RE: [PATCH] x86/APX: optimize MOVBE

["Hu, Lin1" <lin1.hu@intel.com>]

> -----Original Message-----
> From: Jan Beulich <jbeulich@suse.com>
> Sent: Friday, January 12, 2024 8:28 PM
> To: Binutils <binutils@sourceware.org>
> Cc: H.J. Lu <hjl.tools@gmail.com>; Hu, Lin1 <lin1.hu@intel.com>
> Subject: Re: [PATCH] x86/APX: optimize MOVBE
> 
> On 12.01.2024 13:00, Jan Beulich wrote:
> > --- a/opcodes/i386-opc.tbl
> > +++ b/opcodes/i386-opc.tbl
> > @@ -210,6 +210,9 @@ mov, 0xf24, i386&No64, D|RegMem|IgnoreSi  //
> Move
> > after swapping the bytes  movbe, 0x0f38f0, Movbe,
> > D|Modrm|CheckOperandSize|No_bSuf|No_sSuf, {
> > Word|Dword|Qword|Unspecified|BaseIndex, Reg16|Reg32|Reg64 }  movbe,
> > 0x60, Movbe&APX_F,
> D|Modrm|CheckOperandSize|No_bSuf|No_sSuf|EVexMap4,
> > { Reg16|Reg32|Reg64|Unspecified|BaseIndex, Reg16|Reg32|Reg64 }
> > +// This really is BSWAP, but wants "duplicating" here for easy EVEX
> > +-> REX2 // conversion, restarting with the next sequential template.
> > +movbe, 0xfc8, Movbe&APX_F, No_bSuf|No_wSuf|No_sSuf, { Reg32|Reg64 }
> 
> Thinking of it - while not the end of the world, it would certainly be nice if this
> new template wasn't available for "ordinary" use. I'll see if I can come up with
> something.
> 

Maybe just put a flag in _i386_insn and add a new attribute in .tbl can solve the problem. Or add an already existing attribute which can disturb the "ordinary" use, and we change some value after optimization to match the template. Others are OK.

BRs,
Lin