stream

[Sezoir]

Enable building and execution of the stream application with the LLVM Flang compiler.

[Sezoir]

Next steps: I am not planning to take on this project. This is just an initial report found. Please feel free to work on this.

stream:

https://www.cs.virginia.edu/stream/
https://www.cs.virginia.edu/stream/FTP/Code/

Initial report for stream:

• Minimum version of Fortran required using gfortran: F95
• How to compile with gfortran:
  1. First you can download the project by running in bash:

  wget -r -np -nH -l3 -R "index.html*" https://www.cs.virginia.edu/stream/FTP/Code/
  cd stream/FTP/Code
  

  The parameters given are used for the following: -r for recursive download; -R "index.html*" to exclude the unrelated "web" files; -l3 limits recursive search to 3 layers; -np stops ascending to the parent directory; -nh to not create a host directory.
  2. Now you can compile the project by running:

  make FF=gfortran
  

• Does it compile with flang-new: Yes
  You can compile stream, but requires a slight hack, as well as a build of PGMATH. I use classic flang to get a build of PGMATH. To compile:
  1. In mysecond.c file, change the function name mysecond_ to _QPmysecond.
  2. Now you can compile with flang-new with:

  make FF=flang-new CC=clang-12 FFLAGS="-L/path/to/flang/runtime/build/lib -lpgmath"
  

  Where the path to your PGMATH build lib directory is correctly modified.
  An important note for the compilation step is the change in mysecond.c code of mysecond_. In theory we should not need to change this, however we do not currently include the ExternalNameConversion pass in the pass pipeline definition in the driver. We should not require this workaround once the pass is included in the default pass pipeline. Without it, the linker ends up not being able to find the mysecond_ subroutine in stream.f.

Versions:

• gcc version used: 9.3.0
• gfortran version used: 9.3.0
• flang-new, fir-dev branch, HEAD: fb8eb63
• clang-12: 12.0.0
• pgmath built from classic flang, HEAD: e7dedbf9200e740bf38b37892597498fd7dcbb41
• Architecture for compilation: aarch64