Thread: [pocl-devel] OS X build failure

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Hi all,

Any clue what might be triggering this build failure on macOS with LLVM 5?

https://travis-ci.org/conda-forge/pocl-feedstock/builds/351520947

Thanks!
Andreas

Trying to prepare an adaption of POCL 1.1 to FreeBSD (CURRENT at the moment), using
LLVM/CLANG 5, compilation runs well, but the final tests fail due to the initial error in
pocl_version_check, like

[...]
        Start   1: pocl_version_check
  1/197 Test   #1:
pocl_version_check ..................................................................***Failed
Required regular expression not found.Regex=[basic ]
[...]

which then results (for me) in 

[...]
 61% tests passed, 77 tests failed out of 197

Label Time Summary:
EinsteinToolkit    =  11.32 sec*proc (2 tests)
VexCL              =   0.00 sec*proc (27 tests)
clBLAS             =   0.00 sec*proc (46 tests)
clFFT              =   0.00 sec*proc (3 tests)
cuda               =   3.65 sec*proc (49 tests)
hsa                =   0.26 sec*proc (3 tests)
internal           =  11.82 sec*proc (107 tests)
kernel             =   3.33 sec*proc (29 tests)
regression         =   2.56 sec*proc (35 tests)
runtime            =   4.09 sec*proc (21 tests)
spir               =   0.86 sec*proc (1 test)
tce                =   0.69 sec*proc (9 tests)
workgroup          =   1.53 sec*proc (19 tests)

Total Test time (real) = 172.05 sec

The following tests FAILED:
          1 - pocl_version_check (Failed)
        122 - clBLAS_samples_example_chbmv (Not Run)
        123 - clBLAS_samples_example_chemm (Not Run)

[...]
        194 - vexcl_scan_by_key (Not Run)
        195 - vexcl_reduce_by_key (Not Run)
        196 - vexcl_logical (Not Run)
        197 - vexcl_threads (Not Run)
Errors while running CTest
FAILED: CMakeFiles/test.util 
cd /usr/ports/lang/pocl.svn/work/pocl-1.1 && /usr/local/bin/ctest
--force-new-ctest-process ninja: build stopped: subcommand failed.
*** Error code 1
[...]

I haven't looked deeper into it, but I guess the lattice of failed tests above does
result in the initial very first test, no. 1, designated "pocl_version_check", doesn't
it?

What is going wrong here?

Does anybody else face this error?

Thanks in advance,

O. Hartmann

-- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).

Hi,

> Trying to prepare an adaption of POCL 1.1 to FreeBSD (CURRENT at the
> moment), using LLVM/CLANG 5, compilation runs well, but the final
> tests fail due to the initial error in pocl_version_check, like
> 
> pocl_version_check ..................................................................***Failed

> Does anybody else face this error?
> 

That test does one thing: it loads OpenCL and checks that the platform
version matches "OpenCL 1.2 pocl X.Y". Two common ways this could fail
are 1) you have another OpenCL platform installed, or 2) loading
libpocl.so fails (usually due to linking problems).

If you have ocl-icd installed as ICD loader, then an easy check is:

source <path-to-pocl-source-dir>/tools/scripts/devel-envs.sh
clinfo

from pocl's BUILD directory. If it prints 0 platforms then likely you
have linking issues. Or you could build pocl with -DENABLE_ICD=0 and
it'll fail linking the tests, if there are any problems.

Cheers,
-- mb

Am Sat, 10 Mar 2018 11:17:36 +0100
Michal Babej <Fra...@ru...> schrieb:

> Hi,
> 
> > Trying to prepare an adaption of POCL 1.1 to FreeBSD (CURRENT at the
> > moment), using LLVM/CLANG 5, compilation runs well, but the final
> > tests fail due to the initial error in pocl_version_check, like
> > 
> > pocl_version_check ..................................................................***Failed  
> 
> > Does anybody else face this error?
> >   
> 
> That test does one thing: it loads OpenCL and checks that the platform
> version matches "OpenCL 1.2 pocl X.Y". Two common ways this could fail
> are 1) you have another OpenCL platform installed, or 2) loading
> libpocl.so fails (usually due to linking problems).
> 
> If you have ocl-icd installed as ICD loader, then an easy check is:
> 
> source <path-to-pocl-source-dir>/tools/scripts/devel-envs.sh
> clinfo

The script devel-envs.sh is a Bourne shell script, so I guess "sourcing" it like c shell
scripts isn't appropriate here, it fails, executed from POCL's BUILD directory:

source ./tools/scripts/devel-envs.sh
libs_subdir=.: Command not found.
export: Command not found.
export: Command not found.
export: Command not found.
export: Command not found.

[...]

Executing

# (sh ./tools/scripts/devel-envs.sh && clinfo)

gives me 

Unable to find symbol pthread_mutexattr_setkind_np version (null). Aborting.
Abort

This is something I see lately very often due to some weirdness in registering ICDs. This
is with compiling pocl with LLVM 6. 

> 
> from pocl's BUILD directory. If it prints 0 platforms then likely you
> have linking issues. Or you could build pocl with -DENABLE_ICD=0 and
> it'll fail linking the tests, if there are any problems.

Installed (on my test box) is ocl-icd 2.2.12, on another box there is the regular ocl-icd
package/port devel/ocl-icd in the FreeBSD ports tree, which is version 2.2.11. this just
for the record if it matters.

> 
> Cheers,
> -- mb
> 
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> pocl-devel mailing list
> poc...@li...
> https://lists.sourceforge.net/lists/listinfo/pocl-devel

-- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).

Hi,

CMake Error
at /Users/travis/miniconda3/conda-bld/pocl_1520641026167/_b_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_plac/share/cmake-3.10/Modules/TestBigEndian.cmake:49
(message): no suitable type found Call Stack (most recent call first):
  CMakeLists.txt:1137 (TEST_BIG_ENDIAN)

According to a quick google, it's a CMake problem. Did you try building
with Clang as system compiler ?

cmake -DCMAKE_C_COMPILER=<path-to-clang>
-DCMAKE_CXX_COMPILER=<path-to-clang++> <path-to-pocl-sources>

-- mb

Michal Babej <Fra...@ru...> writes:
> CMake Error
> at /Users/travis/miniconda3/conda-bld/pocl_1520641026167/_b_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_plac/share/cmake-3.10/Modules/TestBigEndian.cmake:49
> (message): no suitable type found Call Stack (most recent call first):
>   CMakeLists.txt:1137 (TEST_BIG_ENDIAN)
>
> According to a quick google, it's a CMake problem. Did you try building
> with Clang as system compiler ?
>
> cmake -DCMAKE_C_COMPILER=<path-to-clang>
> -DCMAKE_CXX_COMPILER=<path-to-clang++> <path-to-pocl-sources>

Thanks for the tip! I tried that, it hasn't helped:

https://travis-ci.org/conda-forge/pocl-feedstock/builds/351767014

Andreas

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Am Tue, 13 Mar 2018 08:53:04 +0100
Michal Babej <Fra...@ru...> schrieb:

I'm sorry having such a delay (time ...).

> On Sun, 11 Mar 2018 11:04:45 +0100
> "O. Hartmann" <o.h...@wa...> wrote:
> 
> > The script devel-envs.sh is a Bourne shell script, so I guess
> > "sourcing" it like c shell scripts isn't appropriate here, it fails,
> > executed from POCL's BUILD directory:  
> 
> Ah sorry, forgot about FreeBSD's C shell. That script merely sets up
> 2-3 env variables to run pocl from build directory.
> 
> > Unable to find symbol pthread_mutexattr_setkind_np version (null).
> > Aborting. Abort  
> 
> I'm afraid I have never seen this bug. The closest thing i see in pocl
> source is 

I think this problem is a well known problem with LLVM - and it pops up in new clothings
and at an unexpected place now.

According to this bug

http://lists.llvm.org/pipermail/llvm-bugs/2016-October/051237.html

it seems, that when multiple LLVM backed platforms are present, the error occurs. 

Prior to the checks, with POCL 1.0 or POCL 0.14, the problems occured when trying to run
a pocl program with ocl-icd and other OpenCL platforms present on the system, like clover
or Intel's beignet.

> 
> PTHREAD_CHECK(pthread_mutexattr_settype(&mattr,
> PTHREAD_MUTEX_ERRORCHECK));
> 
> ... but this is in pocl-hsa.c and you're not building HSA-enabled pocl.
> 
> 
> -- mb

Kind regards,

Oliver

- -- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).
-----BEGIN PGP SIGNATURE-----

iLQEARMKAB0WIQQZVZMzAtwC2T/86TrS528fyFhYlAUCWs28TAAKCRDS528fyFhY
lIlzAfdTXoTmHdlk57227PHb1+gxFg8g8J+hCfAqjX+k8pM2n+Lp7ZghaAfAKuR6
vhdvdTaZBUb/c0gGjT51GMop1OAB/1leN+f4WNV+2xfo7YPtTEzp1vN0onoa4HOb
cjvndCe3Yi6+WTtAA3cuQlLorTFIwT63DcxBqEqzukU0zrgOTN8=
=yq4A
-----END PGP SIGNATURE-----

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Am Wed, 11 Apr 2018 09:41:36 +0200
"O. Hartmann" <oha...@wa...> schrieb:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA512
> 
> Am Tue, 13 Mar 2018 08:53:04 +0100
> Michal Babej <Fra...@ru...> schrieb:
> 
> 
> I'm sorry having such a delay (time ...).
> 
> > On Sun, 11 Mar 2018 11:04:45 +0100
> > "O. Hartmann" <o.h...@wa...> wrote:
> >   
> > > The script devel-envs.sh is a Bourne shell script, so I guess
> > > "sourcing" it like c shell scripts isn't appropriate here, it fails,
> > > executed from POCL's BUILD directory:    
> > 
> > Ah sorry, forgot about FreeBSD's C shell. That script merely sets up
> > 2-3 env variables to run pocl from build directory.
> >   
> > > Unable to find symbol pthread_mutexattr_setkind_np version (null).
> > > Aborting. Abort    
> > 
> > I'm afraid I have never seen this bug. The closest thing i see in pocl
> > source is   
> 
> I think this problem is a well known problem with LLVM - and it pops up in new clothings
> and at an unexpected place now.
> 
> According to this bug
> 
> http://lists.llvm.org/pipermail/llvm-bugs/2016-October/051237.html
> 
> it seems, that when multiple LLVM backed platforms are present, the error occurs. 
> 
> Prior to the checks, with POCL 1.0 or POCL 0.14, the problems occured when trying to run
> a pocl program with ocl-icd and other OpenCL platforms present on the system, like
> clover or Intel's beignet.
> 
> > 
> > PTHREAD_CHECK(pthread_mutexattr_settype(&mattr,
> > PTHREAD_MUTEX_ERRORCHECK));
> > 
> > ... but this is in pocl-hsa.c and you're not building HSA-enabled pocl.
> > 
> > 
> > -- mb  
> 
> Kind regards,
> 
> Oliver
[...]

Sorry, there was an issue with our port devel/ocl-icd, the linking against pthread lib
was missing. Now I receive the issue I see since a long time and I was refering to in my
last email (LLVM bug, not fixed):

[...]
Device open failed, aborting...
: CommandLine Error: Option 'enable-value-profiling' registered more than once!
LLVM ERROR: inconsistency in registered CommandLine options
Nothing to output !

- -- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).
-----BEGIN PGP SIGNATURE-----

iLUEARMKAB0WIQQZVZMzAtwC2T/86TrS528fyFhYlAUCWs3M9AAKCRDS528fyFhY
lCLHAgCJYBpXPukDj5Yyhj5yCsIKfvRhkN5BzWEznD1GK5oAqzZkV3GGHEC3I1Ba
4x/spy5tXo0Pc5I4uAhCcD6KG+YTAfwNpq1EpCeKapK35KFc2Me0AbYJX/KvxpQf
RWOWNG7A9oOl5MjuybMhx20mzDUnQaQBHgYZvIAcOgqyJeYItt6r
=oxbS
-----END PGP SIGNATURE-----

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Am Wed, 11 Apr 2018 09:41:36 +0200
"O. Hartmann" <oha...@wa...> schrieb:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA512
> 
> Am Tue, 13 Mar 2018 08:53:04 +0100
> Michal Babej <Fra...@ru...> schrieb:
> 
> 
> I'm sorry having such a delay (time ...).
> 
> > On Sun, 11 Mar 2018 11:04:45 +0100
> > "O. Hartmann" <o.h...@wa...> wrote:
> >   
> > > The script devel-envs.sh is a Bourne shell script, so I guess
> > > "sourcing" it like c shell scripts isn't appropriate here, it fails,
> > > executed from POCL's BUILD directory:    
> > 
> > Ah sorry, forgot about FreeBSD's C shell. That script merely sets up
> > 2-3 env variables to run pocl from build directory.
> >   
> > > Unable to find symbol pthread_mutexattr_setkind_np version (null).
> > > Aborting. Abort    
> > 
> > I'm afraid I have never seen this bug. The closest thing i see in pocl
> > source is   
> 
> I think this problem is a well known problem with LLVM - and it pops up in new clothings
> and at an unexpected place now.
> 
> According to this bug
> 
> http://lists.llvm.org/pipermail/llvm-bugs/2016-October/051237.html
> 
> it seems, that when multiple LLVM backed platforms are present, the error occurs. 
> 
> Prior to the checks, with POCL 1.0 or POCL 0.14, the problems occured when trying to run
> a pocl program with ocl-icd and other OpenCL platforms present on the system, like
> clover or Intel's beignet.
> 
> > 
> > PTHREAD_CHECK(pthread_mutexattr_settype(&mattr,
> > PTHREAD_MUTEX_ERRORCHECK));
> > 
> > ... but this is in pocl-hsa.c and you're not building HSA-enabled pocl.
> > 
> > 
> > -- mb  
> 
> Kind regards,
> 
> Oliver
> 

The problem I reported and the source of the error PTHREAD_CHECK() is a missing linking
against pthread in our devel/ocl-icd port. Fixing that leaves me then with the well known
bug reported here:

 https://github.com/pocl/pocl/issues/474

and, as mentioned above, here

http://lists.llvm.org/pipermail/llvm-bugs/2016-October/051237.html

Regards,

Oliver
- -- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).
-----BEGIN PGP SIGNATURE-----

iLUEARMKAB0WIQQZVZMzAtwC2T/86TrS528fyFhYlAUCWs3VrQAKCRDS528fyFhY
lK9LAf0Swv3er2r3EI8Wv0EerjsVvEvBfkVLqo83shuZhekbEOD60/mw47B9TSer
m6t+oGUDAHrnWC/S72n/0ib80iAuAf9Vbow/J8Thu2bKZDv1oqf46RmHPCYAbuxv
oVMSMvh3zepEJK2+Rp9/Oq6u9BzCl3uNGno1ejm0/bGLD20tPaPU
=dcsI
-----END PGP SIGNATURE-----