diff options
| author | Andrew M. B. Boktor <[email protected]> | 2012-02-22 12:26:47 -0800 |
|---|---|---|
| committer | Andrew Boktor <[email protected]> | 2014-08-14 13:19:03 -0700 |
| commit | 4f2a7a85b94ea6ba2839c022d41369ac7062d050 (patch) | |
| tree | 33b7e9cf4d72d08ec28a09caedd3aadae6d98850 | |
| parent | 1986cc83e77a5fab56d43922bc8375723d7427a2 (diff) | |
Cleaned up the comments I had in the configuration file
[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 11493]
| -rw-r--r-- | configs/Fermi/gpgpusim.config | 16 |
1 files changed, 3 insertions, 13 deletions
diff --git a/configs/Fermi/gpgpusim.config b/configs/Fermi/gpgpusim.config index accc49a..78df079 100644 --- a/configs/Fermi/gpgpusim.config +++ b/configs/Fermi/gpgpusim.config @@ -8,13 +8,12 @@ -gpgpu_n_cores_per_cluster 1 -gpgpu_n_mem 6 -# Fermi +# Fermi clock domains -gpgpu_clock_domains 700.0:700.0:700.0:3696.0 # shader core pipeline config -gpgpu_shader_registers 32768 -# ask wilson about the next two lines -gpgpu_shader_core_pipeline 1024:32:32 -gpgpu_shader_cta 16 -gpgpu_simd_model 1 @@ -22,24 +21,20 @@ # In Fermi, the cache and shared memory could be configured to 16kb:48kb or 48kb:16kb # <nsets>:<bsize>:<assoc>:<rep>:<wr>:<alloc>,<mshr>:<N>:<merge>,<mq> -# mysteriously gives much better correlation (and faster runs) than a configured cache -#-gpgpu_cache:dl1 none -gpgpu_cache:dl1 32:128:4:L:R:f,A:32:8,8 -gpgpu_shmem_size 49152 + +# The alternative configuration for fermi in case more DL1 Cache is selected #-gpgpu_cache:dl1 64:128:6:L:R:f,A:32:8,8 #-gpgpu_shmem_size 16384 # 256 sets, each 512 bits 6-way, this gives 786KB L2 cache -# this entry gives slightly worse correlation that the quadro config -gpgpu_cache:dl2 256:512:6:L:R:m,A:32:4,4 -#-gpgpu_cache:dl2 64:32:8:L:R:m,A:16:4,4 #-gpgpu_cache:dl2_texture_only -gpgpu_cache:il1 4:128:4:L:R:f,A:2:32,4 -gpgpu_tex_cache:l1 4:128:24:L:R:m,F:128:4,128:2 -gpgpu_const_cache:l1 64:64:2:L:R:f,A:2:32,4 -#-gpgpu_tex_cache:l1 none -#-gpgpu_const_cache:l1 none -gpgpu_num_reg_banks 16 @@ -57,7 +52,6 @@ -gpgpu_n_mem_per_ctrlr 2 # for Fermi, bus width is 384bits, this is 24 bytes *2 because of double data rate -# try if making the buswidth smaller affects the correlation, should try the memory frequency too -gpgpu_dram_buswidth 24 -gpgpu_dram_burst_length 4 -gpgpu_mem_address_mask 1 @@ -66,10 +60,6 @@ # ANDREW # GDDR5 timing from hynix H5GQ1H24AFR # {nbk:tCCD:tRRD:tRCD:tRAS:tRP:tRC:CL:WL:tCDLR:tWR} -# CL and WL need to be figured out accurately -# in the GDDR5 datasheet WL= min 3, max 7 -# I couldn't find tCDLR in the datasheet either, what does it mean? ask wilson may be -# This configuration turns out to decrease the cpi correlation by about .03% for Fermi probably increases clk correlation -gpgpu_dram_timing_opt 16:2:5:12:28:12:35:10:7:6:12 # GDDR3 |