Use WebAssembly bulk memory opcodes

Makefile

@@ -22,6 +22,13 @@ BUILD_LIBC_TOP_HALF ?= yes
 # The directory where we're store intermediate artifacts.
 OBJDIR ?= $(CURDIR)/build
 
+# When the length is no larger than this threshold, we consider the
+# overhead of bulk memory opcodes to outweigh the performance benefit,
+# and fall back to the original musl implementation. See
+# https://github.com/WebAssembly/wasi-libc/pull/263 for relevant
+# discussion
+BULK_MEMORY_THRESHOLD ?= 32
+
 # Variables from this point on are not meant to be overridable via the
 # make command-line.
 
@@ -183,6 +190,10 @@ MUSL_PRINTSCAN_SOURCES = \
     $(LIBC_TOP_HALF_MUSL_SRC_DIR)/stdio/vfscanf.c \
     $(LIBC_TOP_HALF_MUSL_SRC_DIR)/stdlib/strtod.c \
     $(LIBC_TOP_HALF_MUSL_SRC_DIR)/stdlib/wcstod.c
+BULK_MEMORY_SOURCES = \
+    $(LIBC_TOP_HALF_MUSL_SRC_DIR)/string/memcpy.c \
+    $(LIBC_TOP_HALF_MUSL_SRC_DIR)/string/memmove.c \
+    $(LIBC_TOP_HALF_MUSL_SRC_DIR)/string/memset.c
 LIBC_TOP_HALF_HEADERS_PRIVATE = $(LIBC_TOP_HALF_DIR)/headers/private
 LIBC_TOP_HALF_SOURCES = $(LIBC_TOP_HALF_DIR)/sources
 LIBC_TOP_HALF_ALL_SOURCES = \
@@ -253,6 +264,7 @@ endif
 MUSL_PRINTSCAN_OBJS = $(call objs,$(MUSL_PRINTSCAN_SOURCES))
 MUSL_PRINTSCAN_LONG_DOUBLE_OBJS = $(patsubst %.o,%.long-double.o,$(MUSL_PRINTSCAN_OBJS))
 MUSL_PRINTSCAN_NO_FLOATING_POINT_OBJS = $(patsubst %.o,%.no-floating-point.o,$(MUSL_PRINTSCAN_OBJS))
+BULK_MEMORY_OBJS = $(call objs,$(BULK_MEMORY_SOURCES))
 LIBWASI_EMULATED_MMAN_OBJS = $(call objs,$(LIBWASI_EMULATED_MMAN_SOURCES))
 LIBWASI_EMULATED_PROCESS_CLOCKS_OBJS = $(call objs,$(LIBWASI_EMULATED_PROCESS_CLOCKS_SOURCES))
 LIBWASI_EMULATED_GETPID_OBJS = $(call objs,$(LIBWASI_EMULATED_GETPID_SOURCES))
@@ -384,6 +396,14 @@ $(MUSL_PRINTSCAN_NO_FLOATING_POINT_OBJS): CFLAGS += \
 	    -D__wasilibc_printscan_no_floating_point \
 	    -D__wasilibc_printscan_floating_point_support_option="\"remove -lc-printscan-no-floating-point from the link command\""
 
+# TODO: apply -mbulk-memory globally, once
+# https://github.com/llvm/llvm-project/issues/52618 is resolved
+$(BULK_MEMORY_OBJS): CFLAGS += \
+        -mbulk-memory
+
+$(BULK_MEMORY_OBJS): CFLAGS += \
+        -DBULK_MEMORY_THRESHOLD=$(BULK_MEMORY_THRESHOLD)
+
 $(LIBWASI_EMULATED_SIGNAL_MUSL_OBJS): CFLAGS += \
 	    -D_WASI_EMULATED_SIGNAL
 

libc-top-half/musl/src/string/memcpy.c

@@ -4,6 +4,10 @@
 
 void *memcpy(void *restrict dest, const void *restrict src, size_t n)
 {
+#if defined(__wasm_bulk_memory__)
+	if (n > BULK_MEMORY_THRESHOLD)
+	  return __builtin_memcpy(dest, src, n);
+#endif
 	unsigned char *d = dest;
 	const unsigned char *s = src;
 

libc-top-half/musl/src/string/memmove.c

@@ -8,6 +8,10 @@ typedef __attribute__((__may_alias__)) size_t WT;
 
 void *memmove(void *dest, const void *src, size_t n)
 {
+#if defined(__wasm_bulk_memory__)
+	if (n > BULK_MEMORY_THRESHOLD)
+		return __builtin_memmove(dest, src, n);
+#endif
 	char *d = dest;
 	const char *s = src;
 

libc-top-half/musl/src/string/memset.c

@@ -3,6 +3,10 @@
 
 void *memset(void *dest, int c, size_t n)
 {
+#if defined(__wasm_bulk_memory__)
+	if (n > BULK_MEMORY_THRESHOLD)
+		return __builtin_memset(dest, c, n);
+#endif
 	unsigned char *s = dest;
 	size_t k;