vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Define GGML_CUDA_DMMV_BLOCK_Y if not defined

Browse source
This commit is contained in:
JohannesGaessler 2023-05-20 19:37:11 +02:00
parent 5d0cf9928b
commit e199938a3a
1 changed files with 5 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml-cuda.cu
View file

@ -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_MUL_BLOCK_SIZE 256
#define CUDA_DEQUANTIZE_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) { 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; const int i = blockDim.x*blockIdx.x + threadIdx.x;