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
413 B19 linesraw
1
2#include "utils.glsl"
3
4// vk_op_sum_rows_push_constants
5layout (push_constant) uniform parameter
6{
7    uint n_cols;
8    uint ne01, ne02;
9    uint nb01, nb02, nb03;
10    uint nb11, nb12, nb13;
11    float weight;
12    uint misalign_offsets;
13    uint ne0_12mp, ne0_12L;
14    uint ne0_1mp, ne0_1L;
15} p;
16
17uint get_aoffset() { return p.misalign_offsets >> 16; }
18uint get_doffset() { return p.misalign_offsets & 0xFFFF; }
19