[ROCM][Perf] Add ROCM_SEGMENTED_ATTN attention-backend works across RDNA3/3.5/4 (#59132)
Signed-off-by: root <jeffaw99@hotmail.com> Signed-off-by: big-yellow-duck <jeffaw99@hotmail.com> Co-authored-by: Tan Pin Siang <1716735+tanpinsiang@users.noreply.github.com> Co-authored-by: vllmellm <190700713+vllmellm@users.noreply.github.com> Co-authored-by: Codex <noreply@openai.com>
This commit is contained in:
co-authored by
Tan Pin Siang
vllmellm
Codex
parent
15c1cc12ac
commit
edfccc9469
@@ -278,6 +278,9 @@ class KernelConfig:
|
||||
enable_flashinfer_autotune: bool = None # type: ignore[assignment]
|
||||
"""If True, run FlashInfer autotuning during kernel warmup."""
|
||||
|
||||
enable_rocm_segmented_attn_autotune: bool = False
|
||||
"""If True, autotune ROCm segmented attention during kernel warmup on RDNA GPUs."""
|
||||
|
||||
# TODO(roberto): Remove after registered CuTeDSL warmups are migrated
|
||||
# to the shared JIT warmup infrastructure.
|
||||
# https://github.com/vllm-project/vllm/pull/47451
|
||||
@@ -406,6 +409,7 @@ class KernelConfig:
|
||||
"enable_cutedsl_warmup",
|
||||
"enable_jit_warmup",
|
||||
"enable_flashinfer_autotune",
|
||||
"enable_rocm_segmented_attn_autotune",
|
||||
"ir_op_priority", # handled separately below
|
||||
}
|
||||
if self.linear_backend_per_quant is None:
|
||||
|
||||
Reference in New Issue
Block a user