lib: fix memcpy() performance with freestanding library builds

With -ffreestanding, for memcpy() to be optimized properly when used for
unaligned accesses, we need to use __builtin_memcpy().

lib/lib_common.h

@@ -37,10 +37,29 @@ void *libdeflate_aligned_malloc(size_t alignment, size_t size);
 void libdeflate_aligned_free(void *ptr);
 
 #ifdef FREESTANDING
+/*
+ * With -ffreestanding, <string.h> may be missing, and we must provide
+ * implementations of memset(), memcpy(), memmove(), and memcmp().
+ * See https://gcc.gnu.org/onlinedocs/gcc/Standards.html
+ *
+ * Also, -ffreestanding disables interpreting calls to these functions as
+ * built-ins.  E.g., calling memcpy(&v, p, WORDBYTES) will make a function call,
+ * not be optimized to a single load instruction.  For performance reasons we
+ * don't want that.  So, declare these functions as macros that expand to the
+ * corresponding built-ins.  This approach is recommended in the gcc man page.
+ * We still need the actual function definitions in case gcc calls them.
+ */
 void *memset(void *s, int c, size_t n);
+#define memset(s, c, n)		__builtin_memset((s), (c), (n))
+
 void *memcpy(void *dest, const void *src, size_t n);
+#define memcpy(dest, src, n)	__builtin_memcpy((dest), (src), (n))
+
 void *memmove(void *dest, const void *src, size_t n);
+#define memmove(dest, src, n)	__builtin_memmove((dest), (src), (n))
+
 int memcmp(const void *s1, const void *s2, size_t n);
+#define memcmp(s1, s2, n)	__builtin_memcmp((s1), (s2), (n))
 #else
 #include <string.h>
 #endif

lib/utils.c

@@ -80,9 +80,11 @@ libdeflate_set_memory_allocator(void *(*malloc_func)(size_t),
 
 /*
  * Implementations of libc functions for freestanding library builds.
- * Not optimized yet.  Normal library builds don't use these.
+ * Normal library builds don't use these.  Not optimized yet; usually the
+ * compiler expands these functions and doesn't actually call them anyway.
  */
 #ifdef FREESTANDING
+#undef memset
 void *memset(void *s, int c, size_t n)
 {
 	u8 *p = s;
@@ -93,6 +95,7 @@ void *memset(void *s, int c, size_t n)
 	return s;
 }
 
+#undef memcpy
 void *memcpy(void *dest, const void *src, size_t n)
 {
 	u8 *d = dest;
@@ -104,6 +107,7 @@ void *memcpy(void *dest, const void *src, size_t n)
 	return dest;
 }
 
+#undef memmove
 void *memmove(void *dest, const void *src, size_t n)
 {
 	u8 *d = dest;
@@ -118,6 +122,7 @@ void *memmove(void *dest, const void *src, size_t n)
 	return dest;
 }
 
+#undef memcmp
 int memcmp(const void *s1, const void *s2, size_t n)
 {
 	const u8 *p1 = s1;