RFR: 8328264: AArch64: remove UseNeon condition in CRC32 intrinsic [v2]
Hamlin Li
mli at openjdk.org
Tue Mar 19 15:32:31 UTC 2024
> Hi,
> Can you review the simple patch?
> Thanks
>
> FYI: Discussed https://github.com/openjdk/jdk/pull/18294#issuecomment-1997727704, this usage of UseNeon flag should be removed, as neon is working by default, and UseNeon could be false, so this means the intrinsic code of Neon can be skipped unexpected.
>
> ## Performance
> Tested jmh `TestCRC32` with "-XX:-UseCRC32 -XX:-UseCryptoPmullForCRC32"
> Before
>
> Benchmark (count) Mode Cnt Score Error Units
> TestCRC32.testCRC32Update 64 avgt 2 53.696 ns/op
> TestCRC32.testCRC32Update 128 avgt 2 104.942 ns/op
> TestCRC32.testCRC32Update 256 avgt 2 207.147 ns/op
> TestCRC32.testCRC32Update 512 avgt 2 411.179 ns/op
> TestCRC32.testCRC32Update 2048 avgt 2 1608.388 ns/op
> TestCRC32.testCRC32Update 16384 avgt 2 12763.513 ns/op
> TestCRC32.testCRC32Update 65536 avgt 2 51024.246 ns/op
>
>
> After
>
> Benchmark (count) Mode Cnt Score Error Units
> TestCRC32.testCRC32Update 64 avgt 2 40.172 ns/op
> TestCRC32.testCRC32Update 128 avgt 2 56.754 ns/op
> TestCRC32.testCRC32Update 256 avgt 2 89.743 ns/op
> TestCRC32.testCRC32Update 512 avgt 2 156.726 ns/op
> TestCRC32.testCRC32Update 2048 avgt 2 579.776 ns/op
> TestCRC32.testCRC32Update 16384 avgt 2 4624.023 ns/op
> TestCRC32.testCRC32Update 65536 avgt 2 18505.180 ns/op
Hamlin Li has updated the pull request incrementally with one additional commit since the last revision:
revert indents
-------------
Changes:
- all: https://git.openjdk.org/jdk/pull/18328/files
- new: https://git.openjdk.org/jdk/pull/18328/files/a5b520ee..6eca8b64
Webrevs:
- full: https://webrevs.openjdk.org/?repo=jdk&pr=18328&range=01
- incr: https://webrevs.openjdk.org/?repo=jdk&pr=18328&range=00-01
Stats: 226 lines in 1 file changed: 101 ins; 101 del; 24 mod
Patch: https://git.openjdk.org/jdk/pull/18328.diff
Fetch: git fetch https://git.openjdk.org/jdk.git pull/18328/head:pull/18328
PR: https://git.openjdk.org/jdk/pull/18328
More information about the hotspot-dev
mailing list