char-fork/llama.cpp

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

Aclyvulkan : incremental shader builds (#16341)e29acf74f

char/aliceai-t5-moe
442 B20 linesraw
1#version 450
2
3#include "dequant_head.glsl"
4
5layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;
6
7layout (binding = 0) readonly buffer A {float data_a[];};
8layout (binding = 1) writeonly buffer D {D_TYPE data_b[];};
9
10void main() {
11    const uint i = gl_GlobalInvocationID.x * 16;
12
13    if (i >= p.nel) {
14        return;
15    }
16
17    [[unroll]] for (uint l = 0; l < 16; l++) {
18        data_b[i + l] = D_TYPE(data_a[i + l]);
19    }
20}