ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-26 17:39:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

iq2_bn(Metal): 710 -> 714 t/s for PP-512

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-10-26 10:54:43 +02:00
parent ca8f9d7e7e
commit 2d0f9b3663
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-metal.metal
View File

@@ -7498,9 +7498,9 @@ void dequantize_iq1_bn(device const block_iq1_bn * xb, short il, thread type4x4
template <typename type4x4>
void dequantize_iq2_bn(device const block_iq2_bn * xb, short il, thread type4x4 & reg) {
// il is in 0...3
constexpr float k_scale[4] = {1.f, 0.25f, 0.0625f, 0.015625f};
constexpr half k_scale[4] = {1.h, 0.25h, 0.0625h, 0.015625h};
constexpr uint8_t k_mask[4] = {0x03, 0x0c, 0x30, 0xc0};
const float d = k_scale[il];
const half d = k_scale[il];
uint8_t mask = k_mask[il];
for (int j = 0; j < 16; ++j) {