Re: [Patch, AVR]: Fix PR46779

[Georg-Johann Lay <avr@gjlay.de>]

Denis Chertykov schrieb:
> 2011/6/24 Richard Henderson <rth@redhat.com>:
> 
>>On 06/23/2011 01:15 PM, Denis Chertykov wrote:
>>
>>>>  text    data     bss     dec     hex filename
>>>> 10032      25       0   10057    2749 bld-avr-orig/gcc/z.o
>>>>  5816      25       0    5841    16d1 bld-avr-new/gcc/z.o
>>>
>>>Richard, can you send me this z.c file ?
>>>Right now I'm notice that new code is worse.
>>
>>That's gcc.c-torture/compile/950612-1.c.
> 
> I have founded that postreload optimizations can't handle results of
> new L_R_A code.
> I think that it's can be handled by CSE (postreload).
> 
> This is a fragment from 950612-1.c:
> 
> ---------- you can skip it I have a short version below -------
> (insn 5186 23 5187 2 (set (reg:HI 30 r30)
>         (const_int 128 [0x80])) c950612-1.c:20 9 {*movhi}
>      (nil))
> 
> (insn 5187 5186 5189 2 (set (reg:HI 30 r30)
>         (plus:HI (reg:HI 30 r30)
>             (reg/f:HI 28 r28))) c950612-1.c:20 21 {*addhi3}
>      (expr_list:REG_EQUIV (plus:HI (reg/f:HI 28 r28)
>             (const_int 128 [0x80]))
>         (nil)))
> 
> (insn 5189 5187 2451 2 (set (mem/c:HI (plus:HI (reg:HI 30 r30)
>                 (const_int 1 [0x1])) [8 %sfp+129 S2 A8])
>         (reg:HI 18 r18)) c950612-1.c:20 9 {*movhi}
>      (nil))
> 
> (note 2451 5189 2537 2 NOTE_INSN_DELETED)
> 
> (note 2537 2451 1651 2 NOTE_INSN_DELETED)
> 
> (note 1651 2537 3180 2 NOTE_INSN_DELETED)
> 
> (note 3180 1651 5191 2 NOTE_INSN_DELETED)
> 
> (insn 5191 3180 5192 2 (set (reg:HI 30 r30)
>         (const_int 128 [0x80])) c950612-1.c:132 9 {*movhi}
>      (nil))
> 
> (insn 5192 5191 5071 2 (set (reg:HI 30 r30)
>         (plus:HI (reg:HI 30 r30)
>             (reg/f:HI 28 r28))) c950612-1.c:132 21 {*addhi3}
>      (expr_list:REG_EQUIV (plus:HI (reg/f:HI 28 r28)
>             (const_int 128 [0x80]))
>         (nil)))
> 
> (insn 5071 5192 5073 2 (set (mem/c:HI (plus:HI (reg:HI 30 r30)
>                 (const_int 3 [0x3])) [8 %sfp+131 S2 A8])
>         (reg/v/f:HI 8 r8 [orig:211 pc ] [211])) c950612-1.c:132 9 {*movhi}
>      (nil))
> ------------------------------------------------------
> 
> Insns 5186,5187 equal to 5191,5192 and 5191,5192 can be removed, but
> reload_cse_regs_1 operate only on one insn.
> 
> 5186 Z=128
> 5187 Z=Y+128       ; REG_EQUIV Z=Y+128
> 5189 HI:[Z+1]=HI:R18
> ...(deleted insns)
> 5191 Z=128
> 5192 Z=Y+128       ; REG_EQUIV Z=Y+128
> 
> (5191,5192) really a one three addressing add Z=Y+128.
> Insns (5191,5192) exists because AVR havn't 3 addressing add.
> Insn 5191 destroy REG_EQUIV (it's right), but reload_cse_regs_1 can't
> optimize insns 5191,5192.

Did you try to add constraint alternative to *addhi3?
Like "*!d,d,n" or even "*!r,r,n"

I saw some code improvement with that alternative.

Johann

> Right now I have only one idea:
> 1. create peephole2 for joining such insns (5191,5192);
> 2. inside machine dependent pass rerun postreload and may be gcse2;
> 3. split joined insns to originals.
> 
> Denis.