[llvm-commits] [LLVM, opt, LoopUnswitch] Compile-time improvements.

Stepan Dyatkovskiy STPWORLD at narod.ru
Tue Jan 10 09:51:05 PST 2012


Ping.

-Stepan.

09.01.2012, 22:00, "Stepan Dyatkovskiy" <STPWORLD at narod.ru>:
> Ping.
>
> -Stepan.
>
> 04.01.2012, 12:16, "Stepan Dyatkovskiy" <STPWORLD at narod.ru>:
>
>>  Ping.
>>  -Stepan
>>
>>  03.01.2012, 22:50, "Stepan Dyatkovskiy" <STPWORLD at narod.ru>:
>>>   Ping.
>>>   -Stepan.
>>>
>>>   02.01.2012, 13:53, "Stepan Dyatkovskiy" <STPWORLD at narod.ru>:
>>>>    ping.
>>>>    -Stepan.
>>>>
>>>>    30.12.2011, 16:14, "Stepan Dyatkovskiy" <stpworld at narod.ru>:
>>>>>     Hi. A made some fixes that improves compile-time:
>>>>>
>>>>>     1. Size heuristics changed. Now we calculate number of unswitching
>>>>>     branches only once per loop.
>>>>>     2. Some checks was moved from UnswitchIfProfitable to
>>>>>     processCurrentLoop, since it is not changed during processCurrentLoop
>>>>>     iteration. It allows decide to skip some loops at an early stage.
>>>>>
>>>>>     I checked the compile-time on test
>>>>>
>>>>>     MultiSource/Benchmarks/Prolangs-C++/shapes/shapes
>>>>>     (there was compile time regression after my previous patch).
>>>>>
>>>>>     Relative to my previous patch the compile-time improved on ~8.5%. Relative
>>>>>     to old revisions (before r146578) the compile time is improved on ~2%.
>>>>>
>>>>>     Please find the patch in attachment for review.
>>>>>
>>>>>     -Stepan.
>>>>>
>>>>>     _______________________________________________
>>>>>     llvm-commits mailing list
>>>>>     llvm-commits at cs.uiuc.edu
>>>>>     http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



More information about the llvm-commits mailing list