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
1.0 KiB37 linesraw
1#ifndef UTILS_COMP
2#define UTILS_COMP
3
4// mod and div are expensive and coordinates/dimensions are often power of 2 or equal to 1
5uint fastmod(uint a, uint b) {
6    if ((b & (b-1)) == 0) {
7        return a & (b-1);
8    }
9    return a % b;
10}
11
12// see init_fastdiv_values in ggml-vulkan.cpp
13uint fastdiv(uint n, uint mp, uint L) {
14    uint msbs, lsbs;
15    // msbs = mulhi(n, mp)
16    umulExtended(n, mp, msbs, lsbs);
17    return (msbs + n) >> L;
18}
19
20uint fastdiv_L(uint packed, uint slot) {
21    return (packed >> (slot * 8)) & 0x3Fu;
22}
23
24uint fastdiv_small(uint a, uint b) {
25    return (a < b) ? 0 : (a / b);
26}
27
28void get_indices(uint idx, out uint i00, out uint i01, out uint i02, out uint i03, uint ne00, uint ne01, uint ne02, uint ne03) {
29    i03 = fastdiv_small(idx, (ne02*ne01*ne00));
30    const uint i03_offset = i03 * ne02*ne01*ne00;
31    i02 = fastdiv_small((idx - i03_offset), (ne01*ne00));
32    const uint i02_offset = i02*ne01*ne00;
33    i01 = (idx - i03_offset - i02_offset) / ne00;
34    i00 = idx - i03_offset - i02_offset - i01*ne00;
35}
36
37#endif // UTILS_COMP