public_git
/
whisper.cpp
mirror of https://github.com/ggml-org/whisper.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

metal : dequantize q8_0 using packed types (llama/27370)

This commit is contained in:
Georgi Gerganov 2026-08-19 14:10:05 +03:00
parent 1c882a8ed0
commit d830bd220c
1 changed files with 5 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
ggml/src/ggml-metal/ggml-metal.metal
View File

@ -656,13 +656,13 @@ void dequantize_q5_1_t4(device const block_q5_1 * xb, short il, thread type4 & r
template <typename type4x4>
void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg) {
device const int8_t * qs = ((device const int8_t *)xb->qs);
device const packed_char4 * qs = (device const packed_char4 *) xb->qs;
const float d = xb->d;
float4x4 reg_f;
for (int i = 0; i < 16; i++) {
reg_f[i/4][i%4] = (qs[i + 16*il] * d);
for (int i = 0; i < 4; ++i) {
reg_f[i] = float4(qs[4*il + i]) * d;
}
reg = (type4x4) reg_f;
@ -670,12 +670,10 @@ void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg
template <typename type4>
void dequantize_q8_0_t4(device const block_q8_0 *xb, short il, thread type4 & reg) {
device const int8_t * qs = ((device const int8_t *)xb->qs);
device const packed_char4 * qs = (device const packed_char4 *) xb->qs;
const float d = xb->d;
for (int i = 0; i < 4; i++) {
reg[i] = (qs[4*(il%4) + i + 16*(il/4)] * d);
}
reg = (type4) (float4(qs[il]) * d);
}
template <typename type4x4>