> Now with patch
>
> On Fri, May 11, 2012 at 8:42 PM, Steven Bosscher <[email protected]>
> wrote:
>>>>> 2012-05-10 Vladimir Makarov<[email protected]>
>>>>>
>>>>> PR rtl-optimization/53125
>>>>> * ira.c (ira): Call find_moveable_pseudos and
>>>>> move_unallocated_pseudos if only ira_conflicts_p is true.
>> And the attached patch fixes the reginfo slowness.
>>
>> The reginfo pass was doing:
>>
>> for each bb:
>> for each insn in bb, from BB_END(bb) to BB_HEAD(bb):
>> for each reg that died between BB_END(bb) and insn: // i.e.
>> register is live at insn
>> REG_LIVE_LENGTH(reg)++
>>
>> With very large basic blocks, and the kind of code for the test case
>> of the PR results in many live registers for SPARC (for x86_64 there
>> are far fewer live registers). For SPARC, there are O(1e5) insns and
>> O(1e4) registers live for most of the basic block. That is effectively
>> almost quadratic behavior in the number of insns per basic block.
>>
>> But the above loop is a bit silly. It is much easier and
>> computationally cheaper to just remember at what insn reg died (last
>> used), and add to REG_LIVE_LENGTH the distance from the insn that sets
>> reg to the insn that used reg.
Yes, this makes a lot of sense. Patch ok with one typo fixed.
>> It turns out that (before or after the patch) partial or conditional
>> sets never kill a register, so that REG_LIVE_LENGTH for registers set
>> by partial or conditional stores is not very accurate.
But this is correct, isn't it? The live range for such registers is
indeed extending before and after the store.
> @@ -310,22 +322,26 @@ regstat_bb_compute_ri (unsigned int bb_i
>
> if (bitmap_set_bit (live, uregno))
> {
> - /* This register is now live. */
> + /* This register is now live. Begin to process it locally.
>
> - /* If we have seen this regno, then it has already been
> - processed correctly with the per insn increment. If
> - we have not seen it we set the bit so that begins to
> - get processed locally. Note that we don't even get
> - here if the variable was live at the end of the block
> - since just a ref inside the block does not effect the
> - calculations. */
> + Note that we don't even get here if the variable was live
> + at the end of the block since just a ref inside the block
> + does not effect the calculations. */
> REG_LIVE_LENGTH (uregno) ++;
> + local_live_last_luid[uregno] = luid;
> bitmap_set_bit (local_live, uregno);
> bitmap_set_bit (local_processed, uregno);
> }
> }
> }
>
> + /* Add the liveness length to all registers that were used somewhere
> + in this bock, but not between that use and the head of this block. */
Typo is here ("bock"->"block").
Paolo
