[Kernel][XPU] Tensor-descriptor operand loads for Triton W8A8 scaled_mm (#47205)

Signed-off-by: oonyshch <xonyshch@gmail.com>
Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
This commit is contained in:
Lena Onyshchenko
2026-08-10 00:56:11 +00:00
committed by GitHub
co-authored by Kunshang Ji
parent d6941300fc
commit 751f2ccdd3
5 changed files with 122 additions and 9 deletions
@@ -368,6 +368,7 @@ _POSSIBLE_INT8_KERNELS: dict[PlatformEnum, list[type[Int8ScaledMMLinearKernel]]]
HummingInt8ScaledMMLinearKernel,
],
PlatformEnum.ROCM: [AiterInt8ScaledMMLinearKernel, TritonInt8ScaledMMLinearKernel],
PlatformEnum.XPU: [TritonInt8ScaledMMLinearKernel],
}
# in priority/performance order (when available)
@@ -29,9 +29,9 @@ class TritonInt8ScaledMMLinearKernel(CutlassInt8ScaledMMLinearKernel):
def is_supported(
cls, compute_capability: int | None = None
) -> tuple[bool, str | None]:
if current_platform.is_cuda_alike():
if current_platform.is_cuda_alike() or current_platform.is_xpu():
return True, None
return False, "requires ROCm or CUDA."
return False, "requires ROCm, CUDA or XPU."
@classmethod
def can_implement(cls, c: Int8ScaledMMLinearLayerConfig) -> tuple[bool, str | None]: