Question about i386.md and 32-bit and 64-bit split and peephole rules

[Uros Bizjak]

Hello!

> As I noticed are some parts in i386.md existing for just for 32-bit,
> but will be used in 64-bit, too.
>
> For example:
>
> (define_split
>   [(set (match_operand:XF 0 "push_operand" "")
> 	(float_extend:XF (match_operand:SF 1 "fp_register_operand" "")))]
>   ""
>   [(set (reg:SI SP_REG) (plus:SI (reg:SI SP_REG) (match_dup 2)))
>    (set (mem:XF (reg:SI SP_REG)) (float_extend:XF (match_dup 1)))]
>   "operands[2] = GEN_INT (TARGET_128BIT_LONG_DOUBLE ? -16 : -12);")
>
> (define_split
>   [(set (match_operand:XF 0 "push_operand" "")
> 	(float_extend:XF (match_operand:SF 1 "fp_register_operand" "")))]
>   "TARGET_64BIT"
>   [(set (reg:DI SP_REG) (plus:DI (reg:DI SP_REG) (match_dup 2)))
>    (set (mem:DF (reg:DI SP_REG)) (float_extend:XF (match_dup 1)))]
>   "operands[2] = GEN_INT (TARGET_128BIT_LONG_DOUBLE ? -16 : -12);")
>
> The first section will be used on 32-bit and on 64-bit, the second
> just for 64-bit. Isn't this wrong? Should in the first define_split
> not explicit said "!TARGET_64BIT"?
> There some other places where this happens too, and for the ESP
> adjustments there seems to be present just the 32-bit version?.
>   

This is wrong, the first splitter should have !TARGET_64BIT. Even better 
approach would be to copy what MIPS does:

;; This mode iterator allows :P to be used for patterns that operate on
;; pointer-sized quantities.  Exactly one of the two alternatives will 
match.
(define_mode_iterator P [(SI "Pmode == SImode") (DI "Pmode == DImode")])

and use P iterator instead of SI and DI in one macroized pattern that 
substitutes both splitters.

Uros.