ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

CUDA: add mean operation (#14313)

Browse source 
* CUDA: add mean operation

* add back sum_rows_f32_cuda

* Review: early exit if col!=0
This commit is contained in:
Aman Gupta 2025-06-22 12:39:54 +08:00 • committed by GitHub
parent aa0ef5c578
commit aa064b2eb7
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 54 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

19
ggml/src/ggml-cuda/mean.cu Normal file
View file

@ -0,0 +1,19 @@
#include "mean.cuh"
void ggml_cuda_op_mean(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
const ggml_tensor * src0 = dst->src[0];
const float * src0_d = (const float *) src0->data;
float * dst_d = (float *) dst->data;
cudaStream_t stream = ctx.stream();
GGML_ASSERT(src0->type == GGML_TYPE_F32);
GGML_ASSERT(dst->type == GGML_TYPE_F32);
GGML_ASSERT(ggml_is_contiguous(src0));
const int64_t ncols = src0->ne[0];
const int64_t nrows = ggml_nrows(src0);
const dim3 block_dims(WARP_SIZE, 1, 1);
const dim3 block_nums(nrows, 1, 1);
reduce_rows_f32</*norm*/ true><<<block_nums, block_dims, 0, stream>>>(src0_d, dst_d, ncols);
}