[ROCm] Optimal MOE Tuning for AMD Radeon Graphics (#3567)

This commit is contained in:
yigex
2025-02-18 09:54:10 +08:00
committed by GitHub
parent 2eab113206
commit ddf39d3fce
2 changed files with 51 additions and 51 deletions

View File

@@ -1,46 +1,46 @@
{
"1": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"2": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"4": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"8": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"16": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
@@ -48,17 +48,17 @@
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"32": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 4,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
@@ -66,8 +66,8 @@
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
@@ -75,8 +75,8 @@
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
@@ -84,77 +84,77 @@
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 4,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"128": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"256": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"512": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"1024": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 8,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"1536": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"num_stages": 2,
"waves_per_eu": 0
},
"2048": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"3072": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"GROUP_SIZE_M": 4,
"num_warps": 4,
"num_stages": 2,
"waves_per_eu": 0
},
"4096": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 4,
"num_warps": 4,