Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

metal : dequantize q8_0 using packed types (#27370)

Browse Source
This commit is contained in:
Georgi Gerganov
2026-08-19 14:10:05 +03:00
committed by GitHub
parent ee4c505a4f
commit 8ef78e644f
1 changed files with 5 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-metal/ggml-metal.metal
+5 -7
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> template <typename type4x4>
void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg) { 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; const float d = xb->d;
float4x4 reg_f; float4x4 reg_f;
for (int i = 0; i < 16; i++) { for (int i = 0; i < 4; ++i) {
reg_f[i/4][i%4] = (qs[i + 16*il] * d); reg_f[i] = float4(qs[4*il + i]) * d;
} }
reg = (type4x4) reg_f; 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> template <typename type4>
void dequantize_q8_0_t4(device const block_q8_0 *xb, short il, thread type4 & reg) { 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; const float d = xb->d;
for (int i = 0; i < 4; i++) { reg = (type4) (float4(qs[il]) * d);
reg[i] = (qs[4*(il%4) + i + 16*(il/4)] * d);
}
} }
template <typename type4x4> template <typename type4x4>