Define GGML_CUDA_DMMV_BLOCK_Y if not defined

Author: JohannesGaessler

ggml-cuda.cu

@@ -85,7 +85,11 @@ static_assert(sizeof(block_q8_0) == sizeof(ggml_fp16_t) + QK8_0, "wrong q8_0 blo
 
 #define CUDA_MUL_BLOCK_SIZE 256
 #define CUDA_DEQUANTIZE_BLOCK_SIZE 256
-#define GGML_CUDA_DMMV_BLOCK_X 32 // dmmv = dequantize_mul_mat_vec
+// dmmv = dequantize_mul_mat_vec
+#define GGML_CUDA_DMMV_BLOCK_X 32
+#ifndef GGML_CUDA_DMMV_BLOCK_Y
+#define GGML_CUDA_DMMV_BLOCK_Y 1 // can by set by compiler option LLAMA_CUDA_BY
+#endif
 
 static __global__ void mul_f32(const float * x, const float * y, float * dst, const int kx, const int ky) {
     const int i = blockDim.x*blockIdx.x + threadIdx.x;