This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: ENTER/BOUND operands order.
- From: "Jan Beulich" <JBeulich at suse dot com>
- To: "Michael V. Zolotukhin" <michael dot v dot zolotukhin at gmail dot com>
- Cc: "H.J. Lu" <hjl dot tools at gmail dot com>, "binutils at sourceware dot org" <binutils at sourceware dot org>, "S??awomir Wojtasiak" <slawomir dot wojtasiak at swksoftware dot pl>
- Date: Thu, 23 Jan 2014 15:55:04 +0000
- Subject: Re: ENTER/BOUND operands order.
- Authentication-results: sourceware.org; auth=none
- References: <52D80C8402000078001144D5 at nat28 dot tlf dot novell dot com> <9f1ffd59bc7111c4ba411898a840f1c6 dot qmail at home dot pl> <20140117070150 dot GA13904 at msticlxl57 dot ims dot intel dot com> <52D8F5FE020000780011470A at nat28 dot tlf dot novell dot com> <CANtU079-g7dcZV8WxkOOUNqRwD0RJu1tTqA5rUxwEaLx4SmktQ at mail dot gmail dot com> <52D90ACA02000078001147A1 at nat28 dot tlf dot novell dot com> <CANtU078QUocaPa9YsNUTrUaRW10xKDx99FJ8LBdFVgAZ_ooG=Q at mail dot gmail dot com> <52D936D402000078001148B5 at nat28 dot tlf dot novell dot com> <20140123103732 dot GA44892 at msticlxl57 dot ims dot intel dot com> <52E102DF0200007800116199 at nat28 dot tlf dot novell dot com> <20140123110528 dot GB44892 at msticlxl57 dot ims dot intel dot com>
>>> On 23.01.14 at 12:05, "Michael V. Zolotukhin" <michael.v.zolotukhin@gmail.com> wrote:
>> I don't think EVEX.W determines the scaling factor here. The
>> input consists of dword elements, hence the scaling ought to
>> be 4. I think the manual even states so (but I didn't re-check).
> No, it's EVEX.W who defines scaling in this case. The manual says that
> VPMULU?DQ has FV (full vector) in 'operand encoding' column, and the table
> 4-6
> gives the scaling factors for different operand encoding depending on
> EVEX.W,
> broadcasting, and vector length. For the given instruction is yields 8.
But then again it says
VPMULDQ zmm1 {k1}{z}, zmm2, zmm3/m512/m32bcst
VPMULUDQ zmm1 {k1}{z}, zmm2, zmm3/m512/m32bcst
Using 8-byte granular scaling with 4-byte broadcast operands
would make little sense.
The table you refer to also says 64bit in the Input column, which
clearly is wrong for these two instructions. Hence I consider this
table inapplicable here, and an explicit note about the scaling
factor would then be desirable.
Jan