[ROCm] better AMD CDNA4 and RDNA4 support for VAE

[Apophis3158]

Better support for AMD latest GPU arches: RDNA4 (gfx1200, gfx1201) and CDNA4 (gfx950), determined based on fp8 support.

Reference: __hip_fp8_e4m3 and __hip_fp8_e5m2 supports in HIP C++ type implementation support table at AMD data types and precision support

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR changes AMD-specific logic to depend on SUPPORT_FP8_OPS. In comfy/model_management.py, pytorch_attention_enabled_vae() now returns False for AMD only when SUPPORT_FP8_OPS is false; if SUPPORT_FP8_OPS is true, it defers to pytorch_attention_enabled(). In comfy/sd.py, VAE_KL_MEM_RATIO in VAE.__init__ is set to 2.73 only when is_amd() is true and SUPPORT_FP8_OPS is false; otherwise it uses the non-AMD value. No public signatures changed.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description check	✅ Passed	The description is directly related to the changeset, explaining the motivation for the changes and referencing AMD documentation about FP8 support.
Title check	✅ Passed	The title accurately reflects the main objective of the PR: adding improved support for AMD GPU architectures CDNA4 and RDNA4 through conditional FP8 operations handling.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Apophis3158]

• Updated inline comments.

These two improvements has already been tested by ROCm users on Windows and Linux long time ago:

• Revert "Disable pytorch attention in VAE for AMD." #8289
• AMD/ROCm - Fix VAE_KL_MEM_RATIO overestimation for modern ROCm #12685

and more.

Most feedback is from gfx120x, so it's better to use SUPPORT_FP8_OPS for the restriction.

ComfyUI/comfy/model_management.py

Lines 409 to 411 in 1de83f9

	if torch_version_numeric >= (2, 7) and rocm_version >= (6, 4):
	if any((a in arch) for a in ["gfx1200", "gfx1201", "gfx950"]): # TODO: more arches, "gfx942" gives error on pytorch nightly 2.10 1013 rocm7.0
	SUPPORT_FP8_OPS = True