2008-05-26 Eric Blake <ebb9@byu.net>

Optimize the generic and x86 memchr.
        * libc/string/memchr.c (memchr) [!__OPTIMIZE_SIZE__]:
        Pre-align pointer so unaligned searches aren't penalized.
        * libc/machine/i386/memchr.S (memchr) [!__OPTIMIZE_SIZE__]: Word
        operations are faster than repnz byte searches.

newlib/ChangeLog

@@ -1,3 +1,11 @@
+2008-05-26  Eric Blake  <ebb9@byu.net>
+
+	Optimize the generic and x86 memchr.
+	* libc/string/memchr.c (memchr) [!__OPTIMIZE_SIZE__]:
+	Pre-align pointer so unaligned searches aren't penalized.
+	* libc/machine/i386/memchr.S (memchr) [!__OPTIMIZE_SIZE__]: Word
+	operations are faster than repnz byte searches.
+					
 2008-05-26  Eric Blake  <ebb9@byu.net>
 
 	Optimize the generic and x86 memset.

newlib/libc/machine/i386/memchr.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
@@ -16,14 +16,16 @@
 SYM (memchr):
 	pushl	ebp
 	movl	esp,ebp
-	pushl 	edi
+	pushl	edi
-	movl 	12(ebp),eax
+	movzbl	12(ebp),eax
-	movl 	16(ebp),ecx
+	movl	16(ebp),ecx
-	movl 	8(ebp),edi
+	movl	8(ebp),edi
 
 	xorl	edx,edx
 	testl	ecx,ecx
-	jz	L1
+	jz	L20
+
+#ifdef __OPTIMIZE_SIZE__
 
 	cld
 	repnz
@@ -31,9 +33,79 @@ SYM (memchr):
 
 	setnz	dl
 	decl	edi
+
+#else /* !__OPTIMIZE_SIZE__ */
+/* Do byte-wise checks until string is aligned.  */
+	testl	$3,edi
+	je	L5
+	cmpb	(edi),al
+	je	L15
+	incl	edi
+	decl	ecx
+	je	L20
+
+	testl	$3,edi
+	je	L5
+	cmpb	(edi),al
+	je	L15
+	incl	edi
+	decl	ecx
+	je	L20
+
+	testl	$3,edi
+	je	L5
+	cmpb	(edi),al
+	je	L15
+	incl	edi
+	decl	ecx
+	je	L20
+
+/* Create a mask, then check a word at a time.  */
+L5:
+	movb	al,ah
+	movl	eax,edx
+	sall	$16,edx
+	orl	edx,eax
+	pushl	ebx
+
+	.p2align 4,,7
+L8:
+	subl	$4,ecx
+	jc	L9
+	movl	(edi),edx
+	addl	$4,edi
+	xorl	eax,edx
+	leal	-16843009(edx),ebx
+	notl	edx
+	andl	edx,ebx
+	testl	$-2139062144,ebx
+	je	L8
+
+	subl	$4,edi
+
+L9:
+	popl	ebx
+	xorl	edx,edx
+	addl	$4,ecx
+	je	L20
+
+/* Final byte-wise checks.  */
+	.p2align 4,,7
+L10:
+	cmpb	(edi),al
+	je	L15
+	incl	edi
+	decl	ecx
+	jne	L10
+
+	xorl	edi,edi
+
+#endif /* !__OPTIMIZE_SIZE__ */
+
+L15:
 	decl	edx
 	andl	edi,edx
-L1:
+L20:
 	movl	edx,eax
 
 	leal	-4(ebp),esp

newlib/libc/string/memchr.c

@@ -20,7 +20,7 @@ DESCRIPTION
 	This function searches memory starting at <<*<[src]>>> for the
 	character <[c]>.  The search only ends with the first
 	occurrence of <[c]>, or after <[length]> characters; in
-	particular, <<NULL>> does not terminate the search.
+	particular, <<NUL>> does not terminate the search.
 
 RETURNS
 	If the character <[c]> is found within <[length]> characters
@@ -64,6 +64,9 @@ QUICKREF
 #error long int is not a 32bit or 64bit byte
 #endif
 
+/* DETECTCHAR returns nonzero if (long)X contains the byte used
+   to fill (long)MASK. */
+#define DETECTCHAR(X,MASK) (DETECTNULL(X ^ MASK))
 
 _PTR
 _DEFUN (memchr, (src_void, c, length),
@@ -71,56 +74,43 @@ _DEFUN (memchr, (src_void, c, length),
 	int c _AND
 	size_t length)
 {
-#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
   _CONST unsigned char *src = (_CONST unsigned char *) src_void;
+  unsigned char d = c;
 
-  c &= 0xff;
+#if !defined(PREFER_SIZE_OVER_SPEED) && !defined(__OPTIMIZE_SIZE__)
+  unsigned long *asrc;
+  unsigned long  mask;
+  int i;
 
-  while (length--)
+  while (UNALIGNED (src))
     {
-      if (*src == c)
+      if (!length--)
-	return (char *) src;
+        return NULL;
+      if (*src == d)
+        return (void *) src;
       src++;
     }
-  return NULL;
-#else
-  _CONST unsigned char *src = (_CONST unsigned char *) src_void;
-  unsigned long *asrc;
-  unsigned long  buffer;
-  unsigned long  mask;
-  int i, j;
 
-  c &= 0xff;
+  if (!TOO_SMALL (length))
-  
-  /* If the size is small, or src is unaligned, then 
-     use the bytewise loop.  We can hope this is rare.  */
-  if (!TOO_SMALL (length) && !UNALIGNED (src)) 
     {
-      /* The fast code reads the ASCII one word at a time and only 
+      /* If we get this far, we know that length is large and src is
+         word-aligned. */
+      /* The fast code reads the source one word at a time and only
          performs the bytewise search on word-sized segments if they
          contain the search character, which is detected by XORing
          the word-sized segment with a word-sized block of the search
-         character and then detecting for the presence of NULL in the
+         character and then detecting for the presence of NUL in the
          result.  */
-      asrc = (unsigned long*) src;
+      asrc = (unsigned long *) src;
-      mask = 0;
+      mask = d << 8 | d;
-      for (i = 0; i < LBLOCKSIZE; i++)
+      mask = mask << 16 | mask;
-        mask = (mask << 8) + c;
+      for (i = 32; i < LBLOCKSIZE * 8; i <<= 1)
+        mask = (mask << i) | mask;
 
       while (length >= LBLOCKSIZE)
         {
-          buffer = *asrc;
+          if (DETECTCHAR (*asrc, mask))
-          buffer ^=  mask;
+            break;
-          if (DETECTNULL (buffer))
-	    {
-              src = (unsigned char*) asrc;
-  	      for ( j = 0; j < LBLOCKSIZE; j++ )
-	        {
-                  if (*src == c)
-                    return (char*) src;
-                  src++;
-	        }
-   	    }
           length -= LBLOCKSIZE;
           asrc++;
         }
@@ -128,16 +118,17 @@ _DEFUN (memchr, (src_void, c, length),
       /* If there are fewer than LBLOCKSIZE characters left,
          then we resort to the bytewise loop.  */
 
-      src = (unsigned char*) asrc;
+      src = (unsigned char *) asrc;
     }
 
+#endif /* not PREFER_SIZE_OVER_SPEED */
+
   while (length--)
     {
-      if (*src == c)
+      if (*src == d)
-        return (char*) src;
+        return (void *) src;
       src++;
     }
 
   return NULL;
-#endif /* not PREFER_SIZE_OVER_SPEED */
 }