ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 19:31:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

One more

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-04-03 13:10:53 +02:00
parent 9c7af40ff6
commit 310bce3c1d
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -15829,9 +15829,9 @@ struct FlashMS {
auto vm2 = vzip2q_u16(vm, vm); auto vm2 = vzip2q_u16(vm, vm);
auto kq = vld1q_f32_x2(cache + k_step*j + 8*l); auto kq = vld1q_f32_x2(cache + k_step*j + 8*l);
vk[2*l+0] = vreinterpretq_f32_u32(vorrq_u32(vandq_u32(vreinterpretq_u32_f32(kq.val[0]), vm1), vk[2*l+0] = vreinterpretq_f32_u32(vorrq_u32(vandq_u32(vreinterpretq_u32_f32(kq.val[0]), vm1),
vbicq_u32(vinf, vm1))); vbicq_u32(vreinterpretq_u32_f32(vinf), vm1)));
vk[2*l+1] = vreinterpretq_f32_u32(vorrq_u32(vandq_u32(vreinterpretq_u32_f32(kq.val[1]), vm2), vk[2*l+1] = vreinterpretq_f32_u32(vorrq_u32(vandq_u32(vreinterpretq_u32_f32(kq.val[1]), vm2),
vbicq_u32(vinf, vm2))); vbicq_u32(vreinterpretq_u32_f32(vinf), vm2)));
} }
float32x4_t vmax = vdupq_n_f32(-INFINITY); float32x4_t vmax = vdupq_n_f32(-INFINITY);
auto vscale32 = vcvt_f32_f16(vget_low_f16(vscale)); auto vscale32 = vcvt_f32_f16(vget_low_f16(vscale));