Benchmarks for find() algorithm #325 #444
Conversation
Boost.Compute, STL and Thrust benchmarks for find() algorithm.
perf/perf_stl_find.cpp
Outdated
There was a problem hiding this comment.
So I tried running this version a few times with varying data sizes but always got the same very small time. This usually means the compiler is optimizing out the call the std::find() entirely because the result is not actually used. I'd add something to store the result in a variable and compare it against host_vector.end().
There was a problem hiding this comment.
I don't have that issue. Are performance tests built with -O3 by default?
Nevertheless, I will commit fix immediately and to be sure it won't happen again with Thrust performance test I will add there the same check.
There was a problem hiding this comment.
Not by default, but when I'm running the performance benchmarks and comparing against the STL I enable optimizations (else the results are not very fair). I usually do this by setting CMAKE_BUILD_TYPE=Release which enables gcc -03.
There was a problem hiding this comment.
Thanks! That's exactly what I need to know. From now on I will use that setting when testing performance, especially against STL.
|
Minor issue with the STL version. After fixing that it should be good to merge. Thanks! |
Making sure that compiler won't cut out std::find() and thrust:find() calls while optimizing the code.
|
It's interesting that Travis build failed. I'm not sure why. |
Benchmarks for find() algorithm #325
|
Looks great! Merged! |
|
And don't worry about the |
2 participants
This addresses issue #325 .
I've added find() algorithm benchmarks for Boost.Compute, STL, Thrust. Benchmarks are based on worst-case scenario (searching for element that isn't in vector).