Move emscripten_futex_wake to native code. NFC

[sbc100]

This is great example of why moving stuff into native code can save on overall size:

jssize: 49696 -> 48861 = -775
wasm size: 16846 -> 16928 = +82

[kleisauke]

Some tests (e.g. browser.test_embind_with_pthreads) are failing with this assertion.

Aborted(Assertion failed: !_emscripten_thread_supports_atomics_wait(), at: /home/kleisauke/emscripten/system/lib/pthread/emscripten_futex_wake.c,37,emscripten_futex_wake)

Which is curious, I tried to solve that with this simplification:

15766.patch

diff --git a/system/lib/pthread/emscripten_futex_wake.c b/system/lib/pthread/emscripten_futex_wake.c
index ac3a42eba..6e87bd5ed 100644
--- a/system/lib/pthread/emscripten_futex_wake.c
+++ b/system/lib/pthread/emscripten_futex_wake.c
@@ -8,12 +8,14 @@
 #include <assert.h>
 #include <errno.h>
 #include <limits.h>
-#include <stdatomic.h>
+#include <atomic.h>
 #include <emscripten/threading.h>
 
 int _emscripten_thread_supports_atomics_wait(void);
-extern void* _Atomic _emscripten_main_thread_futex;
 
+// Stores the memory address that the main thread is waiting on, if any. If
+// the main thread is waiting, we wake it up before waking up any workers.
+EMSCRIPTEN_KEEPALIVE void *volatile _emscripten_main_thread_futex;
 
 // Returns the number of threads (>= 0) woken up, or the value -EINVAL on error.
 // Pass count == INT_MAX to wake up all threads.
@@ -29,19 +31,16 @@ int emscripten_futex_wake(volatile void *addr, int count) {
   // Note that this is not a fair procedure, since we always wake main thread first before any workers, so
   // this scheme does not adhere to real queue-based waiting.
   int main_thread_woken = 0;
-  if (_emscripten_main_thread_futex == addr) {
+  if (a_cas_p(&_emscripten_main_thread_futex, (void *)addr, 0) == addr) {
     // We only use __emscripten_main_thread_futex on the main browser thread,
     // where we cannot block while we wait. Therefore we should only see it set
     // from other threads, and not on the main thread itself. In other words,
     // the main thread must never try to wake itself up!
     assert(!_emscripten_thread_supports_atomics_wait());
-    void* expected = (void*)addr;
-    if (atomic_compare_exchange_strong(&_emscripten_main_thread_futex, &expected, 0)) {
-      --count;
-      main_thread_woken = 1;
-      if (count <= 0) {
-        return 1;
-      }
+    --count;
+    main_thread_woken = 1;
+    if (count <= 0) {
+      return 1;
     }
   }
 
diff --git a/system/lib/pthread/library_pthread.c b/system/lib/pthread/library_pthread.c
index 3d5051f04..612e697d7 100644
--- a/system/lib/pthread/library_pthread.c
+++ b/system/lib/pthread/library_pthread.c
@@ -831,10 +831,6 @@ int emscripten_dispatch_to_thread_async_(pthread_t target_thread,
   return ret;
 }
 
-// Stores the memory address that the main thread is waiting on, if any. If
-// the main thread is waiting, we wake it up before waking up any workers.
-EMSCRIPTEN_KEEPALIVE void* _emscripten_main_thread_futex;
-
 void __emscripten_init_main_thread_js(void* tb);
 
 static void *dummy_tsd[1] = { 0 };

To no avail (same assertion failure).

[sbc100]

Ah, it looks like I had the condition in the assert the wrong way around..

[sbc100]

I like your simplification though! I will apply that..

[sbc100]

I ran the whole of posixtest and posixtest_browser so I'm pretty confident the translation is good.

[tlively]

At some point we might want to audit our code to try to make it work correctly from audio worklets, which also can't block.

[sbc100]

Yeah.. this mechanism assumes only one (busy-looping) main thread I think, so it would need to be revisited.

[Jonathhhan]

I guess I came across the audioworklet issue:

SimpleSequencer.js:1 Uncaught RuntimeError: Aborted(Assertion failed: _emscripten_thread_supports_atomics_wait(), at: 
/home/jonathan/emsdk/upstream/emscripten/system/lib/pthread/emscripten_futex_wake.c,40,emscripten_futex_wake)
    at abort (SimpleSequencer.js:1:23555)
    at ___assert_fail (SimpleSequencer.js:1:92185)
    at emscripten_futex_wake (SimpleSequencer.wasm:0x5fcc66)
    at __wake (SimpleSequencer.wasm:0x5fc80b)
    at __pthread_mutex_unlock (SimpleSequencer.wasm:0x6041a7)
    at sys_unlock (SimpleSequencer.wasm:0x4b4743)
    at ofxPd::audioOut(float*, int, int) (SimpleSequencer.wasm:0x43abb3)
    at ofxOfelia::audioOut(ofSoundBuffer&) (SimpleSequencer.wasm:0x691b3)
    at std::__2::__function::__func<std::__2::__bind<void (ofBaseSoundInput::*)(ofSoundBuffer&), ofBaseSoundInput*&, std::__2::placeholders::__ph<1> const&>, std::__2::allocator<std::__2::__bind<void (ofBaseSoundInput::*)(ofSoundBuffer&), ofBaseSoundInput*&, std::__2::placeholders::__ph<1> const&>>, void (ofSoundBuffer&)>::operator()(ofSoundBuffer&) (SimpleSequencer.wasm:0x63b62)
    at ofxEmscriptenSoundStream::audio_cb(int, int, int, void*) (SimpleSequencer.wasm:0x637e0)

It would be absolutely great, if this gets fixed. Because the app I am using with Emscripten seems to rely on that. And I am a bit lost with fixing it by myself.