musa: use fixed warp size (32) in mul_mat_vec_q_cuda

Signed-off-by: Xiaodong Ye <[email protected]>

Author: yeahdongcn

ggml/src/ggml-cuda/mmvq.cu

@@ -219,7 +219,12 @@ static void mul_mat_vec_q_cuda(
     GGML_ASSERT(ncols_y <= MMVQ_MAX_BATCH_SIZE);
 
     const int device = ggml_cuda_get_device();
+#ifndef GGML_USE_MUSA
     const int warp_size = ggml_cuda_info().devices[device].warp_size;
+#else // GGML_USE_MUSA
+    const int warp_size = WARP_SIZE;
+#endif // GGML_USE_MUSA
+
     const mmvq_parameter_table_id table_id = get_device_table_id(ggml_cuda_info().devices[device].cc);
 
     switch (ncols_y) {