char-fork/llama.cpp

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

Jeff Bolzvulkan: support non-contig unary/glu ops (#24215)1a7718b4c

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
1.3 KiB42 linesraw
1#version 450
2
3#include "types.glsl"
4#include "generic_unary_head.glsl"
5
6#define GGML_TRI_TYPE_UPPER_DIAG 0
7#define GGML_TRI_TYPE_UPPER      1
8#define GGML_TRI_TYPE_LOWER_DIAG 2
9#define GGML_TRI_TYPE_LOWER      3
10
11layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
12
13void main() {
14    const uint idx = get_idx();
15
16    if (idx >= p.ne) {
17        return;
18    }
19
20    const uint i03 = fastdiv(idx, p.ne0_012mp, fastdiv_L(p.ne0_Ls, 0));
21    const uint i03_offset = i03 * p.ne02*p.ne01*p.ne00;
22    const uint i02 = fastdiv(idx - i03_offset, p.ne0_01mp, fastdiv_L(p.ne0_Ls, 1));
23    const uint i02_offset = i02*p.ne01*p.ne00;
24    const uint i01 = fastdiv(idx - i03_offset - i02_offset, p.ne0_0mp, fastdiv_L(p.ne0_Ls, 2));
25    const uint i00 = idx - i03_offset - i02_offset - i01*p.ne00;
26
27    int param = floatBitsToInt(p.param1);
28    bool pass = false;
29    switch (param) {
30    case GGML_TRI_TYPE_UPPER_DIAG: pass = i00 >= i01; break;
31    case GGML_TRI_TYPE_UPPER:      pass = i00 >  i01; break;
32    case GGML_TRI_TYPE_LOWER_DIAG: pass = i00 <= i01; break;
33    case GGML_TRI_TYPE_LOWER:      pass = i00 <  i01; break;
34    }
35
36    if (pass) {
37        const float val = float(data_a[get_aoffset() + src0_idx(idx)]);
38        data_d[get_doffset() + dst_idx(idx)] = D_TYPE(val);
39    } else {
40        data_d[get_doffset() + dst_idx(idx)] = D_TYPE(0);
41    }
42}