Fortran support

[arjo129]

Recently I tried to build GNU octave through emscripten, however it failed because of parts written in Fortran. may

be it would be possible to add fortran, as I understand currently emscripten is translating llvm bytecode to javascript. And in extension, one could add Ada and Obj-C support?

[kripken]

In principle yes. Search the issues and mailing list, I think some of these have work on them done already.

[jandoerntlein]

Here is what I did to enable Fortran support in Emscripten:

• installed DraggonEgg plugin for gfortran (dragonegg.llvm.org)
• introduced SHARED.DRAGONEGG with a path to dragonegg.so in the .emscripten config file
  DRAGONEGG = os.path.expanduser('/usr/lib/gcc/x86_64-linux-gnu/4.6/plugin/dragonegg.so')
• For the sake of clarity added "emfortran" (like em++) as a script for Fortan files
• in the toolchain cmake/Platform/Emscripten.cmake, add "emfortran" as a compiler

(...)
if ("${CMAKE_C_COMPILER}" STREQUAL "")
    set(CMAKE_C_COMPILER "${EMSCRIPTEN_ROOT_PATH}/emcc${EMCC_SUFFIX}")
endif()
if ("${CMAKE_CXX_COMPILER}" STREQUAL "")
    set(CMAKE_CXX_COMPILER "${EMSCRIPTEN_ROOT_PATH}/em++${EMCC_SUFFIX}")
endif()
if ("${CMAKE_Fortran_COMPILER}" STREQUAL "")
    set(CMAKE_Fortran_COMPILER "${EMSCRIPTEN_ROOT_PATH}/emfortran${EMCC_SUFFIX}") # use our new script
endif()
(...)
include(CMakeForceCompiler)
CMAKE_FORCE_C_COMPILER("${CMAKE_C_COMPILER}" Clang)
CMAKE_FORCE_CXX_COMPILER("${CMAKE_CXX_COMPILER}" Clang)
CMAKE_FORCE_Fortran_COMPILER("${CMAKE_Fortran_COMPILER}" gfortran) # use gfortran with dragonegg
(...)
set(CMAKE_C_FLAGS_RELEASE "-DNDEBUG" CACHE STRING "Emscripten-overridden CMAKE_C_FLAGS_RELEASE")
set(CMAKE_C_FLAGS_MINSIZEREL "-DNDEBUG" CACHE STRING "Emscripten-overridden CMAKE_C_FLAGS_MINSIZEREL")
set(CMAKE_C_FLAGS_RELWITHDEBINFO "" CACHE STRING "Emscripten-overridden CMAKE_C_FLAGS_RELWITHDEBINFO")
set(CMAKE_CXX_FLAGS_RELEASE "-DNDEBUG" CACHE STRING "Emscripten-overridden CMAKE_CXX_FLAGS_RELEASE")
set(CMAKE_CXX_FLAGS_MINSIZEREL "-DNDEBUG" CACHE STRING "Emscripten-overridden CMAKE_CXX_FLAGS_MINSIZEREL")
set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "" CACHE STRING "Emscripten-overridden CMAKE_CXX_FLAGS_RELWITHDEBINFO")
set(CMAKE_Fortran_LINK_EXECUTABLE "") # don't link anything
(...)

• finally edited "emcc" to parse Fortran files as well and added functionality to use gfortran and dragonegg

FORTRAN_ENDINGS = ('.f', '.F') # define the endings to identify a Fortran source file
CXX_SUFFIXES = ('.cpp', '.cxx', '.cc')
SOURCE_SUFFIXES = ('.c', '.cpp', '.cxx', '.cc', '.m', '.mm', '.f', '.FF') # accept .f and .FF files

• here is the major change to emcc:

# First, generate LLVM bitcode. For each input file, we get base.o with bitcode
  for input_file in input_files:
    file_suffix = filename_type_suffix(input_file)
    if file_suffix.endswith(SOURCE_SUFFIXES):
      logging.debug('compiling source file: ' + input_file)
      input_file = shared.Building.preprocess(input_file, in_temp(uniquename(input_file)))
      output_file = in_temp(unsuffixed(uniquename(input_file)) + '.o')
      temp_files.append(output_file)
      fortran_llvm = in_temp(unsuffixed(uniquename(input_file)) + '.ll')
      if file_suffix.endswith(FORTRAN_ENDINGS): # create .ll (bitcode) file from .f (Fortran) files
        call    = 'gfortran'
        args    = [input_file, '-flto', '-S', '-m32', '-fplugin='+shared.DRAGONEGG, '-O2', '-o', fortran_llvm] # use the DRAGONEGG variable from the .emscripten config file
      else: 
        args = newargs + ['-emit-llvm', '-c', input_file, '-o', output_file]
        if file_suffix.endswith(CXX_SUFFIXES):
          args += shared.EMSDK_CXX_OPTS
      logging.debug("running: " + call + ' ' + ' '.join(args))
      execute([call] + args) # let compiler frontend print directly, so colors are saved (PIPE kills that)
      if file_suffix.endswith(FORTRAN_ENDINGS):
        shared.Building.llvm_as(fortran_llvm, output_file) # create .o file from .ll
      if not os.path.exists(output_file):
        logging.error('compiler frontend failed to generate LLVM bitcode, halting')
        sys.exit(1)
    else: # bitcode
        (...)

I know that this patch is a very hacky one. It would be great if someone could present a similiar solution that can be merged into the official repo!

[gagern]

I've been using a similar workaround today (in an attempt to compile LAPACK), except I've used a shell script instead of Python code to run gfortran and llvm-as. And I did so without intermediate file, using a stdout to stdin pipeline instead.

I get warnings that the bytecode of the Fortran files is from a different triple than what emcc compils from C. So I guess one would have to compile Dragonegg against the LLVM version of emscripten. Not sure (yet) how to do that.

If I feed a Fortran file to my native clang it compiles that all right. I don't know how it does that, whether it's calling gfortran or gcc behind the scenes, and if so, whether it relies on Dragonegg. I also don't know where that capability gets configured, how to enable that for the emscripten build of LLCM and clang.

Once Fortran can be compiled, we should investigate ways to compile libgfortran as well, to offer a suitable runtime environment. But Fortran sources first.

Search the issues and mailing list, I think some of these have work on them done already.

Some pointers:

• Fortran compilation looks like the most thorough discussion so far, but apparently petered out at some point with no tangible results remaining.
• Emfortran where @zmaril suggested his emfortran script.
• Porting R used f2c to port its Fortran 77 code. Not an option for more recent dialects.
• Port of BLAS, LAPACK, etc? is what I'm currently interested in. Apparently used the f2c approach as well.

[jandoerntlein]

I get warnings that the bytecode of the Fortran files is from a different triple than what emcc compils from C. So I guess one would have to compile Dragonegg against the LLVM version of emscripten. Not sure (yet) how to do that.

I believe you can safely ignore that as long as you make sure that the target architecture (like x86, x64 etc) matches. But still it would be nice to fix this :)

[jakirkham]

Related to issue ( #5553 ).

[stale]

This issue has been automatically marked as stale because there has been no activity in the past year. It will be closed automatically if no further activity occurs in the next 7 days. Feel free to re-open at any time if this issue is still relevant.

[jakirkham]

Yeah I guess this or #5553 should be closed and marked as a duplicate.

[sbc100]

Closing as a duplicate.