Install blas & mkl (#1060)

* Install blas & mkl

These are necessary for the TPU pytorch XLA.
They also provide a significant perforance boost.

They used to be automatically included when we were installing torch
from conda but they are not now that we install torch from pip.

I added a test to prevent regression.

http:https://b/195414519

* separate conda commands to speed up resolution

Dockerfile

@@ -42,6 +42,7 @@ RUN conda config --add channels conda-forge && \
  conda config --add channels nvidia && \
  conda config --add channels rapidsai && \
  # ^ rapidsai is the highest priority channel, default lowest, conda-forge 2nd lowest.
+ conda install mkl blas && \
  # b/161473620#comment7 pin required to prevent resolver from picking pysal 1.x., pysal 2.2.x is also downloading data on import.
  conda install cartopy=0.19 imagemagick=7.0 pyproj==3.1.0 pysal==2.1.0 && \
  /tmp/clean-layer.sh

tests/test_numpy.py

@@ -12,5 +12,5 @@ def test_array(self):
  # Numpy must be linked to the MKL. (Occasionally, a third-party package will muck up the installation
  # and numpy will be reinstalled with an OpenBLAS backing.)
  def test_mkl(self):
- # This will throw an exception if the MKL is not linked correctly.
- get_info("blas_mkl")
+ # This will throw an exception if the MKL is not linked correctly or return an empty dict.
+ self.assertTrue(get_info("blas_mkl"))