From 94be71911faefef98f70aab4e25dd69ded7f71c8 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Johannes=20G=C3=A4=C3=9Fler?= Date: Sat, 6 Dec 2025 13:45:36 +0100 Subject: [PATCH] HIP: fix RDNA3 FP16/BF16 matrix multiplication (llama/17817) --- ggml/src/ggml-cuda/mmf.cu | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/ggml/src/ggml-cuda/mmf.cu b/ggml/src/ggml-cuda/mmf.cu index be2ad1c6..7cf33f0d 100644 --- a/ggml/src/ggml-cuda/mmf.cu +++ b/ggml/src/ggml-cuda/mmf.cu @@ -160,9 +160,9 @@ bool ggml_cuda_should_use_mmf(enum ggml_type type, int cc, int warp_size, const case GGML_TYPE_F32: return ampere_mma_available(cc); case GGML_TYPE_F16: - return volta_mma_available(cc) || turing_mma_available(cc) || amd_wmma_available(cc); + return volta_mma_available(cc) || turing_mma_available(cc) || (amd_wmma_available(cc) && GGML_CUDA_CC_IS_RDNA4(cc)); case GGML_TYPE_BF16: - return ampere_mma_available(cc) || amd_wmma_available(cc); + return ampere_mma_available(cc) || (amd_wmma_available(cc) && GGML_CUDA_CC_IS_RDNA4(cc)); default: return false; }