Issue #249. OCLTrans() generates OpenCL kernels (also fixes fparser2 stmt_fns)

[sergisiso]

This PR (based on issue #249 ) enables the automatic generation of OpenCL kernels referenced by the FortCL PSy layer. This works for the examples/gocean/eg3 and NemoLite2D (both needed a temporal fix for the array assignments wrongly parsed as statement functions by fparser2)

[codecov-io]

Codecov Report

Merging #387 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #387      +/-   ##
==========================================
+ Coverage   97.55%   97.56%   +<.01%     
==========================================
  Files          24       24              
  Lines       10245    10273      +28     
==========================================
+ Hits         9995    10023      +28     
  Misses        250      250

Impacted Files	Coverage Δ
src/psyclone/psyGen.py	95.38% <100%> (+0.04%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4035127...752e2f0. Read the comment docs.

[sergisiso]

@arporter This PR is ready for review. This is a set of small changes for the OCLTrans() transformation to automatically generate the kernels associated with OpenCL PSy-layer. These are:

• rename_and_write now looks for opencl marked kernels and assign .cl names.
• Statement Functions are corrected to assignments when possible.
• updated examples/gocean/eg3 to use GO_ prefix and updated README.
• OpenCL back-end now always use 'e' for scientific notation (instead of Fortran 'd').
• Unit test produces temporary files in a tmpdir directory.
• Updated documentation.

[sergisiso]

@arporter Argh, sorry, 1 more commit to bring this up to master. But it is ready if it passes the tests.

[arporter]

All looks good.
Documentation has been updated and all three guides build OK.
I have some minor comments - see inline.
Main issues are that I get 5 test failures when running the test suite in parallel with compilation testing enabled (--compile and --compileopencl). Also, with Rupert's imminent change to fparser, it will no longer produce Statement_Function objects in the Parse Tree so one test will need updating.
I'll check pylint next time.

[arporter]

Only minor things left now plus a need to update the documentation to be a bit clearer about what happens when creating OpenCL kernels.
Once those things are done, this is ready to go.

[sergisiso]

@arporter Ready for next review

[arporter]

Almost, almost there. I was going to accept it until I discovered that we now have an XPASSing test (a test marked as xfail that now passes). This is test_module_inline_and_compile in gocean1p0_transformations_test.py. Given how hard this was to find I suggest we change the default behaviour of pytest so that an XPASS is reported as a failure. We can do this by adding:

# Ensure that any XPASS ("unexpectedly passing") results are reported
# as failures in the test suite.
[tool:pytest]
xfail_strict=true

to the PSyclone/setup.cfg file. Would you mind doing this here?

There are also a few tiny things that need fixing - see inline.

[sergisiso]

@arporter Ready for the next review, I made the changes we discussed this morning.

[arporter]

All requested changes have been made.
All tests pass or xfail, including with compilation.
One additional test has been marked as xfail because @arporter will address this in #315.