This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH] x86/Intel: accept mandated operand order for vcvt{,u}si2s{d,s}
- From: Kirill Yukhin <kirill dot yukhin at gmail dot com>
- To: Jan Beulich <JBeulich at suse dot com>
- Cc: "H.J. Lu" <hjl dot tools at gmail dot com>, Binutils <binutils at sourceware dot org>, "H. Peter Anvin" <hpa at zytor dot com>
- Date: Tue, 26 May 2015 15:24:18 +0400
- Subject: Re: [PATCH] x86/Intel: accept mandated operand order for vcvt{,u}si2s{d,s}
- Authentication-results: sourceware.org; auth=none
- References: <552FE0630200007800072CD0 at mail dot emea dot novell dot com> <CAMe9rOomWLMwQT6R2qLr1p7_dzmwuNLsz2PEk-6tV3NTt_=24Q at mail dot gmail dot com> <55390A6A0200007800075263 at mail dot emea dot novell dot com> <CAMe9rOrSX3cgXCSJFsbDnKgnyNrX-Y4cYO2VL4uTrKV-O3fcCw at mail dot gmail dot com> <554906C70200007800076D28 at mail dot emea dot novell dot com> <20150525145555 dot GA9967 at msticlxl57 dot ims dot intel dot com> <55644524020000780007DB9A at mail dot emea dot novell dot com>
Hello Jan,
On 26 May 09:04, Jan Beulich wrote:
> >>> On 25.05.15 at 16:55, <kirill.yukhin@gmail.com> wrote:
> >> Thoughts?
> > Wanted to mention couple of points:
> > 1. I agree w/ HJ: SDM is not a source of syntax
> > 2. We have two product already released: ICC (2 version at the moment) and
> > GCC
> > (4.9 and 5) which use existing syntax for emitting those insns
> >
> > Said that, I don't see any reason to introduce (not replace) alternative
> > operand order.
>
> I'm confused - did you perhaps mean to say "I don't see any reason
> not to introduce ..."?
No, this is what I think. I think that having two variants of operands order
is not good idea.
We have those products which use AVX-512:
1. ICC (couple of versions): uses {vreg, vreg, {RC}, GPR} order.
2. Binutils (couple of versions): Same.
3. GCC (4.9, 5): Same.
4. NASM (I tried last ver, 2.11.08): Same.
MASM doesn't support AVX-512 and cannot be source of Intel syntax,
but I'll try to inform them about current variant.
Finally, (I promise, I repeat it last time) as far as SDM is not a source of Intel syntax,
I see no reason to allow another order. IMO, such addition might
even increase confusion about this non-trivial stuff.
--
Thanks, K
>
> Jan
>