char-fork/llama.cpp

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

Jeff Bolzvulkan: fix non-contig rope (#19299)c342c3b93

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
466 B17 linesraw
1#version 450
2
3#include "rope_head.glsl"
4#include "rope_funcs.glsl"
5
6void main() {
7    const uint i0 = 2*gl_GlobalInvocationID.y;
8    const uint row = gl_GlobalInvocationID.x + 32768 * gl_GlobalInvocationID.z;
9    if (row >= pc.nrows) {
10        return;
11    }
12    const uint i3 = row / (pc.ne01*pc.ne02);
13    const uint i2 = (row - i3 * pc.ne01*pc.ne02) / pc.ne01;
14    const uint i1 = (row - i3 * pc.ne01*pc.ne02 - i2 * pc.ne01);
15
16    rope_norm(i0, i1, i2, i3, pc);
17}