Hi Uros, I send you new patch with fixed space/tab alignments.
About your comment.
It is more optimal to put adding of constant before adding of the
register only for case when 3 instructions must be generated to split
lea. In all other cases it does not matter and I left code
unchangeable.
Best regards.
Yuri.
2012/8/14 Uros Bizjak <ubiz...@gmail.com>:
> On Tue, Aug 14, 2012 at 3:35 PM, Yuri Rumyantsev <ysrum...@gmail.com> wrote:
>> Uros,
>>
>> Let me try to explain you why I used such code duplication:
>>
>> Here we have a common case of LEA with 3 different registers - r0
>> (target), r1(base), r2(index) and possible offset.
>> To get the better scheduling we first try to determine what register
>> is prefirable for inititial setting - r1 or r2 through
>> find_nearest_reg_def. And then we generate the following sequence of
>> instructions:
>> r0 = r_best;
>> r0 = $const, r0
>> r0 = r_worse, r0
>> that can save 2 cycles for Atom since first 2 instructions can be hoisted up.
>> I could not find better way for coding it.
>
> If it is important to put adding of const before adding of the
> register, then you can emit similar sequence for other cases, too.
> Something like following:
>
> --cut here--
> ...
> {
> if (regno0 == regno1)
> tmp = parts.index;
> else if (regno0 == regno2)
> tmp = parts.base;
> else
> {
> rtx tmp1;
>
> /* regno1: base, regno2: index */
> if (find_nearest_reg_def (insn, regno1, regno2))
> tmp1 = parts.index, tmp = parts.base;
> else
> tmp1 = parts.base, tmp = parts.index;
>
> emit_insn (gen_rtx_SET (VOIDmode, target, tmp1));
> }
>
> if (parts.disp && parts.disp != const0_rtx)
> ix86_emit_binop (PLUS, mode, target, parts.disp);
> ix86_emit_binop (PLUS, mode, target, tmp);
> return;
> }
> --cut here--
>
>>>> I prepared new patch and ChangeLog. Testing of x32 is in progress.
>
> You didn't fix vertical spaces and tab issues in new patch.
>
> Uros.
lea_split_improve.diff
Description: Binary data
