RFR: 8315970: Big-endian issues after JDK-8310929 [v2]

Mon Sep 11 14:12:52 UTC 2023

On Mon, 11 Sep 2023 13:53:19 GMT, Martin Doerr <mdoerr at openjdk.org> wrote:

> Shouldn't this get optimized by the JIT compilers? Why is `ByteArrayLittleEndian` supposed to be faster? Note that there are still several Big Endian platforms: AIX, s390x and old linux ppc64 (still kept alive)

And none of these are covered by Oracle-internal or GHA testing, sadly.

It'd be interesting to see performance numbers for `putPair` vs `ByteArrayLittleEndian.setInt` in isolation on x86. I.e. before and after this PR. When I've tested a similar thing for #15591 I saw a slight win on a two-byte pair (that might be due improved inlining decisions) but regressions on 4-byte or larger writes.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/15652#issuecomment-1713969279