Optimize strchr for x86.

* libc/machine/i386/strchr.S (strchr): Pre-align data so unaligned
searches aren't penalized.  Special-case searching for 0.

1 files changed, 100 insertions, 13 deletions

diff --git a/newlib/libc/machine/i386/strchr.S b/newlib/libc/machine/i386/strchr.S
index fe425d2d8..1d98b8149 100644
--- a/newlib/libc/machine/i386/strchr.S
+++ b/newlib/libc/machine/i386/strchr.S
@@ -1,6 +1,6 @@
 /*
  * ====================================================
- * Copyright (C) 1998, 2002 by Red Hat Inc. All rights reserved.
+ * Copyright (C) 1998, 2002, 2008 by Red Hat Inc. All rights reserved.
  *
  * Permission to use, copy, modify, and distribute this
  * software is freely granted, provided that this notice
@@ -9,7 +9,7 @@
  */
 
 	#include "i386mach.h"
-	
+
 	.global SYM (strchr)
        SOTYPE_FUNCTION(strchr)
 
@@ -21,14 +21,45 @@ SYM (strchr):
 	pushl ebx
 	xorl ebx,ebx
 	movl 8(ebp),edi
-	movb 12(ebp),bl
+	addb 12(ebp),bl
+
+#ifndef __OPTIMIZE_SIZE__
+/* Special case strchr(p,0).  */
+	je L25
 
-#ifndef __OPTIMIZE_SIZE__	
-/* check if string is aligned, if not do check one byte at a time */
+/* Do byte-wise checks until string is aligned.  */
 	test $3,edi
-	jne L9
+	je L5
+	movl edi,eax
+	movb (eax),cl
+	testb cl,cl
+	je L14
+	cmpb bl,cl
+	je L19
+	incl edi
+
+	test $3,edi
+	je L5
+	movl edi,eax
+	movb (eax),cl
+	testb cl,cl
+	je L14
+	cmpb bl,cl
+	je L19
+	incl edi
+
+	test $3,edi
+	je L5
+	movl edi,eax
+	movb (eax),cl
+	testb cl,cl
+	je L14
+	cmpb bl,cl
+	je L19
+	incl edi
 
 /* create 4 byte mask which is just the desired byte repeated 4 times */
+L5:
 	movl ebx,ecx
 	sall $8,ebx
 	subl $4,edi
@@ -49,15 +80,14 @@ L10:
 	testl $-2139062144,edx
 	jne L9
 
-	movl ebx,eax
-	xorl ecx,eax
-	leal -16843009(eax),edx
-	notl eax
-	andl eax,edx
+	xorl ebx,ecx
+	leal -16843009(ecx),edx
+	notl ecx
+	andl ecx,edx
 	testl $-2139062144,edx
 	je L10
 #endif /* not __OPTIMIZE_SIZE__ */
-		
+
 /* loop while (*s && *s++ != c) */
 L9:
 	leal -1(edi),eax
@@ -69,7 +99,7 @@ L15:
 	je L14
 	cmpb bl,dl
 	jne L15
-	
+
 L14:
 /*  if (*s == c) return address otherwise return NULL */
 	cmpb bl,(eax)
@@ -83,3 +113,60 @@ L19:
 	leave
 	ret
 
+#ifndef __OPTIMIZE_SIZE__
+/* Special case strchr(p,0).  */
+#if 0
+	/* Hideous performance on modern machines.  */
+L25:
+	cld
+	movl $-1,ecx
+	xor eax,eax
+	repnz
+	scasb
+	leal -1(edi),eax
+	jmp L19
+#endif
+L25:
+/* Do byte-wise checks until string is aligned.  */
+	test $3,edi
+	je L26
+	movl edi,eax
+	movb (eax),cl
+	testb cl,cl
+	je L19
+	incl edi
+
+	test $3,edi
+	je L26
+	movl edi,eax
+	movb (eax),cl
+	testb cl,cl
+	je L19
+	incl edi
+
+	test $3,edi
+	je L26
+	movl edi,eax
+	movb (eax),cl
+	testb cl,cl
+	je L19
+	incl edi
+
+L26:
+	subl $4,edi
+
+/* loop performing 4 byte mask checking for desired 0 byte */
+	.p2align 4,,7
+L27:
+	addl $4,edi
+	movl (edi),ecx
+	leal -16843009(ecx),edx
+	movl ecx,eax
+	notl eax
+	andl eax,edx
+	testl $-2139062144,edx
+	je L27
+
+	jmp L9
+
+#endif /* !__OPTIMIZE_SIZE__ */