diff options
| author | Mahmoud Khairy A. Abdallah <[email protected]> | 2021-05-19 22:20:56 -0400 |
|---|---|---|
| committer | Mahmoud Khairy A. Abdallah <[email protected]> | 2021-05-19 22:20:56 -0400 |
| commit | 7fac247e3e1c4326081c3ea4d46da6c5dc83ccb9 (patch) | |
| tree | 6f45c4e4450fb53e55834a37fcc8ad7c3ca4210b | |
| parent | b466afea67e6d6faf49f01ecfe378257fbdb93af (diff) | |
change L1 cache config in Volta+ to be write-through and write-allocate based on recent ubench
| -rw-r--r-- | configs/tested-cfgs/SM75_RTX2060/gpgpusim.config | 2 | ||||
| -rw-r--r-- | configs/tested-cfgs/SM7_QV100/gpgpusim.config | 2 | ||||
| -rw-r--r-- | configs/tested-cfgs/SM7_TITANV/gpgpusim.config | 2 | ||||
| -rw-r--r-- | configs/tested-cfgs/SM86_RTX3070/gpgpusim.config | 3 |
4 files changed, 4 insertions, 5 deletions
diff --git a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config index a63d50f..f715f3a 100644 --- a/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config +++ b/configs/tested-cfgs/SM75_RTX2060/gpgpusim.config @@ -104,7 +104,7 @@ -gpgpu_unified_l1d_size 96 # L1 cache configuration -gpgpu_l1_banks 4 --gpgpu_cache:dl1 S:4:128:64,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 diff --git a/configs/tested-cfgs/SM7_QV100/gpgpusim.config b/configs/tested-cfgs/SM7_QV100/gpgpusim.config index 47bf1c8..5f22a42 100644 --- a/configs/tested-cfgs/SM7_QV100/gpgpusim.config +++ b/configs/tested-cfgs/SM7_QV100/gpgpusim.config @@ -117,7 +117,7 @@ -gpgpu_unified_l1d_size 128 # L1 cache configuration -gpgpu_l1_banks 4 --gpgpu_cache:dl1 S:4:128:64,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 diff --git a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config index 3db64b3..c44563f 100644 --- a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config +++ b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config @@ -118,7 +118,7 @@ -gpgpu_unified_l1d_size 128 # L1 cache configuration -gpgpu_l1_banks 4 --gpgpu_cache:dl1 S:4:128:64,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 diff --git a/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config b/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config index c70cfe8..02cdb9e 100644 --- a/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config +++ b/configs/tested-cfgs/SM86_RTX3070/gpgpusim.config @@ -101,7 +101,6 @@ ## 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 @@ -111,7 +110,7 @@ -gpgpu_unified_l1d_size 128 # Ampere unified cache has four banks -gpgpu_l1_banks 4 --gpgpu_cache:dl1 S:4:128:64,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 |