RFR 8167067: SHA Encoding Fix

[Lupusoru, Razvan A]

Vladimir,

Michael has updated the new webrev for me. Please take a look at the revised patch! And thanks again.
http://cr.openjdk.java.net/~mcberg/8167067/webrev.02/

--Razvan

-----Original Message-----
From: Lupusoru, Razvan A 
Sent: Wednesday, October 05, 2016 10:47 AM
To: 'Vladimir Kozlov' <vladimir.kozlov at oracle.com>; hotspot-compiler-dev at openjdk.java.net
Cc: Berg, Michael C <michael.c.berg at intel.com>
Subject: RE: RFR 8167067: SHA Encoding Fix

Vladimir,

Thanks for the quick reply. :)

So only REX prefix of form REX.R can be used in 64-bit mode to access xmm8-15. There is no VEX nor EVEX encoding form as far as I can see.

I will use your suggestion and will provide updated patch soon! Thanks for the review!

--Razvan

-----Original Message-----
From: Vladimir Kozlov [mailto:vladimir.kozlov at oracle.com]
Sent: Tuesday, October 04, 2016 7:06 PM
To: Lupusoru, Razvan A <razvan.a.lupusoru at intel.com>; hotspot-compiler-dev at openjdk.java.net
Cc: Berg, Michael C <michael.c.berg at intel.com>
Subject: Re: RFR 8167067: SHA Encoding Fix

Hi Razvan,

So only REX can be used? What about EVEX?

I would suggest to use rex_prefix_and_encode() here - we don't generate 0F_3A opcodes directly:

int encode = prefix_and_encode(dst->encoding(), src->encoding(), 0, VEX_OPCODE_0F_3A, /* rex_w */ false);

Thanks,
Vladimir

On 10/4/16 6:04 PM, Lupusoru, Razvan A wrote:
> Hello,
>
>
>
> We have recently uncovered an encoding bug with the SHA x86 
> instructions
> - their encoding should not contain VEX prefix when AVX is enabled. 
> The error manifests as crash due to SIGILL. Thus, we have filed a bug 
> and prepared a solution.
>
>
>
> Bug: https://bugs.openjdk.java.net/browse/JDK-8167067
>
> Webrev: http://cr.openjdk.java.net/~mcberg/8167067/webrev.01/
>
>
>
> Can you please review and help get it integrated? Let me know if there 
> is anything I need to do to facilitate this.
>
>
>
> Thanks so much,
>
> Razvan
>
>
>