Initial support for pthreads with wasm (#5710)

Also adds support for testing with wasm in addition to JS+SAB. To enable testing, set `EMCC_BROWSER_ALSO_WASM=1` and run `browser.test_pthread*`.

Works with EMSCRIPTEN_BROWSER='/path/to/chrome --user-data-dir=clean_dir --js-flags=--experimental-wasm-threads --enable-features=WebAssembly --disable-features=WebAssemblyTrapHandler' with recent (canary) chrome, and also with firefox nightly.

Author: dschuff

emcc.py

@@ -101,6 +101,10 @@
 final = None
 
 
+def exit_with_error(message):
+  logging.error(message)
+  exit(1)
+
 class Intermediate(object):
   counter = 0
 def save_intermediate(name=None, suffix='js'):
@@ -737,8 +741,7 @@ def setting_sub(s):
 
         if not arg.startswith('-'):
           if not os.path.exists(arg):
-            logging.error('%s: No such file or directory ("%s" was expected to be an input file, based on the commandline arguments provided)', arg, arg)
-            exit(1)
+            exit_with_error('%s: No such file or directory ("%s" was expected to be an input file, based on the commandline arguments provided)', arg, arg)
 
           arg_ending = filename_type_ending(arg)
           if arg_ending.endswith(SOURCE_ENDINGS + BITCODE_ENDINGS + DYNAMICLIB_ENDINGS + ASSEMBLY_ENDINGS + HEADER_ENDINGS) or shared.Building.is_ar(arg): # we already removed -o <target>, so all these should be inputs
@@ -766,18 +769,17 @@ def setting_sub(s):
           elif arg_ending.endswith(STATICLIB_ENDINGS):
             if not shared.Building.is_ar(arg):
               if shared.Building.is_bitcode(arg):
-                logging.error(arg + ': File has a suffix of a static library ' + str(STATICLIB_ENDINGS) + ', but instead is an LLVM bitcode file! When linking LLVM bitcode files, use one of the suffixes ' + str(BITCODE_ENDINGS))
+                message = arg + ': File has a suffix of a static library ' + str(STATICLIB_ENDINGS) + ', but instead is an LLVM bitcode file! When linking LLVM bitcode files, use one of the suffixes ' + str(BITCODE_ENDINGS)
               else:
-                logging.error(arg + ': Unknown format, not a static library!')
-              exit(1)
+                message = arg + ': Unknown format, not a static library!'
+              exit_with_error(message)
           else:
             if has_fixed_language_mode:
               newargs[i] = ''
               input_files.append((i, arg))
               has_source_inputs = True
             else:
-              logging.error(arg + ": Input file has an unknown suffix, don't know what to do with it!")
-              exit(1)
+              exit_with_error(arg + ": Input file has an unknown suffix, don't know what to do with it!")
         elif arg.startswith('-L'):
           lib_dirs.append(arg[2:])
           newargs[i] = ''
@@ -862,8 +864,7 @@ def check(input_file):
         input_files = [(i, input_file) for (i, input_file) in input_files if check(input_file)]
 
       if len(input_files) == 0:
-        logging.error('no input files\nnote that input files without a known suffix are ignored, make sure your input files end with one of: ' + str(SOURCE_ENDINGS + BITCODE_ENDINGS + DYNAMICLIB_ENDINGS + STATICLIB_ENDINGS + ASSEMBLY_ENDINGS + HEADER_ENDINGS))
-        exit(1)
+        exit_with_error('no input files\nnote that input files without a known suffix are ignored, make sure your input files end with one of: ' + str(SOURCE_ENDINGS + BITCODE_ENDINGS + DYNAMICLIB_ENDINGS + STATICLIB_ENDINGS + ASSEMBLY_ENDINGS + HEADER_ENDINGS))
 
       newargs = CC_ADDITIONAL_ARGS + newargs
 
@@ -933,8 +934,7 @@ def check(input_file):
         assert shared.Settings.PGO == 0, 'pgo not supported in fastcomp'
         assert shared.Settings.QUANTUM_SIZE == 4, 'altering the QUANTUM_SIZE is not supported'
       except Exception as e:
-        logging.error('Compiler settings are incompatible with fastcomp. You can fall back to the older compiler core, although that is not recommended, see http://kripken.github.io/emscripten-site/docs/building_from_source/LLVM-Backend.html')
-        raise e
+        exit_with_error('Compiler settings are incompatible with fastcomp. You can fall back to the older compiler core, although that is not recommended, see http://kripken.github.io/emscripten-site/docs/building_from_source/LLVM-Backend.html')
 
       assert not shared.Settings.PGO, 'cannot run PGO in ASM_JS mode'
 
@@ -949,8 +949,7 @@ def check(input_file):
       if options.use_closure_compiler:
         shared.Settings.USE_CLOSURE_COMPILER = options.use_closure_compiler
         if not shared.check_closure_compiler():
-          logging.error('fatal: closure compiler is not configured correctly')
-          sys.exit(1)
+          exit_with_error('fatal: closure compiler is not configured correctly')
         if options.use_closure_compiler == 2 and shared.Settings.ASM_JS == 1:
           shared.WarningManager.warn('ALMOST_ASM', 'not all asm.js optimizations are possible with --closure 2, disabling those - your code will be run more slowly')
           shared.Settings.ASM_JS = 2
@@ -1007,8 +1006,7 @@ def check(input_file):
         shared.Settings.NO_FILESYSTEM = 1
         shared.Settings.FETCH = 1
         if not shared.Settings.USE_PTHREADS:
-          logging.error('-s ASMFS=1 requires either -s USE_PTHREADS=1 or -s USE_PTHREADS=2 to be set!')
-          sys.exit(1)
+          exit_with_error('-s ASMFS=1 requires either -s USE_PTHREADS=1 or -s USE_PTHREADS=2 to be set!')
 
       if shared.Settings.FETCH and final_suffix in JS_CONTAINING_SUFFIXES:
         input_files.append((next_arg_index, shared.path_from_root('system', 'lib', 'fetch', 'emscripten_fetch.cpp')))
@@ -1083,24 +1081,18 @@ def check(input_file):
 
       if shared.Settings.USE_PTHREADS:
         if shared.Settings.LINKABLE:
-          logging.error('-s LINKABLE=1 is not supported with -s USE_PTHREADS>0!')
-          exit(1)
+          exit_with_error('-s LINKABLE=1 is not supported with -s USE_PTHREADS>0!')
         if shared.Settings.SIDE_MODULE:
-          logging.error('-s SIDE_MODULE=1 is not supported with -s USE_PTHREADS>0!')
-          exit(1)
+          exit_with_error('-s SIDE_MODULE=1 is not supported with -s USE_PTHREADS>0!')
         if shared.Settings.MAIN_MODULE:
-          logging.error('-s MAIN_MODULE=1 is not supported with -s USE_PTHREADS>0!')
-          exit(1)
+          exit_with_error('-s MAIN_MODULE=1 is not supported with -s USE_PTHREADS>0!')
         if shared.Settings.EMTERPRETIFY:
-          logging.error('-s EMTERPRETIFY=1 is not supported with -s USE_PTHREADS>0!')
-          exit(1)
+          exit_with_error('-s EMTERPRETIFY=1 is not supported with -s USE_PTHREADS>0!')
         if shared.Settings.PROXY_TO_WORKER:
-          logging.error('--proxy-to-worker is not supported with -s USE_PTHREADS>0! Use the option -s PROXY_TO_PTHREAD=1 if you want to run the main thread of a multithreaded application in a web worker.')
-          exit(1)
+          exit_with_error('--proxy-to-worker is not supported with -s USE_PTHREADS>0! Use the option -s PROXY_TO_PTHREAD=1 if you want to run the main thread of a multithreaded application in a web worker.')
       else:
         if shared.Settings.PROXY_TO_PTHREAD:
-          logging.error('-s PROXY_TO_PTHREAD=1 requires -s USE_PTHREADS to work!')
-          exit(1)
+          exit_with_error('-s PROXY_TO_PTHREAD=1 requires -s USE_PTHREADS to work!')
 
       if shared.Settings.OUTLINING_LIMIT:
         if not options.js_opts:
@@ -1120,13 +1112,20 @@ def check(input_file):
           if not DEBUG:
             misc_temp_files.note(asm_target)
 
-      assert shared.Settings.TOTAL_MEMORY >= 16*1024*1024, 'TOTAL_MEMORY must be at least 16MB, was ' + str(shared.Settings.TOTAL_MEMORY)
+      if shared.Settings.TOTAL_MEMORY < 16*1024*1024:
+        exit_with_error('TOTAL_MEMORY must be at least 16MB, was ' + str(shared.Settings.TOTAL_MEMORY))
       if shared.Settings.BINARYEN:
-        assert shared.Settings.TOTAL_MEMORY % 65536 == 0, 'For wasm, TOTAL_MEMORY must be a multiple of 64KB, was ' + str(shared.Settings.TOTAL_MEMORY)
+        if shared.Settings.TOTAL_MEMORY % 65536 != 0:
+          exit_with_error('For wasm, TOTAL_MEMORY must be a multiple of 64KB, was ' + str(shared.Settings.TOTAL_MEMORY))
       else:
-        assert shared.Settings.TOTAL_MEMORY % (16*1024*1024) == 0, 'For asm.js, TOTAL_MEMORY must be a multiple of 16MB, was ' + str(shared.Settings.TOTAL_MEMORY)
-      assert shared.Settings.TOTAL_MEMORY >= shared.Settings.TOTAL_STACK, 'TOTAL_MEMORY must be larger than TOTAL_STACK, was ' + str(shared.Settings.TOTAL_MEMORY) + ' (TOTAL_STACK=' + str(shared.Settings.TOTAL_STACK) + ')'
-      assert shared.Settings.WASM_MEM_MAX == -1 or shared.Settings.WASM_MEM_MAX % 65536 == 0, 'WASM_MEM_MAX must be a multiple of 64KB, was ' + str(shared.Settings.WASM_MEM_MAX)
+        if shared.Settings.TOTAL_MEMORY % (16*1024*1024) != 0:
+          exit_with_error('For asm.js, TOTAL_MEMORY must be a multiple of 16MB, was ' + str(shared.Settings.TOTAL_MEMORY))
+      if shared.Settings.TOTAL_MEMORY < shared.Settings.TOTAL_STACK:
+        exit_with_error('TOTAL_MEMORY must be larger than TOTAL_STACK, was ' + str(shared.Settings.TOTAL_MEMORY) + ' (TOTAL_STACK=' + str(shared.Settings.TOTAL_STACK) + ')')
+      if shared.Settings.WASM_MEM_MAX != -1 and shared.Settings.WASM_MEM_MAX % 65536 != 0:
+        exit_with_error('WASM_MEM_MAX must be a multiple of 64KB, was ' + str(shared.Settings.WASM_MEM_MAX))
+      if shared.Settings.USE_PTHREADS and shared.Settings.WASM and shared.Settings.ALLOW_MEMORY_GROWTH and shared.Settings.WASM_MEM_MAX == -1:
+        exit_with_error('If pthreads and memory growth are enabled, WASM_MEM_MAX must be set')
 
       if shared.Settings.WASM_BACKEND:
         options.js_opts = None
@@ -1150,7 +1149,6 @@ def check(input_file):
         shared.Settings.ASM_JS = 2 # when targeting wasm, we use a wasm Memory, but that is not compatible with asm.js opts
         shared.Settings.GLOBAL_BASE = 1024 # leave some room for mapping global vars
         assert not shared.Settings.SPLIT_MEMORY, 'WebAssembly does not support split memory'
-        assert not shared.Settings.USE_PTHREADS, 'WebAssembly does not support pthreads'
         if shared.Settings.ELIMINATE_DUPLICATE_FUNCTIONS:
           logging.warning('for wasm there is no need to set ELIMINATE_DUPLICATE_FUNCTIONS, the binaryen optimizer does it automatically')
           shared.Settings.ELIMINATE_DUPLICATE_FUNCTIONS = 0
@@ -1337,8 +1335,7 @@ def compile_source_file(i, input_file):
         logging.debug("running: " + ' '.join(shared.Building.doublequote_spaces(args))) # NOTE: Printing this line here in this specific format is important, it is parsed to implement the "emcc --cflags" command
         execute(args) # let compiler frontend print directly, so colors are saved (PIPE kills that)
         if not os.path.exists(output_file):
-          logging.error('compiler frontend failed to generate LLVM bitcode, halting')
-          sys.exit(1)
+          exit_with_error('compiler frontend failed to generate LLVM bitcode, halting')
 
       # First, generate LLVM bitcode. For each input file, we get base.o with bitcode
       for i, input_file in input_files:
@@ -1362,8 +1359,7 @@ def compile_source_file(i, input_file):
             if has_fixed_language_mode:
               compile_source_file(i, input_file)
             else:
-              logging.error(input_file + ': Unknown file suffix when compiling to LLVM bitcode!')
-              sys.exit(1)
+              exit_with_error(input_file + ': Unknown file suffix when compiling to LLVM bitcode!')
 
     # exit block 'bitcodeize inputs'
     log_time('bitcodeize inputs')
@@ -2129,8 +2125,7 @@ def parse_args(newargs):
       elif newargs[i+1].lower() == 'linux':
         options.output_eol = '\n'
       else:
-        logging.error('Invalid value "' + newargs[i+1] + '" to --output_eol!')
-        exit(1)
+        exit_with_error('Invalid value "' + newargs[i+1] + '" to --output_eol!')
       newargs[i] = ''
       newargs[i+1] = ''
 
@@ -2230,8 +2225,7 @@ def binaryen_method_sanity_check():
     valid_methods = ['asmjs', 'native-wasm', 'interpret-s-expr', 'interpret-binary', 'interpret-asm2wasm']
     for m in methods:
       if m.strip() not in valid_methods:
-        logging.error('Unrecognized BINARYEN_METHOD "' + m.strip() + '" specified! Please pass a comma-delimited list containing one or more of: ' + ','.join(valid_methods))
-        sys.exit(1)
+        exit_with_error('Unrecognized BINARYEN_METHOD "' + m.strip() + '" specified! Please pass a comma-delimited list containing one or more of: ' + ','.join(valid_methods))
 
 
 def do_binaryen(final, target, asm_target, options, memfile, wasm_binary_target,
@@ -2266,8 +2260,7 @@ def do_binaryen(final, target, asm_target, options, memfile, wasm_binary_target,
     if shared.Settings.BINARYEN_TRAP_MODE in ('js', 'clamp', 'allow'):
       cmd += ['--trap-mode=' + shared.Settings.BINARYEN_TRAP_MODE]
     else:
-      logging.error('invalid BINARYEN_TRAP_MODE value: ' + shared.Settings.BINARYEN_TRAP_MODE + ' (should be js/clamp/allow)')
-      sys.exit(1)
+      exit_with_error('invalid BINARYEN_TRAP_MODE value: ' + shared.Settings.BINARYEN_TRAP_MODE + ' (should be js/clamp/allow)')
     if shared.Settings.BINARYEN_IGNORE_IMPLICIT_TRAPS:
       cmd += ['--ignore-implicit-traps']
     # pass optimization level to asm2wasm (if not optimizing, or which passes we should run was overridden, do not optimize)
@@ -2292,6 +2285,8 @@ def do_binaryen(final, target, asm_target, options, memfile, wasm_binary_target,
       cmd += ['--no-legalize-javascript-ffi']
     if shared.Building.is_wasm_only():
       cmd += ['--wasm-only'] # this asm.js is code not intended to run as asm.js, it is only ever going to be wasm, an can contain special fastcomp-wasm support
+    if shared.Settings.USE_PTHREADS:
+      cmd += ['--enable-threads']
     if debug_info:
       cmd += ['-g']
     if options.emit_symbol_map or shared.Settings.CYBERDWARF:

src/library_pthread.js

@@ -366,7 +366,12 @@ var LibraryPThread = {
             // object in Module['mainScriptUrlOrBlob'], or a URL to it, so that pthread Workers can
             // independently load up the same main application file.
             urlOrBlob: Module['mainScriptUrlOrBlob'] || currentScriptUrl,
+#if BINARYEN
+            wasmMemory: Module['wasmMemory'],
+            wasmModule: Module['wasmModule'],
+#else
             buffer: HEAPU8.buffer,
+#endif
             tempDoublePtr: tempDoublePtr,
             TOTAL_MEMORY: TOTAL_MEMORY,
             STATICTOP: STATICTOP,

src/preamble.js

@@ -1118,6 +1118,7 @@ if (typeof SharedArrayBuffer === 'undefined' || typeof Atomics === 'undefined')
 #endif
 
 #if USE_PTHREADS
+#if !BINARYEN
 if (typeof SharedArrayBuffer !== 'undefined') {
   if (!ENVIRONMENT_IS_PTHREAD) buffer = new SharedArrayBuffer(TOTAL_MEMORY);
   // Currently SharedArrayBuffer does not have a slice() operation, so polyfill it in.
@@ -1170,6 +1171,18 @@ if (typeof Atomics === 'undefined') {
   Atomics['xor'] = function(t, i, v) { var w = t[i]; t[i] ^= v; return w; }
 }
 
+#else
+if (!ENVIRONMENT_IS_PTHREAD) {
+#if ALLOW_MEMORY_GROWTH
+  Module['wasmMemory'] = new WebAssembly.Memory({ 'initial': TOTAL_MEMORY / WASM_PAGE_SIZE , 'maximum': {{{ WASM_MEM_MAX }}} / WASM_PAGE_SIZE, 'shared': true });
+#else
+  Module['wasmMemory'] = new WebAssembly.Memory({ 'initial': TOTAL_MEMORY / WASM_PAGE_SIZE , 'maximum': TOTAL_MEMORY / WASM_PAGE_SIZE, 'shared': true });
+#endif
+  buffer = Module['wasmMemory'].buffer;
+}
+
+updateGlobalBufferViews();
+#endif // !BINARYEN
 #else // USE_PTHREADS
 
 #if SPLIT_MEMORY == 0
@@ -1185,7 +1198,7 @@ if (Module['buffer']) {
   if (typeof WebAssembly === 'object' && typeof WebAssembly.Memory === 'function') {
 #if ASSERTIONS
     assert(TOTAL_MEMORY % WASM_PAGE_SIZE === 0);
-#endif
+#endif // ASSERTIONS
 #if ALLOW_MEMORY_GROWTH
 #if WASM_MEM_MAX
 #if ASSERTIONS
@@ -1194,19 +1207,19 @@ if (Module['buffer']) {
     Module['wasmMemory'] = new WebAssembly.Memory({ 'initial': TOTAL_MEMORY / WASM_PAGE_SIZE, 'maximum': {{{ WASM_MEM_MAX }}} / WASM_PAGE_SIZE });
 #else
     Module['wasmMemory'] = new WebAssembly.Memory({ 'initial': TOTAL_MEMORY / WASM_PAGE_SIZE });
-#endif
+#endif // BINARYEN_MEM_MAX
 #else
     Module['wasmMemory'] = new WebAssembly.Memory({ 'initial': TOTAL_MEMORY / WASM_PAGE_SIZE, 'maximum': TOTAL_MEMORY / WASM_PAGE_SIZE });
-#endif
+#endif // ALLOW_MEMORY_GROWTH
     buffer = Module['wasmMemory'].buffer;
   } else
-#endif
+#endif // BINARYEN
   {
     buffer = new ArrayBuffer(TOTAL_MEMORY);
   }
 #if ASSERTIONS
   assert(buffer.byteLength === TOTAL_MEMORY);
-#endif
+#endif // ASSERTIONS
 }
 updateGlobalBufferViews();
 #else // SPLIT_MEMORY
@@ -2196,15 +2209,27 @@ function integrateWasmJS() {
     info['env'] = env;
     // handle a generated wasm instance, receiving its exports and
     // performing other necessary setup
-    function receiveInstance(instance) {
+    function receiveInstance(instance, module) {
       exports = instance.exports;
       if (exports.memory) mergeMemory(exports.memory);
       Module['asm'] = exports;
       Module["usingWasm"] = true;
+#if USE_PTHREADS
+      // Keep a reference to the compiled module so we can post it to the workers.
+      Module['wasmModule'] = module;
+      // Instantiation is synchronous in pthreads and we assert on run dependencies.
+      if(!ENVIRONMENT_IS_PTHREAD) removeRunDependency('wasm-instantiate');
+#else
       removeRunDependency('wasm-instantiate');
+#endif
     }
-
-    addRunDependency('wasm-instantiate'); // we can't run yet
+#if USE_PTHREADS
+    if (!ENVIRONMENT_IS_PTHREAD) {
+      addRunDependency('wasm-instantiate'); // we can't run yet (except in a pthread, where we have a custom sync instantiator)
+    }
+#else
+    addRunDependency('wasm-instantiate');
+#endif
 
     // User shell pages can write their own Module.instantiateWasm = function(imports, successCallback) callback
     // to manually instantiate the Wasm module themselves. This allows pages to run the instantiation parallel
@@ -2235,7 +2260,7 @@ function integrateWasmJS() {
       assert(Module === trueModule, 'the Module object should not be replaced during async compilation - perhaps the order of HTML elements is wrong?');
       trueModule = null;
 #endif
-      receiveInstance(output['instance']);
+      receiveInstance(output['instance'], output['module']);
     }
     function instantiateArrayBuffer(receiver) {
       getBinaryPromise().then(function(binary) {

src/pthread-main.js

@@ -54,19 +54,43 @@ Module['print'] = threadPrint;
 Module['printErr'] = threadPrintErr;
 this.alert = threadAlert;
 
+// #if WASM
+Module['instantiateWasm'] = function(info, receiveInstance) {
+  // Instantiate from the module posted from the main thread.
+  // We can just use sync instantiation in the worker.
+  instance = new WebAssembly.Instance(Module['wasmModule'], info);
+  // We don't need the module anymore; new threads will be spawned from the main thread.
+  delete Module['wasmModule'];
+  receiveInstance(instance);
+  return instance.exports;
+}
+//#endif
+
 this.onmessage = function(e) {
   try {
     if (e.data.cmd === 'load') { // Preload command that is called once per worker to parse and load the Emscripten code.
       // Initialize the thread-local field(s):
       tempDoublePtr = e.data.tempDoublePtr;
 
       // Initialize the global "process"-wide fields:
-      buffer = e.data.buffer;
       Module['TOTAL_MEMORY'] = TOTAL_MEMORY = e.data.TOTAL_MEMORY;
       STATICTOP = e.data.STATICTOP;
       DYNAMIC_BASE = e.data.DYNAMIC_BASE;
       DYNAMICTOP_PTR = e.data.DYNAMICTOP_PTR;
 
+
+//#if WASM
+      if (e.data.wasmModule) {
+        // Module and memory were sent from main thread
+        Module['wasmModule'] = e.data.wasmModule;
+        Module['wasmMemory'] = e.data.wasmMemory;
+        buffer = Module['wasmMemory'].buffer;
+      } else {
+//#else
+        buffer = e.data.buffer;
+      }
+//#endif
+
       PthreadWorkerInit = e.data.PthreadWorkerInit;
       if (typeof e.data.urlOrBlob === 'string') {
         importScripts(e.data.urlOrBlob);