mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-07-15 03:30:11 +00:00
cb859854a7
feat: [CK Tile] mxfp8 support for qr async pipeline MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Closes: #6526 ## Motivation Adds MXFP8 support to the QR async FMHA forward pipeline by enabling MX quant scale paths, loosening padding constraints, and updating tests/codegen to cover the new combinations. ## Technical Details Changes: - Extend BlockFmhaPipelineQRKSVSAsync to support MX quant scales (Q/K/V/P scale tiles) and adjust padding/alignment behavior accordingly. - Update FMHA fwd tests to include head-dim adjustment helpers and add a new “General” parameterized test suite. - Update codegen pipeline constraints/selection for qr_async, and adjust tile-window vectorization logic for packed types. ## Test Plan Running CI with extended set of tests ## Test Result CI passing ## Submission Checklist - [ ] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
ck_tile/core
ck_tile/core contains every basic functions and structures to create a GPU kernel using ck_tile. User should only include ck_tile/core.hpp this single header to use all the functionality. Everything is under ck_tile namespace. The coding style under this folder should be similar to std (snake_case for structure/function, Camel for template types...)
algorithm/
coordinate transform and some other reusable algorithm
arch/
contains some basic device building block like mma, buffer addressing, etc...
container/
contains basic container data structure, array/sequence/tuple/...
numeric/
data type, and data type related math
tensor/
tensor descriptors and tile level API
utility/
other utility function for both host/device