mirrors/passt
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

checksum: Interleave lo/hi sums while folding into 128-bit sums, drop TODO

Browse source 
I left a TODO and never checked -- this actually seems to slightly
improve CPIs on AMD Naples (two 128-bit FMA units glued together).

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This commit is contained in:
Stefano Brivio 2021-10-15 16:59:12 +02:00
parent 364cc313ea
commit 74f29d3148
1 changed files with 3 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
checksum.c
View file

@ -217,9 +217,9 @@ static uint32_t csum_avx2(const void *buf, size_t len, uint32_t init)
_mm256_unpacklo_epi32(b, zero));
}
/* Fold four 256bit sums into one 128-bit sum. TODO */
sum256 = _mm256_add_epi64(_mm256_add_epi64(sum_a_hi, sum_a_lo),
_mm256_add_epi64(sum_b_hi, sum_b_lo));
/* Fold four 256bit sums into one 128-bit sum. */
sum256 = _mm256_add_epi64(_mm256_add_epi64(sum_a_hi, sum_b_lo),
_mm256_add_epi64(sum_b_hi, sum_a_lo));
sum128 = _mm_add_epi64(_mm256_extracti128_si256(sum256, 0),
_mm256_extracti128_si256(sum256, 1));