[x265] Suggested patch to fix multithreaded writing of 2-pass stats in Win32 target

[Steve Borho]

On Sat, Jul 25, 2015 at 1:21 PM, Mario Rohkrämer <contact at ligh.de> wrote:
> To clarify: It is supposed to fix a flaw in MinGW using non-atomic legacy
> MSVCRT output routines.
>
>
>
> Am 25.07.2015, 20:17 Uhr, schrieb Mario Rohkrämer <contact at ligh.de>:
>
>> People discussed an additional lock on a stats file output to avoid
>> corruption with many threads:
>>
>> http://forum.doom9.org/showthread.php?p=1731547#post1731547

FWIW: I'm not certain the problem is necessarily in the atomic nature
of the fprintf()
calls. We're not ensuring the order of the last 2-3 frames correctly
so making the fprintf blocking (and thus slow) is mostly hiding our
bug.

-- 
Steve Borho