RFR(XS) 8224558: x86 Fix replicateB encoding
Deshpande, Vivek R
vivek.r.deshpande at intel.com
Wed May 22 17:48:57 UTC 2019
Thanks for looking at it. Yes I will fix the format strings.
I came across this issue with vector API tests.
I tried to write a reproducer using the autovectorizer test, but it always uses the register based rule(instruct Repl32B) instead of memory based rule (instruct Repl32B_mem) and register based rule gives correct result.
This is why it never showed up.
Could you please help me with forcing to use memory based rule with autovectorizer based test.
From: Tobias Hartmann [mailto:tobias.hartmann at oracle.com]
Sent: Wednesday, May 22, 2019 1:13 AM
To: Deshpande, Vivek R <vivek.r.deshpande at intel.com>; 'hotspot-compiler-dev at openjdk.java.net compiler' <hotspot-compiler-dev at openjdk.java.net>
Subject: Re: RFR(XS) 8224558: x86 Fix replicateB encoding
I wonder why this never showed up, could you please add a regression test?
You also need to fix the format strings.
On 22.05.19 02:12, Deshpande, Vivek R wrote:
> Hi All
> The encoding for replicateB in x86.ad uses dst register as one of the
> source without initializing, when the source for the scalar value is memory.
> This leads to wrong replication in the resulting vector.
> I have a fix for the bug in this webrev:
> I have created following JBS Entry:
> Kindly requesting review for the patch.
More information about the hotspot-compiler-dev