HIP: Enable Matrix cores for MMQ Kernels, Enable stream-K for CDNA 3 by deepsek · Pull Request #14624 · ggml-org/llama.cpp

1 min read Original article ↗

and others added 14 commits

May 23, 2025 19:43

@deepsek

@deepsek

@deepsek

JohannesGaessler

blime4 referenced this pull request in blime4/llama.cpp

Feb 5, 2026
…(#14624)

This commit adds support for MFMA instructions to MMQ. CDNA1/GFX908 CDNA2/GFX90a and CDNA3/GFX942 are supported by the MFMA-enabled code path added by this commit. The code path and stream-k is only enabled on CDNA3 for now as it fails to outperform blas in all cases on the other devices.
Blas is currently only consistently outperformed on CDNA3 due to issues in the amd-provided blas libraries.
This commit also improves the awareness of MMQ towards different warp sizes and as a side effect improves the performance of all quant formats besides q4_0 and q4_1, which regress slightly, on GCN gpus.