inv(Matrix{Float64}) makes julia exit without any error

[SimonDanisch]

Using the Color package made Julia exit without any message whatsoever.
I pinned the error down to Color.jl line 616

const CAT97s = [ 0.8562  0.3372 -0.1934
                -0.8360  1.8327  0.0033
                 0.0357 -0.0469  1.0112 ]
const CAT97s_INV = inv(CAT97s)

Same behaviour in the Julia readline.
After playing around with it a little, it seems it doesn't like a 3x3 matrix:
inv(rand(3,3)) makes Julia also fail without any feedback.
The error seems to be somewhere in inv{T}(F::Factorization{T}) at linalg/factorization.jl:795

versioninfo(true) :

Julia Version 0.3.0-prerelease+2591
Commit 550866d (2014-04-10 22:42 UTC)
Platform Info:
System: Windows (x86_64-w64-mingw32)
CPU: Intel(R) Core(TM) i5-4200U CPU @ 1.60GHz
WORD_SIZE: 64
Microsoft Windows [Version 6.3.9600]
uname: MINGW32_NT-6.2 1.0.12(0.46/3/2) 2011-07-20 17:52 i686 unknown
Memory: 7.914325714111328 GB (3898.83984375 MB free)
Uptime: 41389.478131 sec
Load Avg: 0.0 0.0 0.0
Intel(R) Core(TM) i5-4200U CPU @ 1.60GHz:
speed user nice sys idle irq ticks
#1 2295 MHz 1488812 0 757703 38903406 26234 ticks
#2 2295 MHz 1430765 0 815171 38903406 157046 ticks
#3 2295 MHz 1644765 0 702296 38802281 5062 ticks
#4 2295 MHz 1345656 0 588609 39215078 2656 ticks

BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
LAPACK: libopenblas
LIBM: libopenlibm
Environment:
APR_ICONV_PATH = C:\Program Files (x86)\Subversion\iconv
HOMEDRIVE = C:
HOMEPATH = \Users\Sim
JULIA_EDITOR = start
PATHEXT = .COM;.EXE;.BAT;.CMD;.VBS;.VBE;.JS;.JSE;.WSF;.WSH;.MSC
VBOX_INSTALL_PATH = C:\Program Files\Oracle\VirtualBox\

Package Directory: C:\Users\Sim.julia\v0.3
6 required packages:

• Color 0.2.9
• DictUtils 0.0.0
• GLUT 0.3.0
• ImmutableArrays 0.0.4
• Match 0.0.3
• SDL 0.1.0
  12 additional packages:
• BinDeps 0.2.12
• Cartesian 0.1.5
• Events 0.0.0- master (unregistered, dirty)
• GLUtil 0.0.0- master (unregistered, dirty)
• GetC 1.1.1
• Images 0.2.34
• ModernGL 0.0.0- master (unregistered)
• OpenGL 1.1.3
• SIUnits 0.0.1
• TexExtensions 0.0.1
• URIParser 0.0.1
• Zlib 0.1.6

[mbauman]

I happen to be on the exact same commit, but on OS X, and I cannot reproduce (using either inv(rand(3,3)) or inv(CAT97s)).

[arnim]

Commit 742593f* (6 days old master)
x86_64-apple-darwin13.1.0

Works just fine for inv(rand(3,3)) as well.

Maybe you can run the tests and see if there are some other things that fail?

[SimonDanisch]

It says in big friendly letters: SUCCESS.
I hope I did it right, though.
I went to JULIA_HOME/share/julia/test and then I executed $julia runtests.jl

[SimonDanisch]

I tested it a little more. It seems like it happens with all odd dimensions.

[tkelman]

I cannot reproduce this, at the same commit on Windows:

julia> inv(rand(3,3))
3x3 Array{Float64,2}:
  0.296319   1.3562    -1.95925
 -0.51922   -0.320154   2.6635
  2.06626   -0.608049   0.738967

julia> versioninfo()
Julia Version 0.3.0-prerelease+2591
Commit 550866d* (2014-04-10 22:42 UTC)
Platform Info:
  System: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i7-2630QM CPU @ 2.00GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm

[SimonDanisch]

I downloaded the new nightly, same story!

C:\Users\Sim>julia
julia> inv(rand(3,3))
C:\Users\Sim>julia
julia> versioninfo()
Julia Version 0.3.0-prerelease+2599
Commit bf7096c (2014-04-11 23:52 UTC)
Platform Info:
  System: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i5-4200U CPU @ 1.60GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm

[andreasnoack]

Long shot, but could you try blas_set_num_threads(1) before running inv?

[SimonDanisch]

That actually works!

2014-04-12 10:47 GMT+02:00 Andreas Noack Jensen notifications@github.com:

Long shot, but could you try blas_set_num_threads(1) before running inv?

Reply to this email directly or view it on GitHubhttps://github.com//issues/6504#issuecomment-40275399
.

[andreasnoack]

It is not the first time that OpenBLAS has had multithreading bugs on Windows. Could you file an issue at their GitHub site?

[tkelman]

Haswell OpenBlas threading problem then? Probably good idea to try OpenBlas 0.2.9.rc2 before opening an issue, as it may be a bug they fixed already. Do you have a MinGW build environment set up? If not, try replacing libopenblas.dll with the binary from http://sourceforge.net/projects/openblas/files/v0.2.9.rc2/OpenBLAS-v0.2.9.rc2-x86_64-Win.zip nevermind that won't work, that binary isn't using 64-bit integers.

Oh and for future reference, you got most of the test-running process right, but you want to do julia runtests.jl all, not just julia runtests.jl.

[SimonDanisch]

No, I actually don't have anything set up. It's sometimes quite painful to
build with windows, so I rather stay with Linux;)

2014-04-12 11:02 GMT+02:00 Tony Kelman notifications@github.com:

Haswell OpenBlas threading problem then? Probably good idea to try
OpenBlas 0.2.9.rc2 before opening an issue, as it may be a bug they fixed
already. Do you have a MinGW build environment set up?

Reply to this email directly or view it on GitHubhttps://github.com//issues/6504#issuecomment-40275618
.

[tkelman]

The windows build is slow the first time and breaks often, but so far not in unfixable ways.
Cross-compile then?

[SimonDanisch]

Hm, with the x64_86 binaries, Julia says: ERROR: OpenBLAS was not built
with 64bit integer support.

2014-04-12 11:17 GMT+02:00 Tony Kelman notifications@github.com:

The windows build is slow the first time and breaks often, but so far not
in unfixable ways.
Cross-compile then?

Reply to this email directly or view it on GitHubhttps://github.com//issues/6504#issuecomment-40275843
.

[SimonDanisch]

Oh I see... Well here come the errors:

        From worker 4:       * linalg3
        From worker 2:       * linalg1
        From worker 3:       * linalg2
        From worker 5:       * core
        From worker 5:       * keywordargs
        From worker 5:       * numbers
exception on 2: ERROR: assertion failed: |cond(a,Inf) - 375.7017682707787| <= 0.
01
  cond(a,Inf) = 377.32187
  375.7017682707787 = 375.7017682707787
  difference = 1.6201006257056747 > 0.01
 in error at error.jl:22
 in test_approx_eq at test.jl:109
 in anonymous at no file:14
 in runtests at C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\te
st\testdefs.jl:5
 in anonymous at multi.jl:824
 in run_work_thunk at multi.jl:590
 in anonymous at task.jl:824
while loading linalg1.jl, in expression starting on line 9
exception on 4: ERROR: assertion failed: |det($(Expr(:', :(a[:,1:2] * x - b))) *
 (a[:,1:2] * x - b)) - convert(elty,4.437969924812031)| <= 0.00476837158203125
  det($(Expr(:', :(a[:,1:2] * x - b))) * (a[:,1:2] * x - b)) = 4.328434f0 + 1.39
48083f0im
  convert(elty,4.437969924812031) = 4.43797f0 + 0.0f0im
  difference = 1.3991027 > 0.00476837158203125
 in error at error.jl:22
 in test_approx_eq at test.jl:109
 in anonymous at no file:14
 in runtests at C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\te
st\testdefs.jl:5
 in anonymous at multi.jl:824
 in run_work_thunk at multi.jl:590
 in anonymous at task.jl:824
while loading linalg3.jl, in expression starting on line 4
exception on 3: ERROR: assertion failed: |W * v - F * v| <= 0.02384185791015625
  W * v = Complex{Float32}[5.411544f0 + 0.0f0im,5.3362956f0 + 0.0f0im,-1.6706779
f0 + 0.0f0im,-1.1320411f0 + 0.0f0im,-2.3926892f0 + 0.0f0im]
  F * v = Complex{Float32}[4.9138784f0 - 3.4160316f0im,4.4354544f0 - 5.271185f0i
m,-1.3091786f0 + 2.1735759f0im,-1.1513414f0 + 0.1880764f0im,-2.2634819f0 + 0.934
9077f0im]
  difference = 5.3476076 > 0.02384185791015625
 in error at error.jl:22
 in test_approx_eq at test.jl:109
 in anonymous at no file:99
 in runtests at C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\te
st\testdefs.jl:5
 in anonymous at multi.jl:824
 in run_work_thunk at multi.jl:590
 in anonymous at task.jl:824
while loading linalg2.jl, in expression starting on line 23
ERROR: assertion failed: |cond(a,Inf) - 375.7017682707787| <= 0.01
  cond(a,Inf) = 377.32187
  375.7017682707787 = 375.7017682707787
  difference = 1.6201006257056747 > 0.01
 in anonymous at task.jl:1314
while loading linalg1.jl, in expression starting on line 9
while loading C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\test
\runtests.jl, in expression starting on line 35

blas_num_threads=4

[tkelman]

This is with OpenBlas 0.2.9.rc1 (current Julia default for Windows), or 0.2.9.rc2? I'm building a copy of 0.2.9.rc2 right now that I'll upload when it's finished, but I don't have a Haswell machine to test on.

[SimonDanisch]

I haven't changed anything in the install folder so everything should be default:

  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas

[tkelman]

Give this a try: http://sourceforge.net/projects/juliadeps-win/files/openblas-v0.2.9.rc2-x86_64-w64-mingw32.7z

0.2.9.rc2 is passing relevant Blas/Lapack tests on my Sandy Bridge machine.

[SimonDanisch]

Nope, nothing changed :(

julia>

blas_num_threads=4

C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\test> julia runtes
ts.jl all
        From worker 2:       * linalg1
        From worker 4:       * linalg3
        From worker 5:       * core
        From worker 3:       * linalg2
        From worker 5:       * keywordargs
        From worker 5:       * numbers
exception on 2: ERROR: assertion failed: |cond(a,Inf) - 375.7017682707787| <= 0.
01
  cond(a,Inf) = 377.32187
  375.7017682707787 = 375.7017682707787
  difference = 1.6201006257056747 > 0.01
 in error at error.jl:22
 in test_approx_eq at test.jl:109
 in anonymous at no file:14
 in runtests at C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\te
st\testdefs.jl:5
 in anonymous at multi.jl:824
 in run_work_thunk at multi.jl:590
 in anonymous at task.jl:824
while loading linalg1.jl, in expression starting on line 9
exception on 4: ERROR: assertion failed: |det($(Expr(:', :(a[:,1:2] * x - b))) *
 (a[:,1:2] * x - b)) - convert(elty,4.437969924812031)| <= 0.00476837158203125
  det($(Expr(:', :(a[:,1:2] * x - b))) * (a[:,1:2] * x - b)) = 4.328434f0 + 1.39
48083f0im
  convert(elty,4.437969924812031) = 4.43797f0 + 0.0f0im
  difference = 1.3991027 > 0.00476837158203125
 in error at error.jl:22
 in test_approx_eq at test.jl:109
 in anonymous at no file:14
 in runtests at C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\te
st\testdefs.jl:5
 in anonymous at multi.jl:824
 in run_work_thunk at multi.jl:590
 in anonymous at task.jl:824
while loading linalg3.jl, in expression starting on line 4
exception on 3: ERROR: assertion failed: |W * v - F * v| <= 0.02384185791015625
  W * v = Complex{Float32}[5.411544f0 + 0.0f0im,5.3362956f0 + 0.0f0im,-1.6706779
f0 + 0.0f0im,-1.1320411f0 + 0.0f0im,-2.3926892f0 + 0.0f0im]
  F * v = Complex{Float32}[4.9138784f0 - 3.4160316f0im,4.4354544f0 - 5.271185f0i
m,-1.3091786f0 + 2.1735759f0im,-1.1513414f0 + 0.1880764f0im,-2.2634819f0 + 0.934
9077f0im]
  difference = 5.3476076 > 0.02384185791015625
 in error at error.jl:22
 in test_approx_eq at test.jl:109
 in anonymous at no file:99
 in runtests at C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\te
st\testdefs.jl:5
 in anonymous at multi.jl:824
 in run_work_thunk at multi.jl:590
 in anonymous at task.jl:824
while loading linalg2.jl, in expression starting on line 23
ERROR: assertion failed: |cond(a,Inf) - 375.7017682707787| <= 0.01
  cond(a,Inf) = 377.32187
  375.7017682707787 = 375.7017682707787
  difference = 1.6201006257056747 > 0.01
 in anonymous at task.jl:1314
while loading linalg1.jl, in expression starting on line 9
while loading C:\Users\Sim\AppData\Local\Julia 0.3.0-prerelease\share\julia\test
\runtests.jl, in expression starting on line 35

[tkelman]

Okay, now worth reporting to OpenBlas' tracker then. Looks like we have one test case each from linalg1, linalg2, and linalg3, and maybe a fourth if inv(rand(3,3)) still doesn't work with default number of threads?

@andreasnoackjensen do we have some of Julia's Blas/Lapack test cases translated into C or Fortran to make it easier for the OpenBlas devs to reproduce? They may not have a Windows Haswell machine to test on though.

[SimonDanisch]

Yes, inv(rand(3,3)) still doesn't work!

[tkelman]

Hm, just a quick test to be sure you're using the new version of libopenblas.dll - when you start Julia from the command line, then exit(), do you still see the blas_num_threads=4 message? That should be gone in 0.2.9.rc2.

[SimonDanisch]

Yes, that's what I'm doing, and the message is gone ;)

2014-04-12 13:15 GMT+02:00 Tony Kelman notifications@github.com:

Hm, just a quick test to be sure you're using the new version of
libopenblas.dll - when you start Julia from the command line, then close
it, do you still see the blas_num_threads=4 message? That should be gone
in 0.2.9.rc2.

Reply to this email directly or view it on GitHubhttps://github.com//issues/6504#issuecomment-40277905
.

[andreasnoack]

@SimonDanisch Could you try lufact(A) and see if that is enough to make it crash.

[SimonDanisch]

I already did this, that's why I believed, that I pinned it down to
inv{T}(F::Factorization{T})
at linalg/factorization.jl:795

2014-04-12 13:35 GMT+02:00 Andreas Noack Jensen notifications@github.com:

@SimonDanisch https://github.com/SimonDanisch Could you try lufact(A)and see if that is enough to make it crash.

Reply to this email directly or view it on GitHubhttps://github.com//issues/6504#issuecomment-40278221
.

[andreasnoack]

That methods shouldn't get called by inv(rand(3,3)). Just to be sure, if you do

A = rand(3,3);
Alu = lufact(A);
inv(Alu);

Julia crashes at the inv step. Is that right?

[SimonDanisch]

Sorry, I wasn't at my computer anymore
Well yes, it crashes at inv(Alu)

julia> A = rand(3,3);
julia> Alu = lufact(A);
julia> @which inv(Alu)
inv{T}(F::Factorization{T}) at linalg/factorization.jl:795

[andreasnoack]

I just tried the same on a Windows machine and I get the same as you. That is weird. It shouldn't call that method and it doesn't on my Mac. Could you try

invoke(inv, (Base.LinAlg.LU{Float64},), lufact(A))

and see if that works or crashes.

[SimonDanisch]

invoke(inv, (Base.LinAlg.LU http://base.linalg.lu/{Float64},),
lufact(rand(3,3))) - Chrashes :(

2014-04-12 22:08 GMT+02:00 Andreas Noack Jensen notifications@github.com:

I just tried the same on a Windows machine and I get the same as you. That
is weird. It shouldn't call that method and it doesn't on my Mac. Could you
try

invoke(inv, (Base.LinAlg.LU{Float64},), lufact(A))

and see if that works or crashes.

Reply to this email directly or view it on GitHubhttps://github.com//issues/6504#issuecomment-40290415
.

[andreasnoack]

Does A_ldiv_B!(lufact(rand(3,3)),eye(3)) also crash?

[ViralBShah]

What if we just use nehalem kernels with single threading on windows?

[tkelman]

All of the problems have been on Haswell, hopefully we won't need to disable threading completely for all Windows users because of this. That does remind me though, @SimonDanisch can you try at least the 0.2.9 release version with set OPENBLAS_NUM_THREADS=1 (assuming you're running the tests from cmd), see if that makes any difference?

[SimonDanisch]

First run with set OPENBLAS_NUM_THREADS=1

exception on 2: ERROR: assertion failed: |cond(a,Inf) - 375.7017682707787| <= 0. 01

Second run with blas_set_num_threads(1) in first line:

fatal error on 1: ERROR: stream is closed or unusable
 in read at iobuffer.jl:72
 in read at stream.jl:679
 in anonymous at task.jl:829

Might this be related to #7174 ?

Third run (still with blas_set_num_threads(1):

exception on 2: ERROR: assertion failed: |cond(a,Inf) - 375.7017682707787| <= 0. 01

[tkelman]

I built another with OPENBLAS_TARGET_ARCH=NEHALEM here http://sourceforge.net/projects/juliadeps-win/files/openblas-v0.2.9-x86_64-w64-mingw32-nehalem.7z/download - any difference?

Edit: and another http://sourceforge.net/projects/juliadeps-win/files/openblas-53bfa51ee0a47a8b-x86_64-w64-mingw32.7z/download based on OpenMathLib/OpenBLAS#386 that should respond to OPENBLAS_CORETYPE

[spalato]

Hi everyone,

I've just rebuilt julia with a fresh clone, downloaded today, using Cygwin64 on windows 8 using cygwin's gcc. A_ldiv_B!(lufact(rand(3,3)), eye(3)) works. Tests still fail. It seems the behavior has not changed since the end of may, using the openblas release from Apr 22.

$ julia -e "versioninfo()"
Julia Version 0.3.0-prerelease+3679
Commit d0eac34 (2014-06-14 10:06 UTC)
Platform Info:
  System: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i7-4700MQ CPU @ 2.40GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm

Result of a run of tests: https://gist.github.com/spalato/aeeb6e184fd0eab9738f
Identical to results from @SimonDanisch.
Nothing changes when I put blas_set_num_threads(1) on first line of runtests.jl

It seems to be the same issue as previously reported.

[tkelman]

What do you mean openblas release from Apr 22? There was no openblas release on Apr 22. Can you try a few of the alternate openblas binaries I've posted links to above? Just replace your usr/bin/libopenblas.dll with a few of the alternate versions. The 53bfa51ee0a47a8b one should be the most interesting to look at, as it should allow setting OPENBLAS_VERBOSE=2 and changing OPENBLAS_CORETYPE=nehalem, etc.

[spalato]

Hi,

I meant the development version of openblas you linked on Apr 24 (not 22).

I've set OPENBLAS_VERBOSE=2 and OPENBLAS_CORETYPE=nehalem as environment variables (using export) and didn't recompile.

Results from julia runtests.jl linalg from cygwin :

1. With Nehalem: Exactly same errors, no change from previous behavior. No change with blas_set_num_threads(1).
2. 53bfa51ee0a47a8b (later referred to as 53b), with OPENBLAS_CORETYPE=nehalem: SUCCESS.
3. 53b, without OPENBLAS_CORETYPE (undefined): tests fail exactly as previously.

I'm currently testing 53b with blas_set_num_threads(1), but it's slow (and was usually failing pretty quickly).

Hope this helps.

[spalato]

Test result from 53b with OPENLAS_CORETYPE=nehalem and blas_set_num_threads(1): SUCCESS.

It seems the issue is solved for me. Thanks a lot.

[tkelman]

Sounds like we should carry those patches locally then, but also write a Fortran (or C, I suppose) test case to let OpenBLAS know their Haswell kernels still have problems. Also very strange that OPENBLAS_TARGET_ARCH=NEHALEM didn't work but the patched version with OPENBLAS_CORETYPE=nehalem did...

[tkelman]

Bump. OpenBLAS on Haswell Win64 is still evidently broken. @andreasnoackjensen would you be able to translate some of these failing tests into Fortran test cases?

[spalato]

My Make.inc already contains OPENBLAS_TARGET_ARCH=NEHALEM. That could explain why this compile didn't work.

[tkelman]

@spalato, @SimonDanisch could you guys test again with http://sourceforge.net/projects/juliadeps-win/files/openblas-23203d52c1e74199-x86_64-w64-mingw32.7z/download ? Thanks!

[spalato]

Fails as other cases when no environment variables are set. Succeeds after export OPENBLAS_CORETYPE=nehalem.

[tkelman]

@spalato would it be possible for you to give remote access to your machine to @wernsaar, an OpenBLAS developer, for debugging purposes?

[spalato]

I guess we could try. Do you want me to move to OpenMathLib/OpenBLAS#380 (comment) ?

[ViralBShah]

Is this fixed?

[tkelman]

Partially, but not completely. This is from e5b832d on an i7-4702MQ using OpenBlas from xianyi/develop

https://gist.github.com/anonymous/61153d186665d5f0e047

[tkelman]

The upstream openblas branch (wernsaar/develop) looks fixed on Haswell, we can close this as soon as there's a tag (0.2.10.rc1?) to pull in those fixes from.

[wernsaar]

On 28.06.2014 21:29, Tony Kelman wrote:

The upstream openblas branch (wernsaar/develop) looks fixed on Haswell, we can close this as soon as there's a tag (0.2.10.rc1?) to pull in those fixes from.

---

Reply to this email directly or view it on GitHub:
#6504 (comment)
I created the pull request. I think that Xianyi will merge my patches as
soon as possible
and will create OpenBLAS version v0.2.10-rc1.

Best regards

Werner

[ViralBShah]

Can folks here try master and see if this issue is fixed?

[SimonDanisch]

Sorry for the long wait!
It seems to work fine for me!
Most test work, but I got a new error, when runnig the tests:

exception on 2: ERROR: test error during readall(@cmd "echo hello | sort") == "h
ello | sort\n"
could not spawn `echo hello | sort`: no such file or directory (ENOENT)
 in _jl_spawn at process.jl:217
 in spawn at process.jl:348
while loading spawn.jl, in expression starting on line 16
ERROR: test error during readall(@cmd "echo hello | sort") == "hello | sort\n"
could not spawn `echo hello | sort`: no such file or directory (ENOENT)
while loading spawn.jl, in expression starting on line 16
while loading C:\Users\Sim\AppData\Local\Julia-0.3.0-prerelease\share\julia\test
\runtests.jl, in expression starting on line 46

[tkelman]

Thanks @SimonDanisch! That one will be fixed by either #7446 or #7349.

[SimonDanisch]

Cool! =)
Thank you all for the hard work!!
Can't wait for the 0.3.0 release.