ctest failed on fedora s390x

[topazus]

+ /usr/bin/ctest --test-dir redhat-linux-build --output-on-failure --force-new-ctest-process -j2
Internal ctest changing into directory: /builddir/build/BUILD/spectra-1.1.0-build/spectra-1.1.0/redhat-linux-build
Test project /builddir/build/BUILD/spectra-1.1.0-build/spectra-1.1.0/redhat-linux-build
      Start  1: QR
      Start  2: Eigen
 1/28 Test  #1: QR ...............................   Passed    0.03 sec
      Start  3: Schur
 2/28 Test  #3: Schur ............................   Passed    0.57 sec
      Start  4: BKLDLT
 3/28 Test  #4: BKLDLT ...........................   Passed    0.75 sec
      Start  5: Arnoldi
 4/28 Test  #5: Arnoldi ..........................   Passed    0.00 sec
      Start  6: Orthogonalization
 5/28 Test  #6: Orthogonalization ................   Passed    0.01 sec
      Start  7: RitzPairs
 6/28 Test  #7: RitzPairs ........................   Passed    0.00 sec
      Start  8: SearchSpace
 7/28 Test  #8: SearchSpace ......................   Passed    0.00 sec
      Start  9: DenseGenMatProd
 8/28 Test  #9: DenseGenMatProd ..................   Passed    0.00 sec
      Start 10: DenseSymMatProd
 9/28 Test #10: DenseSymMatProd ..................   Passed    0.00 sec
      Start 11: SparseGenMatProd
10/28 Test #11: SparseGenMatProd .................   Passed    0.01 sec
      Start 12: SparseSymMatProd
11/28 Test #12: SparseSymMatProd .................   Passed    0.01 sec
      Start 13: SymEigs
12/28 Test  #2: Eigen ............................   Passed    2.90 sec
      Start 14: SymEigsShift
13/28 Test #14: SymEigsShift .....................   Passed    6.25 sec
      Start 15: HermEigs
14/28 Test #13: SymEigs ..........................   Passed    9.14 sec
      Start 16: GenEigs
15/28 Test #16: GenEigs ..........................   Passed    8.35 sec
      Start 17: GenEigsRealShift
16/28 Test #17: GenEigsRealShift .................   Passed   11.51 sec
      Start 18: GenEigsComplexShift
17/28 Test #15: HermEigs .........................   Passed   29.64 sec
      Start 19: SymGEigsCholesky
18/28 Test #18: GenEigsComplexShift ..............   Passed   28.45 sec
      Start 20: SymGEigsRegInv
19/28 Test #19: SymGEigsCholesky .................   Passed   25.32 sec
      Start 21: SymGEigsShift
20/28 Test #21: SymGEigsShift ....................   Passed    1.02 sec
      Start 22: SVD
21/28 Test #22: SVD ..............................   Passed    0.36 sec
      Start 23: JDSymEigsBase
22/28 Test #23: JDSymEigsBase ....................   Passed    0.00 sec
      Start 24: JDSymEigsDPRConstructor
23/28 Test #24: JDSymEigsDPRConstructor ..........   Passed    0.00 sec
      Start 25: DavidsonSymEigs
24/28 Test #25: DavidsonSymEigs ..................   Passed    3.23 sec
      Start 26: Example1
25/28 Test #26: Example1 .........................***Failed    0.00 sec

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Example1 is a Catch v2.13.8 host application.
Run with -? for options

-------------------------------------------------------------------------------
Example #144, (n, k, m) = (20, 5, 12)
-------------------------------------------------------------------------------
/builddir/build/BUILD/spectra-1.1.0-build/spectra-1.1.0/test/Example1.cpp:110
...............................................................................

/builddir/build/BUILD/spectra-1.1.0-build/spectra-1.1.0/test/Example1.cpp:69: FAILED:
  REQUIRE( diff == Approx(0.0).margin(1e-9) )
with expansion:
  0.5 == Approx( 0.0 )
with messages:
  nconv = 5
  niter = 1
  nops  = 13
  ||AU - UD||_inf = 9.28976e-16
  True eigenvalues =
   4.48574e-17
    0.0489435
    0.0489435
     0.190983
     0.190983
     0.412215
     0.412215
     0.690983
     0.690983
            1
            1
      1.30902
      1.30902
      1.58779
      1.58779
      1.80902
      1.80902
      1.95106
      1.95106
            2
  Estimated =
   1.30902
  1.58779
  1.80902
  1.95106
        2
  diff = 0.5

===============================================================================
test cases:  3 |  2 passed | 1 failed
assertions: 15 | 14 passed | 1 failed


      Start 27: Example2
26/28 Test #27: Example2 .........................   Passed    0.00 sec
      Start 28: Example3
27/28 Test #28: Example3 .........................   Passed    0.00 sec
28/28 Test #20: SymGEigsRegInv ...................   Passed   75.14 sec

96% tests passed, 1 tests failed out of 28

Total Test time (real) = 134.00 sec

The following tests FAILED:
	 26 - Example1 (Failed)

Ref: https://download.copr.fedorainfracloud.org/results/topazus/test/fedora-rawhide-s390x/08466419-spectra/builder-live.log.gz

[ariaydejawad]

My issue is that the HermEigs test takes so long to finish, so I am wondering if it is working – but just taking a little while to finish – or it is failing silently by being stuck in an infinite loop. So, we have similar issues.

I am using an aarch64 Ubuntu base image for building and testing SPECTRA. I using Docker and the Apple Virtualization Framework.

[yixuan]

Hi @topazus, thanks for reporting this issue. Unfortunately I could not reproduce this issue using my own machine, and it may be difficult to debug what was going on. One thing you could help me is to test whether older versions of Spectra would result in this error. If you can locate the commit that causes this issue, it would be very helpful. Thanks!

[yixuan]

@VicBoss1984 HermEigs is indeed very time-consuming, as can be seen in the log above. For local debugging, you can manually edit test/HermEigs.cpp, and comment out any large-scale test cases.

[ariaydejawad]

@yixuan Good news! All the tests ran successfully on my Apple M-series machine, and the longer tests that you mentioned do take longer to run than the others. I had no issues building and bundling spectra from source on my machine using docker, so I have no issues to report.

On an older note, I am still amazed by how quickly HermEigs ran on the OP's computer. OP's computer must be a mainframe of some kind with an insane number of CPU cores for the HermEigs to run that quickly.

[ariaydejawad]

@yixuan I tried running the SPECTRA tests again today, and got a singular failure on the GenEigsRealShift test. The detailed testing logs are as follows:

/workspace/program/gcta/dependencies/spectra/build/test # ./GenEigsRealShift

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
GenEigsRealShift is a Catch v2.13.8 host application.
Run with -? for options

-------------------------------------------------------------------------------
Eigensolver of general real matrix [100x100]
  Largest Real Part
-------------------------------------------------------------------------------
/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:89
...............................................................................

/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:61: FAILED:
  REQUIRE( eigs.info() == CompInfo::Successful )
with expansion:
  2 == 0
with messages:
  nconv = 9
  niter = 501
  nops  = 2708

-------------------------------------------------------------------------------
Eigensolver of general real matrix [100x100]
  Smallest Imaginary Part
-------------------------------------------------------------------------------
/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:105
...............................................................................

/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:50: warning:
  FAILED on this test

nconv = 9
niter = 501
nops  = 1809
-------------------------------------------------------------------------------
Eigensolver of general real matrix [1000x1000]
  Smallest Imaginary Part
-------------------------------------------------------------------------------
/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:105
...............................................................................

/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:50: warning:
  FAILED on this test

nconv = 17
niter = 501
nops  = 6099
-------------------------------------------------------------------------------
Eigensolver of sparse real matrix [10x10]
  Smallest Imaginary Part
-------------------------------------------------------------------------------
/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:105
...............................................................................

/workspace/program/gcta/dependencies/spectra/test/GenEigsRealShift.cpp:50: warning:
  FAILED on this test

nconv = 1
niter = 501
nops  = 1013
===============================================================================
test cases:  6 |  5 passed | 1 failed
assertions: 65 | 64 passed | 1 failed

The rest of the tests all run flawlessly, but this is the only one that reported any failures.

For further details, I have attached the docker build instructions for building the image I used to run these tests in and the full test log files, which include the reports of all the tests that were run.

LastTest.log

Dockerfile.txt