char-fork/llama.cpp

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

Jeff Bolzvulkan: combine duplicated fastdiv functions, rename the one optimizing small divs (#27526)c9ca51c1f

char/aliceai-t5-moe
978 B43 linesraw
1#extension GL_EXT_shader_16bit_storage : require
2
3#include "utils.glsl"
4
5
6layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
7
8layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
9layout (binding = 1) readonly buffer B {A_TYPE data_b[];};
10layout (binding = 2) writeonly buffer D {D_TYPE data_d[];};
11
12layout (push_constant) uniform parameter
13{
14    uint N;
15    uint ne00;
16    uint ne20;
17    uint mode;
18    float alpha;
19    float limit;
20    uint nb00;
21    uint nb01;
22    uint nb02;
23    uint nb03;
24    uint nb10;
25    uint nb11;
26    uint nb12;
27    uint nb13;
28    uint nb20;
29    uint nb21;
30    uint nb22;
31    uint nb23;
32    uint ne21;
33    uint ne22;
34    uint misalign_offsets;
35    uint ne2_012mp; uint ne2_012L;
36    uint ne2_01mp;  uint ne2_01L;
37    uint ne2_0mp;   uint ne2_0L;
38} p;
39
40uint get_aoffset() { return p.misalign_offsets >> 16; }
41uint get_boffset() { return (p.misalign_offsets >> 8) & 0xFF; }
42uint get_doffset() { return p.misalign_offsets & 0xFF; }
43