ggml-cuda : fix unclosed brace in ggml_cuda_get_best_fattn_kernel (#fattn)

Author: Jcfunk

ggml/src/ggml-cuda/fattn.cu

@@ -397,7 +397,7 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
 #ifndef FLASH_ATTN_AVAILABLE
     GGML_UNUSED(device); GGML_UNUSED(dst);
     return BEST_FATTN_KERNEL_NONE;
-#endif// FLASH_ATTN_AVAILABLE
+#endif // FLASH_ATTN_AVAILABLE
 
     const ggml_tensor * KQV   = dst;
     const ggml_tensor * Q     = dst->src[0];
@@ -598,6 +598,8 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const
 
     // Use the WMMA kernel if possible:
     if (ggml_cuda_should_use_wmma_fattn(cc) && K->ne[1] % FATTN_KQ_STRIDE == 0 && Q->ne[0] != 40 && Q->ne[0] != 72 && Q->ne[0] != 512 && Q->ne[0] != 576 && Q->ne[0] != 640) {
+        return BEST_FATTN_KERNEL_WMMA_F16;
+    }
 
     // If there are no tensor cores available, use the generic tile kernel:
     if (can_use_vector_kernel) {