tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-13 11:57:43 -04:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: remove DMMV, consolidate F16 mult mat vec (#10318)

Browse Source
This commit is contained in:
Johannes Gäßler
2024-11-17 09:09:55 +01:00
committed by GitHub
parent 467576b6cc
commit c3ea58aca4
10 changed files with 246 additions and 1000 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
ggml/src/ggml-cuda/mmv.cuh Normal file
View File

@@ -0,0 +1,12 @@
#include "common.cuh"
// maximum number of src0 rows with which to use mul_mat_vec over cuBLAS if FP16 tensor cores are available
#define MMV_MAX_ROWS 512
void ggml_cuda_mul_mat_vec(ggml_backend_cuda_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst);
void ggml_cuda_op_mul_mat_vec(
ggml_backend_cuda_context & ctx,
const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, const char * src0_dd_i, const float * src1_ddf_i,
const char * src1_ddq_i, float * dst_dd_i, const int64_t row_low, const int64_t row_high, const int64_t src1_ncols,
const int64_t src1_padded_row_size, cudaStream_t stream);