Add a machine-specific, optimized implementation of strcpy.

PR: 73111
Submitted by: Ville-Pertti Keinonen <will@iki.fi> (taken from NetBSD)
MFC after: 3 weeks

lib/libc/amd64/string/Makefile.inc

@@ -1,4 +1,4 @@
 # $FreeBSD$
 
 MDSRCS+= bcmp.S bcopy.S bzero.S memcmp.S memcpy.S memmove.S memset.S \
-	strcmp.S
+	strcmp.S strcpy.S

lib/libc/amd64/string/strcpy.S

@@ -0,0 +1,112 @@
+/*
+ * Written by J.T. Conklin <jtc@acorntoolworks.com>
+ * Public domain.
+ */
+
+#include <machine/asm.h>
+__FBSDID("$FreeBSD$");
+
+#if 0
+	RCSID("$NetBSD: strcpy.S,v 1.3 2004/07/19 20:04:41 drochner Exp $")
+#endif
+
+/*
+ * This strcpy implementation copies a byte at a time until the
+ * source pointer is aligned to a word boundary, it then copies by
+ * words until it finds a word containing a zero byte, and finally
+ * copies by bytes until the end of the string is reached.
+ *
+ * While this may result in unaligned stores if the source and
+ * destination pointers are unaligned with respect to each other,
+ * it is still faster than either byte copies or the overhead of
+ * an implementation suitable for machines with strict alignment
+ * requirements.
+ */
+
+ENTRY(strcpy)
+	movq	%rdi,%rax
+	movabsq	$0x0101010101010101,%r8
+	movabsq	$0x8080808080808080,%r9
+
+	/*
+	 * Align source to a word boundary.
+	 * Consider unrolling loop?
+	 */
+	.align	4
+.Lalign:
+	testb	$7,%sil
+	je	.Lword_aligned
+	movb	(%rsi),%dl
+	incq	%rsi
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl
+	jne	.Lalign
+	ret
+
+	.align	4
+.Lloop:
+	movq	%rdx,(%rdi)
+	addq	$8,%rdi
+.Lword_aligned:
+	movq	(%rsi),%rdx
+	movq	%rdx,%rcx
+	addq	$8,%rsi
+	subq	%r8,%rcx
+	testq	%r9,%rcx
+	je	.Lloop
+
+	/*
+	 * In rare cases, the above loop may exit prematurely. We must
+	 * return to the loop if none of the bytes in the word equal 0.
+	 */
+
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 1st byte == 0? */
+	je	.Ldone
+
+	shrq	$8,%rdx
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 2nd byte == 0? */
+	je	.Ldone
+
+	shrq	$8,%rdx
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 3rd byte == 0? */
+	je	.Ldone
+
+	shrq	$8,%rdx
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 4th byte == 0? */
+	je	.Ldone
+
+	shrq	$8,%rdx
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 5th byte == 0? */
+	je	.Ldone
+
+	shrq	$8,%rdx
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 6th byte == 0? */
+	je	.Ldone
+
+	shrq	$8,%rdx
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 7th byte == 0? */
+	je	.Ldone
+
+	shrq	$8,%rdx
+	movb	%dl,(%rdi)
+	incq	%rdi
+	testb	%dl,%dl		/* 8th byte == 0? */
+	jne	.Lword_aligned
+
+.Ldone:
+	ret