simde
simde copied to clipboard
simd-everywhere
Start planning for 0.8 or 1.0 release
I would like to there to be a new SIMDe release once Debian Bullseye is released
517 changed files with 219,189 additions and 13,853 deletions.
https://github.com/simd-everywhere/simde/compare/v0.7.2...master
I was thinking about putting out a 0.8.0 after GSoC is done; development will probably slow down significantly then.
Eventually I'm thinking it might be a good idea to move to a time-based release schedule.
This means that next release version is not 0.7.3 but 0.8 or 1.0, while currently 0.7.3 is actually set? https://github.com/simd-everywhere/simde/commit/36b65e3ea33776accda8cd316e6324ae9d0a9113
@junaruga Correct. The odd 0.0.x releases are development versions, not public releases.
Debian Bullseye has been released! I've uploaded a snapshot of today's SIMDe to Debian Experimental: https://tracker.debian.org/news/1247389/accepted-simde-0730git202108141910103f186a0-1-source-into-experimental/
I just found https://github.com/simd-everywhere/simde/issues/901 , so I've disabled testing with clang on amd64 for the moment; the rest of the builds will be visible at https://buildd.debian.org/status/package.php?p=simde&suite=experimental
I did another snapshot build today: https://buildd.debian.org/status/package.php?p=simde&suite=experimental is of https://github.com/simd-everywhere/simde/commit/90523a281a67a50b0892364bc35f37a5fbfb7479
Mostly GCC 11.2, some clang testing as well
dpbf16 is skipped on i686
https://salsa.debian.org/med-team/simde/-/blob/ca088f16e02a1ddf3a7194bf8610cbe0c0f28a05/debian/rules
I did a new build + test on all of the Debian architectures (official and unofficial) and unfortunately we've got test failures;
assistance in tracking these down would be appreciated ( @wrv are you available? )
https://buildd.debian.org/status/package.php?p=simde&suite=experimental
The top part of the table contains the officially supported architectures in Debian (amd64 through s390x); the bottom portion are the unofficial ports, and tests errors there will not block a release of SIMDe (though I have a soft spot for RISCV64)
Thank you @wrv ; some of the differences are likely due to the extra default compiler options used by Debian. I guess we should add those to the CI to catch things sooner :-)
Here are the extra flags on i386/i686: -O2 -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2
Ah, i686 was being tested on CircleCI; testing with docker shows 292 test failures ; yet Debian's i686 build only had 22 failures (likely to be the result of only a few lines wrong); maybe the -O2 optimized out some wrong paths?
Update from the Debian Med Sprint,
I've built an experimental package of the latest SIMDe code and I'm using it to rebuild over 25 tools across all the Debian architectures. If there are no regressions, then I'll make a 0.7.4 release
The build status with SIMDe 0.7.2
and with the latest dev version of SIMDe
If there are squares on the second link that are red where they are green on the first link, then there is a regression. Only the first 9 architectures are officially supported in Debian, so don't worry about alpha and onward.
(Give it a day or two for the builders to catch up)
How do you recommend I setup my local machine for testing these issues? Is it all in the docker/ directory?
How do you recommend I setup my local machine for testing these issues? Is it all in the docker/ directory?
Yeah, that's the easiest way for compilers that you don't have native system packages for.
I run it with PERSISTENT_BUILD_DIR=$PWD/build ./simde-dev.sh to save the build tree between invocation.
FYI, I'm going to make a 0.7.4 release, as the test builds with the Debian packages all passed; see https://github.com/simd-everywhere/simde/issues/983
Fixes for the other outstanding issues (especially clang) are very much still welcome!