2010/3/23 Konstantin Tokarev <annu...@yandex.ru>:
>
>
> 23.03.10, 20:30, "Tim Vandermeersch" <tim.vandermeer...@gmail.com>:
>
>> On Mon, Mar 22, 2010 at 7:13 PM, Konstantin Tokarev wrote:
>> >
>> >
>> > 22.03.10, 18:56, "Tim Vandermeersch" :
>> >
>> >> On Sun, Mar 21, 2010 at 12:48 PM, Konstantin Tokarev wrote:
>> >> > Are there any perforamnce benchmark scripts for OB? E.g., I'd like to
>> test if OB compiled by Intel is faster than by GCC or if -O3 optimization
>> with Intel is more effective than -O2
>> >> Are there specific parts of the code you want to test? For
>> >> example,OBMol::ConnectTheDots could be made faster for large
>> >> molecules.
>> >
>> > Yes, it's quite slow. But it's a problem of algorithm, not compiler
>> >
>> >>I don't know if anyone has been benchmarking OB code (apart
>> >> from force field code).
>> >> On linux, you can always use the time command to check how long
>> >> something takes. We could also add some macros to obtest.h to do
>> >> benchmarking.
>> >
>> > I know. I wondered if there are some "ready" benchmarks to do less work
>>
>> In openbabel trunk, I added test/obbench.h to provide some macros for
>> performance benchmarking. It provides millisecond precision but
>> converts to minutes/seconds when needed. Using it is very simple:
>>
>> OB_BENCHMARK {
>> // code to benchmark here...
>> }
>>
>> Multiple iterations are performed when a single iteration takes less
>> than 500ms (could be changed, this is an idea taken from Qt's
>> QBENCHMARK).
>>
>> I also added a feature I miss in QBENCHMARK: A way to track benchmarks
>> over time. It is still very basic but the idea is that the benchmark
>> results get saved to a benchmark.txt file. The formatting of the file
>> allows the content to be copied to a spreadsheet for analysis. To use
>> this, a different macro is used:
>>
>> OB_NAMED_BENCHMARK("Some name identifying the benchmark") {
>> // code....
>> }
>>
>> With some cmake magic, it should be possible to record compiler and
>> compiler options. Otherwise, date/time could be recorded.
>>
>> Feedback is welcome. I didn't add this to test/CMakeLists.txt since I
>> didn't want to break anything. It should work though. I've tested here
>> on linux and windows XP.
>>
>> Cheers,
>> Tim
>>
>>
>
> Great! Will try later
> Maybe it's reasonable to run all available tests to check performance of all
> modules. If some module slows down with -O3, "detunings" could be added
The header is working now, see the link below for an example graph
from openoffice calc. I'll see if I can add some more useful
benchmarks but this is just reading a lot of files or generating a
fastsearch index like already suggested. One last remark about -O3: it
is known it can create bugs... That's why we usually use -O2 AFAIK.
link: http://imagebin.org/90060
Cheers,
Tim
>> > --
>> > Regards,
>> > Konstantin
>> >
>>
>>
>
> --
> Regards,
> Konstantin
>
------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_______________________________________________
OpenBabel-Devel mailing list
OpenBabel-Devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/openbabel-devel
