amd64: import asm strlen into libc

Reviewed by:	kib
Differential Revision:	https://reviews.freebsd.org/D28845

(cherry picked from commit 7f06b217c5)

lib/libc/amd64/string/Makefile.inc

@@ -8,4 +8,5 @@ MDSRCS+= \
 	memset.S \
 	strcat.S \
 	strcmp.S \
+	strlen.S \
 	stpcpy.S

lib/libc/amd64/string/strlen.S

@@ -0,0 +1,81 @@
+/*
+ * Written by Mateusz Guzik <mjg@freebsd.org>
+ * Public domain.
+ */
+
+#include <machine/asm.h>
+__FBSDID("$FreeBSD$");
+
+/*
+ * Note: this routine was written with kernel use in mind (read: no simd),
+ * it is only present in userspace as a temporary measure until something
+ * better gets imported.
+ */
+
+#define ALIGN_TEXT      .p2align 4,0x90 /* 16-byte alignment, nop filled */
+
+/*
+ * strlen(string)
+ *	  %rdi
+ *
+ * Uses the ((x - 0x01....01) & ~x & 0x80....80) trick.
+ *
+ * 0x01....01 is replaced with 0x0 - 0x01....01 so that it can be added
+ * with leaq.
+ *
+ * For a description see either:
+ * - "Hacker's Delight" by Henry S. Warren, Jr.
+ * - "Optimizing subroutines in assembly language: An optimization guide for x86 platforms"
+ *   by Agner Fog
+ *
+ * The latter contains a 32-bit variant of the same algorithm coded in assembly for i386.
+ */
+ENTRY(strlen)
+	movabsq	$0xfefefefefefefeff,%r8
+	movabsq	$0x8080808080808080,%r9
+
+	movq	%rdi,%r10
+	movq	%rdi,%rcx
+	testb	$7,%dil
+	jz	2f
+
+	/*
+	 * Handle misaligned reads: align to 8 and fill
+	 * the spurious bytes.
+	 */
+	andq	$~7,%rdi
+	movq	(%rdi),%r11
+	shlq	$3,%rcx
+	movq	$-1,%rdx
+	shlq	%cl,%rdx
+	notq	%rdx
+	orq	%rdx,%r11
+
+	leaq	(%r11,%r8),%rcx
+	notq	%r11
+	andq	%r11,%rcx
+	andq	%r9,%rcx
+	jnz	3f
+
+	/*
+	 * Main loop.
+	 */
+	ALIGN_TEXT
+1:
+	leaq	8(%rdi),%rdi
+2:
+	movq	(%rdi),%r11
+	leaq	(%r11,%r8),%rcx
+	notq	%r11
+	andq	%r11,%rcx
+	andq	%r9,%rcx
+	jz	1b
+3:
+	bsfq	%rcx,%rcx
+	shrq	$3,%rcx
+	leaq	(%rcx,%rdi),%rax
+	subq	%r10,%rax
+	ret
+END(strlen)
+
+	.section .note.GNU-stack,"",%progbits