[kernel][perf] support uncontiguous input for rms_norm kernel (#28103)

Signed-off-by: zhuhaoran <zhuhaoran.zhr@alibaba-inc.com>
Signed-off-by: izhuhaoran <izhuhaoran@qq.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
This commit is contained in:
zhrrr
2025-11-20 19:39:09 -08:00
committed by GitHub
co-authored by Wentao Ye
parent 0e741c12e3
commit a982f5b5ea
4 changed files with 77 additions and 33 deletions
+1 -4
View File
@@ -328,10 +328,7 @@ def rotary_embedding(
def rms_norm(
out: torch.Tensor, input: torch.Tensor, weight: torch.Tensor, epsilon: float
) -> None:
# TODO: Remove this contiguous call when the kernel is updated to support non-contiguous input
# If removed, also need to remove contiguous in MatcherRMSNorm
input_contiguous = input.contiguous()
torch.ops._C.rms_norm(out, input_contiguous, weight, epsilon)
torch.ops._C.rms_norm(out, input, weight, epsilon)
def fused_add_rms_norm(