Re: [PATCH] x86: Remove movsx/movzx with memory operand from AT&T syntax

["H.J. Lu" <hjl.tools@gmail.com>]

On Tue, Feb 11, 2020 at 3:34 PM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Tue, Feb 11, 2020 at 12:11 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > On Tue, Feb 11, 2020 at 9:04 AM H.J. Lu <hjl.tools@gmail.com> wrote:
> > >
> > > On Tue, Feb 11, 2020 at 8:45 AM Jan Beulich <jbeulich@suse.com> wrote:
> > > >
> > > > On 11.02.2020 14:07, H.J. Lu wrote:
> > > > > On Tue, Feb 11, 2020 at 5:04 AM Jan Beulich <jbeulich@suse.com> wrote:
> > > > >>
> > > > >> On 11.02.2020 14:01, H.J. Lu wrote:
> > > > >>> On Tue, Feb 11, 2020 at 4:58 AM Jan Beulich <jbeulich@suse.com> wrote:
> > > > >>>>
> > > > >>>> On 11.02.2020 13:19, H.J. Lu wrote:
> > > > >>>>> On Tue, Feb 11, 2020 at 3:55 AM Jan Beulich <jbeulich@suse.com> wrote:
> > > > >>>>>>
> > > > >>>>>> On 11.02.2020 12:42, H.J. Lu wrote:
> > > > >>>>>>> On Tue, Feb 11, 2020 at 2:25 AM Jan Beulich <jbeulich@suse.com> wrote:
> > > > >>>>>>>>
> > > > >>>>>>>> Some encodings are about to gain a warning - move them from test cases
> > > > >>>>>>>> not expecting any diagnostics to the new, dedicated ones, to allow
> > > > >>>>>>>> better focus on the actual changes in the subsequent patch.
> > > > >>>>>>>>
> > > > >>>>>>>> The new tests added have some wrong expectations right now, which will
> > > > >>>>>>>> be corrected by the next patch. The test is being added here to make
> > > > >>>>>>>> more visible which cases actually were wrong (and hence get changed),
> > > > >>>>>>>> besides demonstrating that in the vast majority of cases the subsequent
> > > > >>>>>>>> change doesn't alter generated code.
> > > > >>>>>>>>
> > > > >>>>>>>> gas/
> > > > >>>>>>>> 2020-02-XX  Jan Beulich  <jbeulich@suse.com>
> > > > >>>>>>>>
> > > > >>>>>>>>         * testsuite/gas/i386/i386.s, testsuite/gas/i386/iamcu-1.s,
> > > > >>>>>>>>         testsuite/gas/i386/ilp32/x86-64.s: Move ambiguous operand size
> > > > >>>>>>>>         tests ...
> > > > >>>>>>>>         * testsuite/gas/i386/noreg16.s, testsuite/gas/i386/noreg32.s,
> > > > >>>>>>>>         testsuite/gas/i386/noreg64.s, testsuite/gas/i386/x86_64.s: ...
> > > > >>>>>>>>         here.
> > > > >>>>>>>>         * testsuite/gas/i386/i386.d, testsuite/gas/i386/i386-intel.d
> > > > >>>>>>>>         testsuite/gas/i386/iamcu-1.d, testsuite/gas/i386/ilp32/x86-64.d,
> > > > >>>>>>>>         testsuite/gas/i386/k1om.d, testsuite/gas/i386/l1om.d,
> > > > >>>>>>>>         testsuite/gas/i386/noreg16.d, testsuite/gas/i386/noreg32.d,
> > > > >>>>>>>>         testsuite/gas/i386/noreg64.d, testsuite/gas/i386/x86_64-intel.d,
> > > > >>>>>>>>         testsuite/gas/i386/x86_64.d: Adjust expectations.
> > > > >>>>>>>>         * testsuite/gas/i386/movx16.s, testsuite/gas/i386/movx16.l,
> > > > >>>>>>>>         testsuite/gas/i386/movx32.s, testsuite/gas/i386/movx32.l,
> > > > >>>>>>>>         testsuite/gas/i386/movx64.s, testsuite/gas/i386/movx64.l: New.
> > > > >>>>>>>>         * testsuite/gas/i386/i386.exp: Run new tests.
> > > > >>>>>>>
> > > > >>>>>>> Please make a separate patch to address MOVSX/MOVZX.
> > > > >>>>>>
> > > > >>>>>> I don't understand what you mean here. This patch simply documents the
> > > > >>>>>> status quo, to make it (much) easier to see what the next patch
> > > > >>>>>> actually adjusts. It doesn't "address" anything. If, for the purpose
> > > > >>>>>> of committing, you'd like to see both patches folded - fine by me. But
> > > > >>>>>> only then, not any earlier.
> > > > >>>>>>
> > > > >>>>>>>  MOVSX and MOVZX
> > > > >>>>>>> should take no suffixes.  AT&T syntax is supported if there is no
> > > > >>>>>>> ambiguity.  AT&T
> > > > >>>>>>> syntax also supports movsXY and movzXY.
> > > > >>>>>>
> > > > >>>>>> Please could you clarify what specifically you'd like to see changed,
> > > > >>>>>> at the very least by pointing out one case each where you think I'm
> > > > >>>>>> moving in the wrong direction (presumably in the next patch really)?
> > > > >>>>>> I'm afraid your response isn't such that I can derive from it what
> > > > >>>>>> exactly you want.
> > > > >>>>>
> > > > >>>>> We support
> > > > >>>>>
> > > > >>>>> movsx %ax, %ecx
> > > > >>>>> movzx %ax, %ecx
> > > > >>>>> movswl %ax, %ecx
> > > > >>>>> movzwl %ax, %ecx
> > > > >>>>>
> > > > >>>>> We disallow
> > > > >>>>>
> > > > >>>>> movsxw %ax, %ecx
> > > > >>>>> movzxw %ax, %ecx
> > > > >>>>
> > > > >>>> We don't (as this patch demonstrates, along with pre-existing tests,
> > > > >>>> unless you mean once again to have an inconsistency between insns
> > > > >>>> with all register operands and similar ones with e memory source),
> > > > >>>> and if you want it to be this way, then please do so yourself, but
> > > > >>>
> > > > >>> I will do it.
> > > > >>>
> > > > >>>> please also only on top of my changes, so I won't need to re-base
> > > > >>>
> > > > >>> Which changes of yours are you referring to?
> > > > >>
> > > > >> This patch and the subsequent one.
> > > > >>
> > > > >
> > > > > Both changes won't be necessary after my changes.
> > > >
> > > > I'm confused. What you want to deal with is - afaict - orthogonal to
> > > > what the next patch in the series here does.
> > > >
> > >
> > > You will see what I mean when I post my patch for review.
> > >
> >
> > AT&T syntax requires suffix to specify memory operand size.  Since
> > movsx and movzx can have different memory operand sizes with the same
> > destination register, this patch removes movsx and movzx with memory
> > operand from AT&T syntax.  Since AT&T syntax uses different mnemonics
> > for movsx and movzx, this change should have little impact on assembly
> > sources.  Tested with Linux kernel 5.5.3 for x86-64 and glibc 2.31 for
> > i686 and x86-64.
> >
>
> Updated patch to add more testcases and allow register operand with
> mov[sz]x[bwl].
>

I found usage of

movzx 4(%edx), %eax

We also need to support:

movsx 4(%edx), %eax

I will update my patch.

-- 
H.J.