emscripten-core · dschuff · Jan 6, 2025 · Oct 30, 2024 · Nov 6, 2024 · Nov 6, 2024
diff --git a/emcc.py b/emcc.py
@@ -386,10 +386,10 @@ def get_clang_flags(user_args):
 
   # In emscripten we currently disable bulk memory by default.
   # This should be removed/updated when we als update the default browser targets.
-  if '-mbulk-memory' not in user_args and '-mno-bulk-memory' not in user_args:
+  #if '-mbulk-memory' not in user_args and '-mno-bulk-memory' not in user_args:
     # Bulk memory may be enabled via threads or directly via -s.
-    if not settings.BULK_MEMORY:
-      flags.append('-mno-bulk-memory')
+  #  if not settings.BULK_MEMORY:
+  #    flags.append('-mno-bulk-memory')
   if '-mnontrapping-fptoint' not in user_args and '-mno-nontrapping-fptoint' not in user_args:
     flags.append('-mno-nontrapping-fptoint')
 

diff --git a/src/library.js b/src/library.js
@@ -406,38 +406,6 @@ addToLibrary({
   // the initial values of the environment accessible by getenv.
   $ENV: {},
 
-  // In -Oz builds, we replace memcpy() altogether with a non-unrolled wasm
-  // variant, so we should never emit _emscripten_memcpy_js() in the build.
-  // In STANDALONE_WASM we avoid the _emscripten_memcpy_js dependency so keep
-  // the wasm file standalone.
-  // In BULK_MEMORY mode we include native versions of these functions based
-  // on memory.fill and memory.copy.
-  // In MAIN_MODULE=1 or EMCC_FORCE_STDLIBS mode all of libc is force included
-  // so we cannot override parts of it, and therefore cannot use libc_optz.
-#if (SHRINK_LEVEL < 2 || LINKABLE || process.env.EMCC_FORCE_STDLIBS) && !STANDALONE_WASM && !BULK_MEMORY
-
-#if MIN_CHROME_VERSION < 45 || MIN_FIREFOX_VERSION < 34 || MIN_SAFARI_VERSION < 100101
-  // https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/TypedArray/copyWithin lists browsers that support TypedArray.prototype.copyWithin, but it
-  // has outdated information for Safari, saying it would not support it.
-  // /~https://github.com/WebKit/webkit/commit/24a800eea4d82d6d595cdfec69d0f68e733b5c52#diff-c484911d8df319ba75fce0d8e7296333R1 suggests support was added on Aug 28, 2015.
-  // Manual testing suggests:
-  //   Safari/601.1 Version/9.0 on iPhone 4s with iOS 9.3.6 (released September 30, 2015) does not support copyWithin.
-  // but the following systems do:
-  //   AppleWebKit/602.2.14 Safari/602.1 Version/10.0 Mobile/14B100 iPhone OS 10_1_1 on iPhone 5s with iOS 10.1.1 (released October 31, 2016)
-  //   AppleWebKit/603.3.8 Safari/602.1 Version/10.0 on iPhone 5 with iOS 10.3.4 (released July 22, 2019)
-  //   AppleWebKit/605.1.15 iPhone OS 12_3_1 Version/12.1.1 Safari/604.1 on iPhone SE with iOS 12.3.1
-  //   AppleWebKit/605.1.15 Safari/604.1 Version/13.0.4 iPhone OS 13_3 on iPhone 6s with iOS 13.3
-  //   AppleWebKit/605.1.15 Version/13.0.3 Intel Mac OS X 10_15_1 on Safari 13.0.3 (15608.3.10.1.4) on macOS Catalina 10.15.1
-  // Hence the support status of .copyWithin() for Safari version range [10.0.0, 10.1.0] is unknown.
-  _emscripten_memcpy_js: `= Uint8Array.prototype.copyWithin
-    ? (dest, src, num) => HEAPU8.copyWithin(dest, src, src + num)
-    : (dest, src, num) => HEAPU8.set(HEAPU8.subarray(src, src+num), dest)`,
-#else
-  _emscripten_memcpy_js: (dest, src, num) => HEAPU8.copyWithin(dest, src, src + num),
-#endif
-
-#endif
-
 #if !STANDALONE_WASM
   // ==========================================================================
   // assert.h

diff --git a/src/library_sigs.js b/src/library_sigs.js
@@ -323,7 +323,6 @@ sigs = {
   _emscripten_get_progname__sig: 'vpi',
   _emscripten_init_main_thread_js__sig: 'vp',
   _emscripten_lookup_name__sig: 'ip',
-  _emscripten_memcpy_js__sig: 'vppp',
   _emscripten_notify_mailbox_postmessage__sig: 'vpp',
   _emscripten_push_main_loop_blocker__sig: 'vppp',
   _emscripten_push_uncounted_main_loop_blocker__sig: 'vppp',

diff --git a/system/lib/libc/emscripten_internal.h b/system/lib/libc/emscripten_internal.h
@@ -30,12 +30,6 @@ _Noreturn void _abort_js(void);
 
 void setThrew(uintptr_t threw, int value);
 
-// An external JS implementation that is efficient for very large copies, using
-// HEAPU8.set()
-void _emscripten_memcpy_js(void* __restrict__ dest,
-                           const void* __restrict__ src,
-                           size_t n) EM_IMPORT(_emscripten_memcpy_js);
-
 void* _emscripten_memcpy_bulkmem(void* __restrict__ dest,
                                  const void* __restrict__ src,
                                  size_t n);

diff --git a/system/lib/libc/emscripten_memcpy.c b/system/lib/libc/emscripten_memcpy.c
@@ -9,20 +9,20 @@
 #include "emscripten_internal.h"
 
 // Use the simple/naive version of memcpy when building with asan
-#if defined(EMSCRIPTEN_OPTIMIZE_FOR_OZ) || __has_feature(address_sanitizer)
+#if __has_feature(address_sanitizer)
 
 static void *__memcpy(void *dest, const void *src, size_t n) {
   unsigned char *d = (unsigned char *)dest;
   const unsigned char *s = (const unsigned char *)src;
-#pragma clang loop unroll(disable)
   while(n--) *d++ = *s++;
   return dest;
 }
 
-#elif defined(__wasm_bulk_memory__)
+#elif defined(__wasm_bulk_memory__) || defined(EMSCRIPTEN_OPTIMIZE_FOR_OZ)
 
 static void *__memcpy(void *restrict dest, const void *restrict src, size_t n) {
-  return n ? _emscripten_memcpy_bulkmem(dest, src, n) : dest;
+  // TODO: Ensure this is inlined with Binaryen or inline asm
+  return _emscripten_memcpy_bulkmem(dest, src, n);
 }
 
 #else
@@ -35,12 +35,11 @@ static void *__memcpy(void *restrict dest, const void *restrict src, size_t n) {
   unsigned char *block_aligned_d_end;
   unsigned char *d_end;
 
-#if !defined(EMSCRIPTEN_STANDALONE_WASM)
   if (n >= 512) {
-    _emscripten_memcpy_js(dest, src, n);
+    // TODO: Re-investigate the size threshold to enable this
+    _emscripten_memcpy_bulkmem(dest, src, n);
     return dest;
   }
-#endif
 
   d_end = d + n;
   if ((((uintptr_t)d) & 3) == (((uintptr_t)s) & 3)) {

diff --git a/system/lib/libc/emscripten_memcpy_bulkmem.S b/system/lib/libc/emscripten_memcpy_bulkmem.S
@@ -7,10 +7,16 @@
 .globl _emscripten_memcpy_bulkmem
 _emscripten_memcpy_bulkmem:
   .functype _emscripten_memcpy_bulkmem (PTR, PTR, PTR) -> (PTR)
-  local.get 0
-  local.get 1
   local.get 2
-  memory.copy 0, 0
+#ifdef __wasm64__
+  i32.wrap_i64
+#endif
+  if
+    local.get 0
+    local.get 1
+    local.get 2
+    memory.copy 0, 0
+  end_if
   local.get 0
   end_function
 

diff --git a/system/lib/libc/emscripten_memset.c b/system/lib/libc/emscripten_memset.c
@@ -9,16 +9,7 @@ __attribute__((no_sanitize("address"))) void *__memset(void *str, int c, size_t
 __attribute__((__weak__)) void *__musl_memset(void *str, int c, size_t n);
 __attribute__((__weak__)) void *__memset(void *str, int c, size_t n);
 
-#ifdef EMSCRIPTEN_OPTIMIZE_FOR_OZ
-
-void *__memset(void *str, int c, size_t n) {
-  unsigned char *s = (unsigned char *)str;
-#pragma clang loop unroll(disable)
-  while(n--) *s++ = c;
-  return str;
-}
-
-#elif defined(__wasm_bulk_memory__)
+#if defined(__wasm_bulk_memory__) || defined(EMSCRIPTEN_OPTIMIZE_FOR_OZ)
 
 void *__memset(void *str, int c, size_t n) {
   return _emscripten_memset_bulkmem(str, c, n);

diff --git a/system/lib/libc/emscripten_memset_bulkmem.S b/system/lib/libc/emscripten_memset_bulkmem.S
@@ -7,10 +7,16 @@
 .globl _emscripten_memset_bulkmem
 _emscripten_memset_bulkmem:
   .functype _emscripten_memset_bulkmem (PTR, i32, PTR) -> (PTR)
-  local.get 0
-  local.get 1
   local.get 2
-  memory.fill 0
+#ifdef __wasm64__
+  i32.wrap_i64
+#endif
+  if
+   local.get 0
+   local.get 1
+   local.get 2
+   memory.fill 0
+  end_if
   local.get 0
   end_function
 

diff --git a/test/other/codesize/test_codesize_cxx_ctors1.imports b/test/other/codesize/test_codesize_cxx_ctors1.imports
@@ -1,6 +1,5 @@
 env.__cxa_throw
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env.emscripten_resize_heap
 wasi_snapshot_preview1.environ_get

diff --git a/test/other/codesize/test_codesize_cxx_ctors1.sent b/test/other/codesize/test_codesize_cxx_ctors1.sent
@@ -1,6 +1,5 @@
 __cxa_throw
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 emscripten_resize_heap
 environ_get

diff --git a/test/other/codesize/test_codesize_cxx_ctors2.imports b/test/other/codesize/test_codesize_cxx_ctors2.imports
@@ -1,5 +1,4 @@
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env.emscripten_resize_heap
 wasi_snapshot_preview1.fd_close

diff --git a/test/other/codesize/test_codesize_cxx_ctors2.sent b/test/other/codesize/test_codesize_cxx_ctors2.sent
@@ -1,6 +1,5 @@
 __cxa_throw
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 emscripten_resize_heap
 environ_get

diff --git a/test/other/codesize/test_codesize_cxx_except.imports b/test/other/codesize/test_codesize_cxx_except.imports
@@ -7,7 +7,6 @@ env.__cxa_throw
 env.__cxa_uncaught_exceptions
 env.__resumeException
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env.emscripten_resize_heap
 env.invoke_diii
@@ -17,7 +16,6 @@ env.invoke_ii
 env.invoke_iii
 env.invoke_iiii
 env.invoke_iiiii
-env.invoke_iiiiii
 env.invoke_iiiiiii
 env.invoke_iiiiiiii
 env.invoke_iiiiiiiiiiii

diff --git a/test/other/codesize/test_codesize_cxx_except.sent b/test/other/codesize/test_codesize_cxx_except.sent
@@ -7,7 +7,6 @@ __cxa_throw
 __cxa_uncaught_exceptions
 __resumeException
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 emscripten_resize_heap
 environ_get

diff --git a/test/other/codesize/test_codesize_cxx_except_wasm.imports b/test/other/codesize/test_codesize_cxx_except_wasm.imports
@@ -1,5 +1,4 @@
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env.emscripten_resize_heap
 wasi_snapshot_preview1.environ_get

diff --git a/test/other/codesize/test_codesize_cxx_except_wasm.sent b/test/other/codesize/test_codesize_cxx_except_wasm.sent
@@ -1,5 +1,4 @@
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 emscripten_resize_heap
 environ_get

diff --git a/test/other/codesize/test_codesize_cxx_except_wasm_exnref.imports b/test/other/codesize/test_codesize_cxx_except_wasm_exnref.imports
@@ -1,5 +1,4 @@
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env.emscripten_resize_heap
 wasi_snapshot_preview1.environ_get

diff --git a/test/other/codesize/test_codesize_cxx_except_wasm_exnref.sent b/test/other/codesize/test_codesize_cxx_except_wasm_exnref.sent
@@ -1,5 +1,4 @@
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 emscripten_resize_heap
 environ_get

diff --git a/test/other/codesize/test_codesize_cxx_mangle.imports b/test/other/codesize/test_codesize_cxx_mangle.imports
@@ -7,7 +7,6 @@ env.__cxa_throw
 env.__cxa_uncaught_exceptions
 env.__resumeException
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env.emscripten_resize_heap
 env.invoke_diii
@@ -17,7 +16,6 @@ env.invoke_ii
 env.invoke_iii
 env.invoke_iiii
 env.invoke_iiiii
-env.invoke_iiiiii
 env.invoke_iiiiiii
 env.invoke_iiiiiiii
 env.invoke_iiiiiiiiiiii

diff --git a/test/other/codesize/test_codesize_cxx_mangle.sent b/test/other/codesize/test_codesize_cxx_mangle.sent
@@ -7,7 +7,6 @@ __cxa_throw
 __cxa_uncaught_exceptions
 __resumeException
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 emscripten_resize_heap
 environ_get

diff --git a/test/other/codesize/test_codesize_cxx_noexcept.imports b/test/other/codesize/test_codesize_cxx_noexcept.imports
@@ -1,6 +1,5 @@
 env.__cxa_throw
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env.emscripten_resize_heap
 wasi_snapshot_preview1.environ_get

diff --git a/test/other/codesize/test_codesize_cxx_noexcept.sent b/test/other/codesize/test_codesize_cxx_noexcept.sent
@@ -1,6 +1,5 @@
 __cxa_throw
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 emscripten_resize_heap
 environ_get

diff --git a/test/other/codesize/test_codesize_cxx_wasmfs.imports b/test/other/codesize/test_codesize_cxx_wasmfs.imports
@@ -1,6 +1,5 @@
 env.__cxa_throw
 env._abort_js
-env._emscripten_memcpy_js
 env._tzset_js
 env._wasmfs_copy_preloaded_file_data
 env._wasmfs_get_num_preloaded_dirs

diff --git a/test/other/codesize/test_codesize_cxx_wasmfs.sent b/test/other/codesize/test_codesize_cxx_wasmfs.sent
@@ -1,6 +1,5 @@
 __cxa_throw
 _abort_js
-_emscripten_memcpy_js
 _tzset_js
 _wasmfs_copy_preloaded_file_data
 _wasmfs_get_num_preloaded_dirs

diff --git a/test/other/codesize/test_codesize_files_wasmfs.imports b/test/other/codesize/test_codesize_files_wasmfs.imports
@@ -11,6 +11,5 @@ j (_wasmfs_get_preloaded_child_path)
 k (_wasmfs_get_num_preloaded_files)
 l (_wasmfs_get_num_preloaded_dirs)
 m (_wasmfs_copy_preloaded_file_data)
-n (_emscripten_memcpy_js)
-o (_abort_js)
-p (random_get)
+n (_abort_js)
+o (random_get)
diff --git a/test/other/codesize/test_codesize_files_wasmfs.sent b/test/other/codesize/test_codesize_files_wasmfs.sent
@@ -11,6 +11,5 @@ j (_wasmfs_get_preloaded_child_path)
 k (_wasmfs_get_num_preloaded_files)
 l (_wasmfs_get_num_preloaded_dirs)
 m (_wasmfs_copy_preloaded_file_data)
-n (_emscripten_memcpy_js)
-o (_abort_js)
-p (random_get)
+n (_abort_js)
+o (random_get)