Re: PATCH: Add SSE4.2 support - pcmpstr part

[Uros Bizjak <ubizjak at gmail dot com>]

H. J. Lu wrote:

> > Also, I see no reason, why we need two new CC modes. CCmode mode is
> > enough to use as a generic comparison mode.
> >     
>
> The new CC modes are due to __builtin_ia32_pcmp?str?o128, which
> checks OF bit, and __builtin_ia32_pcmp?str?s128, which checks SF
> bit. Those aren't covered by the normal CC modes. There are no
> RTX codes for them. That is I invented 2 modes and reuse UNGT/UNLT.
> Besides you can't mix CC modes set by __builtin_ia32_pcmpistr?o128
> and __builtin_ia32_pcmpestr?o128 since they work on different things.
>   

This part is a bit tricky. You propose:

> -   Add CCZ to indicate that only the Zero flag is valid.  */
> +   Add CCZ to indicate that only the Zero flag is valid.
> +
> +   Add CCPCMPESTR/CCPCMPISTR for pcmp[ei]str[im] instructions:
> +   
> +       suffix        rtx_code
> +         a             GTU
> +         c             LTU
> +         o             UNGT    OF == 1
> +         s             UNLT    SF == 1
> +         z             EQ
> + */

But please note, that RTX code (that is present in the CC consumer) 
produces correct Jcc/SETcc insn only in combination with the mode of the 
CC register. We could re-write the table above according to existing 
suffixes from put_condition_code():

suffix        RTX code      mode
  a             GTU         CC
  c             LTU         CC
  o             EQ          CCO   (*)
  s             EQ          CCS   (*)
  z             EQ          CCZ

So, we currently have no mode that could generate "o" or "s" suffix, i.e 
"seto" or "jo".

Builtin expander should then generate pcmp?str pattern with correct 
CCmode and CC consumer with correct comparison code in correct mode. 
Like for example in following code:

--cut here--
int test(int a)
{
 if (a < 0)
   func1();
 else
   func2();
}
--cut here--

that expands to:
--cut here--
(insn 6 5 7 cc.c:3 (set (reg:CCGOC 17 flags)
       (compare:CCGOC (reg/v:SI 59 [ a ])
           (const_int 0 [0x0]))) -1 (nil)
   (nil))

(jump_insn 7 6 0 cc.c:3 (set (pc)
       (if_then_else (ge (reg:CCGOC 17 flags)
               (const_int 0 [0x0]))
           (label_ref 0)
           (pc))) -1 (nil)
   (expr_list:REG_BR_PROB (const_int 7326 [0x1c9e])
       (nil)))
--cut here--

Uros.