Re: [boost] [Fusion] port to c++0x

Subject: Re: [boost] [Fusion] port to c++0x
From: Christopher Schmidt (mr.chr.schmidt_at_[hidden])
Date: 2009-10-03 16:41:12

Larry Evans schrieb:
> On 09/26/09 10:40, Christopher Schmidt wrote:
> [snip]
>> I used gcc 4.4.1 (TDM-1 mingw32) and ptime to measure performance. It 
>> is pretty interesting that the instantiation of the horizontal tuple 
>> type is just slightly slower. The at-operations differ all the more 
>> surprisingly.
>>
>> ===  g++ -std=c++0x -c test.cpp ===
>> Execution time: 5.274 s
>> ===  g++ -std=c++0x -c test.cpp -DAT ===
>> Execution time: 11.821 s
>> ===  g++ -std=c++0x -c test.cpp -DMETA_AT ===
>> Execution time: 12.161 s
>> ===  g++ -std=c++0x -c test.cpp -DAT -DMETA_AT ===
>> Execution time: 19.565 s
>> ===  g++ -std=c++0x -c test.cpp -DVERTICAL ===
>> Execution time: 4.884 s
>> ===  g++ -std=c++0x -c test.cpp -DAT -DVERTICAL ===
>> Execution time: 7.136 s
>> ===  g++ -std=c++0x -c test.cpp -DMETA_AT -DVERTICAL ===
>> Execution time: 5.835 s
>> ===  g++ -std=c++0x -c test.cpp -DAT -DMETA_AT -DVERTICAL ===
>> Execution time: 8.156 s
> [snip]
> There's now a tuple.benchmark.zip in boost vault under the
> variadic_templates directory.  It uses the compiler's
> -ftime-report flag to report times.  The zip file also
> contains a short makefile (suffix .mk).
> 
> The report shows a large difference between gcc4.4 and
> gcc4.5 in the HORIZONTAL implementation.  It shows
> with 4.5 there's not nearly as large a difference
> compared to the VERTICAL implementation.
> 
> -regards,
> Larry

Amazing results. I guess when it comes to gcc 4.5, the overall 
complexity of the code rather than the raw number of template 
instantiations ought be the quantum to measure compile-time performance.
Thanks for all your work!

-Christopher