mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-04-19 22:39:03 +00:00
[rocm-libraries] ROCm/rocm-libraries#5088 (commit 36ca523)
[CK_TILE] Update gfx11 FMHA forward kernel configs ## Motivation Tune gfx11 FMHA codegen to recover performance for mainly PSSK (padded seqlen_q/k) cases. This tuning is based on heuristic search and improves performance in most tested shapes. Performance should be evaluated on top of [`ROCm/rocm-libraries#5018`](https://github.com/ROCm/rocm-libraries/pull/5018) (required baseline). ## Technical Details - Updated gfx11 codegen heuristic choices for tile size and occupancy. - Updated gfx11 pipeline selection: - Disabled the `npad` (`f,f,f,f`) qr entry because it was consistently slower than the `pssk` (`t,t,f,f`) path, and kept `pssk` enabled so npad cases are dispatched to the faster kernel path.` - Kept gfx12 unchanged: with PSSK support from [`ROCm/rocm-libraries#4957`](https://github.com/ROCm/rocm-libraries/pull/4957), existing gfx12 config is already sufficient. - Tuning rationale: - In some cases, higher `kBlockPerCu` lowers register pressure. - On RDNA, this generally aligns with better performance when `waves_per_eu >= 6`. ## Test Plan - test_ck_tile_fmha - tile_example_fmha_fwd: tested this on gfx1100 and gfx1151 ./build/bin/tile_example_fmha_fwd -prec=bf16 -mode={0/1} -b=1 -h=24 -d=128 -s={seqlen} -s_k={seqlen} -lse=0 -iperm={0/1} -operm={0/1} ## Test Result - TFLOPs by sequence length target: `gfx1100` layout: `bhsd` - mode: batch / VGPR usage: 225 vs 214 SeqLen | Baseline | Tuned | Gain -- | -- | -- | -- 1024 | 74.10 | 71.97 | 0.97x 4096 | 66.26 | 77.79 | 1.17x 8192 | 68.18 | 75.88 | 1.11x 12288 | 68.47 | 80.44 | 1.17x 16384 | 59.54 | 79.66 | 1.34x 20480 | 55.78 | 77.91 | 1.40x 24576 | 55.08 | 77.47 | 1.41x 27280 | 47.45 | 77.16 | 1.63x - mode: group / VGPR usage: 256 vs 214 SeqLen | Baseline | Tuned | Gain -- | -- | -- | -- 1024 | 71.47 | 70.6 | 0.99x 4096 | 64.74 | 77.06 | 1.19x 8192 | 64.68 | 75.47 | 1.17x 12288 | 66.43 | 79.95 | 1.20x 16384 | 56.02 | 79.73 | 1.42x 20480 | 50.21 | 78.15 | 1.56x 24576 | 47.29 | 77.53 | 1.64x 27280 | 46.13 | 77.04 | 1.67x - TFLOPs by sequence length target: `gfx1151` layout: `bshd` - mode: batch / VGPR usage: 225 vs 223 Batch | Baseline | Tuned | Gain -- | -- | -- | -- 1024 | 26.85 | 29.17 | 1.09x 4096 | 24.75 | 26.01 | 1.05x 8192 | 25.24 | 25.50 | 1.01x 12288 | 25.18 | 25.00 | 0.99x 16384 | 24.79 | 25.91 | 1.05x 20480 | 25.56 | 25.24 | 0.99x 24576 | 25.13 | 26.20 | 1.04x 27280 | 10.78 | 26.35 | 2.44x - mode: group / VGPR usage: 256 vs 229 Batch | Baseline | Tuned | Gain -- | -- | -- | -- 1024 | 27.44 | 26.71 | 0.97x 4096 | 21.89 | 23.09 | 1.05x 8192 | 22.85 | 24.49 | 1.07x 12288 | 24.33 | 24.42 | 1.00x 16384 | 20.05 | 24.98 | 1.24x 20480 | 14.70 | 25.15 | 1.71x 24576 | 11.30 | 26.31 | 2.33x 27280 | 10.10 | 26.32 | 2.61x ## Submission Checklist - [ ] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
This commit is contained in:
Hosang
committed by
assistant-librarian[bot]
assistant-librarian[bot]
parent
846bcacab4
commit
c800f88911
@@ -1095,7 +1095,10 @@ class KernelComponentFactoryGfx950(
|
||||
|
||||
|
||||
class KernelComponentFactoryGfx11(CompatibilityRuleFactory):
|
||||
arch = ArchTrait("gfx11")
|
||||
arch = ArchTrait(
|
||||
"gfx11",
|
||||
preprocessor_check="defined(__gfx11__) && !defined(__gfx115__)",
|
||||
)
|
||||
|
||||
_DT_FP16_BF16 = ("fp16", "bf16")
|
||||
|
||||
@@ -1109,10 +1112,12 @@ class KernelComponentFactoryGfx11(CompatibilityRuleFactory):
|
||||
return {
|
||||
# bm0, bn0, bk0, bn1, bk1,
|
||||
( 32, 32) : [FmhaFwdTileSize( 64, 64, 16, 32, 32, 32, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1)],
|
||||
( 64, 64) : [FmhaFwdTileSize( 64, 64, 32, 64, 32, 64, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1)],
|
||||
(128, 128) : [FmhaFwdTileSize( 64, 64, 32, 128, 32, 128, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1)],
|
||||
( 64, 64) : [FmhaFwdTileSize( 64, 64, 32, 64, 32, 64, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1, CppConstraint("a.max_seqlen_q < 4096")),
|
||||
FmhaFwdTileSize(128, 64, 32, 64, 32, 64, 8, 1, 1, 8, 1, 1, 16, 16, 16, 16, 16, 16, -1)],
|
||||
(128, 128) : [FmhaFwdTileSize( 64, 64, 32, 128, 32, 128, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1, CppConstraint("a.max_seqlen_q < 4096")),
|
||||
FmhaFwdTileSize(128, 64, 32, 128, 32, 128, 8, 1, 1, 8, 1, 1, 16, 16, 16, 16, 16, 16, 6)],
|
||||
(192, 128) : [FmhaFwdTileSize( 64, 64, 32, 128, 32, 256, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1)],
|
||||
(256, 256) : [FmhaFwdTileSize( 64, 64, 32, 256, 32, 256, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1)],
|
||||
(256, 256) : [FmhaFwdTileSize(128, 64, 32, 256, 32, 256, 8, 1, 1, 8, 1, 1, 16, 16, 16, 16, 16, 16, 6)]
|
||||
} # fmt: skip
|
||||
else:
|
||||
raise ValueError(f"unsupported dtype={dtype}")
|
||||
@@ -1133,12 +1138,25 @@ class KernelComponentFactoryGfx11(CompatibilityRuleFactory):
|
||||
["t", "f"],
|
||||
["t", "f"],
|
||||
):
|
||||
pipelines.append(FmhaFwdPipeline("qr", "row", "f", "f", "f", "f", logits, bias, lse, dropout, qscale, mask, skip, "f", sink)) # fmt: skip
|
||||
# Keep only ttff/tttt for gfx11: ffff path is often similar or worse
|
||||
# pipelines.append(FmhaFwdPipeline("qr", "row", "f", "f", "f", "f", logits, bias, lse, dropout, qscale, mask, skip, "f", sink)) # fmt: skip
|
||||
pipelines.append(FmhaFwdPipeline("qr", "row", "t", "t", "f", "f", logits, bias, lse, dropout, qscale, mask, skip, "f", sink)) # fmt: skip
|
||||
pipelines.append(FmhaFwdPipeline("qr", "row", "t", "t", "t", "t", logits, bias, lse, dropout, qscale, mask, skip, "f", sink)) # fmt: skip
|
||||
return pipelines
|
||||
|
||||
|
||||
class KernelComponentFactoryGfx115(KernelComponentFactoryGfx11):
|
||||
arch = ArchTrait("gfx115")
|
||||
|
||||
@classmethod
|
||||
def get_hdim_tile_size_dict(cls, dtype: str) -> Optional[dict]:
|
||||
result = super().get_hdim_tile_size_dict(dtype)
|
||||
if dtype in cls._DT_FP16_BF16:
|
||||
result[(64, 64)] = [FmhaFwdTileSize( 64, 64, 32, 64, 32, 64, 4, 1, 1, 4, 1, 1, 16, 16, 16, 16, 16, 16, -1)] # fmt: skip
|
||||
result[(256, 256)] = [FmhaFwdTileSize(128, 64, 32, 256, 32, 256, 8, 1, 1, 8, 1, 1, 16, 16, 16, 16, 16, 16, -1)] # fmt: skip
|
||||
return result
|
||||
|
||||
|
||||
class KernelComponentFactoryGfx12(CompatibilityRuleFactory):
|
||||
arch = ArchTrait("gfx12")
|
||||
|
||||
@@ -1230,6 +1248,8 @@ def get_factory(target: str):
|
||||
if target.startswith("gfx9"):
|
||||
return KernelComponentFactoryGfx9
|
||||
|
||||
if target.startswith("gfx115"):
|
||||
return KernelComponentFactoryGfx115
|
||||
if target.startswith("gfx11"):
|
||||
return KernelComponentFactoryGfx11
|
||||
if target.startswith("gfx12"):
|
||||
|
||||
Reference in New Issue
Block a user