[Wasm64] Add support for threads

.circleci/config.yml

@@ -75,24 +75,38 @@ commands:
       node_version:
         description: "version of node to install"
         type: string
+      canary:
+        description: "install a canary version of node"
+        type: boolean
+        default: false
     steps:
       - run:
           name: setup node v<< parameters.node_version >>
           command: |
             cd $HOME
             version=<< parameters.node_version >>
-            wget https://nodejs.org/dist/v${version}/node-v${version}-linux-x64.tar.xz
+            if [[ << parameters.canary >> == "true" ]]; then
+              wget https://nodejs.org/download/v8-canary/v${version}/node-v${version}-linux-x64.tar.xz
+            else
+              wget https://nodejs.org/dist/v${version}/node-v${version}-linux-x64.tar.xz
+            fi
             tar xf node-v${version}-linux-x64.tar.xz
             echo "NODE_JS = [os.path.expanduser('~/node-v${version}-linux-x64/bin/node')]" >> ~/emsdk/.emscripten
             echo "JS_ENGINES = [NODE_JS]" >> ~/emsdk/.emscripten
             echo "if os.path.exists(V8_ENGINE[0]): JS_ENGINES.append(V8_ENGINE)" >> ~/emsdk/.emscripten
             cat ~/emsdk/.emscripten
             echo "export PATH=\"$HOME/node-v${version}-linux-x64/bin:\$PATH\"" >> $BASH_ENV
-  install-latest-node:
+  install-node-latest:
     description: "install latest version of node"
     steps:
       - install-node-version:
          node_version: "19.0.0"
+  install-node-canary:
+    description: "install canary version of node"
+    steps:
+      - install-node-version:
+         node_version: "20.0.0-v8-canary202302081604228b65"
+         canary: true
   install-v8:
     description: "install v8 using jsvu"
     steps:
@@ -532,7 +546,7 @@ jobs:
       - run-tests:
           title: "wasm64l"
           test_targets: "wasm64l"
-      - install-latest-node
+      - install-node-canary
       - run-tests:
           title: "wasm64"
           test_targets: "wasm64"
@@ -548,6 +562,14 @@ jobs:
           command: git submodule update --init
       - pip-install
       - build
+      - install-node-canary
+      - run-tests:
+          title: "selected subset"
+          test_targets: "
+            other.test_gen_struct_info
+            other.test_native_call_before_init
+            other.test_node_unhandled_rejection
+            core2.test_hello_world"
       # Run some basic tests with the minimum version of node that we currently
       # support.
       - install-node-version:
@@ -562,7 +584,7 @@ jobs:
       # Run a few test with the most recent version of node
       # In particular we have some tests that require node flags on older
       # versions of node but not with the most recent version.
-      - install-latest-node
+      - install-node-latest
       - run-tests:
           # Run tests that on older versions of node would require flags, but
           # those flags should not be injected on newer versions.

emscripten.py

@@ -865,6 +865,13 @@ def create_wasm64_wrappers(metadata):
     '__cxa_can_catch': '_ppp',
     '_wasmfs_write_file': '_ppp',
     '__dl_seterr': '_pp',
+    '_emscripten_run_in_main_runtime_thread_js': '___p_',
+    '_emscripten_proxy_execute_task_queue': '_p',
+    '_emscripten_thread_exit': '_p',
+    '_emscripten_thread_init': '_p____',
+    '_emscripten_thread_free_data': '_p',
+    '_emscripten_dlsync_self_async': '_p',
+    '_emscripten_proxy_dlsync_async': '_pp',
   }
 
   wasm64_wrappers = '''

src/library_dylink.js

@@ -1081,11 +1081,11 @@ var LibraryDylink = {
       var filename = UTF8ToString({{{ makeGetValue('handle', C_STRUCTS.dso.name, '*') }}});
       dlSetError('Could not load dynamic lib: ' + filename + '\n' + e);
       {{{ runtimeKeepalivePop() }}}
-      callUserCallback(function () { {{{ makeDynCall('vii', 'onerror') }}}(handle, user_data); });
+      callUserCallback(function () { {{{ makeDynCall('vpp', 'onerror') }}}(handle, user_data); });
     }
     function successCallback() {
       {{{ runtimeKeepalivePop() }}}
-      callUserCallback(function () { {{{ makeDynCall('vii', 'onsuccess') }}}(handle, user_data); });
+      callUserCallback(function () { {{{ makeDynCall('vpp', 'onsuccess') }}}(handle, user_data); });
     }
 
     {{{ runtimeKeepalivePush() }}}
@@ -1115,7 +1115,7 @@ var LibraryDylink = {
 
   // void* dlsym(void* handle, const char* symbol);
   _dlsym_js__deps: ['$dlSetError', '$getFunctionAddress', '$addFunction'],
-  _dlsym_js__sig: 'ppp',
+  _dlsym_js__sig: 'pppp',
   _dlsym_js: function(handle, symbol, symbolIndex) {
     // void *dlsym(void *restrict handle, const char *restrict name);
     // http://pubs.opengroup.org/onlinepubs/009695399/functions/dlsym.html

src/library_pthread.js

@@ -539,6 +539,7 @@ var LibraryPThread = {
     worker.pthread_ptr = 0;
   },
 
+  __emscripten_thread_cleanup__sig: 'vp',
   __emscripten_thread_cleanup: function(thread) {
     // Called when a thread needs to be cleaned up so it can be reused.
     // A thread is considered reusable when it either returns from its
@@ -708,7 +709,7 @@ var LibraryPThread = {
   // allocations from __pthread_create_js we could also remove this.
   __pthread_create_js__noleakcheck: true,
 #endif
-  __pthread_create_js__sig: 'iiiii',
+  __pthread_create_js__sig: 'ipppp',
   __pthread_create_js__deps: ['$spawnThread', 'pthread_self', '$pthreadCreateProxied',
 #if OFFSCREENCANVAS_SUPPORT
     'malloc',
@@ -975,6 +976,7 @@ var LibraryPThread = {
 
   emscripten_receive_on_main_thread_js_callArgs: '=[]',
 
+  emscripten_receive_on_main_thread_js__sig: 'diip',
   emscripten_receive_on_main_thread_js__deps: [
     'emscripten_proxy_to_main_thread_js',
     'emscripten_receive_on_main_thread_js_callArgs'],
@@ -1078,7 +1080,7 @@ var LibraryPThread = {
     // *ThreadMain(void *arg) form, or try linking with the Emscripten linker
     // flag -sEMULATE_FUNCTION_POINTER_CASTS to add in emulation for this x86
     // ABI extension.
-    var result = {{{ makeDynCall('ii', 'ptr') }}}(arg);
+    var result = {{{ makeDynCall('pp', 'ptr') }}}(arg);
 #if STACK_OVERFLOW_CHECK
     checkStackCookie();
 #endif
@@ -1097,6 +1099,7 @@ var LibraryPThread = {
   },
 
 #if MAIN_MODULE
+  _emscripten_thread_exit_joinable__sig: 'vp',
   _emscripten_thread_exit_joinable: function(thread) {
     // Called when a thread exits and is joinable.  We mark these threads
     // as finished, which means that are in state where are no longer actually
@@ -1207,6 +1210,7 @@ var LibraryPThread = {
   },
 
   _emscripten_notify_task_queue__deps: ['$executeNotifiedProxyingQueue'],
+  _emscripten_notify_task_queue__sig: 'vpppp',
   _emscripten_notify_task_queue: function(targetThreadId, currThreadId, mainThreadId, queue) {
     if (targetThreadId == currThreadId) {
       setTimeout(() => executeNotifiedProxyingQueue(queue));

system/lib/compiler-rt/stack_limits.S

@@ -89,9 +89,6 @@ emscripten_wasm_worker_initialize:
   local.get 0
   .globaltype __tls_size, PTR, immutable
   global.get __tls_size
-#ifdef __wasm64__
-  i64.extend_i32_u
-#endif
   PTR.add
   PTR.const 0xf
   PTR.add

system/lib/libc/musl/arch/emscripten/bits/alltypes.h

@@ -93,7 +93,7 @@ typedef struct {
     union {
         int __i[10];
         volatile int __vi[10];
-        unsigned __s[10];
+        unsigned long __s[10];
     } __u;
 #ifdef __EMSCRIPTEN__
     // For canvas transfer implementation in Emscripten, use an extra control field

system/lib/pthread/library_pthread.c

@@ -190,7 +190,7 @@ static void _do_call(void* arg) {
       break;
     case EM_PROXIED_JS_FUNCTION:
       q->returnValue.d =
-        emscripten_receive_on_main_thread_js((int)(size_t)q->functionPtr, q->args[0].i, &q->args[1].d);
+        emscripten_receive_on_main_thread_js((intptr_t)q->functionPtr, q->args[0].i, &q->args[1].d);
       break;
     case EM_FUNC_SIG_V:
       ((em_func_v)q->functionPtr)();
@@ -435,7 +435,7 @@ double _emscripten_run_in_main_runtime_thread_js(int index, int num_args, int64_
   c->calleeDelete = 1-sync;
   c->functionEnum = EM_PROXIED_JS_FUNCTION;
   // Index not needed to ever be more than 32-bit.
-  c->functionPtr = (void*)(size_t)index;
+  c->functionPtr = (void*)(intptr_t)index;
   assert(num_args+1 <= EM_QUEUED_JS_CALL_MAX_ARGS);
   // The types are only known at runtime in these calls, so we store values that
   // must be able to contain any valid JS value, including a 64-bit BigInt if

test/common.py

@@ -513,8 +513,6 @@ def setup_node_pthreads(self):
     self.require_node()
     self.set_setting('USE_PTHREADS')
     self.emcc_args += ['-Wno-pthreads-mem-growth']
-    if self.get_setting('MEMORY64'):
-      self.skipTest('node pthreads not yet supported with MEMORY64')
     if self.get_setting('MINIMAL_RUNTIME'):
       self.skipTest('node pthreads not yet supported with MINIMAL_RUNTIME')
     self.js_engines = [config.NODE_JS]

test/core/pthread/exceptions.cpp

@@ -43,7 +43,7 @@ pthread_t thread[NUM_THREADS];
 
 void CreateThread(int i)
 {
-  int rc = pthread_create(&thread[i], nullptr, ThreadMain, (void*)i);
+  int rc = pthread_create(&thread[i], nullptr, ThreadMain, (void*)(intptr_t)i);
   assert(rc == 0);
 }
 

test/core/pthread/test_pthread_dlopen_many.c

@@ -59,7 +59,7 @@ int main() {
   // start a bunch of threads while holding the lock
   pthread_t threads[NUM_THREADS];
   for (int i = 0; i < NUM_THREADS; i++) {
-    pthread_create(&threads[i], NULL, thread_main, (void*)i);
+    pthread_create(&threads[i], NULL, thread_main, (void*)(intptr_t)i);
   }
 
   // busy wait until all threads are running

test/pthread/test_pthread_cleanup.cpp

@@ -10,25 +10,22 @@
 #include <assert.h>
 #include <unistd.h>
 #include <errno.h>
-#include <emscripten/em_asm.h>
 
 // Stores/encodes the results of calling to cleanup handlers.
-long cleanupState = 1;
+int32_t cleanup_state = 1;
 
 static void cleanup_handler1(void *arg) {
-  cleanupState <<= 2;
+  cleanup_state <<= 2;
   // Perform non-commutative arithmetic to a global var that encodes the cleanup stack order ops.
-  cleanupState *= (long)arg;
-  EM_ASM(console.log('Called clean-up handler 1 with arg ' + $0), arg);
-  //printf("Called clean-up handler 1 with arg %d\n", (long)arg);
+  cleanup_state *= (intptr_t)arg;
+  printf("Called clean-up handler 1 with arg %p (state=%d)\n", arg, cleanup_state);
 }
 
 static void cleanup_handler2(void *arg) {
-  cleanupState <<= 3;
+  cleanup_state <<= 3;
   // Perform non-commutative arithmetic to a global var that encodes the cleanup stack order ops.
-  cleanupState *= (long)arg;
-  EM_ASM(console.log('Called clean-up handler 2 with arg ' + $0), arg);
-  //printf("Called clean-up handler 2 with arg %d\n", (long)arg);
+  cleanup_state *= (intptr_t)arg;
+  printf("Called clean-up handler 2 with arg %p (state=%d)\n", arg, cleanup_state);
 }
 
 static void *thread_start1(void *arg) {
@@ -83,8 +80,8 @@ int main() {
 //   s = pthread_cancel(thr[3]);
 //   assert(s == 0);
   pthread_cleanup_pop(1);
-  printf("Cleanup state variable: %ld\n", cleanupState);
-  assert(cleanupState == 907640832);
+  printf("Cleanup state variable: %d\n", cleanup_state);
+  assert(cleanup_state == 907640832);
 
   pthread_cleanup_pop(1);
   exit(EXIT_SUCCESS);

test/pthread/test_pthread_thread_local_storage.cpp

@@ -83,10 +83,10 @@ int main()
 		{
 			if (thread[i])
 			{
-				int status;
+				intptr_t status;
 				int rc = pthread_join(thread[i], (void**)&status);
 				assert(rc == 0);
-				printf("Main: Joined thread idx %ld with status %d\n", i, status);
+				printf("Main: Joined thread idx %ld with status %lu\n", i, status);
 				assert(status == 0);
 				thread[i] = 0;
 				if (numThreadsToCreate > 0)
@@ -102,10 +102,10 @@ int main()
 	{
 			if (thread[i])
 			{
-				int status = 1;
+				intptr_t status = 1;
 				int rc = pthread_join(thread[i], (void**)&status);
 				assert(rc == 0);
-				printf("Main: Joined thread idx %d with status %d\n", i, status);
+				printf("Main: Joined thread idx %d with status %lu\n", i, status);
 				assert(status == 0);
 			}
 	}

test/test_core.py

@@ -9284,6 +9284,7 @@ def test_pthread_create_embind_stack_check(self):
     self.do_run_in_out_file_test('core/pthread/create.cpp')
 
   @node_pthreads
+  @no_wasm64('MEMORY64 does not yet support exceptions')
   def test_pthread_exceptions(self):
     self.set_setting('PTHREAD_POOL_SIZE', 2)
     self.set_setting('EXIT_RUNTIME')
@@ -9395,6 +9396,7 @@ def test_pthread_dylink_entry_point(self, args):
 
   @needs_dylink
   @node_pthreads
+  @no_wasm64('MEMORY64 does not yet support exceptions')
   def test_pthread_dylink_exceptions(self):
     self.emcc_args.append('-Wno-experimental')
     self.set_setting('USE_PTHREADS')

tools/extract_metadata.py

@@ -90,7 +90,7 @@ def parse_function_for_memory_inits(module, func_index, offset_map):
   call_targets = []
   while module.tell() != end:
     opcode = OpCode(module.read_byte())
-    if opcode in (OpCode.END, OpCode.NOP, OpCode.DROP, OpCode.I32_ADD):
+    if opcode in (OpCode.END, OpCode.NOP, OpCode.DROP, OpCode.I32_ADD, OpCode.I64_ADD):
       pass
     elif opcode in (OpCode.BLOCK,):
       module.read_type()

tools/webassembly.py

@@ -109,7 +109,7 @@ class OpCode(IntEnum):
   F32_CONST = 0x43
   F64_CONST = 0x44
   I32_ADD = 0x6a
-  I64_ADD = 0x6b
+  I64_ADD = 0x7c
   REF_NULL = 0xd0
   ATOMIC_PREFIX = 0xfe
   MEMORY_PREFIX = 0xfc