Update to C++20, spack 1.2.0 and package updates - #1912
Conversation
This reverts commit 86c04bc.
…equires openmp flags, guard old usage
…ave both mfem and caliper enabled
…orking, but requires two in-flight raja PRs
…local nanobind recipe
| "$comment": [ | ||
| "Overlay of the pinned upstream hdf5 manifest used by uberenv's in-place vcpkg ports copy.", | ||
| "Upstream hdf5 defaults are [\"szip\", \"zlib\"]. We intentionally drop \"szip\" from default-features here", | ||
| "but keep the explicit \"szip\" feature below so users can still opt in when needed.", | ||
| "This works around vcpkg pulling libaec through hdf5's default szip feature in Axom's Windows GHA builds.", | ||
| "The problem was not libaec compatibility; the relevant issue was that the libaec source download repeatedly", | ||
| "failed from its upstream archive URL with HTTP 429 in GitHub Actions, making Windows dependency bootstrap flaky." |
There was a problem hiding this comment.
Needed to add a vcpkg.json file for hdf5, was hitting a 429 Error when trying to pull one of hdf5's dependencies.
Error was happening on this branch, as well as develop.
It doesn't look like the dependency was needed for axom, here's the Windows CI run output:
https://github.com/llnl/axom/actions/runs/30129087714
kennyweiss
left a comment
kennyweiss
left a comment
There was a problem hiding this comment.
Thanks for this comprehensive update @bmhan12 !
| mfem: | ||
| require: | ||
| - spec: "@4.9" |
There was a problem hiding this comment.
Update: Fixed! It turns out that we had a comment in the mfem data file used for the test, that was not supported by the reader. Fixed by removing the comment.
| using SkipTag = primal::KnotVector<double>::SkipValidityChecks; | ||
| axom::ArrayView<const double> knots_view(&kv0[0], kv0.Size()); | ||
| const primal::KnotVector<double> kv(knots_view, kv0.GetOrder()); | ||
| const primal::KnotVector<double> kv(knots_view, kv0.GetOrder(), SkipTag {}); | ||
| if(!kv.isValid(true)) | ||
| { | ||
| SLIC_WARNING(axom::fmt::format("MFEM patch {} has an invalid knot vector: {}", patchId, kv)); | ||
| return MFEMReader::READ_FAILED; | ||
| } |
There was a problem hiding this comment.
I rolled back the changes to MFEMReader but added this extra check against a bad knot vector in the input file.
| 2 | ||
|
|
||
| controlpoints | ||
| # points (0,0), (0.5,0), (1,0) with weights (1,2,1) in homogeneous form (x*w, y*w, w) |
There was a problem hiding this comment.
This comment in an mfem mesh test file was the cause of the CUDA config failure -- the mfem reader does not support comments here!
C++20 cuda rehash workaround
4 participants
This PR:
versions.yamlfor list of changes)spack-packages, axom should update these to release versions (these changes should also be mirrored for Windows vcpkg build).rocm 6.4.3(same),rocm 7.2.1,cce@21.0.0for toss 4 craycuda@13.1.1for toss 4Relates to #1855 (still require updated umpire/raja/camp releases)
Closes #1741
Relevant spack-packages PR:
spack/spack-packages#5791