CUDA: fix compilation on CC 6.0 (llama/16091)

diff --git a/src/ggml-cuda/fattn-tile.cu b/src/ggml-cuda/fattn-tile.cu
index a2d9951ea56a729b058682f484d06ce16d9d3646..131a5099a3b07d1d502114abb75cef9aab5a8bfc 100644 (file)
--- a/src/ggml-cuda/fattn-tile.cu
+++ b/src/ggml-cuda/fattn-tile.cu
@@ -35,7 +35,6 @@ static int fattn_tile_get_kq_stride_host(const int D, const int ncols, const int
         switch (D) {
             case 64:
             case 128:
-                return 128;
             case 256:
                 return ncols <= 16 ? 128 : 64;
             default:
@@ -86,7 +85,6 @@ static constexpr __device__ int fattn_tile_get_kq_stride_device(int D, int ncols
     switch (D) {
         case 64:
         case 128:
-            return 128;
         case 256:
             return ncols <= 16 ? 128 : 64;
         default: