tracking issue topics Smit

[rgommers]

Hi @Smit-create, as we talked about on the weekend, let's keep a list of topics and ideas that would be interesting to work on.

• Tracking issue for Meson support of submodules #22 (misc, io, maybe there's more later)
• Removing use of deprecated NumPy C API from SciPy #30
• Investigate if the build time of the sparsetools extension can be improved #33 (this may be a challenging one)
• fixing build warnings in the SciPy code base that we temporarily silenced - bugs could be hiding here: Investigating and fixing build warnings #42
• Try to build with ASAN and document how to get useful output #32
• Following up on the signal.sepfir2d issues: Tackling some sepfir2d improvements #43
• meson: timestamp caching in CI trick. The build/ folder can be saved and reused for future PRs, but the timestamps must be manipulated in a smart way in order to get rebuilds for only the targets that were affected by a PR and commits since the last rebuild. I read a blog post about this that I cannot find right now. (no separate issue yet)
• It would be interesting to learn how BLAS/LAPACK library affects performance on different CPUs, as well as the relative performance of BLAS libraries. There are blog posts about this, in particular for OpenBLAS and MKL on Intel and AMD CPUs. We're getting more interested in aarch64 too. And haven't seen much about arm64 (Mac M1) yet.
• Similar to above, a little easier: how does build time and benchmark performance depend on optimization level (-O0 ... -O3, -g, etc.).
• PROPACK support in Meson build files (finishing CI: add -Werror in Meson job, and sync with upstream master #62).
• Write CI script to verify Meson installed all relevant test files #69
• Verify/update style of meson.build files #72
• update the meson branch in this fork to be up-to-date with SciPy master (see MAINT: Update meson build according to private API changes #93)
• add a dev.py interface for the Meson build: current idea is to import functions that don't need changes from runtests.py and reimplement build_project.
• ensure stats.linalg tests are installed and passing (currently ~90 failures on master)
• write a uarray backend for a small part of scipy.signal and create a Colab notebook which works with CuPy and https://docs.rapids.ai/api/cusignal/nightly/api.html#module-cusignal.filtering.resample
• Build warnings in the Windows Meson build we should fix #120
• Add a new section to the SciPy docs on Meson about development topics, covering building with different optimization levels (closes BENCH: Benchmarks with meson optimization levels #58) and building with ASAN (closes Try to build with ASAN and document how to get useful output #32).
• Related to the above topic: write a new section on how to use several different builds in parallel: gcc, clang regular builds and a gcc debug build on Linux.
• Make Meson work with gcov code coverage: Compiled code coverage with --gcov doesn't work with Meson scipy/scipy#15896
• sdist and wheels support via meson-python

[Smit-create]

Hi Ralf,
I was trying to work on the benchmark part. I am getting the following error:

numpy.distutils.system_info.NotFoundError: No BLAS/LAPACK libraries found.
   To build Scipy from sources, BLAS & LAPACK libraries need to be installed.
   See site.cfg.example in the Scipy source directory and
   https://docs.scipy.org/doc/scipy/reference/building/index.html for details.

I followed the steps mentioned here: https://docs.scipy.org/doc/scipy/reference/building/linux.html#debian-ubuntu to install BLAS/LAPACK

I got this error while running: $asv run

[rgommers]

Do you have libopenblas-dev installed? If so, it looks like the build cannot find it. There's a way to make it find it by using a site.cfg file (https://github.com/scipy/scipy/blob/master/site.cfg.example).

If you're using conda then this should never be needed, but if you installed Python some other way than it's possible that libopenblas and libopenblas-dev got installed in some nonstandard location that numpy.distutils has trouble with.

Please let me know if that didn't fix it!

[Smit-create]

Please let me know if that didn't fix it!

The whole traceback looks like this:

Details

·· Error running /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/bin/python setup.py build (exit status 1)
   STDOUT -------->
   Running scipy/stats/_generate_pyx.py
   Running scipy/linalg/_generate_pyx.py
   Running scipy/special/_generate_pyx.py
   Processing scipy/interpolate/_bspl.pyx
   Processing scipy/interpolate/interpnd.pyx
   Processing scipy/interpolate/_ppoly.pyx
   Processing scipy/spatial/ckdtree.pyx
   Processing scipy/spatial/_hausdorff.pyx
   Processing scipy/spatial/qhull.pyx
   Processing scipy/spatial/_voronoi.pyx
   Processing scipy/spatial/transform/rotation.pyx
   Processing scipy/stats/_sobol.pyx
   Processing scipy/stats/_qmc_cy.pyx
   Processing scipy/stats/biasedurn.pyx
   Processing scipy/stats/_stats.pyx
   Processing scipy/stats/_boost/src/hypergeom_ufunc.pyx
   Processing scipy/stats/_boost/src/binom_ufunc.pyx
   Processing scipy/stats/_boost/src/nbinom_ufunc.pyx
   Processing scipy/stats/_boost/src/beta_ufunc.pyx
   Processing scipy/signal/_peak_finding_utils.pyx
   Processing scipy/signal/_sosfilt.pyx
   Processing scipy/signal/_spectral.pyx
   Processing scipy/signal/_upfirdn_apply.pyx
   Processing scipy/signal/_max_len_seq_inner.pyx
   Processing scipy/linalg/_solve_toeplitz.pyx
   Processing scipy/linalg/_decomp_update.pyx.in
   Processing scipy/linalg/cython_blas.pyx
   Processing scipy/linalg/_matfuncs_sqrtm_triu.pyx
   Processing scipy/linalg/cython_lapack.pyx
   Processing scipy/sparse/_csparsetools.pyx.in
   Processing scipy/sparse/csgraph/_traversal.pyx
   Processing scipy/sparse/csgraph/_tools.pyx
   Processing scipy/sparse/csgraph/_flow.pyx
   Processing scipy/sparse/csgraph/_min_spanning_tree.pyx
   Processing scipy/sparse/csgraph/_matching.pyx
   Processing scipy/sparse/csgraph/_shortest_path.pyx
   Processing scipy/sparse/csgraph/_reordering.pyx
   Processing scipy/_lib/messagestream.pyx
   Processing scipy/_lib/_test_deprecation_call.pyx
   Processing scipy/_lib/_test_deprecation_def.pyx
   Processing scipy/_lib/_ccallback_c.pyx
   Processing scipy/io/matlab/mio5_utils.pyx
   Processing scipy/io/matlab/mio_utils.pyx
   Processing scipy/io/matlab/streams.pyx
   Processing scipy/special/_ellip_harm_2.pyx
   Processing scipy/special/_comb.pyx
   Processing scipy/special/cython_special.pyx
   Processing scipy/special/_ufuncs_cxx.pyx
   Processing scipy/special/_ufuncs.pyx
   Processing scipy/special/_test_round.pyx
   Processing scipy/fftpack/convolve.pyx
   Processing scipy/cluster/_vq.pyx
   Processing scipy/cluster/_optimal_leaf_ordering.pyx
   Processing scipy/cluster/_hierarchy.pyx
   Processing scipy/ndimage/src/_cytest.pyx
   Processing scipy/ndimage/src/_ni_label.pyx
   Processing scipy/optimize/_bglu_dense.pyx
   Processing scipy/optimize/_group_columns.pyx
   Processing scipy/optimize/_highs/cython/src/_highs_wrapper.pyx
   Processing scipy/optimize/_highs/cython/src/_highs_constants.pyx
   Processing scipy/optimize/_lsq/givens_elimination.pyx
   Processing scipy/optimize/_trlib/_trlib.pyx
   Processing scipy/optimize/cython_optimize/_zeros.pyx.in
   Cythonizing sources
   lapack_opt_info:
   lapack_mkl_info:
   customize UnixCCompiler
   C compiler: /home/smit/anaconda3/envs/scipy-dev/bin/x86_64-conda-linux-gnu-cc -Wno-unused-result -Wsign-compare -DNDEBUG -fwrapv -O2 -Wall -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/smit/anaconda3/envs/scipy-dev/include -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/smit/anaconda3/envs/scipy-dev/include -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem /home/smit/anaconda3/envs/scipy-dev/include -DNDEBUG -D_FORTIFY_SOURCE=2 -O2 -isystem /home/smit/anaconda3/envs/scipy-dev/include -fPIC
   
   creating /tmp/tmpizdjdiu3/tmp
   creating /tmp/tmpizdjdiu3/tmp/tmpizdjdiu3
   compile options: '-MMD -MF /tmp/tmpizdjdiu3/file.c.d -c'
   x86_64-conda-linux-gnu-cc: /tmp/tmpizdjdiu3/file.c
     libraries mkl_rt not found in ['/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib', '/usr/local/lib', '/usr/lib', '/usr/lib/']
     NOT AVAILABLE
   
   openblas_lapack_info:
     libraries openblas not found in ['/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib', '/usr/local/lib', '/usr/lib', '/usr/lib/']
     NOT AVAILABLE
   
   openblas_clapack_info:
     libraries openblas,lapack not found in ['/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib', '/usr/local/lib', '/usr/lib', '/usr/lib/']
     NOT AVAILABLE
   
   flame_info:
     libraries flame not found in ['/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib', '/usr/local/lib', '/usr/lib', '/usr/lib/']
     NOT AVAILABLE
   
   accelerate_info:
     NOT AVAILABLE
   
   atlas_3_10_threads_info:
   Setting PTATLAS=ATLAS
     libraries lapack_atlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries tatlas,tatlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries lapack_atlas not found in /usr/local/lib
     libraries tatlas,tatlas not found in /usr/local/lib
     libraries lapack_atlas not found in /usr/lib
     libraries tatlas,tatlas not found in /usr/lib
     libraries lapack_atlas not found in /usr/lib/
     libraries tatlas,tatlas not found in /usr/lib/
   <class 'numpy.distutils.system_info.atlas_3_10_threads_info'>
     NOT AVAILABLE
   
   atlas_3_10_info:
     libraries lapack_atlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries satlas,satlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries lapack_atlas not found in /usr/local/lib
     libraries satlas,satlas not found in /usr/local/lib
     libraries lapack_atlas not found in /usr/lib
     libraries satlas,satlas not found in /usr/lib
     libraries lapack_atlas not found in /usr/lib/
     libraries satlas,satlas not found in /usr/lib/
   <class 'numpy.distutils.system_info.atlas_3_10_info'>
     NOT AVAILABLE
   
   atlas_threads_info:
   Setting PTATLAS=ATLAS
     libraries lapack_atlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries ptf77blas,ptcblas,atlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries lapack_atlas not found in /usr/local/lib
     libraries ptf77blas,ptcblas,atlas not found in /usr/local/lib
     libraries lapack_atlas not found in /usr/lib
     libraries ptf77blas,ptcblas,atlas not found in /usr/lib
     libraries lapack_atlas not found in /usr/lib/
     libraries ptf77blas,ptcblas,atlas not found in /usr/lib/
   <class 'numpy.distutils.system_info.atlas_threads_info'>
     NOT AVAILABLE
   
   atlas_info:
     libraries lapack_atlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries f77blas,cblas,atlas not found in /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib
     libraries lapack_atlas not found in /usr/local/lib
     libraries f77blas,cblas,atlas not found in /usr/local/lib
     libraries lapack_atlas not found in /usr/lib
     libraries f77blas,cblas,atlas not found in /usr/lib
     libraries lapack_atlas not found in /usr/lib/
     libraries f77blas,cblas,atlas not found in /usr/lib/
   <class 'numpy.distutils.system_info.atlas_info'>
     NOT AVAILABLE
   
   lapack_info:
     libraries lapack not found in ['/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib', '/usr/local/lib', '/usr/lib', '/usr/lib/']
     NOT AVAILABLE
   
   lapack_src_info:
     NOT AVAILABLE
   
     NOT AVAILABLE
   
   STDERR -------->
   Running from SciPy source directory.
   /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib/python3.9/site-packages/numpy/distutils/system_info.py:1858: UserWarning: 
       Lapack (http://www.netlib.org/lapack/) libraries not found.
       Directories to search for the libraries can be specified in the
       numpy/distutils/site.cfg file (section [lapack]) or by setting
       the LAPACK environment variable.
     return getattr(self, '_calc_info_{}'.format(name))()
   /home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib/python3.9/site-packages/numpy/distutils/system_info.py:1858: UserWarning: 
       Lapack (http://www.netlib.org/lapack/) sources not found.
       Directories to search for the sources can be specified in the
       numpy/distutils/site.cfg file (section [lapack_src]) or by setting
       the LAPACK_SRC environment variable.
     return getattr(self, '_calc_info_{}'.format(name))()
   Traceback (most recent call last):
     File "/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/project/setup.py", line 631, in <module>
       setup_package()
     File "/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/project/setup.py", line 627, in setup_package
       setup(**metadata)
     File "/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/lib/python3.9/site-packages/numpy/distutils/core.py", line 135, in setup
       config = configuration()
     File "/home/smit/Smitlunagariya/scipy/benchmarks/env/9774801ae38aac7b90590412a25793ef/project/setup.py", line 530, in configuration
       raise NotFoundError(msg)
   numpy.distutils.system_info.NotFoundError: No BLAS/LAPACK libraries found.
   To build Scipy from sources, BLAS & LAPACK libraries need to be installed.
   See site.cfg.example in the Scipy source directory and
   https://docs.scipy.org/doc/scipy/reference/building/index.html for details.

·· Failed to build the project and import the benchmark suite.

it's possible that libopenblas and libopenblas-dev got installed in some nonstandard location

I tried the following too(scipy#9481 (comment)):

$ export LAPACK=/usr/lib/openblas/lib/libopenblas.so
$ export ATLAS=/usr/lib/openblas/lib/libopenblas.so
$ asv run

EDIT:
$ python runtests.py --bench works for me locally.

Also, I edited https://github.com/Smit-create/scipy/blob/meson/.github/workflows/linux.yml and it works fine, see: https://github.com/Smit-create/scipy/runs/3446108473?check_suite_focus=true

[rgommers]

For comparing BLAS libraries on the same machine, here's how to easily switch them out with conda: https://conda-forge.org/docs/maintainer/knowledge_base.html#switching-blas-implementation

[Smit-create]

It produces the following error:

$ conda install "libblas=*=*mkl"
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.

PackagesNotFoundError: The following packages are not available from current channels:

  - libblas[build=*mkl]

Current channels:

  - https://repo.anaconda.com/pkgs/main/linux-64
  - https://repo.anaconda.com/pkgs/main/noarch
  - https://repo.anaconda.com/pkgs/r/linux-64
  - https://repo.anaconda.com/pkgs/r/noarch

To search for alternate channels that may provide the conda package you're
looking for, navigate to

    https://anaconda.org

and use the search bar at the top of the page.

From $ conda list, I get:

# Name                    Version                   Build  Channel

libblas                   3.9.0                9_openblas    conda-forge
libcblas                  3.9.0                9_openblas    conda-forge
libopenblas               0.3.15          pthreads_h8fe5266_1    conda-forge
mkl                       2021.3.0           h06a4308_520  

Should there be something else to change the build type?

[rgommers]

Ah you're installing from defaults (see "Current channels" listed), this feature is specific to conda-forge.

[rgommers]

The solve is slow when done in an existing scipy-dev environment based on the environment.yml in SciPy master, but it does work:

$ conda install "libblas=*=*mkl"
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: done
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: done

## Package Plan ##

  environment location: /home/rgommers/anaconda3/envs/scipy-dev

  added / updated specs:
    - libblas[build=*mkl]


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    binutils_linux-64-2.36     |       hf3e587d_1          23 KB  conda-forge
    gcc_impl_linux-64-9.4.0    |       h03d3576_8        45.7 MB  conda-forge
    gcc_linux-64-9.4.0         |       h391b98a_1          24 KB  conda-forge
    gfortran_impl_linux-64-9.4.0|       h0003116_8        10.0 MB  conda-forge
    gfortran_linux-64-9.4.0    |       hf0ab688_1          23 KB  conda-forge
    gxx_impl_linux-64-9.4.0    |       h03d3576_8        10.6 MB  conda-forge
    gxx_linux-64-9.4.0         |       h0316aca_1          23 KB  conda-forge
    libblas-3.9.0              |   11_linux64_mkl          12 KB  conda-forge
    libcblas-3.9.0             |   11_linux64_mkl          11 KB  conda-forge
    libgcc-devel_linux-64-9.4.0|       hd854feb_8         4.0 MB  conda-forge
    liblapack-3.9.0            |   11_linux64_mkl          11 KB  conda-forge
    libsanitizer-9.4.0         |       h79bfe98_8         6.9 MB  conda-forge
    libstdcxx-devel_linux-64-9.4.0|       hd854feb_8         9.9 MB  conda-forge
    openssl-1.1.1l             |       h7f98852_0         2.1 MB  conda-forge
    ------------------------------------------------------------
                                           Total:        89.4 MB

The following NEW packages will be INSTALLED:

  libsanitizer       conda-forge/linux-64::libsanitizer-9.4.0-h79bfe98_8
  llvm-openmp        conda-forge/linux-64::llvm-openmp-12.0.1-h4bd325d_1
  mkl                conda-forge/linux-64::mkl-2021.3.0-h726a3e6_557
  tbb                conda-forge/linux-64::tbb-2021.3.0-h4bd325d_0

The following packages will be UPDATED:

  gcc_impl_linux-64                       9.3.0-h70c0ae5_19 --> 9.4.0-h03d3576_8
  gcc_linux-64                            9.3.0-hf25ea35_32 --> 9.4.0-h391b98a_1
  gfortran_impl_lin~                      9.3.0-hc4a2995_19 --> 9.4.0-h0003116_8
  gfortran_linux-64                       9.3.0-hdc58fab_32 --> 9.4.0-hf0ab688_1
  gxx_impl_linux-64                       9.3.0-hd87eabc_19 --> 9.4.0-h03d3576_8
  gxx_linux-64                            9.3.0-h3fbe746_32 --> 9.4.0-h0316aca_1
  libblas                                  3.9.0-9_openblas --> 3.9.0-11_linux64_mkl
  libcblas                                 3.9.0-9_openblas --> 3.9.0-11_linux64_mkl
  libgcc-devel_linu~                      9.3.0-h7864c58_19 --> 9.4.0-hd854feb_8
  libgcc-ng                               9.3.0-h2828fa1_19 --> 11.1.0-hc902ee8_8
  libgfortran-ng                          9.3.0-hff62375_19 --> 11.1.0-h69a702a_8
  libgfortran5                            9.3.0-hff62375_19 --> 11.1.0-h6c583b3_8
  libgomp                                 9.3.0-h2828fa1_19 --> 11.1.0-hc902ee8_8
  liblapack                                3.9.0-9_openblas --> 3.9.0-11_linux64_mkl
  libstdcxx-devel_l~                      9.3.0-hb016644_19 --> 9.4.0-hd854feb_8
  libstdcxx-ng                            9.3.0-h6de172a_19 --> 11.1.0-h56837e0_8
  openssl                                 1.1.1k-h7f98852_0 --> 1.1.1l-h7f98852_0

The following packages will be DOWNGRADED:

  _openmp_mutex                                   4.5-1_gnu --> 4.5-1_llvm
  binutils_linux-64                        2.36-hf3e587d_32 --> 2.36-hf3e587d_1

[Smit-create]

Hi,
These are the results on my machine of linalg benchmarks:

Library                                           Real time                       user time              sys time
----------------------------------------------------------------------------------------------------------------------
openblas                                         10m 9.007s                      19m 14.320s            7m 5.249s
mkl                                              7m 46.486s                      10m 54.084s            0m 28.639s
blis                                             12m 2.356s                      17m 40.777s            7m 2.029s
netlib                                           12m 45.545s                     20m 10.659s            7m 47.239s

My CPU details are:

"arch": "x86_64",
"cpu": "Intel(R) Core(TM) i3-4005U CPU @ 1.70GHz",
"num_cpu": "4",
"os": "Linux 5.4.0-81-generic",
"ram": "3950024",

The python script which I used for obtaining the results can be found at: https://gist.github.com/Smit-create/99f7fb2b8781053c39c3f25215e26cbf

[rgommers]

The "deprecate private API in SciPy" issue is coming to an end, we discussed to after that update the meson branch in this fork to be up-to-date with SciPy master.

[rgommers]

Old issue, closing.