Begin using the Emscripten toolchain in clang.

With the Emscripten toolchain, we can make clang do more of what we need
directly rather than in the wrapper scripts.

Author: Dan Gohman

system/include/EGL/eglplatform.h

@@ -75,7 +75,7 @@ typedef HDC     EGLNativeDisplayType;
 typedef HBITMAP EGLNativePixmapType;
 typedef HWND    EGLNativeWindowType;
 
-#elif defined(EMSCRIPTEN)
+#elif defined(__EMSCRIPTEN__)
 
 typedef int EGLNativeDisplayType;
 typedef int EGLNativeWindowType;

system/include/SDL/SDL_config_minimal.h

@@ -33,7 +33,7 @@
 #include <stddef.h>
 #include <stdarg.h>
 
-#if !defined(EMSCRIPTEN) && !defined(_STDINT_H_) && (!defined(HAVE_STDINT_H) || !_HAVE_STDINT_H)
+#if !defined(__EMSCRIPTEN__) && !defined(_STDINT_H_) && (!defined(HAVE_STDINT_H) || !_HAVE_STDINT_H)
 typedef unsigned int size_t;
 typedef signed char int8_t;
 typedef unsigned char uint8_t;

system/include/SDL/SDL_stdinc.h

@@ -65,7 +65,7 @@
 #endif
 #if defined(HAVE_INTTYPES_H)
 # include <inttypes.h>
-#elif defined(EMSCRIPTEN) || defined(HAVE_STDINT_H)
+#elif defined(__EMSCRIPTEN__) || defined(HAVE_STDINT_H)
 # include <stdint.h>
 #endif
 #ifdef HAVE_CTYPE_H

system/include/emscripten/emscripten.h

@@ -14,7 +14,7 @@
 extern "C" {
 #endif
 
-#if !EMSCRIPTEN
+#if !__EMSCRIPTEN__
 #include <SDL/SDL.h> /* for SDL_Delay in async_call */
 #endif
 
@@ -141,7 +141,7 @@ extern void emscripten_async_load_script(const char *script, void (*onload)(), v
  *    you created an object on the stack, it will be cleaned up
  *    before the main loop will be called the first time.
  */
-#if EMSCRIPTEN
+#if __EMSCRIPTEN__
 extern void emscripten_set_main_loop(void (*func)(), int fps, int simulate_infinite_loop);
 extern void emscripten_set_main_loop_arg(void (*func)(void*), void *arg, int fps, int simulate_infinite_loop);
 extern void emscripten_pause_main_loop();
@@ -166,7 +166,7 @@ extern void emscripten_cancel_main_loop();
  * are not counted, do not block the main loop, and can fire
  * at specific time in the future.
  */
-#if EMSCRIPTEN
+#if __EMSCRIPTEN__
 extern void _emscripten_push_main_loop_blocker(void (*func)(void *), void *arg, const char *name);
 extern void _emscripten_push_uncounted_main_loop_blocker(void (*func)(void *), void *arg, const char *name);
 #else
@@ -188,7 +188,7 @@ inline void _emscripten_push_uncounted_main_loop_blocker(void (*func)(void *), v
  * to 10, then push 10 blockers, as they complete the user will
  * see x/10 and so forth.
  */
-#if EMSCRIPTEN
+#if __EMSCRIPTEN__
 extern void emscripten_set_main_loop_expected_blockers(int num);
 #else
 inline void emscripten_set_main_loop_expected_blockers(int num) {}
@@ -203,7 +203,7 @@ inline void emscripten_set_main_loop_expected_blockers(int num) {}
  * If millis is negative, the browser's requestAnimationFrame
  * mechanism is used.
  */
-#if EMSCRIPTEN
+#if __EMSCRIPTEN__
 extern void emscripten_async_call(void (*func)(void *), void *arg, int millis);
 #else
 inline void emscripten_async_call(void (*func)(void *), void *arg, int millis) {
@@ -247,7 +247,7 @@ void emscripten_get_canvas_size(int *width, int *height, int *isFullscreen);
  * absolute time, and is only meaningful in comparison to
  * other calls to this function. The unit is ms.
  */
-#if EMSCRIPTEN
+#if __EMSCRIPTEN__
 double emscripten_get_now();
 #else
 #include <time.h>

system/include/jansson.h

@@ -82,7 +82,7 @@ typedef long long json_int_t;
 typedef long json_int_t;
 #endif /* JSON_INTEGER_IS_LONG_LONG */
 
-#ifdef EMSCRIPTEN
+#ifdef __EMSCRIPTEN__
     extern "C" bool json_typeof(const void *object);
     extern "C" bool json_is_object(const void *object);
     extern "C" bool json_is_array(const void *object);

system/lib/compiler-rt/int_endianness.h

@@ -100,7 +100,7 @@
 
 #endif /* Windows */
 
-#if defined(EMSCRIPTEN)
+#if defined(__EMSCRIPTEN__)
 
 #define _YUGA_LITTLE_ENDIAN 1
 #define _YUGA_BIG_ENDIAN    0

system/lib/dlmalloc.c

@@ -1,6 +1,6 @@
 
 /* XXX Emscripten XXX */
-#if EMSCRIPTEN
+#if __EMSCRIPTEN__
 #define DLMALLOC_EXPORT __attribute__((__weak__, __visibility__("default")))
 /* mmap uses malloc, so malloc can't use mmap */
 #define HAVE_MMAP 0

tests/asmjs-unknown-emscripten.c

@@ -0,0 +1,98 @@
+#ifndef __EMSCRIPTEN__
+#error __EMSCRIPTEN__ is not defined
+#endif
+#ifndef __asmjs__
+#error __asmjs__ is not defined
+#endif
+#ifdef __cplusplus
+#ifndef _GNU_SOURCE
+#error _GNU_SOURCE is not defined in C++
+#endif
+#endif
+#ifndef __unix__
+#error __unix__ is not defined
+#endif
+#ifndef __LITTLE_ENDIAN__
+#error __LITTLE_ENDIAN__ is not defined
+#endif
+
+#ifdef __clang__
+#if __has_feature(pnacl)
+#error has feature pnacl
+#endif
+#endif
+
+#ifdef __native_client__
+#error __native_client__ is defined
+#endif
+#ifdef __pnacl__
+#error __pnacl__ is defined
+#endif
+#ifdef __ELF__
+#error __ELF__ is defined
+#endif
+#ifdef __i386__
+#error __i386__ is defined
+#endif
+#ifdef __i386
+#error __i386 is defined
+#endif
+#ifdef i386
+#error i386 is defined
+#endif
+#ifdef __SSE__
+#error __SSE__ is defined
+#endif
+#ifdef __SSE2__
+#error __SSE2__ is defined
+#endif
+#ifdef __SSE_MATH__
+#error __SSE_MATH__ is defined
+#endif
+#ifdef __SSE2_MATH__
+#error __SSE2_MATH__ is defined
+#endif
+#ifdef __MMX__
+#error __MMX__ is defined
+#endif
+#ifdef __APPLE__
+#error __APPLE__ is defined
+#endif
+#ifdef __linux__
+#error __linux__ is defined
+#endif
+#ifdef __BIG_ENDIAN__
+#error __BIG_ENDIAN__ is defined
+#endif
+#ifdef __LP64__
+#error __LP64__ is defined
+#endif
+
+// We prefer to use __EMSCRIPTEN__, but for compatibility, we define
+// EMSCRIPTEN too.
+#ifndef EMSCRIPTEN
+#error EMSCRIPTEN is not defined
+#endif
+
+#include <stdint.h>
+#include <stddef.h>
+#include <string.h>
+#include <assert.h>
+
+#define STRINGIZE_HELPER(x) #x
+#define STRINGIZE(x) STRINGIZE_HELPER(x)
+
+int main() {
+  assert(sizeof(void*) == 4);
+  assert(sizeof(long) == 4);
+  assert(sizeof(intmax_t) == 8);
+  assert(__alignof(double) == 8);
+  assert(sizeof(long double) == 8);
+  assert(__alignof(long double) == 8);
+  assert(sizeof(intptr_t) == 4);
+  assert(sizeof(size_t) == 4);
+  assert(sizeof(ptrdiff_t) == 4);
+  assert(__FLT_EVAL_METHOD__ == 0);
+  assert(strcmp(STRINGIZE(__USER_LABEL_PREFIX__), "") == 0);
+  return 0;
+}

tests/test_core.py

@@ -504,6 +504,11 @@ def test_sha1(self):
 
     self.do_run(open(path_from_root('tests', 'sha1.c')).read(), 'SHA1=15dd99a1991e0b3826fede3deffc1feba42278e6')
 
+  def test_asmjs_unknown_emscripten(self):
+    if self.emcc_args == None: return self.skip('needs emcc')
+    if not self.is_emscripten_abi(): return self.skip('asmjs-unknown-emscripten needed for asmjs-unknown-emscripten target test')
+    self.do_run(open(path_from_root('tests', 'asmjs-unknown-emscripten.c')).read(), '')
+
   def test_cube2md5(self):
     if self.emcc_args == None: return self.skip('needs emcc')
     if not self.is_emscripten_abi(): return self.skip('asmjs-unknown-emscripten needed for accurate math')

tools/shared.py

@@ -626,20 +626,29 @@ def get_llvm_target():
   COMPILER_OPTS # Can be set in EM_CONFIG, optionally
 except:
   COMPILER_OPTS = []
-COMPILER_OPTS = COMPILER_OPTS + ['-m32', '-DEMSCRIPTEN', '-D__EMSCRIPTEN__',
-                                 '-fno-math-errno',
-                                 #'-fno-threadsafe-statics', # disabled due to issue 1289
+COMPILER_OPTS = COMPILER_OPTS + [#'-fno-threadsafe-statics', # disabled due to issue 1289
                                  '-target', LLVM_TARGET]
 
-# For temporary compatibility, treat 'le32-unknown-nacl' as 'asmjs-unknown-emscripten'.
-if LLVM_TARGET == 'asmjs-unknown-emscripten' or LLVM_TARGET == 'le32-unknown-nacl':
-  COMPILER_OPTS = filter(lambda opt: opt != '-m32', COMPILER_OPTS) # asmjs-unknown-emscripten target is 32-bit anyhow, no need for -m32
-  COMPILER_OPTS += ['-U__native_client__', '-U__pnacl__', '-U__ELF__'] # The nacl target is originally used for Google Native Client. Emscripten is not NaCl, so remove the platform #define, when using their triple.
-
-# Remove various platform specific defines, and set little endian
-COMPILER_STANDARDIZATION_OPTS = ['-U__i386__', '-U__i386', '-Ui386', '-U__STRICT_ANSI__', '-D__IEEE_LITTLE_ENDIAN',
-                                 '-U__SSE__', '-U__SSE_MATH__', '-U__SSE2__', '-U__SSE2_MATH__', '-U__MMX__',
-                                 '-U__APPLE__', '-U__linux__']
+# COMPILER_STANDARDIZATION_OPTS: Options to correct various predefined macro options.
+COMPILER_STANDARDIZATION_OPTS = []
+
+# When we're not using an appropriate target triple, use -m32 to get i386, which we
+# can mostly make work.
+if LLVM_TARGET != 'asmjs-unknown-emscripten' and LLVM_TARGET != 'le32-unknown-nacl':
+  COMPILER_OPTS += ['-m32']
+  COMPILER_STANDARDIZATION_OPTS += ['-U__i386__', '-U__i386', '-Ui386',
+                                    '-U__SSE__', '-U__SSE_MATH__', '-U__SSE2__', '-U__SSE2_MATH__', '-U__MMX__',
+                                    '-U__APPLE__', '-U__linux__']
+
+# With the asmjs-unknown-emscripten target triple, clang sets up language modes
+# and predefined macros properly. When using the other targets, we have to set things
+# up manually.
+if LLVM_TARGET != 'asmjs-unknown-emscripten':
+  COMPILER_OPTS += ['-fno-math-errno']
+  COMPILER_STANDARDIZATION_OPTS += ['-D__IEEE_LITTLE_ENDIAN']
+if LLVM_TARGET == 'le32-unknown-nacl':
+  COMPILER_OPTS += ['-DEMSCRIPTEN', '-D__EMSCRIPTEN__', '-fno-math-errno',
+                    '-U__native_client__', '-U__pnacl__', '-U__ELF__']
 
 USE_EMSDK = not os.environ.get('EMMAKEN_NO_SDK')