Merge pull request #16 from mkhairy/dev

Updating config files and code refactoring of Junuri's code

1 files changed, 9 insertions, 6 deletions

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