optimized C memcpy

unlike the old C memcpy, this version handles word-at-a-time reads and
writes even for misaligned copies. it does not require that the cpu
support misaligned accesses; instead, it performs bit shifts to
realign the bytes for the destination.

essentially, this is the C version of the ARM assembly language
memcpy. the ideas are all the same, and it should perform well on any
arch with a decent number of general-purpose registers that has a
barrel shift operation. since the barrel shifter is an optional cpu
feature on microblaze, it may be desirable to provide an alternate asm
implementation on microblaze, but otherwise the C code provides a
competitive implementation for "generic risc-y" cpu archs that should
alleviate the urgent need for arch-specific memcpy asm.

src/string/memcpy.c

@@ -1,29 +1,124 @@
 #include <string.h>
-#include <stdlib.h>
 #include <stdint.h>
-
-#define SS (sizeof(size_t))
-#define ALIGN (sizeof(size_t)-1)
-#define ONES ((size_t)-1/UCHAR_MAX)
+#include <endian.h>
 
 void *memcpy(void *restrict dest, const void *restrict src, size_t n)
 {
 	unsigned char *d = dest;
 	const unsigned char *s = src;
 
-	if (((uintptr_t)d & ALIGN) != ((uintptr_t)s & ALIGN))
-		goto misaligned;
+#ifdef __GNUC__
 
-	for (; ((uintptr_t)d & ALIGN) && n; n--) *d++ = *s++;
-	if (n) {
-		size_t *wd = (void *)d;
-		const size_t *ws = (const void *)s;
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+#define LS >>
+#define RS <<
+#else
+#define LS <<
+#define RS >>
+#endif
 
-		for (; n>=SS; n-=SS) *wd++ = *ws++;
-		d = (void *)wd;
-		s = (const void *)ws;
-misaligned:
-		for (; n; n--) *d++ = *s++;
+	typedef uint32_t __attribute__((__may_alias__)) u32;
+	uint32_t w, x;
+
+	for (; (uintptr_t)s % 4 && n; n--) *d++ = *s++;
+
+	if ((uintptr_t)d % 4 == 0) {
+		for (; n>=16; s+=16, d+=16, n-=16) {
+			*(u32 *)(d+0) = *(u32 *)(s+0);
+			*(u32 *)(d+4) = *(u32 *)(s+4);
+			*(u32 *)(d+8) = *(u32 *)(s+8);
+			*(u32 *)(d+12) = *(u32 *)(s+12);
+		}
+		if (n&8) {
+			*(u32 *)(d+0) = *(u32 *)(s+0);
+			*(u32 *)(d+4) = *(u32 *)(s+4);
+			d += 8; s += 8;
+		}
+		if (n&4) {
+			*(u32 *)(d+0) = *(u32 *)(s+0);
+			d += 4; s += 4;
+		}
+		if (n&2) {
+			*d++ = *s++; *d++ = *s++;
+		}
+		if (n&1) {
+			*d = *s;
 		}
 		return dest;
 	}
+
+	if (n >= 32) switch ((uintptr_t)d % 4) {
+	case 1:
+		w = *(u32 *)s;
+		*d++ = *s++;
+		*d++ = *s++;
+		*d++ = *s++;
+		n -= 3;
+		for (; n>=17; s+=16, d+=16, n-=16) {
+			x = *(u32 *)(s+1);
+			*(u32 *)(d+0) = (w LS 24) | (x RS 8);
+			w = *(u32 *)(s+5);
+			*(u32 *)(d+4) = (x LS 24) | (w RS 8);
+			x = *(u32 *)(s+9);
+			*(u32 *)(d+8) = (w LS 24) | (x RS 8);
+			w = *(u32 *)(s+13);
+			*(u32 *)(d+12) = (x LS 24) | (w RS 8);
+		}
+		break;
+	case 2:
+		w = *(u32 *)s;
+		*d++ = *s++;
+		*d++ = *s++;
+		n -= 2;
+		for (; n>=18; s+=16, d+=16, n-=16) {
+			x = *(u32 *)(s+2);
+			*(u32 *)(d+0) = (w LS 16) | (x RS 16);
+			w = *(u32 *)(s+6);
+			*(u32 *)(d+4) = (x LS 16) | (w RS 16);
+			x = *(u32 *)(s+10);
+			*(u32 *)(d+8) = (w LS 16) | (x RS 16);
+			w = *(u32 *)(s+14);
+			*(u32 *)(d+12) = (x LS 16) | (w RS 16);
+		}
+		break;
+	case 3:
+		w = *(u32 *)s;
+		*d++ = *s++;
+		n -= 1;
+		for (; n>=19; s+=16, d+=16, n-=16) {
+			x = *(u32 *)(s+3);
+			*(u32 *)(d+0) = (w LS 8) | (x RS 24);
+			w = *(u32 *)(s+7);
+			*(u32 *)(d+4) = (x LS 8) | (w RS 24);
+			x = *(u32 *)(s+11);
+			*(u32 *)(d+8) = (w LS 8) | (x RS 24);
+			w = *(u32 *)(s+15);
+			*(u32 *)(d+12) = (x LS 8) | (w RS 24);
+		}
+		break;
+	}
+	if (n&16) {
+		*d++ = *s++; *d++ = *s++; *d++ = *s++; *d++ = *s++;
+		*d++ = *s++; *d++ = *s++; *d++ = *s++; *d++ = *s++;
+		*d++ = *s++; *d++ = *s++; *d++ = *s++; *d++ = *s++;
+		*d++ = *s++; *d++ = *s++; *d++ = *s++; *d++ = *s++;
+	}
+	if (n&8) {
+		*d++ = *s++; *d++ = *s++; *d++ = *s++; *d++ = *s++;
+		*d++ = *s++; *d++ = *s++; *d++ = *s++; *d++ = *s++;
+	}
+	if (n&4) {
+		*d++ = *s++; *d++ = *s++; *d++ = *s++; *d++ = *s++;
+	}
+	if (n&2) {
+		*d++ = *s++; *d++ = *s++;
+	}
+	if (n&1) {
+		*d = *s;
+	}
+	return dest;
+#endif
+
+	for (; n; n--) *d++ = *s++;
+	return dest;
+}