Pass -fno-threadsafe-statics when not building for pthreads (#16152)

This means the compiler won't generate calls
`__cxa_guard_acquire`/`__cxa_guard_release` around static intitializers.
Even through those funtions don't really do much in non-threads builds
(since locks and atomics don't do much) it seems that neither llvm nor
binaryen can completely remove them.

For example:

```
int foo() {
  return 42;
}

int main() {
  static int a = foo();
  return 0;
}
```

When compiled, without threading normally generates this code:

```
0000ef func[2] <main>:
 0000f0: 02 40                      | block
 0000f2: 41 b8 08                   |   i32.const 1080
 0000f5: 2d 00 00                   |   i32.load8_u 0 0
 0000f8: 41 01                      |   i32.const 1
 0000fa: 71                         |   i32.and
 0000fb: 0d 00                      |   br_if 0
 0000fd: 41 b8 08                   |   i32.const 1080
 000100: 10 05                      |   call 5 <__cxa_guard_acquire>
 000102: 45                         |   i32.eqz
 000103: 0d 00                      |   br_if 0
 000105: 41 b8 08                   |   i32.const 1080
 000108: 10 0d                      |   call 13 <__cxa_guard_release>
 00010a: 0b                         | end
 00010b: 41 00                      | i32.const 0
 00010d: 0b                         | end
```

With `-fno-threadsafe-statics` we get just:

```
0000cb func[1] <main>:
 0000cc: 41 80 08                   | i32.const 1024
 0000cf: 2d 00 00                   | i32.load8_u 0 0
 0000d2: 45                         | i32.eqz
 0000d3: 04 40                      | if
 0000d5: 41 80 08                   |   i32.const 1024
 0000d8: 41 01                      |   i32.const 1
 0000da: 3a 00 00                   |   i32.store8 0 0
 0000dd: 0b                         | end
 0000de: 41 00                      | i32.const 0
 0000e0: 0b                         | end
```

Author: sbc100

emcc.py

@@ -873,12 +873,18 @@ def get_cflags(user_args):
 
   # Set the LIBCPP ABI version to at least 2 so that we get nicely aligned string
   # data and other nice fixes.
-  cflags += [# '-fno-threadsafe-statics', # disabled due to issue 1289
-             '-D__EMSCRIPTEN_major__=' + str(shared.EMSCRIPTEN_VERSION_MAJOR),
+  cflags += ['-D__EMSCRIPTEN_major__=' + str(shared.EMSCRIPTEN_VERSION_MAJOR),
              '-D__EMSCRIPTEN_minor__=' + str(shared.EMSCRIPTEN_VERSION_MINOR),
              '-D__EMSCRIPTEN_tiny__=' + str(shared.EMSCRIPTEN_VERSION_TINY),
              '-D_LIBCPP_ABI_VERSION=2']
 
+  if not settings.USE_PTHREADS:
+    # There is no point in using thread safe static inits in code that cannot
+    # be part of a multi-threaded program.
+    # TODO(sbc): Remove this if/when upstream clang takes care of this
+    # automatically: https://reviews.llvm.org/D118571.
+    cflags += ['-fno-threadsafe-statics']
+
   # Changes to default clang behavior
 
   # Implicit functions can cause horribly confusing function pointer type errors, see #2175

tests/hello_cxx11.cpp

@@ -9,6 +9,14 @@
 #error This file should be compiled with -std=c++11!
 #endif
 
+#ifdef _REENTRANT
+#error Expected to be compiled single-threaded.
+#endif
+
+#ifdef __cpp_threadsafe_static_init
+#error single-threaded builds should not define __cpp_threadsafe_static_init
+#endif
+
 int main( int argc, const char *argv[] ) {
   printf("Hello world!\\n");
   return 0;

tests/other/metadce/hello_libcxx_O2.size

@@ -1 +1 @@
-124608
+124053

tests/other/metadce/hello_libcxx_O2_EVAL_CTORS.size

@@ -1 +1 @@
-124589
+124034

tests/other/metadce/hello_libcxx_O2_EVAL_CTORS_2.size

@@ -1 +1 @@
-121944
+121408

tests/other/metadce/hello_libcxx_O2_fexceptions.size

@@ -1 +1 @@
-166546
+164706

tests/other/metadce/hello_libcxx_O2_fexceptions_DEMANGLE_SUPPORT.size

@@ -1 +1 @@
-226456
+224609