[PATCH] 8217561 : X86: Add floating-point Math.min/max intrinsics
bsrbnd at gmail.com
Sat Feb 9 18:43:02 UTC 2019
Hi Jatin, All,
On Fri, 8 Feb 2019 at 06:55, Bhateja, Jatin <jatin.bhateja at intel.com> wrote:
> Hi Bernard, All,
> Please find below the link to updated patch.
Your patch seems quite good to me.
However, I'd like to suggest the following variant (which is mostly a
copy/paste of your code) focusing on x86_64 as I don't think that
taking care of x86_32 is worth the additional complexity. It uses the
min/max scalar variant which is conceptually better than the packed
one. Regarding the test case, I simply added a main loop along with an
x86_64 test configuration (similar to yours) without modifying the
existing AArch64 parameters:
This variant also includes the following small code improvements:
* copyright years updated where necessary
* useless assertions removed ('vex_prefix_and_encode' already checks
the register encoding in 'legacy_mode')
* constants used to improve readability like 0 -> AVX_128bit, 0x3 -> _false, ...
* several comments fixed like
* 'java.lang.Max()' -> 'java.lang.Math.max()', ...
* 'if b < 0' -> 'if b < +0.0'
* unnecessary trailing spaces removed
Test-tier1 is OK on x86_64 (xeon).
Could we then have a Reviewer feedback (and eventually an approval) for this?
More information about the hotspot-compiler-dev