CUDA: stream-k decomposition for MMQ (#8018)

* CUDA: stream-k decomposition for MMQ

* fix undefined memory reads for small matrices

1 files changed, 1 insertions, 1 deletions

diff --git a/ggml-cuda.cu b/ggml-cuda.cu
index b8298ab2..f914efd7 100644
--- a/ggml-cuda.cu
+++ b/ggml-cuda.cu
@@ -635,7 +635,7 @@ static int64_t get_row_rounding(const std::array<float, GGML_CUDA_MAX_DEVICES> &
         }
 
         const int cc = ggml_cuda_info().devices[id].cc;
-        row_rounding = std::max(row_rounding, (int64_t)get_mmq_y_host(cc, get_mmq_x_max_host(cc)));
+        row_rounding = std::max(row_rounding, (int64_t)get_mmq_y_host(cc));
     }
     return row_rounding;
 }