[XPU] Add tuned Mamba SSU configs for 2 more B70 shapes (#57565)

Signed-off-by: pmanczak <pawel.manczak@intel.com>
This commit is contained in:
Paweł Mańczak
2026-10-08 16:36:05 +08:00
committed by GitHub
parent 0f112d1800
commit 4cf1307e72
4 changed files with 180 additions and 0 deletions
@@ -0,0 +1,47 @@
{
"triton_version": "3.8.0",
"48": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"96": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"192": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"384": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"768": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"1536": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"3072": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"6144": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"12288": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"24576": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"49152": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
}
}
@@ -0,0 +1,39 @@
{
"triton_version": "3.8.0",
"48": {
"BLOCK_SIZE_M": 4,
"num_warps": 2
},
"384": {
"BLOCK_SIZE_M": 8,
"num_warps": 2
},
"768": {
"BLOCK_SIZE_M": 4,
"num_warps": 2
},
"1536": {
"BLOCK_SIZE_M": 4,
"num_warps": 2
},
"3072": {
"BLOCK_SIZE_M": 4,
"num_warps": 2
},
"6144": {
"BLOCK_SIZE_M": 4,
"num_warps": 2
},
"12288": {
"BLOCK_SIZE_M": 8,
"num_warps": 2
},
"24576": {
"BLOCK_SIZE_M": 8,
"num_warps": 2
},
"49152": {
"BLOCK_SIZE_M": 4,
"num_warps": 2
}
}
@@ -0,0 +1,47 @@
{
"triton_version": "3.8.0",
"80": {
"BLOCK_SIZE_M": 16,
"num_warps": 4
},
"640": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"1280": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"2560": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"5120": {
"BLOCK_SIZE_M": 32,
"num_warps": 8
},
"10240": {
"BLOCK_SIZE_M": 32,
"num_warps": 8
},
"20480": {
"BLOCK_SIZE_M": 32,
"num_warps": 8
},
"40960": {
"BLOCK_SIZE_M": 32,
"num_warps": 8
},
"81920": {
"BLOCK_SIZE_M": 32,
"num_warps": 8
},
"122880": {
"BLOCK_SIZE_M": 32,
"num_warps": 8
},
"163840": {
"BLOCK_SIZE_M": 32,
"num_warps": 8
}
}
@@ -0,0 +1,47 @@
{
"triton_version": "3.8.0",
"80": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"640": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"1280": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"2560": {
"BLOCK_SIZE_M": 8,
"num_warps": 4
},
"5120": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"10240": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"20480": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"40960": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"81920": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"122880": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"163840": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
}
}