llama.cpp/ggml/src/ggml-cuda/mean.cu

#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);
}