CUDA: fix matrix multiplication logic for tests (#6667)

ggml-cuda.cu

@@ -1946,7 +1946,7 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
     } else if (!split && !fp16_performance_good && src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && src1->ne[1] == 1) {
         // KQV single-batch
         ggml_cuda_mul_mat_vec_nc(ctx, src0, src1, dst);
-    } else if (!split && fp16_performance_good && src0->type == GGML_TYPE_F16 && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2]*src1->ne[3] > 1) {
+    } else if (!split && src0->type == GGML_TYPE_F16 && (src1->type == GGML_TYPE_F16 || fp16_performance_good) && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2]*src1->ne[3] > 1) {
         // KQ + KQV multi-batch
         ggml_cuda_mul_mat_batched_cublas(ctx, src0, src1, dst);
     } else if (use_dequantize_mul_mat_vec) {