Enable emscripten_futex API for Wasm Workers (#26325)

This involves also including `emscripten_thread_state.S` which provides
APIs such as `emscripten_is_main_runtime_thread` and
emscripten_is_main_browser_thread` which now also work in Wasm Workers.

There is a minor code size hit here of ~80 bytes, but I think its
worth it to have these low level APIs available everywhere.

Depends on: #26309
Fixes: #26314

Author: sbc100

ChangeLog.md

@@ -29,6 +29,12 @@ See docs/process.md for more on how version tagging works.
   default in #22257, and is no longer used by emscripten itself.  It is also
   problematic as it injects a global process.on handler.  It is easy to replace
   with a simple `--pre-js` file for those that require it. (#26326)
+- The following APIs are now available in Wasm Workers:
+   - emscripten_futex_wait
+   - emscripten_futex_wake
+   - emscripten_is_main_runtime_thread
+   - emscripten_is_main_browser_thread
+  (#26325)
 - Several low level emscripten APIs that return success/failure now return the
   C `bool` type rather than `int`.  For example `emscripten_proxy_sync` and
   `emscripten_is_main_runtime_thread`. (#26316)

site/source/docs/api_reference/wasm_workers.rst

@@ -86,6 +86,7 @@ the middle.
 Pthreads and Wasm Workers share several similarities:
 
  * Both can use emscripten_atomic_* Atomics API,
+ * Both can use emscripten_futex_wait/wake API,
  * Both can use GCC __sync_* Atomics API,
  * Both can use C11 and C++11 Atomics APIs,
  * Both types of threads have a local stack.

src/lib/libwasm_worker.js

@@ -25,6 +25,8 @@
 #error "internal error, feature_matrix should not allow this"
 #endif
 
+#endif // ~WASM_WORKERS
+
 {{{
   const workerSupportsFutexWait = () => AUDIO_WORKLET ? "!ENVIRONMENT_IS_AUDIO_WORKLET" : '1';
   const wasmWorkerJs = `
@@ -59,7 +61,6 @@
 }`;
 }}}
 
-#endif // ~WASM_WORKERS
 
 
 addToLibrary({
@@ -342,5 +343,20 @@ if (ENVIRONMENT_IS_WASM_WORKER
       else dispatch(-1/*idx*/, 2/*'timed-out'*/);
     };
     tryAcquireSemaphore();
-  }
+  },
+
+#if !PTHREADS
+  // When pthreads are used we call `__set_thread_state` immediately on worker
+  // creation.  When wasm workers is used without pthreads, we call
+  // `__set_thread_state` lazily to save code size for programs that don't use
+  // the threads state.
+  __do_set_thread_state__deps: ['__set_thread_state'],
+  __do_set_thread_state: (tb) => {
+    ___set_thread_state(
+      /*thread_ptr=*/0,
+      /*is_main_thread=*/!ENVIRONMENT_IS_WORKER,
+      /*is_runtime_thread=*/!ENVIRONMENT_IS_WASM_WORKER,
+      /*supports_wait=*/ENVIRONMENT_IS_WORKER && {{{ workerSupportsFutexWait() }}});
+  },
+#endif
 });

system/include/emscripten/threading.h

@@ -42,11 +42,13 @@ int emscripten_futex_wake(volatile void/*uint32_t*/ * _Nonnull addr, int count);
 
 // Returns true if the current thread is the thread that hosts the Emscripten
 // runtime.
+// Returns false on pthreads and Wasm Workers.
 bool emscripten_is_main_runtime_thread(void);
 
 // Returns true if the current thread is the main browser thread.  In the case
 // that the Emscripten module is started in a worker there will be no thread
 // for which this returns true.
+// Returns false on pthreads and Wasm Workers.
 bool emscripten_is_main_browser_thread(void);
 
 // A temporary workaround to issue

system/lib/pthread/emscripten_thread_state.S

@@ -18,13 +18,12 @@ is_runtime_thread:
 .globaltype supports_wait, i32
 supports_wait:
 
-.section .text,"",@
+#if WASM_WORKERS_ONLY
+.globaltype done_init, i32
+done_init:
+#endif
 
-.globl __get_tp
-__get_tp:
-  .functype __get_tp () -> (PTR)
-  global.get thread_ptr
-  end_function
+.section .text,"",@
 
 .globl __set_thread_state
 __set_thread_state:
@@ -39,23 +38,59 @@ __set_thread_state:
   global.set supports_wait
   end_function
 
+#if WASM_WORKERS_ONLY
+// With Wasm Workers we do lazy initializtion of the thread
+// state so that only workers that call these APIs actually
+// initializes their state.
+.functype __do_set_thread_state () -> ()
+
+lazy_init_thread_state:
+  .functype lazy_init_thread_state () -> ()
+  block
+    global.get done_init
+    br_if 0
+    call __do_set_thread_state
+    i32.const 1
+    global.set done_init
+  end_block
+  end_function
+#endif
+
+.globl __get_tp
+__get_tp:
+  .functype __get_tp () -> (PTR)
+#if WASM_WORKERS_ONLY
+  call lazy_init_thread_state
+#endif
+  global.get thread_ptr
+  end_function
+
 # Semantically the same as testing "!ENVIRONMENT_IS_PTHREAD" in JS
 .globl emscripten_is_main_runtime_thread
 emscripten_is_main_runtime_thread:
   .functype emscripten_is_main_runtime_thread () -> (i32)
+#if WASM_WORKERS_ONLY
+  call lazy_init_thread_state
+#endif
   global.get is_runtime_thread
   end_function
 
 # Semantically the same as testing "!ENVIRONMENT_IS_WORKER" in JS
 .globl emscripten_is_main_browser_thread
 emscripten_is_main_browser_thread:
   .functype emscripten_is_main_browser_thread () -> (i32)
+#if WASM_WORKERS_ONLY
+  call lazy_init_thread_state
+#endif
   global.get is_main_thread
   end_function
 
 # Semantically the same as testing "!ENVIRONMENT_IS_WEB && !ENVIRONMENT_IS_AUDIO_WORKLET" in JS
 .globl _emscripten_thread_supports_atomics_wait
 _emscripten_thread_supports_atomics_wait:
   .functype _emscripten_thread_supports_atomics_wait () -> (i32)
+#if WASM_WORKERS_ONLY
+  call lazy_init_thread_state
+#endif
   global.get supports_wait
   end_function

system/lib/pthread/library_pthread_stub.c

@@ -21,6 +21,10 @@ bool emscripten_has_threading_support() { return false; }
 
 int emscripten_num_logical_cores() { return 1; }
 
+#ifndef __EMSCRIPTEN_WASM_WORKERS__
+// These low level primites are defined in both pthreads and wasm workers
+// builds.
+
 int emscripten_futex_wait(volatile void /*uint32_t*/* addr,
                           uint32_t val,
                           double maxWaitMilliseconds) {
@@ -39,10 +43,20 @@ int emscripten_futex_wake(volatile void /*uint32_t*/* addr, int count) {
 }
 
 bool emscripten_is_main_runtime_thread() {
-  // TODO: We probably shouldn't be returning true here in WASM_WORKERS builds.
   return true;
 }
 
+void __wait(volatile int *addr, volatile int *waiters, int val, int priv) {}
+
+void __lock(void* ptr) {}
+
+void __unlock(void* ptr) {}
+
+void __acquire_ptc() {}
+
+void __release_ptc() {}
+#endif
+
 void emscripten_main_thread_process_queued_calls() {
   // nop
 }
@@ -385,16 +399,6 @@ int sem_destroy(sem_t *sem) {
   return 0;
 }
 
-void __wait(volatile int *addr, volatile int *waiters, int val, int priv) {}
-
-void __lock(void* ptr) {}
-
-void __unlock(void* ptr) {}
-
-void __acquire_ptc() {}
-
-void __release_ptc() {}
-
 // When pthreads is not enabled, we can't use the Atomics futex api to do
 // proper sleeps, so simulate a busy spin wait loop instead.
 void emscripten_thread_sleep(double msecs) {

test/codesize/test_codesize_cxx_ctors1.json

@@ -2,9 +2,9 @@
   "a.out.js": 19555,
   "a.out.js.gz": 8102,
   "a.out.nodebug.wasm": 132828,
-  "a.out.nodebug.wasm.gz": 49874,
+  "a.out.nodebug.wasm.gz": 49876,
   "total": 152383,
-  "total_gz": 57976,
+  "total_gz": 57978,
   "sent": [
     "__cxa_throw",
     "_abort_js",

test/codesize/test_codesize_cxx_ctors2.json

@@ -2,9 +2,9 @@
   "a.out.js": 19532,
   "a.out.js.gz": 8087,
   "a.out.nodebug.wasm": 132248,
-  "a.out.nodebug.wasm.gz": 49531,
+  "a.out.nodebug.wasm.gz": 49533,
   "total": 151780,
-  "total_gz": 57618,
+  "total_gz": 57620,
   "sent": [
     "__cxa_throw",
     "_abort_js",

test/codesize/test_codesize_cxx_except.json

@@ -2,9 +2,9 @@
   "a.out.js": 23216,
   "a.out.js.gz": 9081,
   "a.out.nodebug.wasm": 172758,
-  "a.out.nodebug.wasm.gz": 57395,
+  "a.out.nodebug.wasm.gz": 57391,
   "total": 195974,
-  "total_gz": 66476,
+  "total_gz": 66472,
   "sent": [
     "__cxa_begin_catch",
     "__cxa_end_catch",

test/codesize/test_codesize_cxx_except_wasm.json

@@ -2,9 +2,9 @@
   "a.out.js": 19366,
   "a.out.js.gz": 8022,
   "a.out.nodebug.wasm": 148153,
-  "a.out.nodebug.wasm.gz": 55275,
+  "a.out.nodebug.wasm.gz": 55273,
   "total": 167519,
-  "total_gz": 63297,
+  "total_gz": 63295,
   "sent": [
     "_abort_js",
     "_tzset_js",