[PATCH 04/10] Support Intel CMPccXADD

[H.J. Lu]

On Sat, Oct 15, 2022 at 11:25 PM Jan Beulich <jbeulich@suse.com> wrote:
>
> On 14.10.2022 20:27, H.J. Lu wrote:
> >  On Fri, Oct 14, 2022 at 6:46 AM Jan Beulich <jbeulich@suse.com> wrote:
> >> On 14.10.2022 11:12, Haochen Jiang wrote:
> >>> --- a/opcodes/i386-opc.tbl
> >>> +++ b/opcodes/i386-opc.tbl
> >>> @@ -3296,3 +3296,24 @@ vpdpbsud, 0xf350, None, CpuAVX_VNNI_INT8, Modrm|Vex|Space0F38|VexVVVV|VexW0|Chec
> >>>  vpdpbsuds, 0xf351, None, CpuAVX_VNNI_INT8, Modrm|Vex|Space0F38|VexVVVV|VexW0|CheckRegSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM|RegYMM|Unspecified|BaseIndex, RegXMM|RegYMM, RegXMM|RegYMM }
> >>>
> >>>  // AVX_VNNI_INT8 instructions end.
> >>> +
> >>> +// CMPCCXADD instructions.
> >>> +
> >>> +cmpbexadd, 0x66e6, None, CpuCMPCCXADD|Cpu64, Modrm|Vex128|Space0F38|VexVVVV=1|SwapSources|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { Reg32|Reg64, Reg32|Reg64, Dword|Qword|Unspecified|BaseIndex }
> >>
> >> Along the lines of the earlier comment - you want to use the <cc>
> >> template here, eliminating the need for 16 almost identical lines _and_
> >> supplying all condition code representation in one go.
> >>
> >> Apart from that you forgot CheckRegSize here afaict. And please again
> >> VexVVVV alone, without =1. Also for non-vector insns perhaps better plain
> >> Vex instead of Vex128. Further these insns should allow for l and q
> >> suffixes in AT&T mode.
> >
> > l and q suffixes here are totally unnecessary.  For new instructions,
> > suffixes should be required only if needed.
>
> I disagree, and I'd like to ask that decisions like this not be made behind
> closed doors at Intel. As said many times - there needs to be consistency,
> which is a prereq for predictability. A programmer shall not be required to
> read gas sources to know for which insns suffixes are permitted and for

We can improve documentation.  A programmer should use SDM first.
Suffixes should be used only when SDM doesn't address AT&T syntax
needs.

> which they aren't. Plus do you realize that "new insns" also covers RAO-INT,
> where suffixes are permitted (as expected)?

There is a discrepancy.

> Jan



-- 
H.J.