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

promote aux16 into a vector.

Browse source
This commit is contained in:
Julia Longtin 2024-03-23 16:24:11 +00:00
parent 2f0a949ae0
commit 66d26d4914
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml-phi-knc-dot_q5_K_q8_K.c
View file

@ -58,7 +58,7 @@ void ggml_vec_dot_q5_K_q8_K(int n, float * restrict s, size_t bs, const void * r
uint32_t utmp[4]; uint32_t utmp[4];
int8_t aux8[QK_K]; int8_t aux8[QK_K];
// int16_t aux16[16]; // int16_t aux16[16];
int16x16_t aux16; int16x16_t aux16 __attribute__((aligned(64)));
float32x8_t sums __attribute__((aligned(64))); float32x8_t sums __attribute__((aligned(64)));
/* use a vector operation to clear these floats. */ /* use a vector operation to clear these floats. */
@ -87,8 +87,8 @@ void ggml_vec_dot_q5_K_q8_K(int n, float * restrict s, size_t bs, const void * r
for (int j = 0; j < QK_K/16; ++j) { for (int j = 0; j < QK_K/16; ++j) {
const float dl = d * sc[j]; const float dl = d * sc[j];
for (int l = 0; l < 16; ++l) ((int16 *)&aux16)[l] = q8[l] * a[l]; for (int l = 0; l < 16; ++l) ((int16_t *)&aux16)[l] = q8[l] * a[l];
for (int l = 0; l < 8; ++l) ((float *)&sums)[l] += dl * (((int16 *)&aux16)[l] + ((int16 *)&aux16)[8+l]); for (int l = 0; l < 8; ++l) ((float *)&sums)[l] += dl * (((int16_t *)&aux16)[l] + ((int16_t *)&aux16)[8+l]);
q8 += 16; a += 16; q8 += 16; a += 16;
} }
} }