Regression test and OpenFAST updates

[ghost]

Regression Test

• New failure criteria

if channelRange < 1 {
	norm = InfinityNorm
} else {
	norm = InfinityNorm / channelRange
}

• Optional error plotting and results summary provided with matplotlib generated files embedded into html
• Improved results output when running the regression test manually
• Improved control when running the regression test manually - single case execution, generate plots without running the case (great for debugging)
• Added abstraction and improved the API's to the custom python infrastructure

OpenFAST

• Export unpacked binary output file
• Git commit is embedded in the binary when built with CMake; VS integration is on the way
  -- Including the git info in the module specific outputs is more involved since each module handles its own I/O and the I/O code is tucked into larger subroutines - saving for future work
• Fix BeamDyn, BeamDyn driver, and OrcaFlex driver precision-type bugs
• Replace unicode with ascii characters in output
• Additional meta data is printed in the OpenFAST log:

**************************************************************************************************
 OpenFAST

 Copyright (C) 2017 National Renewable Energy Laboratory

 This program comes with ABSOLUTELY NO WARRANTY. See the "license.txt" file distributed with this
 software for details.
 **************************************************************************************************

 OpenFAST v8.17.01c-dev-dc586c7*
 Compile Info:
  - Architecture: 64 bit
  - Precision: double
 Execution Info:
  - Date: 08/28/2017
  - Time: 19:55:51-0600

 OpenFAST input file heading:
     FAST Certification Test #06: AOC 15/50 with many DOFs with gen start, loss of grid, and
     tip-brake shutdown. Many parameters are pure fiction.

 Running ElastoDyn.
 Running AeroDyn14.
 Running InflowWind.
 Running ServoDyn.
  Timestep: 0 of 35 seconds.

TO DO BEFORE MERGE

• Verify on peregrine & windows
• Update r-test baseline solutions for all machine/compiler combinations
• Include OpenFAST 0.1.0 - OpenFAST 1.0.0 comparison
• Update the versions for glue codes and modules

[sayerhs]

Is it possible to add the original repo from where you got this file? Also are the licensing terms of the original repo/file compatible with moving it into OpenFAST repo?

[ghost]

@sayerhs good point. that file is from this widely used github repo: https://github.com/rpavlik/cmake-modules. It is licensed under Boost Version 1.0. The original author and a link to the license are included in the docstring of the files included in our repo so I think we are in the clear legally. For reference, the first paragraph of the license is included below. And I'll add a link to the repo in that file's docstring.

Boost Software License - Version 1.0 - August 17th, 2003

Permission is hereby granted, free of charge, to any person or organization
obtaining a copy of the software and accompanying documentation covered by
this license (the "Software") to use, reproduce, display, distribute,
execute, and transmit the Software, and to prepare derivative works of the
Software, and to permit third-parties to whom the Software is furnished to
do so, all subject to the following:

[sayerhs]

@rafaelmudafort Let's add a link to the git repo in the CMake file comments. That should suffice.

[sayerhs]

Same issue with this file.

[ghost]

see above

[sayerhs]

Is this an OpenFAST specific modification? Standard CMake variable is not BLASLIB, I think that is Intel MKL variable.

[ghost]

@gantech was this BLASLIB variable only used for your script fast-build.sh?

[gantech]

I think I used it only for fast-build.sh and specifically to compile with mkl. I'm not sure if it was required. Is there another path forward?

   -DCMAKE_INSTALL_PREFIX=$openfast_dir/install/ \
   -DCMAKE_BUILD_TYPE=RELEASE \
   -DBUILD_FAST_CPP_API=ON \
   -DYAML_ROOT:PATH=$yaml_install_dir \
   -DHDF5_USE_STATIC_LIBRARIES=ON \
   -DHDF5_ROOT:PATH=$hdf5_install_dir \
   -DLAPACK_LIBRARIES="$BLASLIB" \
   -DBLAS_LIBRARIES="$BLASLIB" \
   -DFPE_TRAP_ENABLED=OFF \
   $EXTRA_ARGS \
../ &> log.cmake```

[ghost]

@gantech that script can use the BLASLIB variable, but I think @sayerhs comment is that the line in the instructions is not accurate since the cmake or spack build processes do not involve explicitly setting that BLASLIB variable.

[ghost]

How about this edit:

OpenFAST has the following dependencies:

- LAPACK libraries
- For the optional C++ API, `HDF5 <https://support.hdfgroup.org/HDF5/>`__ (provided by ``HDF5_ROOT``) and `yaml-cpp <https://github.com/jbeder/yaml-cpp>`__ (provided by ``YAML_ROOT``)
- For the optional testing framework, Python 3+

[gantech]

Yup... I agree. What is the Cmake standard for Lapack libraries though? May be I should be setting LAPACK_DIR=${BLASLIB} in the sample install scripts.

[sayerhs]

Intel's BLASLIB includes both -L flags and -l. My recommendation would be to clarify in the documentation that the users should set BLAS_LIBRARIES and LAPACK_LIBRARIES appropriately if the library isn't found in standard paths. And use BLASLIB as an example when using Intel MKL.

[gantech]

ok... Can I submit a separate pull request reg. this then?

[sayerhs]

Push to this branch/pull request (assuming Raf gives you write access) it is a simple doc fix anyway.

[ghost]

@gantech you should be able to submit edits to this pull request. lmk if not

[michaelasprague]

@raf what is your plan for "Include OpenFAST 0.1.0 - OpenFAST 1.0.0 comparison"

[ghost]

@michaelasprague that will be a document listing the test cases that failed the regression test, documenting some of the major differences in solutions, and listing the procedure used for the regression test. This is very similar to the document I've prepared (though not yet fully shared) on the FAST v8 - OpenFAST 0.1.0 comparison.

[HaymanConsulting]

There will be some further tidying up of the GIT hash-related code, but otherwise, I'm fine with these changes.