This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH v2] x86: fix handling of 64-bit operand size VPCMPESTR{I,M}
- From: "Jan Beulich" <JBeulich at suse dot com>
- To: "H.J. Lu" <hjl dot tools at gmail dot com>
- Cc: "Binutils" <binutils at sourceware dot org>
- Date: Fri, 24 Feb 2017 02:28:49 -0700
- Subject: Re: [PATCH v2] x86: fix handling of 64-bit operand size VPCMPESTR{I,M}
- Authentication-results: sourceware.org; auth=none
- References: <58AEE65B020000780013D38E@prv-mh.provo.novell.com> <CAMe9rOp==xeCNf3mv18fA+VdyZ9r3wMxW0F7_p0DW7ZgFQK9ww@mail.gmail.com> <58AF1834020000780013D540@prv-mh.provo.novell.com> <CAMe9rOqT0VRckVXhVWWuh4L=tGcYDRpqx2+bYxazc=22CLvpgQ@mail.gmail.com>
>>> On 23.02.17 at 17:20, <hjl.tools@gmail.com> wrote:
> On Thu, Feb 23, 2017 at 8:13 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>>> On 23.02.17 at 16:58, <hjl.tools@gmail.com> wrote:
>>> On Thu, Feb 23, 2017 at 4:40 AM, Jan Beulich <JBeulich@suse.com> wrote:
>>>> Just like REX.W affects operand size of the implicit rAX/rDX inputs to
>>>> PCMPESTR{I,M}, VEX.W does for VPCMPESTR{I,M}. Allow Q or L suffixes on
>>>> the instructions for AT&T mode.
>>>>
>>>> For Intel mode, however, suffixes aren't the preferred choice, so
>>>> build_vex_prefix() needs to not only honor internally specified REX
>>>> flags (converting them to their VEX equivalents), but also such
>>>> resulting from explicit REX* prefix uses. Also adjust
>>>> build_evex_prefix() for consistency.
>>>
>>> Please don't use EXISTING REX prefixes for this purpose. A new
>>> prefix to control encoding is better, but it probably belongs to another
>>> patch.
>>
>> I'm sorry, but I don't follow. If you want to go that route, we'll need
>> more than just one new prefix, and it would be far from clear where
>> to draw the boundary. Granted this would allow to address the
>> 32-/16-bit side of things at once, but no, I don't think I have the
>> time or energy to go that route (or else I would have done so right
>> away, instead of adding a remark to the description in this regard).
>
> We shouldn't change the meaning of existing REX prefixes. Let's
> leave Intel mode alone for now.
While I disagree, in the interest of getting the bug here fixed at
least in some way, I've submitted v3 with the REX parts dropped.
Jan