bitset count performance about stdlib HOT 10 CLOSED

fortran intrinsic function popcnt looks better. https://gcc.gnu.org/onlinedocs/gfortran/POPCNT.html

from stdlib.

Looks like, but then why is it not used in bit_count_large()?

from stdlib.

If I remember well, bitset_large was developed as an externsion of bitset_64. For bitset_64, popcnt cannot be used because some bits may not be used. The solution with btest() is therefore legitimate. The extension of bit_count for bitset_large from bitset_64 was straighfoward.

However, I agree that @PierUgit 's solution or using popcnt in this loop would be a more efficient implementation.

@PierUgit , do you still have such an implementation (i recognize that my answer is quite late!)? If so, would you like to open a PR with your solution/optimization?

from stdlib.

@jvdp1 I don't have a code ready in Github, I did some tests independently on a code of mine... To be honest I have first to learn how to use fpm and build the stdlib before being able to submit PRs here...

from stdlib.

@jvdp1 I don't have a code ready in Github, I did some tests independently on a code of mine...

Is it ok for you if I submit a PR and ask you for reviewing/testing it?

To be honest I have first to learn how to use fpm and build the stdlib before being able to submit PRs here...

I usually use CMake for building stdlib. However, I agree that the doc is not clear enough for building stdlib with fpm. I believe that @jalvesz will submit soon a PR with more detailed information for building stdlib with fpm. Your review of this proposal (if you want of course) could be useful too.

from stdlib.

It was actually much easier than expected! I could build stdlib with fpm, modify the code, and add a test case (this routine was not tested). I will submit a PR soon.

I did it from the stdlib-fpm branch, though, and it looks quite behind the master branch...

from stdlib.

It was actually much easier than expected! I could build stdlib with fpm, modify the code, and add a test case (this routine was not tested). I will submit a PR soon.

I did it from the stdlib-fpm branch, though, and it looks quite behind the master branch...

Actually development should be always from the master branch. The branch stdlib-fpm is just generated for the users. It contains all .f90 generated from the .fypp files, because fpm does not support the fypp preprocessor yet. But it seems fine for your PR.

from stdlib.

Probably I should close the PR and restart from the master branch...

from stdlib.

Probably I should close the PR and restart from the master branch...

Ok. Indeed, probably better and easier.

from stdlib.

Regarding the build with fpm, what I do and want to include as info in the readme is the following:

source ./ci/fpm-deployment.sh
cd stdlib-fpm/
fpm test --profile release

The fpm-deployment.sh scrit takes care of processing the files and creating the stdlib-fpm which contains only .f90

BTW, in my PR regarding str2num I included a change to this script as we realized that the .fypp files in the test folder were not processed.

from stdlib.