[AMD] optimize Kimi K2.5 fused_moe_triton performance by tuning (#19228)

This commit is contained in:
RoyWang
2026-02-27 03:50:13 +08:00
committed by GitHub
parent 288300aafd
commit a1ef8e2cc0
5 changed files with 486 additions and 23 deletions

View File

@@ -0,0 +1,164 @@
{
"1": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"2": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 1,
"num_stages": 2,
"waves_per_eu": 0
},
"4": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 1,
"num_stages": 2,
"waves_per_eu": 0
},
"8": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"16": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"24": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 1,
"num_stages": 2,
"waves_per_eu": 0
},
"32": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"48": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"64": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"96": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"128": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"256": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"512": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"1024": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"1536": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"2048": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"3072": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"4096": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
}
}

View File

@@ -0,0 +1,164 @@
{
"1": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"2": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 1,
"num_stages": 2,
"waves_per_eu": 0
},
"4": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 1,
"num_stages": 2,
"waves_per_eu": 0
},
"8": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"16": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"24": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 1,
"num_stages": 2,
"waves_per_eu": 0
},
"32": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"48": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"64": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"96": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"128": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 8,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"256": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"512": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"1024": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"1536": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"2048": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"3072": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
},
"4096": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 2,
"waves_per_eu": 0
}
}