vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

kompute : fix basic Q6_K get_rows, 26 -> 24 failures

Browse source
This commit is contained in:
Jared Van Bortel 2024-01-25 15:38:39 -05:00
parent de9fba0d39
commit 445a3734b7
1 changed files with 2 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

7
kompute-shaders/common.comp
View file

@ -73,9 +73,6 @@ struct block_q6_k {
}; };
mat4 dequantize_q6_k(const block_q6_k xb, uint il) { mat4 dequantize_q6_k(const block_q6_k xb, uint il) {
const float16_t d_all = xb.d; const float16_t d_all = xb.d;
uint8_t ql[QK_K/2];
uint8_t qh[QK_K/4];
int8_t scales[QK_K/16];
const uint qlIndex = 64*(il/8) + 32*((il/2)&1) + 16*(il&1); const uint qlIndex = 64*(il/8) + 32*((il/2)&1) + 16*(il&1);
const uint qhIndex = 32*(il/8) + 16*(il&1); const uint qhIndex = 32*(il/8) + 16*(il&1);
@ -89,8 +86,8 @@ mat4 dequantize_q6_k(const block_q6_k xb, uint il) {
const float16_t dl = float16_t(d_all * sc * coef); const float16_t dl = float16_t(d_all * sc * coef);
mat4 reg; mat4 reg;
for (int i = 0; i < 16; ++i) { for (int i = 0; i < 16; ++i) {
const float16_t q = (il&1) != 0 ? ((ql[qlIndex + i] & kmask2) | ((qh[qhIndex + i] & kmask1) << 2)) const float16_t q = (il&1) != 0 ? ((xb.ql[qlIndex + i] & kmask2) | ((xb.qh[qhIndex + i] & kmask1) << 2))
: ((ql[qlIndex + i] & kmask2) | ((qh[qhIndex + i] & kmask1) << 4)); : ((xb.ql[qlIndex + i] & kmask2) | ((xb.qh[qhIndex + i] & kmask1) << 4));
reg[i/4][i%4] = dl * q - ml; reg[i/4][i%4] = dl * q - ml;
} }
return reg; return reg;