Move emscripten_futex_wake to native code. NFC (#15766)

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

Author: sbc100

src/library_pthread.js

@@ -951,42 +951,6 @@ var LibraryPThread = {
     return 0;
   },
 
-  // Returns the number of threads (>= 0) woken up, or the value -EINVAL on error.
-  // Pass count == INT_MAX to wake up all threads.
-  emscripten_futex_wake: function(addr, count) {
-    if (addr <= 0 || addr > HEAP8.length || addr&3 != 0 || count < 0) return -{{{ cDefine('EINVAL') }}};
-    if (count == 0) return 0;
-    // Waking (at least) INT_MAX waiters is defined to mean wake all callers.
-    // For Atomics.notify() API Infinity is to be passed in that case.
-    if (count >= {{{ cDefine('INT_MAX') }}}) count = Infinity;
-
-    // See if main thread is waiting on this address? If so, wake it up by resetting its wake location to zero.
-    // 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.
-    var mainThreadWaitAddress = Atomics.load(HEAP32, __emscripten_main_thread_futex >> 2);
-    var mainThreadWoken = 0;
-    if (mainThreadWaitAddress == addr) {
-#if ASSERTIONS
-      // 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(!ENVIRONMENT_IS_WEB);
-#endif
-      var loadedAddr = Atomics.compareExchange(HEAP32, __emscripten_main_thread_futex >> 2, mainThreadWaitAddress, 0);
-      if (loadedAddr == mainThreadWaitAddress) {
-        --count;
-        mainThreadWoken = 1;
-        if (count <= 0) return 1;
-      }
-    }
-
-    // Wake any workers waiting on this address.
-    var ret = Atomics.notify(HEAP32, addr >> 2, count);
-    if (ret >= 0) return ret + mainThreadWoken;
-    throw 'Atomics.notify returned an unexpected value ' + ret;
-  },
-
   __atomic_is_lock_free: function(size, ptr) {
     return size <= 4 && (size & (size-1)) == 0 && (ptr&(size-1)) == 0;
   },

system/lib/pthread/emscripten_futex_wake.c

@@ -0,0 +1,52 @@
+/*
+ * Copyright 2021 The Emscripten Authors.  All rights reserved.
+ * Emscripten is available under two separate licenses, the MIT license and the
+ * University of Illinois/NCSA Open Source License.  Both these licenses can be
+ * found in the LICENSE file.
+ */
+
+#include <assert.h>
+#include <errno.h>
+#include <limits.h>
+#include <atomic.h>
+#include <emscripten/threading.h>
+
+int _emscripten_thread_supports_atomics_wait(void);
+
+// 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.
+void* _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.
+int emscripten_futex_wake(volatile void *addr, int count) {
+  if (!addr || (((intptr_t)addr) & 3) != 0 || count < 0) {
+    return -EINVAL;
+  }
+  if (count == 0) {
+    return 0;
+  }
+
+  // See if main thread is waiting on this address? If so, wake it up by resetting its wake location to zero.
+  // 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 (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 (that should always support waiting), 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());
+    --count;
+    main_thread_woken = 1;
+    if (count <= 0) {
+      return 1;
+    }
+  }
+
+  // Wake any workers waiting on this address.
+  int ret = __builtin_wasm_memory_atomic_notify((int*)addr, count);
+  assert(ret >= 0);
+  return ret + main_thread_woken;
+}

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 };

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.exports

@@ -8,7 +8,7 @@ G
 H
 I
 J
-K
+p
 q
 r
 s

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.funcs

@@ -33,6 +33,7 @@ $emscripten_async_run_in_main_thread
 $emscripten_current_thread_process_queued_calls
 $emscripten_dispatch_to_thread_
 $emscripten_futex_wait
+$emscripten_futex_wake
 $emscripten_main_thread_process_queued_calls
 $emscripten_proxy_main
 $emscripten_run_in_main_runtime_thread_js

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.imports

@@ -13,4 +13,3 @@ a.l
 a.m
 a.n
 a.o
-a.p

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.jssize

@@ -1 +1 @@
-49636
+48861

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.sent

@@ -13,4 +13,3 @@ l
 m
 n
 o
-p

tests/other/metadce/minimal_main_Oz_USE_PTHREADS_PROXY_TO_PTHREAD.size

@@ -1 +1 @@
-16846
+16911

tools/system_libs.py

@@ -828,6 +828,7 @@ def get_files(self):
           'emscripten_proxy_main.c',
           'emscripten_thread_state.S',
           'emscripten_futex_wait.c',
+          'emscripten_futex_wake.c',
         ])
     else:
       ignore += ['thread']