Merge branch 'main' into bulkmem

Author: dschuff

embuilder.py

@@ -113,7 +113,7 @@
     'crtbegin',
     'libsanitizer_common_rt',
     'libubsan_rt',
-    'libwasm_workers_stub-debug',
+    'libwasm_workers-debug-stub',
     'libfetch',
     'libfetch-mt',
     'libwasmfs',

src/library_glemu.js

@@ -2047,7 +2047,7 @@ var LibraryGLEmulation = {
         GLImmediate.rendererComponents[name] = 1;
 #if ASSERTIONS
         if (GLImmediate.enabledClientAttributes[name]) {
-          out("Warning: glTexCoord used after EnableClientState for TEXTURE_COORD_ARRAY for TEXTURE0. Disabling TEXTURE_COORD_ARRAY...");
+          warnOnce("Warning: glTexCoord used after EnableClientState for TEXTURE_COORD_ARRAY for TEXTURE0. Disabling TEXTURE_COORD_ARRAY...");
         }
 #endif
         GLImmediate.enabledClientAttributes[name] = true;
@@ -2980,11 +2980,10 @@ var LibraryGLEmulation = {
           var attr = attributes[i];
           var srcStride = Math.max(attr.sizeBytes, attr.stride);
           if ((srcStride & 3) == 0 && (attr.sizeBytes & 3) == 0) {
-            var size4 = attr.sizeBytes>>2;
-            var srcStride4 = Math.max(attr.sizeBytes, attr.stride)>>2;
             for (var j = 0; j < count; j++) {
-              for (var k = 0; k < size4; k++) { // copy in chunks of 4 bytes, our alignment makes this possible
-                HEAP32[((start + attr.offset + bytes*j)>>2) + k] = HEAP32[(attr.pointer>>2) + j*srcStride4 + k];
+              for (var k = 0; k < attr.sizeBytes; k+=4) { // copy in chunks of 4 bytes, our alignment makes this possible
+                var val = {{{ makeGetValue('attr.pointer', 'j*srcStride + k', 'i32') }}};
+                {{{ makeSetValue('start + attr.offset', 'bytes*j + k', 'val', 'i32') }}};
               }
             }
           } else {
@@ -3533,7 +3532,7 @@ var LibraryGLEmulation = {
     if (!GLctx.currentElementArrayBufferBinding) {
       assert(type == GLctx.UNSIGNED_SHORT); // We can only emulate buffers of this kind, for now
     }
-    out("DrawElements doesn't actually prepareClientAttributes properly.");
+    warnOnce("DrawElements doesn't actually prepareClientAttributes properly.");
 #endif
     GLImmediate.prepareClientAttributes(count, false);
     GLImmediate.mode = mode;

src/library_webgpu.js

@@ -410,9 +410,9 @@ var LibraryWebGPU = {
       'mapped': 3,
     },
     Int_CompilationMessageType : {
-      'error': 0,
-      'warning': 1,
-      'info': 2,
+      'error': 1,
+      'warning': 2,
+      'info': 3,
     },
     Int_DeviceLostReason: {
       'undefined': 1,

system/lib/compiler-rt/stack_limits.S

@@ -15,14 +15,17 @@
 #endif
 
 .globaltype __stack_pointer, PTR
+.globl __stack_pointer
 
 .section .globals,"",@
 
 # TODO(sbc): It would be nice if these we initialized directly
 # using PTR.const rather than using the `emscripten_stack_init`
 .globaltype __stack_end, PTR
 __stack_end:
+.globl __stack_end
 .globaltype __stack_base, PTR
+.globl __stack_base
 __stack_base:
 
 .section .text,"",@
@@ -82,57 +85,6 @@ emscripten_stack_get_free:
   PTR.sub
   end_function
 
-#ifdef __EMSCRIPTEN_WASM_WORKERS__
-# TODO: Relocate the following to its own file wasm_worker.S, but need to figure out how to reference
-# __stack_base and __stack_end globals from a separate file as externs in order for that to work.
-.globl _emscripten_wasm_worker_initialize
-_emscripten_wasm_worker_initialize:
-  .functype _emscripten_wasm_worker_initialize (PTR /*stackLowestAddress*/, i32 /*stackSize*/) -> ()
-
-  // __stack_end = stackLowestAddress + (__builtin_wasm_tls_size() + 15) & -16;
-  local.get 0
-  .globaltype __tls_size, PTR, immutable
-  global.get __tls_size
-  PTR.add
-  PTR.const 0xf
-  PTR.add
-  PTR.const -0x10
-  PTR.and
-  global.set __stack_end
-
-  // __stack_base = stackLowestAddress + stackSize;
-  local.get 0
-  local.get 1
-#ifdef __wasm64__
-  i64.extend_i32_u
-#endif
-  PTR.add
-  global.set __stack_base
-
-// TODO: We'd like to do this here to avoid JS side calls to __set_stack_limits.
-//       (or even better, we'd like to avoid duplicate versions of the stack variables)
-// See https://github.com/emscripten-core/emscripten/issues/16496
-//  global.get __stack_base
-//  global.get __stack_end
-//  .functype __set_stack_limits (PTR, PTR) -> ()
-//  call __set_stack_limits
-
-  // __wasm_init_tls(stackLowestAddress);
-  local.get 0
-  .functype __wasm_init_tls (PTR) -> ()
-  call __wasm_init_tls
-
-  // N.b. The function __wasm_init_tls above does not need
-  // __stack_pointer initialized, and it destroys the value it was set to.
-  // So we must initialize __stack_pointer only *after* completing __wasm_init_tls:
-
-  // __stack_pointer = __stack_base;
-  global.get __stack_base
-  global.set __stack_pointer
-
-  end_function
-#endif
-
 # Add emscripten_stack_init to static ctors
 .section .init_array.1,"",@
 .p2align ALIGN

system/lib/dlmalloc.c

@@ -21,13 +21,12 @@
 /* XXX Emscripten Tracing API. This defines away the code if tracing is disabled. */
 #include <emscripten/trace.h>
 
-#ifdef __EMSCRIPTEN_WASM_WORKERS__
+#ifdef __EMSCRIPTEN_SHARED_MEMORY__
 #define USE_LOCKS 1
 #endif
 
 /* Make malloc() and free() threadsafe by securing the memory allocations with pthread mutexes. */
 #if __EMSCRIPTEN_PTHREADS__
-#define USE_LOCKS 1
 #define USE_SPIN_LOCKS 0 // Ensure we use pthread_mutex_t.
 #endif
 

system/lib/libc/musl/src/errno/__errno_location.c

@@ -1,22 +1,19 @@
 #include <errno.h>
 #include "pthread_impl.h"
 
-#if __EMSCRIPTEN_PTHREADS__
-// for pthreads, use the proper location on the thread info, so each
-// thread has its own errno
-int *__errno_location(void)
-{
-	return &__pthread_self()->errno_val;
-}
-#else
-// for single-threaded mode, avoid linking in pthreads support code
-// just for this
-static int __errno_storage = 0;
+#if __EMSCRIPTEN__
+// For emscripten we use TLS here instead of `__pthread_self`, so that in single
+// threaded builds this gets lowered away to normal global variable.
+static _Thread_local int __errno_storage = 0;
+#endif
 
 int *__errno_location(void)
 {
+#if __EMSCRIPTEN__
 	return &__errno_storage;
-}
+#else
+	return &__pthread_self()->errno_val;
 #endif
+}
 
 weak_alias(__errno_location, ___errno_location);

system/lib/wasm_worker/wasm_worker_initialize.S

@@ -0,0 +1,64 @@
+.extern __stack_pointer
+.extern __stack_base
+.extern __stack_end
+
+#ifdef __wasm64__
+#define PTR i64
+#define ALIGN 3
+#define PTRSTORE .int64
+#else
+#define PTR i32
+#define ALIGN 2
+#define PTRSTORE .int32
+#endif
+
+.globaltype __stack_pointer, PTR
+.globaltype __stack_end, PTR
+.globaltype __stack_base, PTR
+
+.globl _emscripten_wasm_worker_initialize
+_emscripten_wasm_worker_initialize:
+  .functype _emscripten_wasm_worker_initialize (PTR /*stackLowestAddress*/, i32 /*stackSize*/) -> ()
+
+  // __stack_end = stackLowestAddress + (__builtin_wasm_tls_size() + 15) & -16;
+  local.get 0
+  .globaltype __tls_size, PTR, immutable
+  global.get __tls_size
+  PTR.add
+  PTR.const 0xf
+  PTR.add
+  PTR.const -0x10
+  PTR.and
+  global.set __stack_end
+
+  // __stack_base = stackLowestAddress + stackSize;
+  local.get 0
+  local.get 1
+#ifdef __wasm64__
+  i64.extend_i32_u
+#endif
+  PTR.add
+  global.set __stack_base
+
+// TODO: We'd like to do this here to avoid JS side calls to __set_stack_limits.
+//       (or even better, we'd like to avoid duplicate versions of the stack variables)
+// See https://github.com/emscripten-core/emscripten/issues/16496
+//  global.get __stack_base
+//  global.get __stack_end
+//  .functype __set_stack_limits (PTR, PTR) -> ()
+//  call __set_stack_limits
+
+  // __wasm_init_tls(stackLowestAddress);
+  local.get 0
+  .functype __wasm_init_tls (PTR) -> ()
+  call __wasm_init_tls
+
+  // N.b. The function __wasm_init_tls above does not need
+  // __stack_pointer initialized, and it destroys the value it was set to.
+  // So we must initialize __stack_pointer only *after* completing __wasm_init_tls:
+
+  // __stack_pointer = __stack_base;
+  global.get __stack_base
+  global.set __stack_pointer
+
+  end_function

test/code_size/hello_wasm_worker_wasm.json

@@ -5,8 +5,8 @@
   "a.js.gz": 455,
   "a.ww.js": 115,
   "a.ww.js.gz": 127,
-  "a.wasm": 1850,
-  "a.wasm.gz": 1050,
-  "total": 3248,
-  "total_gz": 2016
+  "a.wasm": 1894,
+  "a.wasm.gz": 1077,
+  "total": 3292,
+  "total_gz": 2043
 }

test/core/test_dlfcn_self.exports

@@ -9,7 +9,9 @@ __progname_full
 __sig_actions
 __sig_pending
 __signgam
+__stack_base
 __stack_chk_guard
+__stack_end
 __threwValue
 __timezone
 __tzname

test/core/test_em_asm_unicode.out

@@ -1 +1 @@
-hello world…
+hello world…

test/core/test_longjmp.out

@@ -1,4 +1,4 @@
 first
 result: 1 1
 second
-result: 2 -1
+result: 2 -1

test/test_browser.py

@@ -3050,8 +3050,6 @@ def test_sdl2_gl_read(self):
     # SDL, OpenGL, readPixels
     self.btest_exit('test_sdl2_gl_read.c', args=['-sUSE_SDL=2'])
 
-  @no_4gb('https://github.com/libsdl-org/SDL/issues/9052')
-  @no_2gb('https://github.com/libsdl-org/SDL/issues/9052')
   @requires_graphics_hardware
   def test_sdl2_glmatrixmode_texture(self):
     self.reftest('test_sdl2_glmatrixmode_texture.c', 'test_sdl2_glmatrixmode_texture.png',
@@ -3105,8 +3103,6 @@ def test_sdl2_fog_linear(self):
   def test_sdl2_unwasteful(self):
     self.btest_exit('test_sdl2_unwasteful.c', args=['-sUSE_SDL=2', '-O1'])
 
-  @no_2gb('https://github.com/libsdl-org/SDL/issues/9052')
-  @no_4gb('https://github.com/libsdl-org/SDL/issues/9052')
   def test_sdl2_canvas_write(self):
     self.btest_exit('test_sdl2_canvas_write.c', args=['-sUSE_SDL=2'])
 
@@ -5011,9 +5007,13 @@ def test_wasm_worker_futex_wait(self, args):
 
   # Tests Wasm Worker thread stack setup
   @also_with_minimal_runtime
-  def test_wasm_worker_thread_stack(self):
-    for mode in (0, 1, 2):
-      self.btest('wasm_worker/thread_stack.c', expected='0', args=['-sWASM_WORKERS', f'-sSTACK_OVERFLOW_CHECK={mode}'])
+  @parameterized({
+    '0': (0,),
+    '1': (1,),
+    '2': (2,),
+  })
+  def test_wasm_worker_thread_stack(self, mode):
+    self.btest('wasm_worker/thread_stack.c', expected='0', args=['-sWASM_WORKERS', f'-sSTACK_OVERFLOW_CHECK={mode}'])
 
   # Tests emscripten_malloc_wasm_worker() and emscripten_current_thread_is_wasm_worker() functions
   @also_with_minimal_runtime

test/wasm_worker/c11__Thread_local.c

@@ -5,17 +5,15 @@
 
 _Thread_local int __attribute__((aligned(64))) tls = 1;
 
-void main_thread_func()
-{
+void main_thread_func() {
   assert(!emscripten_current_thread_is_wasm_worker());
   EM_ASM(out($0), tls);
 #ifdef REPORT_RESULT
   REPORT_RESULT(tls);
 #endif
 }
 
-void worker_main()
-{
+void worker_main() {
   assert(emscripten_current_thread_is_wasm_worker());
   assert(((intptr_t)&tls % 64) == 0);
   assert(tls != 42);
@@ -27,8 +25,7 @@ void worker_main()
 
 char stack[1024];
 
-int main()
-{
+int main() {
   EM_ASM(out($0), tls);
   assert(((intptr_t)&tls % 64) == 0);
   assert(!emscripten_current_thread_is_wasm_worker());

test/wasm_worker/cancel_all_wait_asyncs.c

@@ -9,23 +9,20 @@ volatile int32_t addr = 1;
 
 bool testSucceeded = 1;
 
-void asyncWaitFinishedShouldNotBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData)
-{
+void asyncWaitFinishedShouldNotBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData) {
   emscripten_console_log("asyncWaitFinishedShouldNotBeCalled");
   testSucceeded = 0;
   assert(0); // We should not reach here
 }
 
-void asyncWaitFinishedShouldBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData)
-{
+void asyncWaitFinishedShouldBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData) {
   emscripten_console_log("asyncWaitFinishedShouldBeCalled");
 #ifdef REPORT_RESULT
   REPORT_RESULT(testSucceeded);
 #endif
 }
 
-int main()
-{
+int main() {
   emscripten_console_log("Async waiting on address should give a wait token");
   ATOMICS_WAIT_TOKEN_T ret = emscripten_atomic_wait_async((int32_t*)&addr, 1, asyncWaitFinishedShouldNotBeCalled, (void*)42, EMSCRIPTEN_WAIT_ASYNC_INFINITY);
   assert(EMSCRIPTEN_IS_VALID_WAIT_TOKEN(ret));

test/wasm_worker/cancel_all_wait_asyncs_at_address.c

@@ -9,23 +9,20 @@ volatile int32_t addr = 1;
 
 bool testSucceeded = 1;
 
-void asyncWaitFinishedShouldNotBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData)
-{
+void asyncWaitFinishedShouldNotBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData) {
   emscripten_console_log("asyncWaitFinishedShouldNotBeCalled");
   testSucceeded = 0;
   assert(0); // We should not reach here
 }
 
-void asyncWaitFinishedShouldBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData)
-{
+void asyncWaitFinishedShouldBeCalled(int32_t *ptr, uint32_t val, ATOMICS_WAIT_RESULT_T waitResult, void *userData) {
   emscripten_console_log("asyncWaitFinishedShouldBeCalled");
 #ifdef REPORT_RESULT
   REPORT_RESULT(testSucceeded);
 #endif
 }
 
-int main()
-{
+int main() {
   emscripten_console_log("Async waiting on address should give a wait token");
   ATOMICS_WAIT_TOKEN_T ret = emscripten_atomic_wait_async((int32_t*)&addr, 1, asyncWaitFinishedShouldNotBeCalled, (void*)42, EMSCRIPTEN_WAIT_ASYNC_INFINITY);
   assert(EMSCRIPTEN_IS_VALID_WAIT_TOKEN(ret));