| commit | 3c9980698988ef64072f1fac339b180f52792faf | |
| author | Danila Kutenin <danilak@google.com> | |
| Mon, 27 Jun 2022 16:12:13 +0000 (27 16:12 +0000) | ||
| committer | Szabolcs Nagy <szabolcs.nagy@arm.com> | |
| Wed, 6 Jul 2022 08:26:20 +0000 (6 09:26 +0100) | ||
| tree | 3c32dabb3fcbfa564647fcedd9be5c7674a30fc2 | treesnapshot (tar.gz zip) |
| parent | bd0b58837c7df091046e7531642f379a52e1e157 | commitdiff |
aarch64: Optimize string functions with shrn instruction
We found that string functions were using AND+ADDP
to find the nibble/syndrome mask but there is an easier
opportunity through `SHRN dst.8b, src.8h, 4` (shift
right every 2 bytes by 4 and narrow to 1 byte) and has
same latency on all SIMD ARMv8 targets as ADDP. There
are also possible gaps for memcmp but that's for
another patch.
We see 10-20% savings for small-mid size cases (<=128)
which are primary cases for general workloads.
We found that string functions were using AND+ADDP
to find the nibble/syndrome mask but there is an easier
opportunity through `SHRN dst.8b, src.8h, 4` (shift
right every 2 bytes by 4 and narrow to 1 byte) and has
same latency on all SIMD ARMv8 targets as ADDP. There
are also possible gaps for memcmp but that's for
another patch.
We see 10-20% savings for small-mid size cases (<=128)
which are primary cases for general workloads.