new sign/zero extension elimination pass

[Tom de Vries]

Paolo Bonzini wrote:
> On 10/21/2010 12:24 PM, Paolo Bonzini wrote:
>> On 10/18/2010 05:42 PM, Andrew Pinski wrote:
>>>>> For MIPS, compilation results in the following insns.
>>>>>
>>>>> (set (reg/v:SI 199)
>>>>> (sign_extend:SI (subreg:HI (reg:SI 200) 2)))
>>>>>
>>>>> ...
>>>>>
>>>>> (set (reg:QI 203)
>>>>> (subreg:QI (reg/v:SI 199) 3))
>>>>>
>>>>> These insns are the only def and the only use of reg 199, each 
>>>> located in a
>>>>> different bb.
>>> This sounds like a job for GCSE to do.
>> Actually, fwprop should _already_ do that if assuming simplify-rtx.c 
>> does the simplification of (subreg:QI (sign_extend:SI (subreg:HI (reg:SI 
>> 200) 2))) 3).
> 
> ... which this code should do in simplify_subreg:
> 
>   /* Optimize SUBREG truncations of zero and sign extended values.  */
>   if ((GET_CODE (op) == ZERO_EXTEND
>        || GET_CODE (op) == SIGN_EXTEND)
>       && GET_MODE_BITSIZE (outermode) < GET_MODE_BITSIZE (innermode))
>     {
>       unsigned int bitpos = subreg_lsb_1 (outermode, innermode, byte);
> 
>       /* If we're requesting the lowpart of a zero or sign extension,
>          there are three possibilities.  If the outermode is the same
>          as the origmode, we can omit both the extension and the subreg.
>          If the outermode is not larger than the origmode, we can apply
>          the truncation without the extension.  Finally, if the outermode
>          is larger than the origmode, but both are integer modes, we
>          can just extend to the appropriate mode.  */
>       if (bitpos == 0)
>         {
>           enum machine_mode origmode = GET_MODE (XEXP (op, 0));
>           if (outermode == origmode)
>             return XEXP (op, 0);
>           if (GET_MODE_BITSIZE (outermode) <= GET_MODE_BITSIZE (origmode))
>             return simplify_gen_subreg (outermode, XEXP (op, 0), origmode,
>                                         subreg_lowpart_offset (outermode,
>                                                                origmode));
>           if (SCALAR_INT_MODE_P (outermode))
>             return simplify_gen_unary (GET_CODE (op), outermode,
>                                        XEXP (op, 0), origmode);
>         }
> 
> However, the def of pseudo 200 is "complex enough" that fwprop will not want
> to propagate it unless it simplifies to a constant.
> 
> It should be enough to tell fwprop that such propagations are always fine
> if the destination pseudo has one use only.  In this case register pressure
> cannot increase.
> 
> Paolo

Paolo,

Thanks for the pointer to fwprop. I agree with you that for this example, it
would make sense to have this done in fwprop.

But, I still think we need the new pass. I don't see how fwprop would help in
the example mentioned in
http://gcc.gnu.org/ml/gcc-patches/2010-10/msg01796.html, where a superfluous
zero_extend:
...
(insn 10 9 11 2 ext13.c:5 (set (reg:SI 204 [ D.1987+-2 ])
        (zero_extend:SI (subreg:HI (reg:SI 213) 2))) 186 {*zero_extendhisi2}
(expr_list:REG_DEAD (reg:SI 213)
        (nil)))
...

is used by:
...
(insn 15 14 16 2 ext13.c:5 (set (reg:SI 217)
        (plus:SI (reg:SI 205 [ D.1988+-2 ])
            (reg:SI 204 [ D.1987+-2 ]))) 10 {*addsi3} (nil))

(insn 17 16 18 2 ext13.c:6 (set (reg:SI 218)
        (minus:SI (reg:SI 204 [ D.1987+-2 ])
            (reg:SI 205 [ D.1988+-2 ]))) 23 {subsi3} (expr_list:REG_DEAD (reg:SI
205 [ D.1988+-2 ])
        (expr_list:REG_DEAD (reg:SI 204 [ D.1987+-2 ])
            (nil))))
...

- Tom