Failing tests when built with openblas on OSX

[jbaum98]

As described in NixOS/nixpkgs#45013, testing fflas-ffpack fails on OSX when built against openblas:

==============================================
   FFLAS-FFPACK 2.3.2: tests/test-suite.log
==============================================

# TOTAL: 33
# PASS:  11
# SKIP:  0
# XFAIL: 0
# FAIL:  22
# XPASS: 0
# ERROR: 0

.. contents:: :depth: 2

FAIL: test-fger
===============

FAIL test-fger (exit status: 139)

FAIL: test-ftrsv
================

Checking with Modular<double> mod 36289873
FAIL test-ftrsv (exit status: 139)

FAIL: test-ftrtri
=================

Checking with Modular<double> mod 131
Checking FTRTRI_Lower_Unit....................PASSED (0.031471)
Checking FTRTRI_Upper_Unit....................PASSED (0.017059)
Checking FTRTRI_Lower_NonUnit.................PASSED (0.023665)
Checking FTRTRI_Upper_NonUnit.................PASSED (0.021203)
Checking with Modular<double> mod 13
Checking FTRTRI_Lower_Unit....................PASSED (0.018172)
Checking FTRTRI_Upper_Unit....................PASSED (0.020317)
Checking FTRTRI_Lower_NonUnit.................PASSED (0.016548)
Checking FTRTRI_Upper_NonUnit.................PASSED (0.00784)
Checking with Modular<double> mod 33811
Checking FTRTRI_Lower_Unit....................PASSED (0.004757)
Checking FTRTRI_Upper_Unit....................PASSED (0.004347)
Checking FTRTRI_Lower_NonUnit.................PASSED (0.004994)
Checking FTRTRI_Upper_NonUnit.................PASSED (0.007178)
Checking with ModularBalanced<double> mod 131
FAIL test-ftrtri (exit status: 139)

FAIL: test-ftrmv
================

Checking with Modular<double> mod 646771
Checking FTRMV_Lower_NoTrans_Unit............PASSED (0.004181)
FAIL test-ftrmv (exit status: 139)

FAIL: test-ftrsm
================

Checking with Modular<double> mod 1547437
FAIL test-ftrsm (exit status: 139)

FAIL: test-ftrsm-check
======================

FAIL test-ftrsm-check (exit status: 139)

FAIL: test-ftrmm
================

Checking with Modular<double> mod 61
FAIL test-ftrmm (exit status: 139)

FAIL: test-fgemm
================

Checking ..........................Modular<double> mod 1747 ...  ** On entry to DGEMM  parameter number 10 had an illegal value
 ** On entry to DGEMM  parameter number 10 had an illegal value
 ** On entry to DGEMM  parameter number 10 had an illegal value
 ** On entry to DGEMM  parameter number 10 had an illegal value
 ** On entry to DGEMM  parameter number 10 had an illegal value
 ** On entry to DGEMM  parameter number 10 had an illegal value
FAIL
a   :1, b   : 0
m   :16, n   : 44, k   : 36
ldA :48, ldB : 47, ldC : 45
Error C[0,0]=813 D[0,0]=202
Error C[0,1]=205 D[0,1]=612
Error C[0,2]=256 D[0,2]=1713
Error C[0,3]=102 D[0,3]=494
Error C[0,4]=872 D[0,4]=631
Error C[0,5]=165 D[0,5]=699
Error C[0,6]=1299 D[0,6]=1459
Error C[0,7]=246 D[0,7]=70
Error C[0,8]=1737 D[0,8]=1069
Error C[0,9]=592 D[0,9]=1088
Error C[0,10]=401 D[0,10]=637
Error C[0,11]=396 D[0,11]=1058
Error C[0,12]=1587 D[0,12]=848
Error C[0,13]=1498 D[0,13]=838
Error C[0,14]=1695 D[0,14]=1059
Error C[0,15]=196 D[0,15]=974
Error C[0,16]=290 D[0,16]=1518
Error C[0,17]=1634 D[0,17]=1479
Error C[0,18]=213 D[0,18]=1132
Error C[0,19]=769 D[0,19]=1695
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
FAILED
FAIL test-fgemm (exit status: 1)

FAIL: test-fgemm-check
======================

FAIL test-fgemm-check (exit status: 139)

FAIL: test-lu
=============

FAIL test-lu (exit status: 139)

FAIL: test-pluq-check
=====================

FAIL test-pluq-check (exit status: 139)

FAIL: test-fsyrk
================

Checking with Modular<double> mod 5
FAIL test-fsyrk (exit status: 139)

FAIL: test-fsytrf
=================

FAIL test-fsytrf (exit status: 139)

FAIL: test-invert-check
=======================

 -q 131071 -n 1000 -i 3 -s 1534291159349551
FAIL test-invert-check (exit status: 139)

FAIL: test-rankprofiles
=======================

FAIL test-rankprofiles (exit status: 139)

FAIL: test-det-check
====================

init: 0.00863194s (0.02807 cpu) [1]
FAIL test-det-check (exit status: 139)

FAIL: test-echelon
==================

FAIL test-echelon (exit status: 139)

FAIL: test-charpoly
===================

FAIL test-charpoly (exit status: 139)

FAIL: test-charpoly-check
=========================

FAIL test-charpoly-check (exit status: 139)

FAIL: test-minpoly
==================

FAIL test-minpoly (exit status: 139)

FAIL: test-solve
================

FAIL test-solve (exit status: 139)

FAIL: test-fgemv
================

FAIL test-fgemv (exit status: 139)

============================================================================
Testsuite summary for FFLAS-FFPACK 2.3.2
============================================================================
# TOTAL: 33
# PASS:  11
# SKIP:  0
# XFAIL: 0
# FAIL:  22
# XPASS: 0
# ERROR: 0

[ClementPernet]

Hi,
Thanks for the report and sorry for the delay.
it took me a while to get a OSX box running. Unfortunately, I can not reproduce the bug.
I tried both the last releases (which are the one packaged in SageMath) and upstream givaro and fflas-ffpack. The whole test-suite passes on those two cases.

I'm using upstream Openblas (at a71923514fb9ad611fb87345b8cfa4383404b54f), and g++ provided by LLVM on a Darwin 13.4.0.

ciosx:tests ci$ uname -a
Darwin ciosx 13.4.0 Darwin Kernel Version 13.4.0: Wed Mar 18 16:20:14 PDT 2015; root:xnu-2422.115.14~1/RELEASE_X86_64 x86_64
ciosx:tests ci$ g++ -v
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/usr/include/c++/4.2.1
Apple LLVM version 6.0 (clang-600.0.51) (based on LLVM 3.5svn)
Target: x86_64-apple-darwin13.4.0
Thread model: posix

Could you give more details on your compilation environment?

[vinklein]

Hi,
I get one of these errors on OSX with fflas-ffpack and openblas.

$ more test-suite.log
==============================================
   FFLAS-FFPACK 2.3.2: tests/test-suite.log
==============================================

# TOTAL: 34
# PASS:  33
# SKIP:  0
# XFAIL: 0
# FAIL:  1
# XPASS: 0
# ERROR: 0

.. contents:: :depth: 2

FAIL: test-fgemm-check
======================

Checking ..........................Modular<double> mod 1693 ... PASSED with seed = 1541770760090127
Checking .........................Modular<double> mod 12421 ... PASSED with seed = 1541770760090128
Checking ...........................Modular<double> mod 139 ... PASSED with seed = 1541770760090129
Checking ..................ModularBalanced<double> mod 1693 ... PASSED with seed = 1541770760090127
Checking .................ModularBalanced<double> mod 12421 ... PASSED with seed = 1541770760090128
Checking ...................ModularBalanced<double> mod 139 ... PASSED with seed = 1541770760090129
Checking ...........................Modular<float> mod 1693 ... PASSED with seed = 1541770760090127
Checking .............................Modular<float> mod 47 ... FAILED
FAILED with seed = 1541770760090128
FAIL test-fgemm-check (exit status: 1)

Config
fflas_ffpack: 2.3.2
openblas: 0.2.20 (with some patches)

$ uname -a
Darwin laptop-147-210-128-171.labri.fr 17.7.0 Darwin Kernel Version 17.7.0: Thu Jun 21 22:53:14 PDT 2018; root:xnu-4570.71.2~1/RELEASE_X86_64 x86_64
$ g++ -v
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/Users/doctorant/sage/local/libexec/gcc/x86_64-apple-darwin17.7.0/7.2.0/lto-wrapper
Target: x86_64-apple-darwin17.7.0
Configured with: ../src/configure --prefix=/Users/doctorant/sage/local --with-local-prefix=/Users/doctorant/sage/local --with-gmp=/Users/doctorant/sage/local --with-mpfr=/Users/doctorant/sage/local --with-mpc=/Users/doctorant/sage/local --with-system-zlib --disable-multilib --disable-nls --enable-languages=c,c++,fortran --disable-libitm --with-build-config=bootstrap-debug --without-isl --without-cloog  
Thread model: posix
gcc version 7.2.0 (GCC) 

Link to the sage's ticket #26130.

[timokau]

I just accidentally reproduced the test-fgemm-check failure on linux by building openblas without openmp. NixOS defaults to no openmp on darwin, so that may be the issue or one of the issues.

[timokau]

Okay this may just have been a coincidence. With the exact same environment, the second build succeeded. I'm going to try a couple more times to see if it re-occurs (and if openmp makes any difference).

[timokau]

The failure also occurs with openmp enabled. It just occurs a lot ofter when disabled: 3/10 times without openmp, 1/20 times with.

Of course that could just be coincidence or load may be a factor.

[timokau]

Looks like that (and probably also @vinklein's issue) is actually #146, which is fixed on master but not in a released version yet.

[P1K]

Hi,
I thought I'd run some tests myself as I have an OSX install, and I have no failed tests on Darwin with openblas. In particular, I let test-fgemm-check run for several minutes without catching anything.

[ClementPernet]

Ok this seems to be a fixed issue. Closing.