RE: [PATCH 1/3] x86: Split REX/REX2 old registers judgment.

["Cui, Lili" <lili.cui@intel.com>]

> On 20.05.2024 08:22, Cui, Lili wrote:
> > --- a/gas/config/tc-i386.c
> > +++ b/gas/config/tc-i386.c
> > @@ -4303,22 +4303,20 @@ static void establish_rex (void)
> >    /* Respect a user-specified REX prefix.  */
> >    i.rex |= i.prefix[REX_PREFIX] & REX_OPCODE;
> >
> > -  /* For 8 bit registers we need an empty rex prefix.  Also if the
> > -     instruction already has a prefix, we need to convert old
> > -     registers to new ones.  */
> > -
> > -  if ((i.types[first].bitfield.class == Reg && i.types[first].bitfield.byte
> > -       && ((i.op[first].regs->reg_flags & RegRex64) != 0 || i.rex != 0
> > -	   || i.rex2 != 0))
> > -      || (i.types[last].bitfield.class == Reg && i.types[last].bitfield.byte
> > -	  && ((i.op[last].regs->reg_flags & RegRex64) != 0 || i.rex != 0
> > -	      || i.rex2 != 0)))
> > -    {
> > -      unsigned int x;
> > -
> > -      if (!is_apx_rex2_encoding () && !is_any_vex_encoding(&i.tm))
> > -	i.rex |= REX_OPCODE;
> > -      for (x = first; x <= last; x++)
> > +  /* For 8 bit registers without a prefix, we need an empty rex
> > + prefix.  */
> 
> As you're touching this comment, can you please also correct it? It's not all 8-
> bit registers which are affected here (as expressed by the
> RegRex64 checks).
> 
Done.

> > +  if (((i.types[first].bitfield.class == Reg && i.types[first].bitfield.byte
> > +	&& ((i.op[first].regs->reg_flags & RegRex64) != 0 ))
> 
> Nit: Please parenthesize this line and ...
> 
> > +       || (i.types[last].bitfield.class == Reg && i.types[last].bitfield.byte
> > +	   && (i.op[last].regs->reg_flags & RegRex64) != 0 ))
> 
> ... this one the same way, preferably like the latter. Also please omit the blanks
> before the closing parentheses.
> 
Done.

> > +      && !is_apx_rex2_encoding () && !is_any_vex_encoding (&i.tm) &&
> > + !i.rex)
> 
> Is the !i.rex part needed? It wasn't there before. If it's needed, it would imo
> better come ahead of the APX/VEX checks.
> 
Yes, I added it because there is already i.rex and we don't need to overwrite it. And I will put it ahead of the APX/VEX checks.

> > +    i.rex |= REX_OPCODE;
> > +
> > +  /* For REX/REX2 prefix instructions, we need to convert old registers
> > +     (AL, CL, DL and BL) to new ones (AXL, CXL, DXL and BXL) and report bad
> > +     for AH, CH, DH and BH.  */
> 
> "report bad for" is a little odd; how about simply "reject"?
> 
Done.

> Okay with respective adjustments.

Thanks,
Lili.