Here is the corrected version of change. Also, I think, I need
write-after-approval access to commit the change.
thanks, Dinar,
PR rtl-optimization/57268
* sched-deps.c (sched_analyze_2): Flush dependence lists if
the sum of the read and write lists exceeds MAX_PENDING_LIST_LENGTH.
On Wed, May 29, 2013 at 6:49 PM, Jeff Law <[email protected]> wrote:
> On 05/29/2013 06:52 AM, Steven Bosscher wrote:
>>
>> Hello,
>>
>> On Wed, May 29, 2013 at 2:01 PM, Dinar Temirbulatov wrote:
>>>
>>> Hi,
>>> I noticed that the scheduler created long dependence list about ~9000
>>> elements long and slowed compilation time for about an hour. Attached
>>> patch flushes the dependence list is case of it is longer than
>>> MAX_PENDING_LIST_LENGTH. Tested with gcc testsite on x86_64-linux-gnu
>>> with c and c++ enabled. Ok for trunk?
>>> thanks, Dinar.
>>>
>>> 2013-05-28 Dinar Temirbulatov <dinar at kugelworks dot com>
>>>
>>> PR rtl-optimization/57268
>>> * sched-deps.c (sched_analyze_2): Flush dependence list
>>> then it is longer than MAX_PENDING_LIST_LENGTH.
>>
>>
>>
>> * sched-deps.c (sched_analyze_2): Flush dependence lists if
>> the sum of the read and write lists exceeds
>> MAX_PENDING_LIST_LENGTH.
>>
>>
>>
>>> if (!deps->readonly)
>>> - add_insn_mem_dependence (deps, true, insn, x);
>>> + {
>>> + if ((deps->pending_read_list_length +
>>> deps->pending_write_list_length)
>>> + > MAX_PENDING_LIST_LENGTH)
>>> + flush_pending_lists (deps, insn, true, true);
>>> + add_insn_mem_dependence (deps, true, insn, x);
>>> + }
>>
>>
>> The "flush_pending_lists", "add_insn_mem_dependence" and "}" lines are
>> not indented correctly. The if (...+...) line is too long (max. 80
>> characters per line). The GCC style would be
>>
>> if (!deps->readonly)
>> {
>> if ((deps->pending_read_list_length
>> + deps->pending_write_list_length)
>> > MAX_PENDING_LIST_LENGTH)
>> flush_pending_lists (deps, insn, true, true);
>> add_insn_mem_dependence (deps, true, insn, x);
>> }
>>
>> (The aesthetics of GCC code style is a matter for debate, but not here
>> and now ;-)
>
> And just to be clear, with Steven's suggested changes, this patch is OK.
>
> jeff
