Updating Branch from develop #4794
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
@trilinos/tpetra @trilinos/belos Trilinos' snapshot of kokkos-kernels in order to fix #4630. Thanks to Kyungjoo and Victor for feedback.
Currently, VerboseObject provides an option to allow output to be redirected to a file. Unfortunately, this feature results in every MPI process opening/writing the file, which is not typically what an OStream is used to accomplish. For file output, this implementation leads to extreme performance degradation with large parallel jobs. This patch accomplishes: It restricts file opening/writing to a single MPI process. The implementation of this is problematic, because VerboseObject's sublist reader has no access to any Comm. Ideally, the interface should be refactored to allow a comm to be passed, so that the logic can use the com to enforce which rank writes output. We will have to address the comm/ostream issues later.
In this case, user-facing executables were being added as tests, and the unit-test executable wasn't being added at all. Also fix a case where stk-balance was calling a couple of deprecated exodus functions. Also restrict the unit-tests that are run, since we don't yet have sufficient guards in place for insisting on what packages and tpls need to have been enabled...
@trilinos/tpetra Add functions in the Tpetra::Details namespace, that get a View of statically allocated memory. This is related to #4626, and is part of the work-around for the Intel compiler issues observed in PR #4648. I also added tests that build and pass on CUDA. The tests include a work-around for a Kokkos issue: kokkos/kokkos#2051.
@trilinos/tpetra
Add the following overloads of Tpetra::deep_copy:
- Copying from Teuchos::SerialDenseMatrix<int, ST> to
Tpetra::MultiVector<ST, LO, GO, NT>.
- Copying from Tpetra::MultiVector<ST, LO, GO, NT> to
Teuchos::SerialDenseMatrix<int, ST>.
These are useful for #4626. PR #4648 had functions that did the same
thing (but didn't work for nonconstant-stride MultiVectors, which
could matter in Belos). This commit formalizes the functions in
Tpetra, and adds tests. The tests exercise the constant-stride and
nonconstant-stride use cases, and build and pass with CUDA.
NOTE: This commit adds to Tpetra an optional TeuchosNumerics
subpackage dependency. Tpetra has implicitly assumed TeuchosNumerics
for a long time, so I don't think this will have practical effect.
However, it's better for us to state dependencies explicitly.
Comment out parameter names in function definitions to avoid `-Wunused-parameter` warnings.
…arameter-warnings Automatically Merged using Trilinos Pull Request AutoTester PR Title: Ifpack2: Remove Unused Parameter Warnings PR Author: jmgate
@trilinos/tpetra
@trilinos/tpetra
the --force-clean flag had two conflicting names and the final make command (make -C %s clean) failed in an infinite recursion. I just modified it to cd to the directory first and then run 'make clean'
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Another step towards correct cmake files for stk... PR Author: alanw0
Comment out parameter names in function definitions to avoid `-Wunused-parameter` warnings.
This is attempting to fix the error Albany is seeing.
Comment out parameter names in function definitions to avoid `-Wunused-parameter` warnings.
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Teuchos: VerboseObject fix filestreams PR Author: jjellio
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Tpetra: Fix #4729 PR Author: mhoemmen
…rameter-warnings Automatically Merged using Trilinos Pull Request AutoTester PR Title: Pamgen: Remove Unused Parameter Warnings PR Author: jmgate
@trilinos/belos Use "default" to make destructors and (some, where it makes sense) constructors for Tpetra-specific solver classes.
@trilinos/tpetra
@trilinos/tpetra - Fix how MultiVector computes strides. - Fix #4633 (use create_mirror_view, not create_mirror, in the one MultiVector constructor that takes a View instead of a DualView) - Clean up code.
@trilinos/tpetra Tpetra::MultiVector::reduce now gets the right answer when the MultiVector was created from a Kokkos::DualView with stride(1) > extent(0). I added unit tests for this.
This reduces the overall ninja parallel build level from 64 to 32 processes and the parallel link level from 64 to 16.
This reduces the overall ninja parallel build level from 64 to 32 processes and the parallel link level from 64 to 16.
Fix warning that will be generated on the next stk snapshot.
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Tempus fix fpes PR Author: ikalash
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Tpetra: Start working around Intel compiler issues noticed in PR #4648 PR Author: mhoemmen
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Panzer: another intel warning PR Author: rppawlo
…t square sense.
the system matrix can in fact be close to singular.
This should fix issue #4738
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Merge atdm-nightly to 'develop' to avoid future merge commits the other way (3/29/2019) PR Author: bartlettroscoe
@trilinos/belos @trilinos/tpetra Make the Tpetra specialization of Belos::MultiVecTraits use static CUDA allocations for temporary local MultiVectors (e.g., for Tpetra::MultiVector::multiply results). We've seen this improve GPU performance in solves. The fix always creates contiguous allocations, so it does not depend on the fix for #4639 in PR #4747 (merged into develop yesterday).
…ob-pool-limits Automatically Merged using Trilinos Pull Request AutoTester PR Title: Use ninja job pool limits to fix cuda+rdc+static builds on 'ride' and 'waterman' (#4502) PR Author: bartlettroscoe
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Intrepid2: Use LAPACK GELS instead in GESV for HDIV projection (issue #4738) PR Author: mperego
Now that the system automatically adjusts for all Primary Tested packages with the '-pt' and '_pt' matching, we don't need the file ATDMDevEnvAllPtPackages.cmake anymore. Also, we don't need the build: * Trilinos-atdm-white-ride-cuda-9.2-gnu-7.2.0-rdc-release-debug-pt-pbp.sh anymore because the all-at-once build will pass now.
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Belos: Fix #4626 (Tpetra specialization of MultiVecTraits) PR Author: mhoemmen
…ix-ctest-s-pt-build Automatically Merged using Trilinos Pull Request AutoTester PR Title: Remove file ATDMDevEnvAllPtPackages.cmake (#4502) PR Author: bartlettroscoe
Needed for, e.g., unit-tests in codes that need to recreate the parameter library every test, but use a singleton library.
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Set Werror for Intrepid2 in ICC build PR Author: ZUUL42
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Sacado: Add method to clear parameter library. PR Author: etphipp
Previously the LOCA Stepper didn't compute any eigenvalues on the final target step when LOCA attempts to hit the parameter bound. I added another parameter, "Compute Eigenvalues On Target Step", to enable this. I didn't just use the existing parameter, "Compute Eigenvalues", because that would change LOCA's default behavior when computing eigenvalues, and was unsure of the consequences for existing applications.
…get_step Automatically Merged using Trilinos Pull Request AutoTester PR Title: LOCA: Add parameter and logic to compute eigenvalues on target step. PR Author: etphipp
Automatically Merged using Trilinos Pull Request AutoTester PR Title: change the atdm sems-rhel7 env to use mkl-18.0.5 PR Author: fryeguy52
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Werror domi fix PR Author: ZUUL42
Automatically Merged using Trilinos Pull Request AutoTester PR Title: Kokkos-kernels: Fix #4630 PR Author: mhoemmen
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.