#!/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, emranlib etc. 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/emcc") SET(CMAKE_CXX_LINKER "PATH/emcc") SET(CMAKE_C_LINK_EXECUTABLE "PATH/emcc") SET(CMAKE_CXX_LINK_EXECUTABLE "PATH/emcc") 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.dirname(os.path.dirname(os.path.abspath(__file__))) def path_from_root(*pathelems): return os.path.join(__rootpath__, *pathelems) sys.path += [path_from_root('')] from tools.shared import * 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, subprocess, shlex from subprocess import PIPE, STDOUT from tools import shared from tools.shared import Compression, execute, suffix, unsuffixed, unsuffixed_basename # Mapping of emcc opt levels to llvm opt levels. We use llvm opt level 3 in emcc opt # levels 2 and 3 (emcc 3 is unsafe opts, so unsuitable for the only level to get # llvm opt level 3, and speed-wise emcc level 2 is already the slowest/most optimizing # level) LLVM_OPT_LEVEL = { 0: 0, 1: 1, 2: 3, 3: 3, } 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 and libc++ cannot be # added in. Also, LLVM optimizations will not be done, nor dead code elimination AUTODEBUG = os.environ.get('EMCC_AUTODEBUG') # If set to 1, we will run the autodebugger (the automatic debugging tool, see tools/autodebugger). # Note that this will disable inclusion of libraries. This is useful because including # dlmalloc makes it hard to compare native and js builds EMCC_CFLAGS = os.environ.get('EMCC_CFLAGS') # Additional compiler flags that we treat as if they were passed to us on the commandline if DEBUG: print >> sys.stderr, '\nemcc invocation: ', ' '.join(sys.argv), (' + ' + EMCC_CFLAGS if EMCC_CFLAGS else '') if EMCC_CFLAGS: sys.argv.append(EMCC_CFLAGS) if DEBUG and LEAVE_INPUTS_RAW: print >> sys.stderr, 'emcc: leaving inputs raw' stdout = PIPE if not DEBUG else None # suppress output of child processes stderr = PIPE if not DEBUG else None # unless we are in DEBUG mode 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) 2012 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 LLVM -O1 optimizations, and no runtime assertions or C++ exception catching (to re-enable C++ exception catching, use -s DISABLE_EXCEPTION_CATCHING=0 ). (For details on the affects of different opt levels, see apply_opt_level() in tools/shared.py and also src/settings.js.) Note: Optimizations are only done when compiling to JavaScript, not to intermediate bitcode. -O2 As -O1, plus the relooper (loop recreation), plus closure compiler advanced opts, plus LLVM -O2 optimizations 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. For the available options, see src/settings.js --typed-arrays 0: No typed arrays 1: Parallel typed arrays 2: Shared (C-like) typed arrays (default) --llvm-opts 0: No LLVM optimizations (default in -O0) 1: -O1 LLVM optimizations (default in -O1) 2: -O2 LLVM optimizations 3: -O3 LLVM optimizations (default in -O2+) --llvm-lto 0: No LLVM LTO (default in -O0) 1: LLVM LTO (default in -O1+) Note: If LLVM optimizations are not run (see --llvm-opts), setting this to 1 has no effect. --closure 0: No closure compiler (default in -O0, -O1) 1: Run closure compiler (default in -O2, -O3) --js-transform will be called on the generated code before it is optimized. This lets you modify the JavaScript, for example adding some code or removing some code, in a way that those modifications will be optimized together with the generated code properly. will be called with the filename of the generated code as a parameter; to modify the code, you can read the original data and then append to it or overwrite it with the modified data. is interpreted as a space-separated list of arguments, for example, of "python processor.py" will cause a python script to be run. --pre-js A file whose contents are added before the generated code. This is done *before* optimization, so it will be minified properly if closure compiler is run. --post-js A file whose contents are added after the generated code This is done *before* optimization, so it will be minified properly if closure compiler is run. --embed-file A file to embed inside the generated JavaScript. The compiled code will be able to access the file in the current directory with the same name as given here. So if you do --embed-file dir/file.dat, then (1) dir/file.dat must exist relative to where you run emcc, and (2) your compiled code will be able to find the file by reading that same path, dir/file.dat. If a directory is passed here, its entire contents will be embedded. --preload-file A file to preload before running the compiled code asynchronously. Otherwise similar to --embed-file, except that this option is only relevant when generating HTML (it uses asynchronous binary XHRs), or JS that will be used in a web page. If a directory is passed here, its entire contents will be preloaded. Preloaded files are stored in filename.data, where filename.html is the main file you are compiling to. To run your code, you will need both the .html and the .data. --compression Compress both the compiled code and embedded/ preloaded files. should be a triple, ,, where native_encoder is a native executable that compresses stdin to stdout (the simplest possible interface), js_decoder is a JavaScript file that implements a decoder, and js_name is the name of the function to call in the decoder file (which should receive an array/typed array and return an array/typed array. Compression only works when generating HTML. When compression is on, all filed specified to be preloaded are compressed in one big archive, which is given the same name as the output HTML but with suffix .data.compress --minify 0: Do not minify the generated JavaScript's whitespace (default if closure compiler will not be run) 1: Minify the generated JavaScript's whitespace (default if closure compiler will be run). Note that this by itself will not minify the code (closure does that) --ignore-dynamic-linking Normally emcc will treat dynamic linking like static linking, by linking in the code from the dynamic library. This fails if the same dynamic library is linked more than once. With this option, dynamic linking is ignored, which allows the build system to proceed without errors. However, you will need to manually link to the shared libraries later on yourself. --shell-file The path name to a skeleton HTML file used when generating HTML output. The shell file used needs to have this token inside it: {{{ SCRIPT_CODE }}} Note that this argument is ignored if a target other than HTML is specified using the -o option. --js-library A JavaScript library to use in addition to those in Emscripten's src/library_* -v Turns on verbose output. This will pass -v to Clang, and also enable EMCC_DEBUG to details emcc's operations --remove-duplicates If set, will remove duplicate symbols when linking. This can be useful because llvm-link's behavior is not as permissive as ld is. --clear-cache Manually clears the cache of compiled emscripten system libraries (libc++, libc++abi, dlmalloc). This is normally handled automatically, but if you update llvm in-place (instead of having a different directory for a new version), the caching mechanism can get confused. Clearing the cache can fix weird problems related to cache incompatibilities, like clang failing to link with library files. The target file, if specified (-o ), defines what will be generated: .js JavaScript (default) .html HTML with embedded JavaScript .bc LLVM bitcode .o LLVM bitcode (same as .bc) 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. emcc is affected by several environment variables. For details, view the source of emcc (search for 'os.environ'). ''' % (this, this, this) exit(0) elif len(sys.argv) == 2 and sys.argv[1] == '-v': # -v with no inputs print 'emcc (Emscripten GCC-like replacement) 2.0' exit(subprocess.call([shared.CLANG, '-v'])) # 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') or 'conftest.c' in sys.argv 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 + ['-DEMSCRIPTEN'] + sys.argv[1:] if DEBUG: print >> sys.stderr, 'emcc, just configuring: ', ' '.join(cmd) exit(subprocess.call(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 += shlex.split(EMMAKEN_CFLAGS) # ---------------- Utilities --------------- SOURCE_SUFFIXES = ('.c', '.cpp', '.cxx', '.cc') BITCODE_SUFFIXES = ('.bc', '.o', '.obj') DYNAMICLIB_SUFFIXES = ('.dylib', '.so', '.dll') STATICLIB_SUFFIXES = ('.a',) ASSEMBLY_SUFFIXES = ('.ll',) LIB_PREFIXES = ('', 'lib') seen_names = {} def uniquename(name): if name not in seen_names: seen_names[name] = str(len(seen_names)) return unsuffixed(name) + '_' + seen_names[name] + (('.' + suffix(name)) if suffix(name) else '') # ---------------- 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 if '-M' in sys.argv or '-MM' in sys.argv: # Just output dependencies, do not compile. Warning: clang and gcc behave differently with -MF! (clang seems to not recognize it) cmd = [CC] + shared.COMPILER_OPTS + sys.argv[1:] if DEBUG: print >> sys.stderr, 'emcc, just dependencies: ', ' '.join(cmd) exit(subprocess.call(cmd)) # 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 len(sys.argv) >= 3: # if there is a source and a target, then copy, otherwise do nothing sys.argv = filter(lambda arg: not arg.startswith('-I'), sys.argv) if DEBUG: print >> sys.stderr, 'Just copy:', sys.argv[-1], target shutil.copy(sys.argv[-1], target) else: if DEBUG: print >> sys.stderr, 'No-op.' 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, os.path.basename(name)) try: call = CXX if use_cxx else CC ## Parse args newargs = sys.argv[1:] opt_level = 0 llvm_opts = None llvm_lto = None closure = None js_transform = None pre_js = '' post_js = '' minify_whitespace = None preload_files = [] embed_files = [] compression = None ignore_dynamic_linking = False shell_path = shared.path_from_root('src', 'shell.html') js_libraries = [] remove_duplicates = False 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_opts = eval(newargs[i+1]) newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--llvm-lto'): check_bad_eq(newargs[i]) llvm_lto = eval(newargs[i+1]) 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].startswith('--js-transform'): check_bad_eq(newargs[i]) js_transform = newargs[i+1] newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--pre-js'): check_bad_eq(newargs[i]) pre_js += open(newargs[i+1]).read() + '\n' newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--post-js'): check_bad_eq(newargs[i]) post_js += open(newargs[i+1]).read() + '\n' newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--minify'): check_bad_eq(newargs[i]) minify_whitespace = int(newargs[i+1]) newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--embed-file'): check_bad_eq(newargs[i]) embed_files.append(newargs[i+1]) newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--preload-file'): check_bad_eq(newargs[i]) preload_files.append(newargs[i+1]) newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--compression'): check_bad_eq(newargs[i]) parts = newargs[i+1].split(',') assert len(parts) == 3, '--compression requires specifying native_encoder,js_decoder,js_name - see emcc --help. got: %s' % newargs[i+1] Compression.encoder = parts[0] Compression.decoder = parts[1] Compression.js_name = parts[2] assert os.path.exists(Compression.encoder), 'native encoder %s does not exist' % Compression.encoder assert os.path.exists(Compression.decoder), 'js decoder %s does not exist' % Compression.decoder Compression.on = True newargs[i] = '' newargs[i+1] = '' elif newargs[i] == '--ignore-dynamic-linking': ignore_dynamic_linking = True newargs[i] = '' elif newargs[i] == '-v': shared.COMPILER_OPTS += ['-v'] DEBUG = 1 newargs[i] = '' elif newargs[i].startswith('--shell-file'): check_bad_eq(newargs[i]) shell_path = newargs[i+1] newargs[i] = '' newargs[i+1] = '' elif newargs[i].startswith('--js-library'): check_bad_eq(newargs[i]) js_libraries.append(newargs[i+1]) newargs[i] = '' newargs[i+1] = '' elif newargs[i] == '--remove-duplicates': remove_duplicates = True newargs[i] = '' elif newargs[i] == '--clear-cache': newargs[i] = '' print >> sys.stderr, 'emcc: clearing cache' shared.Cache.erase() elif newargs[i].startswith(('-I/', '-L/')): print >> sys.stderr, 'emcc: warning: -I or -L of an absolute path encountered. If this is to a local system header/library, it may cause problems (local system files make sense for compiling natively on your system, but not necessarily to JavaScript)' # Of course an absolute path to a non-system-specific library or header is fine, and you can ignore this warning. The danger are system headers that are e.g. x86 specific and nonportable. The emscripten bundled headers are modified to be portable, local system ones are generally not newargs = [ arg for arg in newargs if arg is not '' ] if llvm_opts is None: llvm_opts = LLVM_OPT_LEVEL[opt_level] if llvm_lto is None: llvm_lto = llvm_opts > 0 if closure is None: closure = 1 if opt_level >= 2 else 0 if minify_whitespace is None: minify_whitespace = closure # if closure is run, minify whitespace 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': if i+1 < len(newargs) and '=' in newargs[i+1]: # -s OPT=VALUE is for us, -s by itself is a linker option settings_changes.append(newargs[i+1]) newargs[i] = newargs[i+1] = '' else: print >> sys.stderr, 'emcc: warning: treating -s as linker option and not as -s OPT=VALUE for js compilation' 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 '' ] # Find input files input_files = [] has_source_inputs = False lib_dirs = [shared.path_from_root('system', 'local', 'lib'), shared.path_from_root('system', 'lib')] libs = [] 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 i > 0: prev = newargs[i-1] if prev in ['-MT', '-install_name']: continue # ignore this gcc-style argument if arg.endswith(SOURCE_SUFFIXES + BITCODE_SUFFIXES + DYNAMICLIB_SUFFIXES + ASSEMBLY_SUFFIXES) or shared.Building.is_ar(arg): # we already removed -o , 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(ASSEMBLY_SUFFIXES) or shared.Building.is_bitcode(arg): input_files.append(arg) elif arg.endswith(STATICLIB_SUFFIXES + DYNAMICLIB_SUFFIXES): # if it's not, and it's a library, just add it to libs to find later l = unsuffixed_basename(arg) for prefix in LIB_PREFIXES: if not prefix: continue if l.startswith(prefix): l = l[len(prefix):] break; libs.append(l) newargs[i] = '' else: print >> sys.stderr, 'emcc: %s: warning: Not valid LLVM bitcode' % arg else: print >> sys.stderr, 'emcc: %s: warning: No such file or directory' % arg elif arg.startswith('-L'): lib_dirs.append(arg[2:]) newargs[i] = '' elif arg.startswith('-l'): libs.append(arg[2:]) newargs[i] = '' original_input_files = input_files[:] newargs = [ arg for arg in newargs if arg is not '' ] # Find library files for lib in libs: if DEBUG: print >> sys.stderr, 'emcc: looking for library "%s"' % lib found = False for prefix in LIB_PREFIXES: for suff in STATICLIB_SUFFIXES + DYNAMICLIB_SUFFIXES: name = prefix + lib + suff for lib_dir in lib_dirs: path = os.path.join(lib_dir, name) if os.path.exists(path): if DEBUG: print >> sys.stderr, 'emcc: found library "%s" at %s' % (lib, path) input_files.append(path) found = True break if found: break if found: break if ignore_dynamic_linking: input_files = filter(lambda input_file: not input_file.endswith(DYNAMICLIB_SUFFIXES), input_files) if len(input_files) == 0: print >> sys.stderr, 'emcc: no input files' print >> sys.stderr, 'note that input files without a known suffix are ignored, make sure your input files end with one of: ' + str(SOURCE_SUFFIXES + BITCODE_SUFFIXES + DYNAMICLIB_SUFFIXES + STATICLIB_SUFFIXES + ASSEMBLY_SUFFIXES) exit(0) 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' if '.' in target: final_suffix = target.split('.')[-1] else: final_suffix = '' assert not (Compression.on and final_suffix != 'html'), 'Compression only works when generating HTML' # 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' temp_files = [] # 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(uniquename(input_file)) + '.o') temp_files.append(output_file) args = newargs + ['-emit-llvm', '-c', input_file, '-o', output_file] if DEBUG: print >> sys.stderr, "emcc running:", call, ' '.join(args) execute([call] + args) # 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(BITCODE_SUFFIXES): if DEBUG: print >> sys.stderr, 'emcc: copying bitcode file: ', input_file temp_file = in_temp(unsuffixed(uniquename(input_file)) + '.o') shutil.copyfile(input_file, temp_file) temp_files.append(temp_file) elif input_file.endswith(DYNAMICLIB_SUFFIXES) or shared.Building.is_ar(input_file): if DEBUG: print >> sys.stderr, 'emcc: copying library file: ', input_file temp_file = in_temp(uniquename(input_file)) shutil.copyfile(input_file, temp_file) temp_files.append(temp_file) else: #.ll if not LEAVE_INPUTS_RAW: # Note that by assembling the .ll file, then disassembling it later, we will # remove annotations which is a good thing for compilation time if DEBUG: print >> sys.stderr, 'emcc: assembling assembly file: ', input_file temp_file = in_temp(unsuffixed(uniquename(input_file)) + '.o') shared.Building.llvm_as(input_file, temp_file) temp_files.append(temp_file) if not LEAVE_INPUTS_RAW: assert len(temp_files) == len(input_files) # If we were just asked to generate bitcode, stop there if final_suffix not in ['js', 'html']: if llvm_opts > 0: print >> sys.stderr, 'emcc: warning: -Ox flags ignored, since not generating JavaScript' if not specified_target: for input_file in input_files: shutil.move(in_temp(unsuffixed(uniquename(input_file)) + '.o'), unsuffixed_basename(input_file) + '.' + final_suffix) else: if len(input_files) == 1: shutil.move(in_temp(unsuffixed(uniquename(input_files[0])) + '.o'), specified_target) else: assert len(original_input_files) == 1 or not has_dash_c, 'fatal error: cannot specify -o with -c with multiple files' + str(sys.argv) + ':' + str(original_input_files) # We have a specified target (-o ), which is not JavaScript or HTML, and # we have multiple files: Link them if DEBUG: print >> sys.stderr, 'emcc: link: ' + str(temp_files) shared.Building.link(temp_files, specified_target, remove_duplicates=remove_duplicates) exit(0) ## Continue on to create JavaScript if DEBUG: print >> sys.stderr, 'emcc: will generate JavaScript' extra_files_to_link = [] if not LEAVE_INPUTS_RAW and not AUTODEBUG: # Check if we need to include some libraries that we compile. (We implement libc ourselves in js, but # compile a malloc implementation and stdlibc++.) # Note that we assume a single symbol is enough to know if we have/do not have dlmalloc etc. If you # include just a few symbols but want the rest, this will not work. # dlmalloc def create_dlmalloc(): if DEBUG: print >> sys.stderr, 'emcc: building dlmalloc for cache' execute(shared.ENV_PREFIX + ['python', shared.EMCC, shared.path_from_root('system', 'lib', 'dlmalloc.c'), '-g', '-o', in_temp('dlmalloc.o')], stdout=stdout, stderr=stderr) # we include the libc++ new stuff here, so that the common case of using just new/delete is quick to link execute(shared.ENV_PREFIX + ['python', shared.EMXX, shared.path_from_root('system', 'lib', 'libcxx', 'new.cpp'), '-g', '-o', in_temp('new.o')], stdout=stdout, stderr=stderr) shared.Building.link([in_temp('dlmalloc.o'), in_temp('new.o')], in_temp('dlmalloc_full.o')) return in_temp('dlmalloc_full.o') def fix_dlmalloc(): # 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. # XXX We also need to add libc symbols that use malloc, for example strdup. It's very rare to use just them and not # a normal malloc symbol (like free, after calling strdup), so we haven't hit this yet, but it is possible. dlmalloc_symbols = open(shared.path_from_root('system', 'lib', 'dlmalloc.symbols')).read().split('\n') # libcxx def create_libcxx(): if DEBUG: print >> sys.stderr, 'emcc: building libcxx for cache' shared.Building.build_library('libcxx', shared.EMSCRIPTEN_TEMP_DIR, shared.EMSCRIPTEN_TEMP_DIR, ['libcxx.bc'], configure=None, copy_project=True, source_dir=shared.path_from_root('system', 'lib', 'libcxx')) return os.path.join(shared.EMSCRIPTEN_TEMP_DIR, 'libcxx', 'libcxx.bc') def fix_libcxx(): assert shared.Settings.QUANTUM_SIZE == 4, 'We do not support libc++ with QUANTUM_SIZE == 1' # libcxx might need corrections, so turn them all on. TODO: check which are actually needed shared.Settings.CORRECT_SIGNS = shared.Settings.CORRECT_OVERFLOWS = shared.Settings.CORRECT_ROUNDINGS = 1 #print >> sys.stderr, 'emcc: info: using libcxx turns on CORRECT_* options' libcxx_symbols = map(lambda line: line.strip().split(' ')[1], open(shared.path_from_root('system', 'lib', 'libcxx', 'symbols')).readlines()) libcxx_symbols = filter(lambda symbol: symbol not in dlmalloc_symbols, libcxx_symbols) libcxx_symbols = set(libcxx_symbols) # libcxxabi - just for dynamic_cast for now def create_libcxxabi(): if DEBUG: print >> sys.stderr, 'emcc: building libcxxabi for cache' shared.Building.build_library('libcxxabi', shared.EMSCRIPTEN_TEMP_DIR, shared.EMSCRIPTEN_TEMP_DIR, ['libcxxabi.bc'], configure=None, copy_project=True, source_dir=shared.path_from_root('system', 'lib', 'libcxxabi')) return os.path.join(shared.EMSCRIPTEN_TEMP_DIR, 'libcxxabi', 'libcxxabi.bc') def fix_libcxxabi(): assert shared.Settings.QUANTUM_SIZE == 4, 'We do not support libc++abi with QUANTUM_SIZE == 1' #print >> sys.stderr, 'emcc: info: using libcxxabi, this may need CORRECT_* options' #shared.Settings.CORRECT_SIGNS = shared.Settings.CORRECT_OVERFLOWS = shared.Settings.CORRECT_ROUNDINGS = 1 libcxxabi_symbols = map(lambda line: line.strip().split(' ')[1], open(shared.path_from_root('system', 'lib', 'libcxxabi', 'symbols')).readlines()) libcxxabi_symbols = filter(lambda symbol: symbol not in dlmalloc_symbols, libcxxabi_symbols) libcxxabi_symbols = set(libcxxabi_symbols) force = False # If we have libcxx, we must force inclusion of dlmalloc, since libcxx uses new internally. Note: this is kind of hacky for name, create, fix, library_symbols in [('libcxx', create_libcxx, fix_libcxx, libcxx_symbols), ('libcxxabi', create_libcxxabi, fix_libcxxabi, libcxxabi_symbols), ('dlmalloc', create_dlmalloc, fix_dlmalloc, dlmalloc_symbols)]: need = [] has = [] for temp_file in temp_files: symbols = shared.Building.llvm_nm(temp_file) for library_symbol in library_symbols: if library_symbol in symbols.undefs: need.append(library_symbol) if library_symbol in symbols.defs: has.append(library_symbol) if DEBUG: print >> sys.stderr, 'emcc: considering including %s: we need |%s| and have |%s|' % (name, str(need), str(has)) if force or (need and not has): # We need to build and link the library in if DEBUG: print >> sys.stderr, 'emcc: including %s' % name extra_files_to_link.append(shared.Cache.get(name, create)) force = True if fix: fix() # First, combine the bitcode files if there are several. We must also link if we have a singleton .a if len(input_files) + len(extra_files_to_link) > 1 or \ (not LEAVE_INPUTS_RAW and not (suffix(temp_files[0]) in BITCODE_SUFFIXES or suffix(temp_files[0]) in DYNAMICLIB_SUFFIXES) and shared.Building.is_ar(temp_files[0])): linker_inputs = temp_files + extra_files_to_link if DEBUG: print >> sys.stderr, 'emcc: linking: ', linker_inputs shared.Building.link(linker_inputs, in_temp(target_basename + '.bc'), remove_duplicates=remove_duplicates) final = in_temp(target_basename + '.bc') else: if not LEAVE_INPUTS_RAW: shutil.move(temp_files[0], in_temp(target_basename + '.bc')) final = in_temp(target_basename + '.bc') else: final = input_files[0] if DEBUG: print >> sys.stderr, 'emcc: saving intermediate processing steps to %s' % shared.EMSCRIPTEN_TEMP_DIR intermediate_counter = 0 def save_intermediate(name=None, suffix='js'): global intermediate_counter shutil.copyfile(final, os.path.join(shared.EMSCRIPTEN_TEMP_DIR, 'emcc-%d%s.%s' % (intermediate_counter, '' if name is None else '-' + name, suffix))) intermediate_counter += 1 if not LEAVE_INPUTS_RAW: save_intermediate('basebc', 'bc') # Optimize, if asked to if llvm_opts > 0 and not LEAVE_INPUTS_RAW: if DEBUG: print >> sys.stderr, 'emcc: LLVM -O%d' % llvm_opts shared.Building.llvm_opt(in_temp(target_basename + '.bc'), llvm_opts) if DEBUG: save_intermediate('opt', 'bc') # Do LTO in a separate pass to work around LLVM bug XXX (see failure e.g. in cubescript) if llvm_lto and shared.Building.can_use_unsafe_opts() and shared.Building.can_build_standalone(): lto_opts = [] if not shared.Building.can_inline(): lto_opts.append('-disable-inlining') lto_opts.append('-std-link-opts') if DEBUG: print >> sys.stderr, 'emcc: LLVM LTO:', lto_opts shared.Building.llvm_opt(in_temp(target_basename + '.bc'), lto_opts) if DEBUG: save_intermediate('lto', 'bc') else: # If possible, remove dead functions etc., this potentially saves a lot in the size of the generated code (and the time to compile it) if not LEAVE_INPUTS_RAW and shared.Building.can_build_standalone(): if DEBUG: print >> sys.stderr, 'emcc: LLVM dead globals elimination' shared.Building.llvm_opt(in_temp(target_basename + '.bc'), ['-internalize', '-globaldce']) if DEBUG: save_intermediate('dce', 'bc') # Prepare .ll for Emscripten if not LEAVE_INPUTS_RAW: final = shared.Building.llvm_dis(final, final + '.ll') else: assert len(input_files) == 1 if DEBUG: save_intermediate('ll', 'll') if AUTODEBUG: if DEBUG: print >> sys.stderr, 'emcc: autodebug' execute(shared.ENV_PREFIX + ['python', shared.AUTODEBUGGER, final, final + '.ad.ll']) final += '.ad.ll' if DEBUG: save_intermediate('autodebug', 'll') # Emscripten if DEBUG: print >> sys.stderr, 'emcc: LLVM => JS' extra_args = [] if not js_libraries else ['--libraries', ','.join(map(os.path.abspath, js_libraries))] final = shared.Building.emscripten(final, append_ext=False, extra_args=extra_args) if DEBUG: save_intermediate('original') # Embed and preload files if len(preload_files) + len(embed_files) > 0: if DEBUG: print >> sys.stderr, 'emcc: setting up files' file_args = [] if len(preload_files) > 0: file_args.append('--preload') file_args += preload_files if len(embed_files) > 0: file_args.append('--embed') file_args += embed_files if Compression.on: file_args += ['--compress', Compression.encoder, Compression.decoder, Compression.js_name] code = execute(shared.ENV_PREFIX + ['python', shared.FILE_PACKAGER, unsuffixed(target) + '.data'] + file_args, stdout=PIPE)[0] src = open(final).read().replace('// {{PRE_RUN_ADDITIONS}}', code) final += '.files.js' open(final, 'w').write(src) if DEBUG: save_intermediate('files') # Apply pre and postjs files if pre_js or post_js: if DEBUG: print >> sys.stderr, 'emcc: applying pre/postjses' src = open(final).read() final += '.pp.js' open(final, 'w').write(pre_js + src + post_js) if DEBUG: save_intermediate('pre-post') # Apply a source code transformation, if requested if js_transform: shutil.copyfile(final, final + '.tr.js') final += '.tr.js' posix = True if not shared.WINDOWS else False if DEBUG: print >> sys.stderr, 'emcc: applying transform: %s' % js_transform execute(shlex.split(js_transform, posix=posix) + [os.path.abspath(final)]) if DEBUG: save_intermediate('transformed') # It is useful to run several js optimizer passes together, to save on unneeded unparsing/reparsing js_optimizer_queue = [] def flush_js_optimizer_queue(): global final, js_optimizer_queue if len(js_optimizer_queue) > 0: if not DEBUG: final = shared.Building.js_optimizer(final, js_optimizer_queue) else: for name in js_optimizer_queue: print >> sys.stderr, 'emcc: applying js optimization pass:', name final = shared.Building.js_optimizer(final, [name]) save_intermediate(name) js_optimizer_queue = [] if opt_level >= 1: if DEBUG: print >> sys.stderr, 'emcc: running pre-closure post-opts' if DEBUG: # Clean up the syntax a bit final = shared.Building.js_optimizer(final, []) if DEBUG: save_intermediate('pretty') if shared.Settings.RELOOP: js_optimizer_queue += ['hoistMultiples', 'loopOptimizer'] flush_js_optimizer_queue() if DEBUG: print >> sys.stderr, 'emcc: running variable eliminator' final = shared.Building.eliminator(final) if DEBUG: save_intermediate('eliminator') js_optimizer_queue += ['simplifyExpressionsPre'] if shared.Settings.RELOOP: js_optimizer_queue += ['optimizeShiftsAggressive'] # aggressive shifts optimization requires loops, it breaks on switches flush_js_optimizer_queue() final = shared.Building.eliminator(final) # aggressive shifts optimization introduces some new variables, remove ones that we can if DEBUG: save_intermediate('eliminator') if closure: flush_js_optimizer_queue() if DEBUG: print >> sys.stderr, 'emcc: running closure' final = shared.Building.closure_compiler(final) if DEBUG: save_intermediate('closure') if opt_level >= 1: if DEBUG: print >> sys.stderr, 'emcc: running post-closure post-opts' js_optimizer_queue += ['simplifyExpressionsPost'] if minify_whitespace: js_optimizer_queue += ['compress'] flush_js_optimizer_queue() # If we were asked to also generate HTML, do that if final_suffix == 'html': if DEBUG: print >> sys.stderr, 'emcc: generating HTML' shell = open(shell_path).read() html = open(target, 'w') if not Compression.on: html.write(shell.replace('{{{ SCRIPT_CODE }}}', open(final).read())) else: # Compress the main code js_target = unsuffixed(target) + '.js' shutil.move(final, js_target) Compression.compress(js_target) # Run the decompressor in a worker, and add code to # 1. download the compressed file # 2. decompress to a typed array # 3. convert to a string of source code # 4. insert a script element with that source code (more effective than eval) decoding = ''' var decompressWorker = new Worker('decompress.js'); var decompressCallbacks = []; var decompressions = 0; Module["decompress"] = function(data, callback) { var id = decompressCallbacks.length; decompressCallbacks.push(callback); decompressWorker.postMessage({ data: data, id: id }); if (Module['setStatus']) { decompressions++; Module['setStatus']('Decompressing...'); } }; decompressWorker.onmessage = function(event) { decompressCallbacks[event.data.id](event.data.data); decompressCallbacks[event.data.id] = null; if (Module['setStatus']) { decompressions--; if (decompressions == 0) { Module['setStatus'](''); } } }; var compiledCodeXHR = new XMLHttpRequest(); compiledCodeXHR.open('GET', '%s', true); compiledCodeXHR.responseType = 'arraybuffer'; compiledCodeXHR.onload = function() { var arrayBuffer = compiledCodeXHR.response; if (!arrayBuffer) throw('Loading compressed code failed.'); var byteArray = new Uint8Array(arrayBuffer); Module.decompress(byteArray, function(decompressed) { var source = Array.prototype.slice.apply(decompressed).map(function(x) { return String.fromCharCode(x) }).join(''); // createObjectURL instead? var scriptTag = document.createElement('script'); scriptTag.setAttribute('type', 'text/javascript'); scriptTag.innerHTML = source; document.body.appendChild(scriptTag); }); }; compiledCodeXHR.send(null); ''' % Compression.compressed_name(js_target) html.write(shell.replace('{{{ SCRIPT_CODE }}}', decoding)) # Add decompressor with web worker glue code decompressor = open('decompress.js', 'w') decompressor.write(open(Compression.decoder).read()) decompressor.write(''' onmessage = function(event) { postMessage({ data: %s(event.data.data), id: event.data.id }); }; ''' % Compression.js_name) decompressor.close() html.close() else: # copy final JS to output shutil.move(final, target) finally: if not TEMP_DIR: try: shutil.rmtree(temp_dir) except: pass else: print >> sys.stderr, 'emcc saved files are in:', temp_dir