mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-04-28 18:32:04 +00:00
Fix AVX2
This commit is contained in:
@@ -6660,10 +6660,12 @@ static void repack_q16_k(int nrows, int n_per_row, const block_q8_K * x, block_q
|
|||||||
for (int i = 0; i < 4; ++i) y[ibl].qs[64*ib + 4*k + i] = x16[k][ibl].qs[4*ib+i];
|
for (int i = 0; i < 4; ++i) y[ibl].qs[64*ib + 4*k + i] = x16[k][ibl].qs[4*ib+i];
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
#ifdef HAVE_FANCY_SIMD
|
||||||
for (int l = 0; l < 64; ++l) {
|
for (int l = 0; l < 64; ++l) {
|
||||||
auto v = _mm512_xor_si512(_mm512_loadu_si512((const __m512i *)y[ibl].qs + l), _mm512_set1_epi8(-128));
|
auto v = _mm512_xor_si512(_mm512_loadu_si512((const __m512i *)y[ibl].qs + l), _mm512_set1_epi8(-128));
|
||||||
_mm512_storeu_si512((__m512i *)y[ibl].qs + l, v);
|
_mm512_storeu_si512((__m512i *)y[ibl].qs + l, v);
|
||||||
}
|
}
|
||||||
|
#endif
|
||||||
}
|
}
|
||||||
x += 16*nblock;
|
x += 16*nblock;
|
||||||
y += nblock;
|
y += nblock;
|
||||||
|
|||||||
Reference in New Issue
Block a user