Merge pull request #16 from mkhairy/dev

Updating config files and code refactoring of Junuri's code

4 files changed, 41 insertions, 33 deletions

diff --git a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config
index d7573ab..f715f3a 100644
--- a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config
+++ b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config
@@ -99,19 +99,22 @@
 
 # <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
--gpgpu_adaptive_cache_config 0
--gpgpu_shmem_option 0,8,16,32,64,100
--gpgpu_unified_l1d_size 128
+-gpgpu_adaptive_cache_config 1
+-gpgpu_shmem_option 32,64
+-gpgpu_unified_l1d_size 96
+# L1 cache configuration
 -gpgpu_l1_banks 4
--gpgpu_cache:dl1  S:1:128:512,L:L:m:N:L,A:512:8,16:0,32
+-gpgpu_cache:dl1  S:4:128:64,L:T:m:L:L,A:512:8,16:0,32
+-gpgpu_l1_cache_write_ratio 25
+-gpgpu_l1_latency 20
+-gpgpu_gmem_skip_L1D 0
+-gpgpu_n_cluster_ejection_buffer_size 32
+-gpgpu_flush_l1_cache 1
+# shared memory configuration
 -gpgpu_shmem_size 65536
 -gpgpu_shmem_sizeDefault 65536
 -gpgpu_shmem_per_block 65536
--gpgpu_gmem_skip_L1D 0
--gpgpu_n_cluster_ejection_buffer_size 32
--gpgpu_l1_latency 20
 -gpgpu_smem_latency 20
--gpgpu_flush_l1_cache 1
 
 # 64 sets, each 128 bytes 16-way for each memory sub partition (128 KB per memory sub partition). This gives us 3MB L2 cache
 -gpgpu_cache:dl2 S:64:128:16,L:B:m:L:P,A:192:4,32:0,32
diff --git a/configs/tested-cfgs/SM7_QV100/gpgpusim.config b/configs/tested-cfgs/SM7_QV100/gpgpusim.config
index 59c7f43..5f22a42 100644
--- a/configs/tested-cfgs/SM7_QV100/gpgpusim.config
+++ b/configs/tested-cfgs/SM7_QV100/gpgpusim.config
@@ -94,7 +94,7 @@
 -gpgpu_shmem_num_banks 32
 -gpgpu_shmem_limited_broadcast 0
 -gpgpu_shmem_warp_parts 1
--gpgpu_coalesce_arch 60
+-gpgpu_coalesce_arch 70
 
 # Volta has four schedulers per core
 -gpgpu_num_sched_per_core 4
@@ -113,20 +113,21 @@
 # For more info, see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory-7-x 
 # disable this mode in case of multi kernels/apps execution
 -gpgpu_adaptive_cache_config 1
-# Volta unified cache has four banks
+-gpgpu_shmem_option 0,8,16,32,64,96
+-gpgpu_unified_l1d_size 128
+# L1 cache configuration
 -gpgpu_l1_banks 4
--gpgpu_cache:dl1  S:1:128:256,L:L:m:N:L,A:512:8,16:0,32
+-gpgpu_cache:dl1  S:4:128:64,L:T:m:L:L,A:512:8,16:0,32
+-gpgpu_l1_cache_write_ratio 25
+-gpgpu_l1_latency 20
+-gpgpu_gmem_skip_L1D 0
+-gpgpu_flush_l1_cache 1
+-gpgpu_n_cluster_ejection_buffer_size 32
+# shared memory configuration
 -gpgpu_shmem_size 98304
 -gpgpu_shmem_sizeDefault 98304
 -gpgpu_shmem_per_block 65536
--gpgpu_gmem_skip_L1D 0
--gpgpu_n_cluster_ejection_buffer_size 32
--gpgpu_l1_latency 20
 -gpgpu_smem_latency 20
--gpgpu_flush_l1_cache 1
--gpgpu_l1_cache_write_ratio 25
--gpgpu_shmem_option 0,12,24,48,96
--gpgpu_unified_l1d_size 128
 
 # 32 sets, each 128 bytes 24-way for each memory sub partition (96 KB per memory sub partition). This gives us 6MB L2 cache
 -gpgpu_cache:dl2 S:32:128:24,L:B:m:L:P,A:192:4,32:0,32
diff --git a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config
index 32245d7..c44563f 100644
--- a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config
+++ b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config
@@ -114,20 +114,21 @@
 # For more info, see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory-7-x 
 # disable this mode in case of multi kernels/apps execution
 -gpgpu_adaptive_cache_config 1
-# Volta unified cache has four banks
+-gpgpu_shmem_option 0,8,16,32,64,96
+-gpgpu_unified_l1d_size 128
+# L1 cache configuration
 -gpgpu_l1_banks 4
--gpgpu_cache:dl1  S:1:128:256,L:L:m:N:L,A:512:8,16:0,32
+-gpgpu_cache:dl1  S:4:128:64,L:T:m:L:L,A:512:8,16:0,32
+-gpgpu_l1_cache_write_ratio 25
+-gpgpu_gmem_skip_L1D 0
+-gpgpu_l1_latency 20
+-gpgpu_flush_l1_cache 1
+-gpgpu_n_cluster_ejection_buffer_size 32
+# shared memory configuration
 -gpgpu_shmem_size 98304
 -gpgpu_shmem_sizeDefault 98304
 -gpgpu_shmem_per_block 65536
--gpgpu_gmem_skip_L1D 0
--gpgpu_n_cluster_ejection_buffer_size 32
--gpgpu_l1_latency 20
 -gpgpu_smem_latency 20
--gpgpu_flush_l1_cache 1
--gpgpu_l1_cache_write_ratio 25
--gpgpu_shmem_option 0,12,24,48,96
--gpgpu_unified_l1d_size 128
 
 # 32 sets, each 128 bytes 24-way for each memory sub partition (96 KB per memory sub partition). This gives us 4.5MB L2 cache
 -gpgpu_cache:dl2 S:32:128:24,L:B:m:L:P,A:192:4,32:0,32
diff --git a/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config b/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config
index f5418ad..02cdb9e 100644
--- a/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config
+++ b/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config
@@ -101,23 +101,26 @@
 ## L1/shared memory configuration
 # <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
-# Default config is 28KB DL1 and 100KB shared memory
 # In Ampere, we assign the remaining shared memory to L1 cache 
 # if the assigned shd mem = 0, then L1 cache = 128KB
 # For more info, see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#global-memory-8-x
 # disable this mode in case of multi kernels/apps execution
 -gpgpu_adaptive_cache_config 1
+-gpgpu_shmem_option 0,8,16,32,64,100
+-gpgpu_unified_l1d_size 128
 # Ampere unified cache has four banks
 -gpgpu_l1_banks 4
--gpgpu_cache:dl1  S:1:128:256,L:L:m:N:L,A:512:8,16:0,32
+-gpgpu_cache:dl1  S:4:128:64,L:T:m:L:L,A:512:8,16:0,32
+-gpgpu_l1_cache_write_ratio 25
+-gpgpu_gmem_skip_L1D 0
+-gpgpu_l1_latency 20
+-gpgpu_n_cluster_ejection_buffer_size 32
+-gpgpu_flush_l1_cache 1
+# shared memory configuration
 -gpgpu_shmem_size 102400
 -gpgpu_shmem_sizeDefault 102400
 -gpgpu_shmem_per_block 102400
--gpgpu_gmem_skip_L1D 0
--gpgpu_n_cluster_ejection_buffer_size 32
--gpgpu_l1_latency 20
 -gpgpu_smem_latency 20
--gpgpu_flush_l1_cache 1
 
 # 32 sets, each 128 bytes 24-way for each memory sub partition (96 KB per memory sub partition). This gives us 3MB L2 cache
 -gpgpu_cache:dl2 S:32:128:24,L:B:m:L:P,A:192:4,32:0,32