haraldwolff
/
whisper.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ggml : add ggml_vdotq_s32 alias (llama/4715) 

ggml-ci
pull/1725/head
Georgi Gerganov 2023-12-31 11:43:31 +02:00
parent a28dacec65
commit d5673af79f
1 changed files with 61 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

118
ggml-quants.c
View File

@ -410,13 +410,17 @@ inline static ggml_int8x16x4_t ggml_vld1q_s8_x4(const int8_t * ptr) {
#if !defined(__ARM_FEATURE_DOTPROD)
inline static int32x4_t vdotq_s32(int32x4_t acc, int8x16_t a, int8x16_t b) {
inline static int32x4_t ggml_vdotq_s32(int32x4_t acc, int8x16_t a, int8x16_t b) {
const int16x8_t p0 = vmull_s8(vget_low_s8 (a), vget_low_s8 (b));
const int16x8_t p1 = vmull_s8(vget_high_s8(a), vget_high_s8(b));
return vaddq_s32(acc, vaddq_s32(vpaddlq_s16(p0), vpaddlq_s16(p1)));
}
#else
#define ggml_vdotq_s32(a, b, c) vdotq_s32(a, b, c)
#endif
#endif
@ -2481,8 +2485,8 @@ void ggml_vec_dot_q4_0_q8_0(int n, float * restrict s, const void * restrict vx,
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
// dot product into int32x4_t
const int32x4_t p_0 = vdotq_s32(vdotq_s32(vdupq_n_s32(0), v0_0ls, v1_0l), v0_0hs, v1_0h);
const int32x4_t p_1 = vdotq_s32(vdotq_s32(vdupq_n_s32(0), v0_1ls, v1_1l), v0_1hs, v1_1h);
const int32x4_t p_0 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_0ls, v1_0l), v0_0hs, v1_0h);
const int32x4_t p_1 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_1ls, v1_1l), v0_1hs, v1_1h);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(p_0), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(p_1), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
@ -2769,8 +2773,8 @@ void ggml_vec_dot_q4_1_q8_1(const int n, float * restrict s, const void * restri
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
// dot product into int32x4_t
const int32x4_t p_0 = vdotq_s32(vdotq_s32(vdupq_n_s32(0), v0_0l, v1_0l), v0_0h, v1_0h);
const int32x4_t p_1 = vdotq_s32(vdotq_s32(vdupq_n_s32(0), v0_1l, v1_1l), v0_1h, v1_1h);
const int32x4_t p_0 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_0l, v1_0l), v0_0h, v1_0h);
const int32x4_t p_1 = ggml_vdotq_s32(ggml_vdotq_s32(vdupq_n_s32(0), v0_1l, v1_1l), v0_1h, v1_1h);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(p_0), GGML_FP16_TO_FP32(x0->d)*y0->d);
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(p_1), GGML_FP16_TO_FP32(x1->d)*y1->d);
@ -2936,11 +2940,11 @@ void ggml_vec_dot_q5_0_q8_0(const int n, float * restrict s, const void * restri
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(vaddq_s32(
vdotq_s32(vdupq_n_s32(0), v0_0lf, v1_0l),
vdotq_s32(vdupq_n_s32(0), v0_0hf, v1_0h))), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
ggml_vdotq_s32(vdupq_n_s32(0), v0_0lf, v1_0l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_0hf, v1_0h))), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(vaddq_s32(
vdotq_s32(vdupq_n_s32(0), v0_1lf, v1_1l),
vdotq_s32(vdupq_n_s32(0), v0_1hf, v1_1h))), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
ggml_vdotq_s32(vdupq_n_s32(0), v0_1lf, v1_1l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_1hf, v1_1h))), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
}
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1);
@ -3228,11 +3232,11 @@ void ggml_vec_dot_q5_1_q8_1(const int n, float * restrict s, const void * restri
const int8x16_t v1_1h = vld1q_s8(y1->qs + 16);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(vaddq_s32(
vdotq_s32(vdupq_n_s32(0), v0_0lf, v1_0l),
vdotq_s32(vdupq_n_s32(0), v0_0hf, v1_0h))), GGML_FP16_TO_FP32(x0->d)*y0->d);
ggml_vdotq_s32(vdupq_n_s32(0), v0_0lf, v1_0l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_0hf, v1_0h))), GGML_FP16_TO_FP32(x0->d)*y0->d);
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(vaddq_s32(
vdotq_s32(vdupq_n_s32(0), v0_1lf, v1_1l),
vdotq_s32(vdupq_n_s32(0), v0_1hf, v1_1h))), GGML_FP16_TO_FP32(x1->d)*y1->d);
ggml_vdotq_s32(vdupq_n_s32(0), v0_1lf, v1_1l),
ggml_vdotq_s32(vdupq_n_s32(0), v0_1hf, v1_1h))), GGML_FP16_TO_FP32(x1->d)*y1->d);
}
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1) + summs0 + summs1;
@ -3483,12 +3487,12 @@ void ggml_vec_dot_q8_0_q8_0(const int n, float * restrict s, const void * restri
const int8x16_t y1_1 = vld1q_s8(y1->qs + 16);
sumv0 = vmlaq_n_f32(sumv0, vcvtq_f32_s32(vaddq_s32(
vdotq_s32(vdupq_n_s32(0), x0_0, y0_0),
vdotq_s32(vdupq_n_s32(0), x0_1, y0_1))), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
ggml_vdotq_s32(vdupq_n_s32(0), x0_0, y0_0),
ggml_vdotq_s32(vdupq_n_s32(0), x0_1, y0_1))), GGML_FP16_TO_FP32(x0->d)*GGML_FP16_TO_FP32(y0->d));
sumv1 = vmlaq_n_f32(sumv1, vcvtq_f32_s32(vaddq_s32(
vdotq_s32(vdupq_n_s32(0), x1_0, y1_0),
vdotq_s32(vdupq_n_s32(0), x1_1, y1_1))), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
ggml_vdotq_s32(vdupq_n_s32(0), x1_0, y1_0),
ggml_vdotq_s32(vdupq_n_s32(0), x1_1, y1_1))), GGML_FP16_TO_FP32(x1->d)*GGML_FP16_TO_FP32(y1->d));
}
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1);
@ -3598,8 +3602,8 @@ void ggml_vec_dot_q2_K_q8_K(const int n, float * restrict s, const void * restri
// We use this macro instead of a function call because for some reason
// the code runs 2-3% slower, even if the function is declared inline
#define MULTIPLY_ACCUM_WITH_SCALE(index)\
isum += vaddvq_s32(vdotq_s32(vzero, q2bytes.val[0], q8bytes.val[0])) * aux[is+(index)];\
isum += vaddvq_s32(vdotq_s32(vzero, q2bytes.val[1], q8bytes.val[1])) * aux[is+1+(index)];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[0], q8bytes.val[0])) * aux[is+(index)];\
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[1], q8bytes.val[1])) * aux[is+1+(index)];
#define SHIFT_MULTIPLY_ACCUM_WITH_SCALE(shift, index)\
q8bytes = ggml_vld1q_s8_x2(q8); q8 += 32;\
@ -3973,10 +3977,10 @@ void ggml_vec_dot_q2_K_q8_K(const int n, float * restrict s, const void * restri
q2bytes.val[2] = vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q2bits, 4), m3));
q2bytes.val[3] = vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q2bits, 6), m3));
isum1 += vaddvq_s32(vdotq_s32(vzero, q2bytes.val[0], q8bytes.val[0])) * scales[0];
isum2 += vaddvq_s32(vdotq_s32(vzero, q2bytes.val[1], q8bytes.val[1])) * scales[1];
isum1 += vaddvq_s32(vdotq_s32(vzero, q2bytes.val[2], q8bytes.val[2])) * scales[2];
isum2 += vaddvq_s32(vdotq_s32(vzero, q2bytes.val[3], q8bytes.val[3])) * scales[3];
isum1 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[0], q8bytes.val[0])) * scales[0];
isum2 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[1], q8bytes.val[1])) * scales[1];
isum1 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[2], q8bytes.val[2])) * scales[2];
isum2 += vaddvq_s32(ggml_vdotq_s32(vzero, q2bytes.val[3], q8bytes.val[3])) * scales[3];
sum += d * (isum1 + isum2);
}
@ -4256,10 +4260,10 @@ void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restri
q3bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[0], 2), m3b)), vreinterpretq_s8_u8(q3h.val[2]));
q3bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[1], 2), m3b)), vreinterpretq_s8_u8(q3h.val[3]));
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[0], q8bytes_1.val[0])) * scale[0];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[1], q8bytes_1.val[1])) * scale[1];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[2], q8bytes_1.val[2])) * scale[2];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[3], q8bytes_1.val[3])) * scale[3];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[0], q8bytes_1.val[0])) * scale[0];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[1], q8bytes_1.val[1])) * scale[1];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[2], q8bytes_1.val[2])) * scale[2];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[3], q8bytes_1.val[3])) * scale[3];
scale += 4;
@ -4273,10 +4277,10 @@ void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restri
q3bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[0], 6), m3b)), vreinterpretq_s8_u8(q3h.val[2]));
q3bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vandq_u8(vshrq_n_u8(q3bits.val[1], 6), m3b)), vreinterpretq_s8_u8(q3h.val[3]));
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[0], q8bytes_2.val[0])) * scale[0];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[1], q8bytes_2.val[1])) * scale[1];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[2], q8bytes_2.val[2])) * scale[2];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[3], q8bytes_2.val[3])) * scale[3];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[0], q8bytes_2.val[0])) * scale[0];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[1], q8bytes_2.val[1])) * scale[1];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[2], q8bytes_2.val[2])) * scale[2];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[3], q8bytes_2.val[3])) * scale[3];
scale += 4;
@ -4757,10 +4761,10 @@ void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restri
q3bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(vshrq_n_u8(q3bits, 4), m3b), q3h.val[2]));
q3bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q3bits, 6), q3h.val[3]));
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[0], q8bytes.val[0])) * scales[0];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[1], q8bytes.val[1])) * scales[2];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[2], q8bytes.val[2])) * scales[1];
isum += vaddvq_s32(vdotq_s32(vzero, q3bytes.val[3], q8bytes.val[3])) * scales[3];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[0], q8bytes.val[0])) * scales[0];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[1], q8bytes.val[1])) * scales[2];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[2], q8bytes.val[2])) * scales[1];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q3bytes.val[3], q8bytes.val[3])) * scales[3];
sum += d * isum;
@ -5109,14 +5113,14 @@ void ggml_vec_dot_q4_K_q8_K(const int n, float * restrict s, const void * restri
q4bytes.val[0] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[0], m4b));
q4bytes.val[1] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[1], m4b));
const int32x4_t p1 = vdotq_s32(vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
const int32x4_t p1 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
sumi1 += vaddvq_s32(p1) * scales[2*j+0];
q8bytes = ggml_vld1q_s8_x2(q8); q8 += 32;
q4bytes.val[0] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[0], 4));
q4bytes.val[1] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[1], 4));
const int32x4_t p2 = vdotq_s32(vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
const int32x4_t p2 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
sumi2 += vaddvq_s32(p2) * scales[2*j+1];
}
@ -5449,13 +5453,13 @@ void ggml_vec_dot_q4_K_q8_K(const int n, float * restrict s, const void * restri
q4bytes.val[0] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[0], m4b));
q4bytes.val[1] = vreinterpretq_s8_u8(vandq_u8 (q4bits.val[1], m4b));
const int32x4_t p1 = vdotq_s32(vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
const int32x4_t p1 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[0]), q4bytes.val[1], q8bytes.val[1]);
const int32_t sumi1 = vaddvq_s32(p1) * scales[0];
q4bytes.val[0] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[0], 4));
q4bytes.val[1] = vreinterpretq_s8_u8(vshrq_n_u8(q4bits.val[1], 4));
const int32x4_t p2 = vdotq_s32(vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[2]), q4bytes.val[1], q8bytes.val[3]);
const int32x4_t p2 = ggml_vdotq_s32(ggml_vdotq_s32(mzero, q4bytes.val[0], q8bytes.val[2]), q4bytes.val[1], q8bytes.val[3]);
const int32_t sumi2 = vaddvq_s32(p2) * scales[1];
sumf += d * (sumi1 + sumi2);
@ -5722,8 +5726,8 @@ void ggml_vec_dot_q5_K_q8_K(const int n, float * restrict s, const void * restri
q5bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q5bits.val[0], 4), q5h.val[2]));
q5bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q5bits.val[1], 4), q5h.val[3]));
sumi += vaddvq_s32(vdotq_s32(vdotq_s32(mzero, q5bytes.val[0], q8bytes.val[0]), q5bytes.val[1], q8bytes.val[1])) * *scales++;
sumi += vaddvq_s32(vdotq_s32(vdotq_s32(mzero, q5bytes.val[2], q8bytes.val[2]), q5bytes.val[3], q8bytes.val[3])) * *scales++;
sumi += vaddvq_s32(ggml_vdotq_s32(ggml_vdotq_s32(mzero, q5bytes.val[0], q8bytes.val[0]), q5bytes.val[1], q8bytes.val[1])) * *scales++;
sumi += vaddvq_s32(ggml_vdotq_s32(ggml_vdotq_s32(mzero, q5bytes.val[2], q8bytes.val[2]), q5bytes.val[3], q8bytes.val[3])) * *scales++;
}
sumf += d * sumi - dmin * sumi_mins;
@ -6112,10 +6116,10 @@ void ggml_vec_dot_q5_K_q8_K(const int n, float * restrict s, const void * restri
q5bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vshrq_n_u8(q5bits.val[0], 4)), vreinterpretq_s8_u8(q5h.val[2]));
q5bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vshrq_n_u8(q5bits.val[1], 4)), vreinterpretq_s8_u8(q5h.val[3]));
int32_t sumi1 = sc[0] * vaddvq_s32(vdotq_s32(mzero, q5bytes.val[0], q8bytes.val[0]));
int32_t sumi2 = sc[1] * vaddvq_s32(vdotq_s32(mzero, q5bytes.val[1], q8bytes.val[1]));
int32_t sumi3 = sc[2] * vaddvq_s32(vdotq_s32(mzero, q5bytes.val[2], q8bytes.val[2]));
int32_t sumi4 = sc[3] * vaddvq_s32(vdotq_s32(mzero, q5bytes.val[3], q8bytes.val[3]));
int32_t sumi1 = sc[0] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[0], q8bytes.val[0]));
int32_t sumi2 = sc[1] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[1], q8bytes.val[1]));
int32_t sumi3 = sc[2] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[2], q8bytes.val[2]));
int32_t sumi4 = sc[3] * vaddvq_s32(ggml_vdotq_s32(mzero, q5bytes.val[3], q8bytes.val[3]));
sumf += d * (sumi1 + sumi2 + sumi3 + sumi4);
}
@ -6399,10 +6403,10 @@ void ggml_vec_dot_q6_K_q8_K(const int n, float * restrict s, const void * restri
q6bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[2], m4b), q6h.val[2]));
q6bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vandq_u8(q6bits.val[3], m4b), q6h.val[3]));
isum += vaddvq_s32(vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
scale += 4;
@ -6426,10 +6430,10 @@ void ggml_vec_dot_q6_K_q8_K(const int n, float * restrict s, const void * restri
q6bytes.val[2] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[2], 4), q6h.val[2]));
q6bytes.val[3] = vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[3], 4), q6h.val[3]));
isum += vaddvq_s32(vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
scale += 4;
}
//sum += isum * d_all * y[i].d;
@ -6816,10 +6820,10 @@ void ggml_vec_dot_q6_K_q8_K(const int n, float * restrict s, const void * restri
q6bytes.val[2] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[0], 4), q6h.val[2])), m32s);
q6bytes.val[3] = vsubq_s8(vreinterpretq_s8_u8(vorrq_u8(vshrq_n_u8(q6bits.val[1], 4), q6h.val[3])), m32s);
isum += vaddvq_s32(vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
isum += vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[0], q8bytes.val[0])) * scale[0] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[1], q8bytes.val[1])) * scale[1] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[2], q8bytes.val[2])) * scale[2] +
vaddvq_s32(ggml_vdotq_s32(vzero, q6bytes.val[3], q8bytes.val[3])) * scale[3];
sum += isum * d_all * y[i].d;