[PATH] AMD MWAITX enablement

[Pawar, Amit]

-----Original Message-----
From: H.J. Lu [mailto:hjl.tools@gmail.com] 
Sent: Friday, June 26, 2015 8:34 PM
To: Pawar, Amit
Cc: binutils@sourceware.org; Jan Beulich
Subject: Re: [PATH] AMD MWAITX enablement

Please don't top post.
[Pawar, Amit] Thank you and will take care. Please check inline comments.

On Fri, Jun 26, 2015 at 8:00 AM, Pawar, Amit <Amit.Pawar@amd.com> wrote:
>
> Comments are fixed and please suggest how to allow following instructions without operands
>
> --- Required ----
> label:
> monitorx
> mwaitx
> --------------------
>
> ---- Accepted----
> monitorx %eax, %ecx, %edx
> mwaitx %eax, %ecx, %ebx
> --------------------
> by adding following code
>
> if (i.operands != 3)
>   abort ();
>
>
> -----Original Message-----
> From: H.J. Lu [mailto:hjl.tools@gmail.com]
> Sent: Friday, June 26, 2015 5:26 PM
> To: Pawar, Amit
> Cc: binutils@sourceware.org; Jan Beulich
> Subject: Re: [PATH] AMD MWAITX enablement
>
> On Fri, Jun 26, 2015 at 4:09 AM, Pawar, Amit <Amit.Pawar@amd.com> wrote:
>>
>> Operand check is required at that line as instructions MONITORX/MWAITX are accepted without operands also.
>
> Add

You have

if (i.tm.cpu_flags.bitfield.cpumwaitx && i.operands > 0)
  {
....
  }
[Pawar, Amit] This condition was included to handle mwaitx cpu instructions with more than zero 
operands and skips for zero operands.

If i.operands == 0, you won't reach
>[Pawar, Amit] Zero operands instruction are allowed but should not enter here as it will get 
handled by normal code.

if (i.operands != 3)
  abort ();
[Pawar, Amit] Will throw error for zero operand instructions, as this is inside if case and which is 
not required here.

Am I missing something?
[Pawar, Amit] Please let me know if you need any more information.

> if (i.operands != 3)
> abort ();
>
> +      for (x = 0; x < 2; x++)
> +        {
>           ^^^ Remove it.
>
> +          if (register_number (i.op[x].regs) != x)
>
> Use
>                 goto bad_register_operand;
>
> +            {
> +              as_bad (_("can't use register '%s%s' as operand %d in '%s'."),
> +                      register_prefix, i.op[x].regs->reg_name, x + 1,
> +                      i.tm.name);
> +            }
> +        }
> +
> +      /* Check for third operand for mwaitx/monitorx insn*/
> +      if ( register_number (i.op[2].regs) != 2 +
> (i.tm.extension_opcode == 0xfb))
>             ^ Remove extra space.
>
> This line is too lone. Please break it into 2 lines:
>
> if (register_number (i.op[x].regs)
>     != x + (i.tm.extension_opcode == 0xfb))
>
> +        {
>
> Add a label here
>
> bad_register_operand:
> +          as_bad (_("can't use register '%s%s' as operand %d in '%s'."),
> +                  register_prefix, i.op[x].regs->reg_name, 3,
>
>         ^^^ Change it to x + 1.
> +                  i.tm.name);
> +        }
> +
> +      i.operands = 0;
> +    }
>
>>
>> -----Original Message-----
>> From: H.J. Lu [mailto:hjl.tools@gmail.com]
>> Sent: Friday, June 26, 2015 3:34 PM
>> To: Pawar, Amit
>> Cc: binutils@sourceware.org; Jan Beulich
>> Subject: Re: [PATH] AMD MWAITX enablement
>>
>> On Thu, Jun 25, 2015 at 10:46 PM, Pawar, Amit <Amit.Pawar@amd.com> wrote:
>>> PFA MWAITX fixed patch. OK to apply?
>>>
>>
>> Please change the operand check to
>>
>> if (i.tm.cpu_flags.bitfield.cpumwaitx)
>> {
>>   if ( i.operands != 3)
>> abort
>>
>> for (i = 0, i< 2; i++)
>> ...
>>
>> if (register_number (i.op[2].regs) != 2 + (opcode ==mwaitx ) ...
>> }
>>
>> --
>> H.J.
>
>
>
> --
> H.J.



-- 
H.J.