--- /dev/null
+/* $NetBSD: memchr.S,v 1.6 2014/03/22 19:16:34 jakllsch Exp $ */
+
+/*-
+ * Copyright (c) 2009 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by David Laight.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <machine/asm.h>
+
+#if defined(LIBC_SCCS)
+ RCSID("$NetBSD: memchr.S,v 1.6 2014/03/22 19:16:34 jakllsch Exp $")
+#endif
+
+/*
+ * The instruction sequences used try to avoid data dependencies
+ * between adjacent instructions (to allow parallel execution).
+ * The 'imul' for %r9 could be put into the delay following the
+ * memory read (ie inside the loop) at no obvious cost - except
+ * that the loop is currently exactly 32 bytes - 2 fetch blocks!.
+ *
+ * I don't think aligning any of the other branch targets is useful.
+ */
+
+ENTRY3(memchr)
+ movabsq $0x0101010101010101,%r8
+ lea (%rdi,%rdx),%r10 /* limit of buffer to scan */
+ movzbq %sil,%rsi /* mask high bits! */
+
+ /* 'directpath' imuls can execute 3 at a time ... (amd) */
+ imul %r8,%rsi /* search byte replicated in word */
+ imul $0x80,%r8,%r9 /* 0x8080808080808080 */
+ test $7,%dil
+ jnz 20f /* jump if misaligned */
+ jmp 1f /* jump to avoid 4 nops (13 bytes) in gap */
+
+ _ALIGN_TEXT /* entire loop now in 32 aligned bytes */
+1:
+ cmpq %r10,%rdi /* end of buffer ? */
+ jae 30f /* jump if so */
+
+ movq (%rdi),%rax /* value to check */
+ addq $8,%rdi
+ xorq %rsi,%rax /* now looking for zeros */
+2:
+ mov %rax,%rcx
+ subq %r8,%rax /* x - 0x01 */
+ not %rcx
+ andq %r9,%rax /* (x - 0x01) & 0x80 */
+ andq %rcx,%rax /* ((x - 0x01) & 0x80) & ~x */
+ je 1b /* jump if not found */
+
+/* Found byte in word, get its address */
+ bsf %rax,%rax
+ shr $3,%eax
+ lea -8(%rax,%rdi),%rax
+ cmpq %r10,%rax /* need to check not beyond buffer */
+ jae 30f
+ rep
+ ret /* amd - no ret after jmp */
+
+/* Input misaligned, read aligned and make low bytes invalid */
+20:
+ mov %dil,%cl /* misalignment amount 1..7 (+high bits )*/
+ and $~7,%dil /* %rdi now start of word */
+ test %rdx,%rdx /* zero length, don't read */
+ jz 30f
+
+ neg %cl /* 7..1 (+high bits) */
+ mov (%rdi),%rax /* word containing first byte */
+ addq $8,%rdi
+ and $7,%cl /* 7..1 */
+
+ mov %r8,%r11 /* any value with bits in each byte */
+ shl $3,%cl /* 56..8 */
+ xorq %rsi,%rax /* now looking for zeros */
+
+ /* Set low bytes non-zero */
+ shr %cl,%r11 /* non-zero in unwanted bytes */
+ or %r11,%rax /* low bytes now set */
+ jmp 2b
+
+/* Not found */
+30: xorq %rax,%rax
+ ret
+END(memchr)