ptestlong failures Sage-9.3

[Richard_L]

'make ptestlong' gave the following error summary on a new build from source:

----------------------------------------------------------------------
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/calculus/riemann.pyx  # Killed due to illegal instruction
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/schemes/riemann_surfaces/riemann_surface.py  # Killed due to illegal instruction
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/matrix/matrix2.pyx  # Killed due to illegal instruction
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/misc/functional.py  # Killed due to illegal instruction
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/calculus/interpolators.pyx  # Killed due to illegal instruction
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/matrix/matrix_double_dense.pyx  # Killed due to illegal instruction
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/schemes/affine/affine_homset.py  # Killed due to illegal instruction
sage -t --long --warn-long 103.3 --random-seed=0 src/sage/modular/modform/numerical.py  # Killed due to illegal instruction
----------------------------------------------------------------------
Total time for all tests: 3514.1 seconds
    cpu time: 22425.1 seconds
    cumulative wall time: 27133.4 seconds
make: *** [Makefile:201: ptestlong] Error 16

I allowed the package to build Python 3.9.2. The OS is openSUSE Leap 15.2. There were 8 threads active during testing.

[Dima Pasechnik]

On Sun, May 16, 2021 at 4:26 PM Richard_L <Rich...@lozestech.com> wrote:
>
> 'make ptestlong' gave the following error summary on a new build from source:
>
> ----------------------------------------------------------------------
> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/calculus/riemann.pyx # Killed due to illegal instruction

this kind of error indicates that a component is using CPU
instructions not understood by your CPU.
It could be either something you installed from the system, or
something that you built from source as a part of Sage.
You can look in logs/ptestlong.log to locate more precise errors/stack dumps.

> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/schemes/riemann_surfaces/riemann_surface.py # Killed due to illegal instruction
> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/matrix/matrix2.pyx # Killed due to illegal instruction
> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/misc/functional.py # Killed due to illegal instruction
> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/calculus/interpolators.pyx # Killed due to illegal instruction
> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/matrix/matrix_double_dense.pyx # Killed due to illegal instruction
> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/schemes/affine/affine_homset.py # Killed due to illegal instruction
> sage -t --long --warn-long 103.3 --random-seed=0 src/sage/modular/modform/numerical.py # Killed due to illegal instruction
> ----------------------------------------------------------------------
> Total time for all tests: 3514.1 seconds
> cpu time: 22425.1 seconds
> cumulative wall time: 27133.4 seconds
> make: *** [Makefile:201: ptestlong] Error 16
>
> I allowed the package to build Python 3.9.2. The OS is openSUSE Leap 15.2. There were 8 threads active during testing.
>

> --
> You received this message because you are subscribed to the Google Groups "sage-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/f499990e-b662-4ace-b2ee-74a4cec2bc7cn%40googlegroups.com.

[Richard_L]

What I find in common for the 7 failures is:

"/home/rllozes/sage-9.3/local/lib/libopenblas.so.0(zgemm_kernel_n+0x9d)[0x7f2b0f8c549d]", and further

"0x00007f2b0f8c549d in zgemm_kernel_n () at ../kernel/x86_64/zgemm_kernel_1x4_nehalem.S:214"

indicating an openblas error.

The install.log shows that we built openblas with correct detection of AVX (yes), AVX2 (no), and AVX-512 (no), which was once a common root cause for openblas failures (ca. 2012). OTOH, the system already supplies an openblas in /usr/lib64, which configure did not find.

[Dima Pasechnik]

On Mon, 17 May 2021, 16:42 Richard_L, <Rich...@lozestech.com> wrote:

What I find in common for the 7 failures is:

"/home/rllozes/sage-9.3/local/lib/libopenblas.so.0(zgemm_kernel_n+0x9d)[0x7f2b0f8c549d]", and further

"0x00007f2b0f8c549d in zgemm_kernel_n () at ../kernel/x86_64/zgemm_kernel_1x4_nehalem.S:214"

indicating an openblas error.

The install.log shows that we built openblas with correct detection of AVX (yes), AVX2 (no), and AVX-512 (no), which was once a common root cause for openblas failures (ca. 2012). OTOH, the system already supplies an openblas in /usr/lib64, which configure did not find.

you might need the development headers etc for system openblas installed.

Or, perhaps, they are installed, but pkg-config does not know about it.

Do you have system wide pkg-config installed?

To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/0d3ad5a5-891b-4005-b75a-07c378406d7dn%40googlegroups.com.

[Richard_L]

To your first point, the development headers are not installed. That can be remedied easily.

To your second point, yes, pkg-config is installed. There is no openblas.pc, apart from that in the sage-9.3 directory tree.

BTW, I had no problem with Sage-9.2, which also built openblas (0.3.9) from source.

[Dima Pasechnik]

please check the last part of the output of ./configure 

among other things it lists system libraries to install (rather than build from source).

it particular it would print the contents of

https://github.com/sagemath/sage/blob/develop/build/pkgs/openblas/distros/opensuse.txt

which is

openblas-devel

Most probably the pc file for openblas will be there, and we would not have to try solving this puzzle for you 

first do

make openblas-clean

to remove (broken?) Sage installation of 

openblas.

To view this discussion on the web visit https://groups.google.com/d/msgid/sage-devel/f4a85ad8-770f-4a25-9e9f-110c0479b08fn%40googlegroups.com.

[Richard_L]

Thanks, Dima.

Installing openblas-devel did the trick. Confusingly, the system default installation is libopenblas_pthreads0 and its companion header file,, while the headers and libraries Sage wants are found in openblas-devel. Also, the system default is older (0.3.7) even than the version included in SageMath-9.2 (0.3.9). But now I get no errors from ptestlong.

I'll probably have to revisit this in a few months. openSUSE 15.3 will drop soon, and doubtless will include a much newer openblas.