Always use bulkmemory memcpy and memset when available (#20144)

Unlike the JS versions of these function there is no need to only use
these for small inputs.

Results of running the test_memcpy_128b test before and after this
change:

```
v8-bulk: mean: 1.536 (+-0.071) secs  median: 1.495  range: 1.471-1.650  (noise: 4.630%)  (5 runs)
        size:   149291, compressed:    54249
```

->

```
v8-bulk: mean: 1.489 (+-0.117) secs  median: 1.535  range: 1.268-1.606  (noise: 7.871%)  (5 runs)-
        size:   148387, compressed:    53813-
```

See comments in #19128

Author: sbc100

src/library.js

@@ -385,8 +385,8 @@ addToLibrary({
   $ENV: {},
 
   // In -Oz builds, we replace memcpy() altogether with a non-unrolled wasm
-  // variant, so we should never emit emscripten_memcpy_big() in the build.
-  // In STANDALONE_WASM we avoid the emscripten_memcpy_big dependency so keep
+  // 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.
@@ -407,11 +407,11 @@ addToLibrary({
   //   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_big: `= Uint8Array.prototype.copyWithin
+  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_big: (dest, src, num) => HEAPU8.copyWithin(dest, src, src + num),
+  emscripten_memcpy_js: (dest, src, num) => HEAPU8.copyWithin(dest, src, src + num),
 #endif
 
 #endif

src/library_sigs.js

@@ -690,7 +690,7 @@ sigs = {
   emscripten_math_sqrt__sig: 'dd',
   emscripten_math_tan__sig: 'dd',
   emscripten_math_tanh__sig: 'dd',
-  emscripten_memcpy_big__sig: 'vppp',
+  emscripten_memcpy_js__sig: 'vppp',
   emscripten_navigator_hardware_concurrency__sig: 'i',
   emscripten_notify_memory_growth__sig: 'vp',
   emscripten_num_logical_cores__sig: 'i',

system/lib/libc/emscripten_internal.h

@@ -28,10 +28,14 @@ extern "C" {
 
 // An external JS implementation that is efficient for very large copies, using
 // HEAPU8.set()
-void emscripten_memcpy_big(void* __restrict__ dest,
+void emscripten_memcpy_js(void* __restrict__ dest,
                            const void* __restrict__ src,
-                           size_t n) EM_IMPORT(emscripten_memcpy_big);
-void emscripten_memset_big(void* ptr, char value, size_t n);
+                           size_t n) EM_IMPORT(emscripten_memcpy_js);
+
+void* emscripten_memcpy_bulkmem(void* __restrict__ dest,
+                                const void* __restrict__ src,
+                                size_t n);
+void* emscripten_memset_bulkmem(void* ptr, char value, size_t n);
 
 void emscripten_notify_memory_growth(size_t memory_index);
 

system/lib/libc/emscripten_memcpy.c

@@ -19,6 +19,12 @@ static void *__memcpy(void *dest, const void *src, size_t n) {
   return dest;
 }
 
+#elif defined(__wasm_bulk_memory__)
+
+static void *__memcpy(void *restrict dest, const void *restrict src, size_t n) {
+  return emscripten_memcpy_bulkmem(dest, src, n);
+}
+
 #else
 
 static void *__memcpy(void *restrict dest, const void *restrict src, size_t n) {
@@ -29,9 +35,9 @@ 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) || defined(__wasm_bulk_memory__)
+#if !defined(EMSCRIPTEN_STANDALONE_WASM)
   if (n >= 512) {
-    emscripten_memcpy_big(dest, src, n);
+    emscripten_memcpy_js(dest, src, n);
     return dest;
   }
 #endif

system/lib/libc/emscripten_memcpy_big.S renamed to system/lib/libc/emscripten_memcpy_bulkmem.S

@@ -4,11 +4,12 @@
 #define PTR i32
 #endif
 
-.globl emscripten_memcpy_big
-emscripten_memcpy_big:
-  .functype emscripten_memcpy_big (PTR, PTR, PTR) -> ()
+.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
+  local.get 0
   end_function

system/lib/libc/emscripten_memset.c

@@ -20,16 +20,8 @@ void *__memset(void *str, int c, size_t n) {
 
 #elif defined(__wasm_bulk_memory__)
 
-#define memset __musl_memset
-#include "musl/src/string/memset.c"
-#undef memset
-
 void *__memset(void *str, int c, size_t n) {
-  if (n >= 512) {
-    emscripten_memset_big(str, c, n);
-    return str;
-  }
-  return __musl_memset(str, c, n);
+  return emscripten_memset_bulkmem(str, c, n);
 }
 
 #else

system/lib/libc/emscripten_memset_big.S renamed to system/lib/libc/emscripten_memset_bulkmem.S

@@ -4,11 +4,12 @@
 #define PTR i32
 #endif
 
-.globl emscripten_memset_big
-emscripten_memset_big:
-  .functype emscripten_memset_big (PTR, i32, PTR) -> ()
+.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
+  local.get 0
   end_function

system/lib/standalone/standalone.c

@@ -142,11 +142,6 @@ weak int getentropy(void* buffer, size_t length) {
 
 // Emscripten additions
 
-// Should never be called in standalone mode
-weak void emscripten_memcpy_big(void *restrict dest, const void *restrict src, size_t n) {
-  __builtin_unreachable();
-}
-
 size_t emscripten_get_heap_max() {
   // In standalone mode we don't have any wasm instructions to access the max
   // memory size so the best we can do (without calling an import) is return

test/other/metadce/test_metadce_cxx_ctors1.imports

@@ -1,6 +1,6 @@
 env.__cxa_throw
 env.abort
-env.emscripten_memcpy_big
+env.emscripten_memcpy_js
 env.emscripten_resize_heap
 env.strftime_l
 wasi_snapshot_preview1.environ_get

test/other/metadce/test_metadce_cxx_ctors1.jssize

@@ -1 +1 @@
-25060
+25059