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

cuBLAS: refactor and optimize f16 mat mul performance (#1259)

Browse source 
* cuBLAS: refactor, convert fp16 to fp32 on device

* cuBLAS: use multiple streams, choose smartly between mul_mat_q and mul_mat_f16

* fix build

* cuBLAS: update block_q5_1
This commit is contained in:
slaren 2023-05-01 18:11:07 +02:00 • committed by GitHub
parent ea3a0ad6b6
commit 58b367c2d7
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 480 additions and 259 deletions
Download patch file Download diff file Expand all files Collapse all files

47
ggml-cuda.h
View file

@ -1,54 +1,19 @@
#include <cublas_v2.h>
#include <cuda_runtime.h>
#include "ggml.h"
#ifdef __cplusplus
extern "C" {
#endif
#define CUDA_CHECK(err) \
do { \
cudaError_t err_ = (err); \
if (err_ != cudaSuccess) { \
fprintf(stderr, "CUDA error %d at %s:%d: %s\n", err_, __FILE__, __LINE__, \
cudaGetErrorString(err_)); \
exit(1); \
} \
} while (0)
#define CUBLAS_CHECK(err) \
do { \
cublasStatus_t err_ = (err); \
if (err_ != CUBLAS_STATUS_SUCCESS) { \
fprintf(stderr, "cuBLAS error %d at %s:%d\n", err_, __FILE__, __LINE__); \
exit(1); \
} \
} while (0)
extern cublasHandle_t g_cublasH;
extern cudaStream_t g_cudaStream;
extern cudaStream_t g_cudaStream2;
extern cudaEvent_t g_cudaEvent;
void ggml_init_cublas(void);
bool ggml_cuda_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
size_t ggml_cuda_mul_mat_get_wsize(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
void ggml_cuda_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst, void * wdata, size_t wsize);
// TODO: export these with GGML_API
void * ggml_cuda_host_malloc(size_t size);
void ggml_cuda_host_free(void * ptr);
void * ggml_cuda_pool_malloc(size_t size, size_t * actual_size);
void ggml_cuda_pool_free(void * ptr, size_t size);
void dequantize_row_q4_0_cuda(const void * vx, float * y, int k, cudaStream_t stream);
void dequantize_row_q4_1_cuda(const void * vx, float * y, int k, cudaStream_t stream);
void dequantize_row_q4_2_cuda(const void * vx, float * y, int k, cudaStream_t stream);
void dequantize_row_q5_0_cuda(const void * vx, float * y, int k, cudaStream_t stream);
void dequantize_row_q5_1_cuda(const void * vx, float * y, int k, cudaStream_t stream);
void dequantize_row_q8_0_cuda(const void * vx, float * y, int k, cudaStream_t stream);
cudaError_t ggml_cuda_h2d_tensor_2d(void * dst, const struct ggml_tensor * src, uint64_t i3, uint64_t i2, cudaStream_t stream);
typedef void (*dequantize_row_q_cuda_t)(const void * x, float * y, int k, cudaStream_t stream);
dequantize_row_q_cuda_t ggml_get_dequantize_row_q_cuda(enum ggml_type type);
#ifdef __cplusplus
}
#endif