Skip to content
This repository has been archived by the owner on Oct 11, 2024. It is now read-only.

Commit

Permalink
[Kernel][ROCm][AMD] fused_moe Triton configs v2 for mi300X (vllm-proj…
Browse files Browse the repository at this point in the history 
…ect#5932)
  • Loading branch information
@divakar-amd @robertgshaw2-redhat
divakar-amd authored and robertgshaw2-redhat committed Jul 1, 2024
1 parent 209a147 commit 4d5e0b9
Show file tree
Hide file tree
Showing 4 changed files with 500 additions and 194 deletions.
164 changes: 118 additions & 46 deletions .../model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
View file
Original file line number Diff line number Diff line change
@@ -1,128 +1,200 @@
{
"1": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 1,
"num_stages": 0
"num_warps": 2,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 1
},
"2": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_stages": 0
"num_warps": 2,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"4": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 64,
"num_stages": 1
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"8": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 32,
"num_stages": 1
"GROUP_SIZE_M": 1,
"num_warps": 1,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"16": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 8,
"num_stages": 1
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"24": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 64,
"num_stages": 1
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_warps": 1,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"32": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 8,
"num_stages": 1
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 1
},
"48": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 16,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_stages": 0
"GROUP_SIZE_M": 4,
"num_warps": 2,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"64": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_stages": 0
"GROUP_SIZE_M": 4,
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"96": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"num_stages": 0
"GROUP_SIZE_M": 4,
"num_warps": 4,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"128": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_stages": 0
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 4,
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"256": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 8,
"num_stages": 0
"GROUP_SIZE_M": 4,
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 1
},
"512": {
"BLOCK_SIZE_M": 256,
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 8,
"num_stages": 0
"GROUP_SIZE_M": 4,
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"1024": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_stages": 0
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 32,
"kpack": 2
},
"1536": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_stages": 0
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"2048": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_stages": 0
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 2
},
"3072": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_stages": 0
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 1
},
"4096": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_stages": 0
"num_warps": 8,
"num_stages": 0,
"waves_per_eu": 0,
"matrix_instr_nonkdim": 16,
"kpack": 1
}
}
Loading

0 comments on commit 4d5e0b9

Please sign in to comment.