add mxfp8 support

[solinzby1]

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[Copilot]

Pull request overview

Adds MXFP8 (E4M3) support to the existing preshuffle GEMM path by extending the MXFP4 preshuffle kernel to accept FP8 B operands, providing a per-1x32 FP8 quantization helper, and adding an A8W8 correctness/perf test (gfx950-only).

Changes:

• Add per_1x32_f8_quant to generate MXFP8 (E4M3) codes plus per-32 E8M0 scales.
• Extend kernels/gemm/mxfp4_preshuffle.py:launch_gemm with a b_dtype constexpr and FP8-B load/layout handling.
• Add a new A8W8 (MXFP8×MXFP8) preshuffle GEMM test and plumb b_dtype through the test launcher.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
tests/kernels/utils/fp4_utils.py	Adds per-1x32 MXFP8 quantization helper for generating FP8 codes + E8M0 scales.
tests/kernels/test_preshuffle_gemm.py	Adds an A8W8 FP8 MX preshuffle GEMM test and threads b_dtype into the launcher.
kernels/gemm/mxfp4_preshuffle.py	Extends the preshuffle GEMM kernel to support FP8 B (fragment layout, view/strides, load path).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[coderfeli]

LGTM. perf numbers? @solinzby1

[coderfeli]

conflict @solinzby1

[solinzby1]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Merged main into this branch. No conflicts were encountered — the merge resolved automatically. Commit: Merge origin/main into mxfp8_dsl to resolve conflicts.