search

evaleev / libint

Libint: high-performance library for computing Gaussian integrals in quantum mechanics
Other
218 stars 96 forks source link

2.8.0 release #298

Closed evaleev closed 4 months ago

evaleev commented 6 months ago

this is to track issues with the just-released 2.8.0.

cc: @loriab

loriab commented 6 months ago

So far all's well. My toy conda builds all check out, and the production unix ones are building now.

Only thing I've noticed is that https://github.com/evaleev/libint/blob/master/export/cmake/CMakeLists.txt.export#L220 should be set back to TRUE. According to the boxed note in https://cmake.org/cmake/help/latest/prop_tgt/UNITY_BUILD.html#unity-build, it should be left under user control. But given it can speed up builds by 4x, I'm for imposing it on.

loriab commented 6 months ago

Also seem to have lost the definition of ncoords here https://github.com/evaleev/libint/blob/master/tests/hartree-fock/hartree-fock%2B%2B.cc#L692 as const auto ncoords = 3 * atoms.size(); . That only shows up for non-toy configurations.

evaleev commented 6 months ago

Only thing I've noticed is that https://github.com/evaleev/libint/blob/master/export/cmake/CMakeLists.txt.export#L220 should be set back to TRUE. According to the boxed note in https://cmake.org/cmake/help/latest/prop_tgt/UNITY_BUILD.html#unity-build, it should be left under user control. But given it can speed up builds by 4x, I'm for imposing it on.

ouch, indeed, this looks like an accidental regression. Thanks, will fix

JonathonMisiewicz commented 6 months ago

Only thing I've noticed is that https://github.com/evaleev/libint/blob/master/export/cmake/CMakeLists.txt.export#L220 should be set back to TRUE. According to the boxed note in https://cmake.org/cmake/help/latest/prop_tgt/UNITY_BUILD.html#unity-build, it should be left under user control. But given it can speed up builds by 4x, I'm for imposing it on.

ouch, indeed, this looks like an accidental regression. Thanks, will fix

That's right, that get changed but didn't get restored when we were trying to get σ working on Windows.

evaleev commented 6 months ago

Also seem to have lost the definition of ncoords here https://github.com/evaleev/libint/blob/master/tests/hartree-fock/hartree-fock%2B%2B.cc#L692 as const auto ncoords = 3 * atoms.size(); . That only shows up for non-toy configurations.

ouch, fixed.

Progressing to 2.8.1 ...

loriab commented 6 months ago

All OS packages (linux, mac-intel, mac-silicon, windows) built cleanly at conda-forge. L, M, W conda packages were clean for partial Psi4 testing. Psi4 full tests with L package was clean. In summary, no further issues found for v2.8.0. I'll generate a high AM pkg overnight.

JonathonMisiewicz commented 6 months ago

My attempts to compile MPQC with 2.8.0 failed with include errors. Need #302 merged in to fix them.

loriab commented 6 months ago

Continuing this issue for the v2.8.1 release: