char-fork/llama.cpp

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

Giuseppe Scrivanovulkan: implement ADD1, ARANGE, FILL, SOFTPLUS, STEP, ROUND, CEIL, FLOOR, TRUNC (#17319)7d77f0732

char/aliceai-t5-moe
702 B28 linesraw
1#version 450
2
3#extension GL_EXT_shader_16bit_storage : require
4
5#include "types.glsl"
6#include "generic_binary_head.glsl"
7
8const uint num_threads = 256;
9
10layout(local_size_x = num_threads, local_size_y = 1, local_size_z = 1) in;
11
12void main() {
13    uint idx = get_idx();
14
15    const uint num_iter = 2;
16
17    [[unroll]] for (uint i = 0; i < num_iter; ++i) {
18        if (idx >= p.ne) {
19            continue;
20        }
21        uint i00, i01, i02, i03;
22        get_indices(idx, i00, i01, i02, i03);
23
24        data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]) + FLOAT_TYPE(data_b[get_boffset()]));
25
26        idx += num_threads;
27    }
28}