From f545f4df756e6ec981e5e346bcc7e7a0ee2cb643 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Fri, 22 Aug 2025 17:36:29 +0800
Subject: [PATCH] with forced wmma for cu11

---
 ggml/src/ggml-cuda/fattn.cu | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/ggml/src/ggml-cuda/fattn.cu b/ggml/src/ggml-cuda/fattn.cu
index c435bf5a7..3440ed287 100644
--- a/ggml/src/ggml-cuda/fattn.cu
+++ b/ggml/src/ggml-cuda/fattn.cu
@@ -396,9 +396,9 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
         }
 
         //kcpp: use wmma to fix cu11 incoherence
-        // if (fp16_mma_available(cc) && (ggml_cuda_highest_compiled_arch(cc) <= GGML_CUDA_CC_TURING || cc == GGML_CUDA_CC_TURING)) {
-        //     return BEST_FATTN_KERNEL_WMMA_F16;
-        // }
+        if (fp16_mma_available(cc) && (ggml_cuda_highest_compiled_arch(cc) <= GGML_CUDA_CC_TURING || cc == GGML_CUDA_CC_TURING)) {
+            return BEST_FATTN_KERNEL_WMMA_F16;
+        }
         return BEST_FATTN_KERNEL_MMA_F16;
     }