emscripten/emcc

#!/usr/bin/env python

'''
emcc - compiler helper script
=============================

emcc is a drop-in replacement for a compiler like gcc or clang.

Tell your build system to use this instead of the compiler, and similarly
use emar, emld and emranlib instead of the same command without 'em'.

Example uses:

 * For configure, instead of ./configure, cmake, etc., run emconfigure.py
   with that command as an argument, for example

    emconfigure.py ./configure [options]
  
   emconfigure.py is a tiny script that just sets some environment vars
   as a convenience. The command just shown is equivalent to

    EMMAKEN_JUST_CONFIGURE=1 RANLIB=PATH/emranlib AR=PATH/emar CXX=PATH/em++ CC=PATH/emcc ./configure [options]

   where PATH is the path to this file.

   EMMAKEN_JUST_CONFIGURE tells emcc that it is being run in ./configure,
   so it should relay everything to gcc/g++. You should not define that when
   running make, of course.

 * With CMake, the same command will work (with cmake instead of ./configure). You may also be
   able to do the following in your CMakeLists.txt:

    SET(CMAKE_C_COMPILER "PATH/emcc")
    SET(CMAKE_CXX_COMPILER "PATH/em++")
    SET(CMAKE_LINKER "PATH/emld")
    SET(CMAKE_CXX_LINKER "PATH/emld")
    SET(CMAKE_C_LINK_EXECUTABLE "PATH/emld")
    SET(CMAKE_CXX_LINK_EXECUTABLE "PATH/emld")
    SET(CMAKE_AR "PATH/emar")
    SET(CMAKE_RANLIB "PATH/emranlib")

 * For SCons the shared.py can be imported like so:
    __file__ = str(Dir('#/project_path_to_emscripten/dummy/dummy'))
    __rootpath__ = os.path.abspath(os.path.dirname(os.path.dirname(__file__)))
    def path_from_root(*pathelems):
      return os.path.join(__rootpath__, *pathelems)
    exec(open(path_from_root('tools', 'shared.py'), 'r').read())
    
   For using the Emscripten compilers/linkers/etc. you can do:
    env = Environment()
    ...
    env.Append(CCFLAGS = COMPILER_OPTS)
    env.Replace(LINK = LLVM_LD)
    env.Replace(LD   = LLVM_LD)
   TODO: Document all relevant setup changes

After setting that up, run your build system normally.

Note the appearance of em++ instead of emcc
for the C++ compiler. This is needed for cases where we get
a C++ file with a C extension, in which case CMake can be told
to run g++ on it despite the .c extension, see

  https://github.com/kripken/emscripten/issues/6

(If a similar situation occurs with ./configure, you can do the same there too.)

emcc can be influenced by a few environment variables:

  EMMAKEN_NO_SDK - Will tell emcc *not* to use the emscripten headers. Instead
                   your system headers will be used.

  EMMAKEN_COMPILER - The compiler to be used, if you don't want the default clang.
'''

import os, sys, shutil, tempfile
from subprocess import Popen, PIPE, STDOUT
from tools import shared

DEBUG = os.environ.get('EMCC_DEBUG')
TEMP_DIR = os.environ.get('EMCC_TEMP_DIR')
LEAVE_INPUTS_RAW = os.environ.get('EMCC_LEAVE_INPUTS_RAW') # Do not compile .ll files into .bc, just compile them with emscripten directly
                                                           # Not recommended, this is mainly for the test runner, or if you have some other
                                                           # specific need.
                                                           # One major limitation with this mode is that dlmalloc will not be added in.
                                                           # LLVM optimizations will also not be done.

if DEBUG: print >> sys.stderr, 'emcc: ', ' '.join(sys.argv)

shared.check_sanity()

# Handle some global flags

if len(sys.argv) == 1:
  print 'emcc: no input files'
  exit(0)

if sys.argv[1] == '--version':
  print '''emcc (Emscripten GCC-like replacement) 2.0
Copyright (C) 2011 the Emscripten authors.
This is free and open source software under the MIT license.
There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  '''
  exit(0)
elif sys.argv[1] == '--help':
  this = os.path.basename('em++' if os.environ.get('EMMAKEN_CXX') else 'emcc')

  print '''%s [options] file...

Most normal gcc/g++ options will work, for example:
  --help                   Display this information
  --version                Display compiler version information

Options that are modified or new in %s include:
  -O0                      No optimizations (default)
  -O1                      Simple optimizations, including safe LLVM
                           optimizations, and no runtime assertions
                           or C++ exception catching (to re-enable
                           C++ exception catching, use
                           -s DISABLE_EXCEPTION_CATCHING=0 )
  -O2                      As -O1, plus the relooper (loop recreation),
                           plus closure compiler advanced opts
                           Warning: Compiling with this takes a long time!
  -O3                      As -O2, plus dangerous optimizations that may
                           break the generated code! If that happens, try
                           -O2 and then adding dangerous optimizations one
                           by one.
  -s OPTION=VALUE          JavaScript code generation option passed
                           into the emscripten compiler
  --typed-arrays <mode>    0: No typed arrays
                           1: Parallel typed arrays
                           2: Shared (C-like) typed arrays (default)
  --llvm-opts <level>      0: No LLVM optimizations (default in -O0)
                           1: Safe/portable LLVM optimizations
                              (default in -O1 and above)
                           2: Full, unsafe/unportable LLVM optimizations;
                              this will almost certainly break the
                              generated code!
  --closure <on>           0: No closure compiler (default in -O0, -O1)
                           1: Run closure compiler (default in -O2, -O3)

The target file, if specified (-o <target>), defines what will
be generated:

  <name>.js                JavaScript (default)
  <name>.html              HTML with embedded JavaScript
  <name>.bc                LLVM bitcode
  <name>.o                 LLVM bitcode

The -c option (which tells gcc not to run the linker) will
cause LLVM bitcode to be generated, as %s only generates
JavaScript in the final linking stage of building.

The input file(s) can be either source code files that
Clang can handle (C or C++), LLVM bitcode in binary form,
or LLVM assembly files in human-readable form.

''' % (this, this, this)
  exit(0)

# If this is a configure-type thing, do not compile to JavaScript, instead use clang
# to compile to a native binary (using our headers, so things make sense later)
CONFIGURE_CONFIG = os.environ.get('EMMAKEN_JUST_CONFIGURE')
CMAKE_CONFIG = 'CMakeFiles/cmTryCompileExec.dir' in ' '.join(sys.argv)# or 'CMakeCCompilerId' in ' '.join(sys.argv)
if CONFIGURE_CONFIG or CMAKE_CONFIG:
  compiler = shared.CLANG
  if not ('CXXCompiler' in ' '.join(sys.argv) or os.environ.get('EMMAKEN_CXX')):
    compiler = shared.to_cc(compiler)
  cmd = [compiler] + shared.EMSDK_OPTS + sys.argv[1:]
  if DEBUG: print >> sys.stderr, 'emcc, just configuring: ', compiler, cmd
  exit(os.execvp(compiler, cmd))

if os.environ.get('EMMAKEN_COMPILER'):
  CXX = os.environ['EMMAKEN_COMPILER']
else:
  CXX = shared.CLANG

CC = shared.to_cc(CXX)

# If we got here from a redirection through emmakenxx.py, then force a C++ compiler here
if os.environ.get('EMMAKEN_CXX'):
  CC = CXX

CC_ADDITIONAL_ARGS = shared.COMPILER_OPTS # + ['-g']?

EMMAKEN_CFLAGS = os.environ.get('EMMAKEN_CFLAGS')
if EMMAKEN_CFLAGS: CC_ADDITIONAL_ARGS += EMMAKEN_CFLAGS.split(' ')

# ---------------- Utilities ---------------

SOURCE_SUFFIXES = ('.c', '.cpp', '.cxx', '.cc')
BITCODE_SUFFIXES = ('.bc', '.o', '.ll')

def unsuffixed(name):
  return '.'.join(name.split('.')[:-1])

def unsuffixed_basename(name):
  return os.path.basename(unsuffixed(name))

LLVM_INTERNAL_OPT_LEVEL = 2

# ---------------- End configs -------------

if len(sys.argv) == 1 or sys.argv[1] in ['x', 't']:
  # noop ar
  if DEBUG: print >> sys.stderr, 'emcc, just ar'
  sys.exit(0)

use_cxx = True
header = False # pre-compiled headers. We fake that by just copying the file

for i in range(1, len(sys.argv)):
  arg = sys.argv[i]
  if not arg.startswith('-'):
    if arg.endswith('.c'):
      use_cxx = False
    if arg.endswith('.h') and sys.argv[i-1] != '-include':
      header = True

# Check if a target is specified
target = None
for i in range(len(sys.argv)-1):
  if sys.argv[i].startswith('-o='):
    raise Exception('Invalid syntax: do not use -o=X, use -o X')

  if sys.argv[i] == '-o':
    target = sys.argv[i+1]
    sys.argv = sys.argv[:i] + sys.argv[i+2:]
    break

if header: # header or such
  if DEBUG: print >> sys.stderr, 'Just copy.'
  shutil.copy(sys.argv[-1], sys.argv[-2])
  exit(0)

if TEMP_DIR:
  temp_dir = TEMP_DIR
  if os.path.exists(temp_dir):
    shutil.rmtree(temp_dir) # clear it
  os.makedirs(temp_dir)
else:
  temp_dir = tempfile.mkdtemp()

def in_temp(name):
  return os.path.join(temp_dir, name)

try:
  call = CXX if use_cxx else CC

  ## Parse args

  newargs = sys.argv[1:]

  opt_level = 0
  llvm_opt_level = None
  closure = None

  def check_bad_eq(arg):
    assert '=' not in arg, 'Invalid parameter (do not use "=" with "--" options)'

  for i in range(len(newargs)):
    if newargs[i].startswith('-O'):
      try:
        opt_level = int(newargs[i][2])
        assert 0 <= opt_level <= 3
      except:
        raise Exception('Invalid optimization level: ' + newargs[i])
      newargs[i] = ''
    elif newargs[i].startswith('--llvm-opts'):
      check_bad_eq(newargs[i])
      llvm_opt_level = eval(newargs[i+1])
      assert 0 <= llvm_opt_level <= 1, 'Only two levels of LLVM optimizations are supported so far, 0 (none) and 1 (safe)'
      newargs[i] = ''
      newargs[i+1] = ''
    elif newargs[i].startswith('--closure'):
      check_bad_eq(newargs[i])
      closure = int(newargs[i+1])
      newargs[i] = ''
      newargs[i+1] = ''
    elif newargs[i] == '-MF': # clang cannot handle this, so we fake it
      f = open(newargs[i+1], 'w')
      f.write('\n')
      f.close()
      newargs[i] = ''
      newargs[i+1] = ''
  newargs = [ arg for arg in newargs if arg is not '' ]

  if llvm_opt_level is None: llvm_opt_level = 1 if opt_level >= 1 else 0
  if closure is None: closure = 1 if opt_level >= 2 else 0

  if closure:
    assert os.path.exists(shared.CLOSURE_COMPILER), 'emcc: fatal: Closure compiler (%s) does not exist' % shared.CLOSURE_COMPILER

  settings_changes = []
  for i in range(len(newargs)):
    if newargs[i] == '-s':
      assert '=' in newargs[i+1], 'Incorrect syntax for -s (use -s OPT=VAL): ' + newargs[i+1]
      settings_changes.append(newargs[i+1])
      newargs[i] = newargs[i+1] = ''
    elif newargs[i].startswith('--typed-arrays'):
      assert '=' not in newargs[i], 'Invalid typed arrays parameter (do not use "=")'
      settings_changes.append('USE_TYPED_ARRAYS=' + newargs[i+1])
      newargs[i] = ''
      newargs[i+1] = ''
  newargs = [ arg for arg in newargs if arg is not '' ]

  input_files = []
  has_source_inputs = False
  for i in range(len(newargs)): # find input files XXX this a simple heuristic. we should really analyze based on a full understanding of gcc params,
                                # right now we just assume that what is left contains no more |-x OPT| things
    arg = newargs[i]
    if arg.endswith(SOURCE_SUFFIXES + BITCODE_SUFFIXES): # we already removed -o <target>, so all these should be inputs
      newargs[i] = ''
      if os.path.exists(arg):
        if arg.endswith(SOURCE_SUFFIXES):
          input_files.append(arg)
          has_source_inputs = True
        else:
          # this should be bitcode, make sure it is valid
          if arg.endswith('.ll') or shared.Building.is_bitcode(arg):
            input_files.append(arg)
          else:
            print >> sys.stderr, 'emcc: %s: Not valid LLVM bitcode' % arg
      else:
        print >> sys.stderr, 'emcc: %s: No such file or directory' % arg
  newargs = [ arg for arg in newargs if arg is not '' ]

  assert len(input_files) > 0, 'emcc: no input files'

  newargs += CC_ADDITIONAL_ARGS

  specified_target = target
  target = specified_target if specified_target is not None else 'a.out.js' # specified_target is the user-specified one, target is what we will generate

  target_basename = unsuffixed_basename(target)

  # -c means do not link in gcc, and for us, the parallel is to not go all the way to JS, but stop at bitcode
  has_dash_c = '-c' in newargs
  if has_dash_c:
    assert has_source_inputs, 'Must have source code inputs to use -c'
    target = target_basename + '.o'

  final_suffix = target.split('.')[-1]

  # Apply optimization level settings
  shared.Settings.apply_opt_level(opt_level, noisy=True)

  # Apply -s settings in newargs here (after optimization levels, so they can override them)
  for change in settings_changes:
    key, value = change.split('=')
    exec('shared.Settings.' + key + ' = ' + value)

  ## Compile source code to bitcode

  if DEBUG: print >> sys.stderr, 'emcc: compiling to bitcode'

  # First, generate LLVM bitcode. For each input file, we get base.o with bitcode
  for input_file in input_files:
    if input_file.endswith(SOURCE_SUFFIXES):
      if DEBUG: print >> sys.stderr, 'emcc: compiling source file: ', input_file
      output_file = in_temp(unsuffixed_basename(input_file) + '.o')
      args = newargs + ['-emit-llvm', '-c', input_file, '-o', output_file]
      if DEBUG: print >> sys.stderr, "emcc running:", call, ' '.join(args)
      Popen([call] + args).communicate() # let compiler frontend print directly, so colors are saved (PIPE kills that)
      if not os.path.exists(output_file):
        print >> sys.stderr, 'emcc: compiler frontend failed to generate LLVM bitcode, halting'
        sys.exit(1)
    else: # bitcode
      if input_file.endswith(('.bc', '.o')):
        if DEBUG: print >> sys.stderr, 'emcc: copying bitcode file: ', input_file
        shutil.copyfile(input_file, in_temp(unsuffixed_basename(input_file) + '.o'))
      else: #.ll
        if not LEAVE_INPUTS_RAW:
          if DEBUG: print >> sys.stderr, 'emcc: assembling assembly file: ', input_file
          shared.Building.llvm_as(input_file, in_temp(unsuffixed_basename(input_file) + '.o'))

  # Optimize, if asked to
  if llvm_opt_level > 0 and not LEAVE_INPUTS_RAW:
    if DEBUG: print >> sys.stderr, 'emcc: LLVM opts'
    for input_file in input_files:
      try:
        shared.Building.llvm_opt(in_temp(unsuffixed_basename(input_file) + '.o'), LLVM_INTERNAL_OPT_LEVEL, safe=llvm_opt_level < 2)
      except:
        # This might be an invalid input, which will get ignored during linking later anyhow
        print >> sys.stderr, 'emcc: warning: LLVM opt failed to run on %s, continuing without optimization' % input_file

  # If we were just asked to generate bitcode, stop there
  if final_suffix not in ['js', 'html']:
    if not specified_target:
      for input_file in input_files:
        shutil.move(in_temp(unsuffixed_basename(input_file) + '.o'), unsuffixed_basename(input_file) + '.' + final_suffix)
    else:
      if len(input_files) == 1:
        shutil.move(in_temp(unsuffixed_basename(input_files[0]) + '.o'), specified_target)
      else:
        assert not has_dash_c, 'fatal error: cannot specify -o with -c with multiple files' + str(sys.argv)
        # We have a specified target (-o <target>), which is not JavaScript or HTML, and
        # we have multiple files: Link them TODO: llvm link-time opts?
        ld_args = map(lambda input_file: in_temp(unsuffixed_basename(input_file) + '.o'), input_files) + \
                  ['-o', specified_target]
                  #[arg.split('-Wl,')[1] for arg in filter(lambda arg: arg.startswith('-Wl,'), sys.argv)]
        if DEBUG: print >> sys.stderr, 'emcc: link: ' + str(ld_args)
        Popen([shared.LLVM_LINK] + ld_args).communicate()
    exit(0)

  ## Continue on to create JavaScript

  if DEBUG: print >> sys.stderr, 'emcc: generating JavaScript'

  extra_files_to_link = []

  if not LEAVE_INPUTS_RAW:
    # Check if we need to include dlmalloc. Note that we assume a single symbol is enough to know if we have/do not have dlmalloc. If you
    # include just a few symbols but want the rest, this will not work.
    need_dlmalloc = False
    has_dlmalloc = False
    for input_file in input_files:
      symbols = shared.Building.llvm_nm(in_temp(unsuffixed_basename(input_file) + '.o'))
      for malloc_def in ['malloc', 'free', 'calloc', 'memalign', 'realloc', 'valloc', 'pvalloc', 'mallinfo', 'mallopt', 'malloc_trim', 'malloc_stats', 'malloc_usable_size', 'malloc_footprint', 'malloc_max_footprint', 'independent_calloc', 'independent_comalloc']:
        if malloc_def in symbols.undefs:
          need_dlmalloc = True
        if malloc_def in symbols.defs:
          has_dlmalloc = True
    if need_dlmalloc and not has_dlmalloc:
      # We need to build and link dlmalloc in
      if DEBUG: print >> sys.stderr, 'emcc: including dlmalloc'
      Popen([shared.EMCC, shared.path_from_root('src', 'dlmalloc.c'), '-g', '-o', in_temp('dlmalloc.o')], stdout=PIPE, stderr=PIPE).communicate()
      if llvm_opt_level > 0:
        shared.Building.llvm_opt(in_temp('dlmalloc.o'), LLVM_INTERNAL_OPT_LEVEL, safe=llvm_opt_level < 2)
      extra_files_to_link.append(in_temp('dlmalloc.o'))

      # dlmalloc needs some sign correction. # If we are in mode 0, switch to 2. We will add our lines
      try:
        if shared.Settings.CORRECT_SIGNS == 0: raise Exception('we need to change to 2')
      except: # we fail if equal to 0 - so we need to switch to 2 - or if CORRECT_SIGNS is not even in Settings
        shared.Settings.CORRECT_SIGNS = 2
      if shared.Settings.CORRECT_SIGNS == 2:
        shared.Settings.CORRECT_SIGNS_LINES = [shared.path_from_root('src', 'dlmalloc.c') + ':' + str(i+4) for i in [4816, 4191, 4246, 4199, 4205, 4235, 4227]]
      # If we are in mode 1, we are correcting everything anyhow. If we are in mode 3, we will be corrected
      # so all is well anyhow too.

  # First, combine the bitcode files if there are several
  if len(input_files) + len(extra_files_to_link) > 1:
    linker_inputs = map(lambda input_file: in_temp(unsuffixed_basename(input_file) + '.o'), input_files) + extra_files_to_link
    if DEBUG: print >> sys.stderr, 'emcc: linking: ', linker_inputs
    shared.Building.link(linker_inputs,
                         in_temp(target_basename + '.bc'))
    # TODO: LLVM link-time opts? here and/or elsewhere?
  else:
    if not LEAVE_INPUTS_RAW:
      shutil.move(in_temp(unsuffixed_basename(input_files[0]) + '.o'), in_temp(target_basename + '.bc'))

  # Emscripten
  try:
    if shared.Settings.RELOOP:
      print >> sys.stderr, 'Warning: The relooper optimization can be very slow.'
  except:
    pass

  if not LEAVE_INPUTS_RAW:
    emscripten_input = in_temp(target_basename + '.bc')
  else:
    assert len(input_files) == 1
    emscripten_input = input_files[0]
  final = shared.Building.emscripten(emscripten_input, append_ext=False)

  # Apply a source code transformation, if requested
  source_transform = os.environ.get('EMCC_JS_PROCESSOR')
  if source_transform:
    exec source_transform in locals()
    shutil.copyfile(final, final + '.tr.js')
    final += '.tr.js'
    process(final)

  if DEBUG:
    print >> sys.stderr, 'emcc: saving intermediate processing steps to %s' % shared.EMSCRIPTEN_TEMP_DIR

    intermediate_counter = 0
    def save_intermediate(name=None):
      global intermediate_counter
      shutil.copyfile(final, os.path.join(shared.EMSCRIPTEN_TEMP_DIR, 'emcc-%d%s.js' % (intermediate_counter, '' if name is None else '-' + name)))
      intermediate_counter += 1

    save_intermediate('original')
    final = shared.Building.js_optimizer(final, []) # Clean up the syntax a bit, so comparisons to later passes are simpler
    save_intermediate('pretty')

  if opt_level >= 1:
    # js optimizer
    if DEBUG: print >> sys.stderr, 'emcc: running pre-closure post-opts'
    final = shared.Building.js_optimizer(final, ['hoistMultiples'])
    if DEBUG: save_intermediate('hoistMultiples')
    final = shared.Building.js_optimizer(final, ['loopOptimizer'])
    if DEBUG: save_intermediate('loopOptimizer')

    # eliminator
    final = shared.Building.eliminator(final)
    if DEBUG: save_intermediate('eliminator')

    # js optimizer pre-pass
    final = shared.Building.js_optimizer(final, 'simplifyExpressionsPre')
    if DEBUG: save_intermediate('simplifyExpressionsPre')
    final = shared.Building.js_optimizer(final, 'optimizeShiftsConservative')
    if DEBUG: save_intermediate('optimizeShiftsConservative')
    #final = shared.Building.js_optimizer(final, 'optimizeShiftsAggressive')
    #if DEBUG: save_intermediate('optimizeShiftsAggressive')
    #final = shared.Building.eliminator(final) # aggressive shifts optimization introduces some new variables, remove ones that we can
    #if DEBUG: save_intermediate('eliminator')

  if closure:
    if DEBUG: print >> sys.stderr, 'emcc: running closure'
    final = shared.Building.closure_compiler(final)
    if DEBUG: save_intermediate('closure')

  if opt_level >= 1:
    # js optimizer post-pass
    if DEBUG: print >> sys.stderr, 'emcc: running post-closure post-opts'
    final = shared.Building.js_optimizer(final, 'simplifyExpressionsPost')
    if DEBUG: save_intermediate('simplifyExpressionsPost')

  # If we were asked to also generate HTML, do that
  if final_suffix == 'html':
    if DEBUG: print >> sys.stderr, 'emcc: generating HTML'
    shell = open(shared.path_from_root('src', 'shell.html')).read()
    html = open(target_basename + '.html', 'w')
    html.write(shell.replace('{{{ SCRIPT_CODE }}}', open(final).read()))
    html.close()
  else:
    # copy final JS to output
    shutil.move(final, target_basename + '.js')

finally:
  if not TEMP_DIR:
    try:
      shutil.rmtree(temp_dir)
    except:
      pass
  else:
    print >> sys.stderr, 'emcc saved files are in:', temp_dir