char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

shaofeiqiopencl: add FILL op support (#18682)568371a72

char/aliceai-t5-moe
465 B17 linesraw
1#pragma OPENCL EXTENSION cl_khr_fp16 : enable
2
3//------------------------------------------------------------------------------
4// fill
5//------------------------------------------------------------------------------
6__kernel void kernel_fill_f32(
7        __global float *dst,
8        ulong offsetd,
9        float v,
10        int n
11
12) {
13    dst = (global float*)((global char*)dst + offsetd);
14    if(get_global_id(0) < n){
15        dst[get_global_id(0)] = v;
16    }
17}