[ILikeIneine]

Purpose

I adjust the supports_fp8() in each platform according to the code semantics for all backend in TritonExperts. This make the logic more general in TritonExperts.

Need further check by maintainers of rocm and xpu.

Test Plan

Test Result

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[gemini-code-assist]

Code Review

This pull request refactors FP8 support detection by centralizing the logic within platform-specific classes. Review feedback highlights that the ROCm implementation introduces a regression in detection accuracy compared to the previous prefix-based check, and the XPU implementation is too broad, recommending a more specific check for data center GPUs.

[gemini-code-assist]

The change to on_gfx9() or on_gfx12x() is a regression in accuracy for FP8 support detection. on_gfx9() (defined on line 189) includes gfx90a (MI200 series), which does not have native hardware support for FP8 compute. Additionally, on_gfx9() as currently defined only checks for gfx942 and gfx950, missing other GFX94x variants like gfx940 (MI300A) which were covered by the previous prefix-based check. It is recommended to revert to the more accurate explicit check.

Suggested change

	return on_gfx9() or on_gfx12x()
	return any(gfx in _GCN_ARCH for gfx in ["gfx94", "gfx95", "gfx12"])

[gemini-code-assist]

Hardcoding supports_fp8 to True is too broad for the XPU platform. This includes GPUs like Intel Arc (DG2/Alchemist) which do not support FP8. While the previous logic in fused_moe.py was also broad, moving this to the platform class is an opportunity to provide a more accurate check. Since get_device_capability is not implemented for XPU, checking if the device is a Data Center GPU (e.g., PVC) is a better approximation for FP8 support.

Suggested change

	)
	return "vllm.distributed.device_communicators.xpu_communicator.XpuCommunicator" # noqa
	def supports_fp8(cls) -> bool:
	return cls.is_data_center_gpu()

[xuechendi]

LGTM

[ILikeIneine]

xpu seems ok, Could anyone review the rocm platform file?

cc: @mgoin @tjtanaa

[tjtanaa]

@xinyu-intel @xuechendi can you help to take alook if this logic for XPU is correct? Can we assume all XPU supports fp8?

[xinyu-intel]

I think the logic is align with the original:

        device_supports_fp8 = (
            is_rocm_on_gfx9
            or is_rocm_on_gfx12x
            or (p.is_cuda() and p.has_device_capability((8, 9)))
            or p.is_xpu()
        )

[ILikeIneine]

@xinyu-intel Hi, the intel ci failed on TestTritonTopkTopp. Seems that it's not introduced by this pr?

[jikunshang]

intel ci doesn't gate pr merge now. you can ignore failure.

[ILikeIneine]

@tjtanaa @mgoin Hi, everything seems ok. Could you approve these file and trigger auto merge?

[NickLucche]

It looks like supports_fp8 was only used in tests amd side.
If tests are good amd side we're good to merge here even without amd stamp