Unverified Commit a781e84e authored by Michael Goin's avatar Michael Goin Committed by GitHub
Browse files

[Perf] Tune configs for triton block fp8 gemm H100/H200 (#23748)


Signed-off-by: default avatarmgoin <mgoin64@gmail.com>
parent 1b7b161a
{ {
"1": { "1": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 32,
"num_warps": 8, "num_warps": 8,
"num_stages": 4 "num_stages": 5
}, },
"2": { "2": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32, "GROUP_SIZE_M": 32,
"num_warps": 4, "num_warps": 4,
"num_stages": 5 "num_stages": 3
}, },
"4": { "4": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64, "GROUP_SIZE_M": 32,
"num_warps": 4, "num_warps": 4,
"num_stages": 5 "num_stages": 4
}, },
"8": { "8": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16, "GROUP_SIZE_M": 64,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 3
}, },
"16": { "16": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32, "GROUP_SIZE_M": 16,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 3
}, },
"24": { "24": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 5
}, },
"32": { "32": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 3
}, },
"48": { "48": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32, "GROUP_SIZE_M": 16,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 4
}, },
...@@ -69,21 +69,21 @@ ...@@ -69,21 +69,21 @@
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16, "GROUP_SIZE_M": 16,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 3
}, },
"96": { "96": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 16,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 5
}, },
"128": { "128": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64, "GROUP_SIZE_M": 32,
"num_warps": 4, "num_warps": 4,
"num_stages": 4 "num_stages": 4
}, },
...@@ -91,7 +91,7 @@ ...@@ -91,7 +91,7 @@
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32, "GROUP_SIZE_M": 16,
"num_warps": 4, "num_warps": 4,
"num_stages": 3 "num_stages": 3
}, },
...@@ -99,13 +99,13 @@ ...@@ -99,13 +99,13 @@
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32, "GROUP_SIZE_M": 16,
"num_warps": 4, "num_warps": 4,
"num_stages": 3 "num_stages": 3
}, },
"1024": { "1024": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32, "GROUP_SIZE_M": 32,
"num_warps": 4, "num_warps": 4,
...@@ -115,7 +115,7 @@ ...@@ -115,7 +115,7 @@
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 32,
"num_warps": 4, "num_warps": 4,
"num_stages": 3 "num_stages": 3
}, },
...@@ -123,7 +123,7 @@ ...@@ -123,7 +123,7 @@
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 16,
"num_warps": 4, "num_warps": 4,
"num_stages": 3 "num_stages": 3
}, },
...@@ -131,15 +131,15 @@ ...@@ -131,15 +131,15 @@
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 64,
"num_warps": 4, "num_warps": 4,
"num_stages": 3 "num_stages": 4
}, },
"4096": { "4096": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 64,
"num_warps": 4, "num_warps": 4,
"num_stages": 3 "num_stages": 3
} }
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment