From 0c17f7d6bb0a5fc92bdf80223130d78ca422ea8f Mon Sep 17 00:00:00 2001
From: uvos <carl@uvos.xyz>
Date: Mon, 28 Sep 2026 13:47:29 +0200
Subject: [PATCH] HIP: fix template skip for DKQ > 256 mfma kernels
 (llama/29559)

---
 ggml/src/ggml-cuda/fattn-mma-f16.cuh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ggml/src/ggml-cuda/fattn-mma-f16.cuh b/ggml/src/ggml-cuda/fattn-mma-f16.cuh
index fa0d347af..083d3228a 100644
--- a/ggml/src/ggml-cuda/fattn-mma-f16.cuh
+++ b/ggml/src/ggml-cuda/fattn-mma-f16.cuh
@@ -1875,7 +1875,7 @@ static __global__ void flash_attn_ext_f16(
 #endif // defined(AMD_WMMA_AVAILABLE)
 
 #if defined(AMD_MFMA_AVAILABLE)
-    if (ncols1*ncols2 < 16 || (DKQ > 256 && ncols1*ncols2 < 64)) {
+    if (ncols1*ncols2 < 16 || (DKQ > 256 && ncols1*ncols2 < 32)) {
         NO_DEVICE_CODE;
         return;
     }