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

iq1s_blocks16: very slightly faster TG on Metal

Browse source 
Still pathetic at 37 t/s
This commit is contained in:
Iwan Kawrakow 2024-03-08 17:51:08 +01:00
parent 15acc7923b
commit 8561139a48
1 changed files with 12 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

23
ggml-metal.metal
View file

@ -4338,19 +4338,18 @@ void kernel_mul_mv_iq1_s_f32_impl(
device const block_iq1_s * x = (device const block_iq1_s *) src0 + ib_row + offset0; device const block_iq1_s * x = (device const block_iq1_s *) src0 + ib_row + offset0;
device const float * y = (device const float *) src1 + r1*ne10 + im*ne00*ne1; device const float * y = (device const float *) src1 + r1*ne10 + im*ne00*ne1;
float yl[16]; float yl[32];
float sumf[N_DST]={0.f}, all_sum; float sumf[N_DST]={0.f}, all_sum;
const int nb32 = nb * (QK_K / 32); const int nb32 = nb * (QK_K / 32);
const int ix = tiisg/2; const int ix = tiisg;
const int il = tiisg%2;
device const float * y4 = y + 32 * ix + 16 * il; device const float * y4 = y + 32 * ix;
for (int ib32 = ix; ib32 < nb32; ib32 += 16) { for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
for (int i = 0; i < 16; ++i) { for (int i = 0; i < 32; ++i) {
yl[i] = y4[i]; yl[i] = y4[i];
} }
@ -4358,18 +4357,20 @@ void kernel_mul_mv_iq1_s_f32_impl(
const int ib = ib32 % (QK_K / 32); const int ib = ib32 % (QK_K / 32);
device const block_iq1_s * xr = x + ibl; device const block_iq1_s * xr = x + ibl;
device const uint8_t * qs = xr->qs + 4 * ib + 2 * il; device const uint8_t * qs = xr->qs + 4 * ib;
device const uint16_t * qh = xr->qh + ib; device const uint16_t * qh = xr->qh + ib;
device const half * dh = &xr->d; device const half * dh = &xr->d;
for (int row = 0; row < N_DST; row++) { for (int row = 0; row < N_DST; row++) {
constant int8_t * grid1 = (constant int8_t *)(iq1s_grid + (qs[0] | (((qh[0] >> (6*il+0)) & 7) << 8))); constant int8_t * grid1 = (constant int8_t *)(iq1s_grid + (qs[0] | ((qh[0] << 8) & 0x700)));
constant int8_t * grid2 = (constant int8_t *)(iq1s_grid + (qs[1] | (((qh[0] >> (6*il+3)) & 7) << 8))); constant int8_t * grid2 = (constant int8_t *)(iq1s_grid + (qs[1] | ((qh[0] << 5) & 0x700)));
constant int8_t * grid3 = (constant int8_t *)(iq1s_grid + (qs[2] | ((qh[0] << 2) & 0x700)));
constant int8_t * grid4 = (constant int8_t *)(iq1s_grid + (qs[3] | ((qh[0] >> 1) & 0x700)));
float sum = 0; float sum = 0;
for (int j = 0; j < 8; ++j) { for (int j = 0; j < 8; ++j) {
sum += yl[j+0] * grid1[j] + yl[j+8] * grid2[j]; sum += yl[j+0] * grid1[j] + yl[j+8] * grid2[j] + yl[j+16] * grid3[j] + yl[j+24] * grid4[j];
} }
sumf[row] += (float)dh[0] * sum * (2*(qh[0] >> 12) + 1); sumf[row] += (float)dh[0] * sum * (2*(qh[0] >> 12) + 1);
@ -4378,7 +4379,7 @@ void kernel_mul_mv_iq1_s_f32_impl(
qh += nb*sizeof(block_iq1_s)/2; qh += nb*sizeof(block_iq1_s)/2;
} }
y4 += 16 * 32; y4 += 32 * 32;
} }
for (int row = 0; row < N_DST; ++row) { for (int row = 0; row < N_DST; ++row) {