+            	 total_access[type]+= m_stats[type][status];
         }
     }
     for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) {
@@ -962,7 +994,7 @@ void baseline_cache::send_read_request(new_addr_type addr, new_addr_type block_a
         m_mshrs.add(mshr_addr,mf);
         m_extra_mf_fields[mf] = extra_mf_fields(mshr_addr,mf->get_addr(),cache_index, mf->get_data_size(), m_config);
         mf->set_data_size( m_config.get_atom_sz() );
-        mf->set_addr( block_addr );
+        mf->set_addr( mshr_addr );
         m_miss_queue.push_back(mf);
         mf->set_status(m_miss_queue_status,time);
         if(!wa)
@@ -1432,7 +1464,7 @@ data_cache::process_tag_probe( bool wr,
             access_status = (this->*m_wr_hit)( addr,
                                       cache_index,
                                       mf, time, events, probe_status );
-        }else if ( probe_status != RESERVATION_FAIL ) {
+        }else if ( (probe_status != RESERVATION_FAIL) || (probe_status == RESERVATION_FAIL && m_config.m_write_alloc_policy == NO_WRITE_ALLOCATE) ) {
             access_status = (this->*m_wr_miss)( addr,
                                        cache_index,
                                        mf, time, events, probe_status );
diff --git a/src/gpgpu-sim/gpu-cache.h b/src/gpgpu-sim/gpu-cache.h
index 0d07878..d2b7757 100644
--- a/src/gpgpu-sim/gpu-cache.h
+++ b/src/gpgpu-sim/gpu-cache.h
@@ -39,7 +39,7 @@
 #include <iostream>
 
 enum cache_block_state {
-    INVALID,
+    INVALID=0,
     RESERVED,
     VALID,
     MODIFIED
@@ -125,6 +125,7 @@ struct cache_block_t {
     virtual unsigned get_modified_size() = 0;
     virtual void set_m_readable(bool readable, mem_access_sector_mask_t sector_mask)=0;
     virtual bool is_readable(mem_access_sector_mask_t sector_mask)=0;
+    virtual void print_status()=0;
     virtual ~cache_block_t() {}
 
 
@@ -144,7 +145,7 @@ struct line_cache_block: public cache_block_t  {
 	        m_set_modified_on_fill = false;
 	        m_readable = true;
 	    }
-	    void allocate( new_addr_type tag, new_addr_type block_addr, unsigned time, mem_access_sector_mask_t sector_mask )
+	    void allocate( new_addr_type tag, new_addr_type block_addr, unsigned time, mem_access_sector_mask_t sector_mask)
 	    {
 	        m_tag=tag;
 	        m_block_addr=block_addr;
@@ -220,6 +221,9 @@ struct line_cache_block: public cache_block_t  {
 		virtual bool is_readable(mem_access_sector_mask_t sector_mask) {
 			return m_readable;
 		}
+		virtual void print_status() {
+			 printf("m_block_addr is %u, status = %u\n", m_block_addr, m_status);
+		}
 
 
 private:
@@ -401,6 +405,11 @@ struct sector_cache_block : public cache_block_t {
 		return modified * SECTOR_SIZE;
 	}
 
+    virtual void print_status() {
+    	 printf("m_block_addr is %u, status = %u %u %u %u\n", m_block_addr, m_status[0], m_status[1], m_status[2], m_status[3]);
+    }
+
+
 private:
     unsigned m_sector_alloc_time[SECTOR_CHUNCK_SIZE];
     unsigned m_last_sector_access_time[SECTOR_CHUNCK_SIZE];
@@ -455,8 +464,10 @@ enum mshr_config_t {
 };
 
 enum set_index_function{
-    FERMI_HASH_SET_FUNCTION = 0,
-    LINEAR_SET_FUNCTION,
+	LINEAR_SET_FUNCTION = 0,
+	BITWISE_XORING_FUNCTION,
+	HASH_IPOLY_FUNCTION,
+	FERMI_HASH_SET_FUNCTION,
     CUSTOM_SET_FUNCTION
 };
 
@@ -578,6 +589,7 @@ public:
 
         switch(sif){
         case 'H': m_set_index_function = FERMI_HASH_SET_FUNCTION; break;
+        case 'P': m_set_index_function = HASH_IPOLY_FUNCTION; break;
         case 'C': m_set_index_function = CUSTOM_SET_FUNCTION; break;
         case 'L': m_set_index_function = LINEAR_SET_FUNCTION; break;
         default: exit_parse_error();
@@ -1030,7 +1042,7 @@ protected:
     std::string m_name;
     cache_config &m_config;
     tag_array*  m_tag_array;
-    mshr_table m_mshrs;
+	mshr_table m_mshrs;
     std::list<mem_fetch*> m_miss_queue;
     enum mem_fetch_status m_miss_queue_status;
     mem_fetch_interface *m_memport;
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 17f1714..c5d4464 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -36,6 +36,7 @@
 
 
 #include "shader.h"
+#include "shader_trace.h"
 #include "dram.h"
 #include "mem_fetch.h"
 
@@ -87,6 +88,14 @@ 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;
 
 /* Clock Domains */
 
@@ -245,7 +254,7 @@ void shader_core_config::reg_options(class OptionParser * opp)
                    "per-shader L1 data cache config "
                    " {<nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>,<mshr>:<N>:<merge>,<mq> | none}",
                    "none" );
-    option_parser_register(opp, "-gpgpu_cache:dl1PreShared", OPT_CSTR, &m_L1D_config.m_config_stringPrefShared,
+    option_parser_register(opp, "-gpgpu_cache:dl1PrefShared", OPT_CSTR, &m_L1D_config.m_config_stringPrefShared,
                    "per-shader L1 data cache config "
                    " {<nsets>:<bsize>:<assoc>,<rep>:<wr>:<alloc>:<wr_alloc>,<mshr>:<N>:<merge>,<mq> | none}",
                    "none" );
@@ -268,6 +277,9 @@ void shader_core_config::reg_options(class OptionParser * opp)
     option_parser_register(opp, "-gpgpu_shader_registers", OPT_UINT32, &gpgpu_shader_registers, 
                  "Number of registers per shader core. Limits number of concurrent CTAs. (default 8192)",
                  "8192");
+    option_parser_register(opp, "-gpgpu_ignore_resources_limitation", OPT_BOOL, &gpgpu_ignore_resources_limitation,
+                 "gpgpu_ignore_resources_limitation (default 0)",
+                 "0");
     option_parser_register(opp, "-gpgpu_shader_cta", OPT_UINT32, &max_cta_per_core, 
                  "Maximum number of concurrent CTAs in shader (default 8)",
                  "8");
@@ -783,6 +795,10 @@ void gpgpu_sim::init()
     gpu_sim_insn = 0;
     last_gpu_sim_insn = 0;
     m_total_cta_launched=0;
+    partiton_reqs_in_parallel = 0;
+    partiton_replys_in_parallel = 0;
+    partiton_reqs_in_parallel_util = 0;
+    gpu_sim_cycle_parition_util = 0;
 
     reinit_clock_domains();
     set_param_gpgpu_num_shaders(m_config.num_shader());
@@ -819,8 +835,16 @@ void gpgpu_sim::update_stats() {
     gpu_tot_sim_cycle += gpu_sim_cycle;
     gpu_tot_sim_insn += gpu_sim_insn;
     gpu_tot_issued_cta += m_total_cta_launched;
+    partiton_reqs_in_parallel_total += partiton_reqs_in_parallel;
+    partiton_replys_in_parallel_total += partiton_replys_in_parallel;
+    partiton_reqs_in_parallel_util_total += partiton_reqs_in_parallel_util;
+    gpu_tot_sim_cycle_parition_util += gpu_sim_cycle_parition_util ;
 
     gpu_sim_cycle = 0;
+    partiton_reqs_in_parallel = 0;
+    partiton_replys_in_parallel = 0;
+    partiton_reqs_in_parallel_util = 0;
+    gpu_sim_cycle_parition_util = 0;
     gpu_sim_insn = 0;
     m_total_cta_launched = 0;
 }
@@ -1004,6 +1028,21 @@ void gpgpu_sim::gpu_print_stat()
    printf("gpu_stall_dramfull = %d\n", gpu_stall_dramfull);
    printf("gpu_stall_icnt2sh    = %d\n", gpu_stall_icnt2sh );
 
+   printf("partiton_reqs_in_parallel = %lld\n", partiton_reqs_in_parallel);
+   printf("partiton_reqs_in_parallel_total    = %lld\n", partiton_reqs_in_parallel_total );
+   printf("partiton_level_parallism = %12.4f\n", (float)partiton_reqs_in_parallel / gpu_sim_cycle);
+   printf("partiton_level_parallism_total  = %12.4f\n", (float)(partiton_reqs_in_parallel+partiton_reqs_in_parallel_total) / (gpu_tot_sim_cycle+gpu_sim_cycle) );
+   printf("partiton_reqs_in_parallel_util = %lld\n", partiton_reqs_in_parallel_util);
+   printf("partiton_reqs_in_parallel_util_total    = %lld\n", partiton_reqs_in_parallel_util_total );
+   printf("gpu_sim_cycle_parition_util = %lld\n", gpu_sim_cycle_parition_util);
+   printf("gpu_tot_sim_cycle_parition_util    = %lld\n", gpu_tot_sim_cycle_parition_util );
+   printf("partiton_level_parallism_util = %12.4f\n", (float)partiton_reqs_in_parallel_util / gpu_sim_cycle_parition_util);
+   printf("partiton_level_parallism_util_total  = %12.4f\n", (float)(partiton_reqs_in_parallel_util+partiton_reqs_in_parallel_util_total) / (gpu_sim_cycle_parition_util+gpu_tot_sim_cycle_parition_util) );
+   printf("partiton_replys_in_parallel = %lld\n", partiton_replys_in_parallel);
+   printf("partiton_replys_in_parallel_total    = %lld\n", partiton_replys_in_parallel_total );
+   printf("L2_BW  = %12.4f GB/Sec\n", ((float)(partiton_replys_in_parallel * 32) / (gpu_sim_cycle * m_config.icnt_period)) / 1000000000);
+   printf("L2_BW_total  = %12.4f GB/Sec\n", ((float)((partiton_replys_in_parallel+partiton_replys_in_parallel_total) * 32) / ((gpu_tot_sim_cycle+gpu_sim_cycle) * m_config.icnt_period)) / 1000000000 );
+
    time_t curr_time;
    time(&curr_time);
    unsigned long long elapsed_time = MAX( curr_time - g_simulation_starttime, 1 );
@@ -1218,8 +1257,8 @@ bool shader_core_ctx::occupy_shader_resource_1block(kernel_info_t & k, bool occu
        m_occupied_regs += (padded_cta_size * ((kernel_info->regs+3)&~3));
        m_occupied_ctas++;
 
-       printf("GPGPU-Sim uArch: Shader %d occupied %d threads, %d shared mem, %d registers, %d ctas\n",
-            m_sid, m_occupied_n_threads, m_occupied_shmem, m_occupied_regs, m_occupied_ctas);  
+      SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Occupied %d threads, %d shared mem, %d registers, %d ctas\n",
+            m_occupied_n_threads, m_occupied_shmem, m_occupied_regs, m_occupied_ctas);  
    }
 
    return true;
@@ -1344,8 +1383,8 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel )
     m_n_active_cta++;
 
     shader_CTA_count_log(m_sid, 1);
-    printf("GPGPU-Sim uArch: core:%3d, cta:%2u, start_tid:%4u, end_tid:%4u, initialized @(%lld,%lld)\n", 
-        m_sid, free_cta_hw_id, start_thread, end_thread, gpu_sim_cycle, gpu_tot_sim_cycle );
+    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 );
 
 }
 
@@ -1409,6 +1448,7 @@ void gpgpu_sim::cycle()
       for (unsigned i=0;i<m_shader_config->n_simt_clusters;i++) 
          m_cluster[i]->icnt_cycle(); 
    }
+    unsigned partiton_replys_in_parallel_per_cycle = 0;
     if (clock_mask & ICNT) {
         // pop from memory controller to interconnect
         for (unsigned i=0;i<m_memory_config->m_n_mem_sub_partition;i++) {
@@ -1421,6 +1461,7 @@ void gpgpu_sim::cycle()
                     mf->set_status(IN_ICNT_TO_SHADER,gpu_sim_cycle+gpu_tot_sim_cycle);
                     ::icnt_push( m_shader_config->mem2device(i), mf->get_tpc(), mf, response_size );
                     m_memory_sub_partition[i]->pop();
+                    partiton_replys_in_parallel_per_cycle++;
                 } else {
                     gpu_stall_icnt2sh++;
                 }
@@ -1429,6 +1470,7 @@ void gpgpu_sim::cycle()
             }
         }
     }
+    partiton_replys_in_parallel += partiton_replys_in_parallel_per_cycle;
 
    if (clock_mask & DRAM) {
       for (unsigned i=0;i<m_memory_config->m_n_mem;i++){
@@ -1441,6 +1483,7 @@ void gpgpu_sim::cycle()
    }
 
    // L2 operations follow L2 clock domain
+   unsigned partiton_reqs_in_parallel_per_cycle = 0;
    if (clock_mask & L2) {
        m_power_stats->pwr_mem_stat->l2_cache_stats[CURRENT_STAT_IDX].clear();
       for (unsigned i=0;i<m_memory_config->m_n_mem_sub_partition;i++) {
@@ -1452,11 +1495,17 @@ void gpgpu_sim::cycle()
           } else {
               mem_fetch* mf = (mem_fetch*) icnt_pop( m_shader_config->mem2device(i) );
               m_memory_sub_partition[i]->push( mf, gpu_sim_cycle + gpu_tot_sim_cycle );
+              partiton_reqs_in_parallel_per_cycle++;
           }
           m_memory_sub_partition[i]->cache_cycle(gpu_sim_cycle+gpu_tot_sim_cycle);
           m_memory_sub_partition[i]->accumulate_L2cache_stats(m_power_stats->pwr_mem_stat->l2_cache_stats[CURRENT_STAT_IDX]);
        }
    }
+   partiton_reqs_in_parallel += partiton_reqs_in_parallel_per_cycle;
+   if(partiton_reqs_in_parallel_per_cycle > 0){
+	   partiton_reqs_in_parallel_util += partiton_reqs_in_parallel_per_cycle;
+	   gpu_sim_cycle_parition_util++;
+   }
 
    if (clock_mask & ICNT) {
       icnt_transfer();
@@ -1543,7 +1592,8 @@ void gpgpu_sim::cycle()
             hrs     = elapsed_time/3600 - 24*days;
             minutes = elapsed_time/60 - 60*(hrs + 24*days);
             sec = elapsed_time - 60*(minutes + 60*(hrs + 24*days));
-            printf("GPGPU-Sim uArch: cycles simulated: %lld  inst.: %lld (ipc=%4.1f) sim_rate=%u (inst/sec) elapsed = %u:%u:%02u:%02u / %s", 
+
+            DPRINTF(LIVENESS, "GPGPU-Sim uArch: cycles simulated: %lld  inst.: %lld (ipc=%4.1f) sim_rate=%u (inst/sec) elapsed = %u:%u:%02u:%02u / %s", 
                    gpu_tot_sim_cycle + gpu_sim_cycle, gpu_tot_sim_insn + gpu_sim_insn, 
                    (double)gpu_sim_insn/(double)gpu_sim_cycle,
                    (unsigned)((gpu_tot_sim_insn+gpu_sim_insn) / elapsed_time),
diff --git a/src/gpgpu-sim/mem_latency_stat.cc b/src/gpgpu-sim/mem_latency_stat.cc
index 35d6d84..c5452b9 100644
--- a/src/gpgpu-sim/mem_latency_stat.cc
+++ b/src/gpgpu-sim/mem_latency_stat.cc
@@ -230,7 +230,9 @@ void memory_stats_t::memlatstat_print( unsigned n_mem, unsigned gpu_mem_n_bk )
       if (num_mfs) {
          printf("averagemflatency = %lld \n", mf_total_lat/num_mfs);
          printf("avg_icnt2mem_latency = %lld \n", tot_icnt2mem_latency/num_mfs);
-         printf("avg_mrq_latency = %lld \n", tot_mrq_latency/tot_mrq_num);
+         if(tot_mrq_num)
+        	 printf("avg_mrq_latency = %lld \n", tot_mrq_latency/tot_mrq_num);
+
          printf("avg_icnt2sh_latency = %lld \n", tot_icnt2sh_latency/num_mfs);
       }
       printf("mrq_lat_table:");
diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index bf482fb..d2f40a1 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -2088,12 +2088,12 @@ void shader_core_ctx::register_cta_thread_exit( unsigned cta_num, kernel_info_t
       m_barriers.deallocate_barrier(cta_num);
       shader_CTA_count_unlog(m_sid, 1);
 
-      printf("GPGPU-Sim uArch: Shader %d finished CTA #%d (%lld,%lld), %u CTAs running\n", m_sid, cta_num, gpu_sim_cycle, gpu_tot_sim_cycle,
-             m_n_active_cta );
+     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);
 
       if( m_n_active_cta == 0 ) {
-          printf("GPGPU-Sim uArch: Shader %u empty (last released kernel %u \'%s\').\n", m_sid, kernel->get_uid(),
-                 kernel->name().c_str() );
+        SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Empty (last released kernel %u \'%s\').\n",
+            kernel->get_uid(), kernel->name().c_str());
           fflush(stdout);
 
           //Shader can only be empty when no more cta are dispatched
@@ -2108,8 +2108,10 @@ void shader_core_ctx::register_cta_thread_exit( unsigned cta_num, kernel_info_t
       kernel->dec_running();
       if( !m_gpu->kernel_more_cta_left(kernel) ) {
           if( !kernel->running() ) {
-              printf("GPGPU-Sim uArch: GPU detected kernel %u \'%s\' finished on shader %u.\n", kernel->get_uid(),
-                kernel->name().c_str(), m_sid );
+              SHADER_DPRINTF(LIVENESS,
+                "GPGPU-Sim uArch: GPU detected kernel %u \'%s\' finished on shader %u.\n", kernel->get_uid(),
+                kernel->name().c_str(), m_sid);
+
               if(m_kernel == kernel)
                 m_kernel = NULL;
               m_gpu->set_kernel_done( kernel );
@@ -2614,6 +2616,10 @@ unsigned int shader_core_config::max_cta( const kernel_info_t &k ) const
     assert( result <= MAX_CTA_PER_SHADER );
     if (result < 1) {
        printf ("GPGPU-Sim uArch: ERROR ** Kernel requires more resources than shader has.\n");
+       if(gpgpu_ignore_resources_limitation) {
+    	   printf ("GPGPU-Sim uArch: gpgpu_ignore_resources_limitation is set, ignore the ERROR!\n");
+    	   return 1;
+       }
        abort();
     }
 
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 5b41c06..ae22eaa 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -1386,6 +1386,7 @@ struct shader_core_config : public core_config
     unsigned gpgpu_num_reg_banks;
     bool gpgpu_reg_bank_use_warp_id;
     bool gpgpu_local_mem_map;
+    bool gpgpu_ignore_resources_limitation;
     
     unsigned max_sp_latency;
     unsigned max_sfu_latency;
diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc
index ad4587a..52e2f5e 100644
--- a/src/gpgpusim_entrypoint.cc
+++ b/src/gpgpusim_entrypoint.cc
@@ -93,8 +93,15 @@ bool g_sim_active = false;
 bool g_sim_done = true;
 bool break_limit = false;
 
+static void termination_callback()
+{
+    printf("GPGPU-Sim: *** exit detected ***\n");
+    fflush(stdout);
+}
+
 void *gpgpu_sim_thread_concurrent(void*)
 {
+    atexit(termination_callback);
     // concurrent kernel execution simulation thread
     do {
         if(g_debug_execution >= 3) {
@@ -156,8 +163,8 @@ void *gpgpu_sim_thread_concurrent(void*)
            printf("GPGPU-Sim: ** STOP simulation thread (no work) **\n");
            fflush(stdout);
         }
-        g_the_gpu->print_stats();
         if(sim_cycles) {
+            g_the_gpu->print_stats();
             g_the_gpu->update_stats();
             print_simulation_time();
         }
@@ -165,10 +172,10 @@ void *gpgpu_sim_thread_concurrent(void*)
         g_sim_active = false;
         pthread_mutex_unlock(&g_sim_lock);
     } while( !g_sim_done );
-    if(g_debug_execution >= 3) {
-       printf("GPGPU-Sim: *** simulation thread exiting ***\n");
-       fflush(stdout);
-    }
+
+    printf("GPGPU-Sim: *** simulation thread exiting ***\n");
+    fflush(stdout);
+
     if(break_limit) {
     	printf("GPGPU-Sim: ** break due to reaching the maximum cycles (or instructions) **\n");
     	exit(1);
diff --git a/src/gpuwattch/makefile b/src/gpuwattch/makefile
index ab718cc..354c9ec 100644
--- a/src/gpuwattch/makefile
+++ b/src/gpuwattch/makefile
@@ -11,10 +11,10 @@ opt: $(TAR).mk obj_opt
 	@$(MAKE) TAG=opt -C . -f $(TAR).mk
 
 obj_dbg:
-	mkdir $@
+	mkdir -p $@
 
 obj_opt:
-	mkdir $@
+	mkdir -p $@
 
 depend:
 	@$(MAKE) TAG=opt -C . -f $(TAR).mk depend
diff --git a/src/trace_streams.tup b/src/trace_streams.tup
index 3455a00..074c7c8 100644
--- a/src/trace_streams.tup
+++ b/src/trace_streams.tup
@@ -31,5 +31,6 @@ TS_TUP_BEGIN( trace_streams_type )
     TS_TUP( MEMORY_PARTITION_UNIT ),
     TS_TUP( MEMORY_SUBPARTITION_UNIT ),
     TS_TUP( INTERCONNECT ),
+    TS_TUP( LIVENESS ),
     TS_TUP( NUM_TRACE_STREAMS )
 TS_TUP_END( trace_streams_type )