[PATCH] 8217561 : X86: Add floating-point Math.min/max intrinsics
bsrbnd at gmail.com
Sat Feb 2 16:37:39 UTC 2019
On Fri, 1 Feb 2019 at 22:58, Viswanathan, Sandhya
<sandhya.viswanathan at intel.com> wrote:
> Hi All,
> Please find below the updated patch from Jatin:
> This patch fixes the issue reported by Nils.
Could you clarify the problem and its solution?
I don't see any USE_KILL you mentioned earlier, or maybe the test case
wasn't correct as Jatin said previously?
> The compiler jtreg tests on Haswell and SKX pass successfully.
The fix also affects x86_32.ad, could you confirm that at least
test-tier1 is successful with these older families too?
> It shows about 25% gain on Haswell desktop and about 75% on SKX server.
> Changing the maxps to maxss and cmpps to cmpss didn’t show any performance difference.
Conceptually, I dislike useless computation even if well parallelized
which might increase the processor's temperature and reduce
performance on long runs (but this is a personal opinion).
Regarding copyright headers, I note that the year hasn't been updated
to 2019 on some of them and on the test file:
3 * Copyright (c) 2018, Arm Limited. All rights reserved.
Is 'Arm Limitied' intended?
> Best Regards,
More information about the hotspot-compiler-dev