[Pachi] win32 build performance
paulsteyn1 at afrihost.co.za
Sat Jan 21 16:24:12 CET 2012
I've also been working on a windows build of Pachi using MinGW.
Unfortunately I haven't had much time for it for a while now, and there
were several issues that needed coding around, which I haven't resolved
I'd be quite interested to know how you solved a few issues, namely:
1) Did you use -mthreads option on MinGW, or any other threading
I'm still talking on the MinGW mailing list to determine exactly what
threading options one should use to port pthreads; there is no clear
consensus right now. (I'm building another ported app with MinGW, and
can't recall if -mthreads was necessary for Pachi or not, I know it was
used on at least one of them, if not both)
2) what did you use for inet_ntop?
Windows has an InetNtop function, but MinGW doesn't implement this. I
know Petr said this is not that important, but did you just remove it,
or recode it, or what?
3) What did you use for nanosleep?
There is no nanosleep on Windows. Petr again indicated that just using
Sleep should be fine. Is this what you used, or did you use
timeBeginPeriod or one of the other higher granularity alternatives?
What other sort of portability modifications did you do?
Are you planning on writing a HOWTO for it?
I was planning to look at this again, although now I might not need to.
Let me know if there is any other stuff you need help with, maybe
between us we can get more/better results.
On Sat, 21 Jan 2012 22:43:03 +0800 (CST)
linfk <linfk at 163.com> wrote:
> I managed to build a win32 version of Pachi today. With some
> portability modification to the code, I successfuly built the 9.0.1
> version with MinGW. Now It works on windows, but still has some
> performance problems. It runs on a pc with intel Q6600 cpu, 4G ram.
> With the following params: -t _1500 threads=4, max_tree_size=2048 the
> win32 version only reaches about 2000 games/s (500 games/s/thread).
> On the same PC, I also have a linux version of Pachi running in a
> VMware virtual machine. With the same params, even in a vm, it
> reaches 4000 games/s (1000 games/s/thread). I tried to tune up the
> optimization level option of the compiler, however any optimization
> make pachi crash when it's thinking. I also tried to compile with
> MSVC, but it seems MSVC lacks proper C99 support.
More information about the Pachi