yann@1365: Original patch from: gentoo/src/patchsets/glibc/2.9/1020_all_glibc-2.9-strlen-hack.patch
yann@1365: 
yann@1365: -= BEGIN original header =-
yann@1365: http://sourceware.org/bugzilla/show_bug.cgi?id=5807
yann@1365: http://www.cl.cam.ac.uk/~am21/progtricks.html
yann@1365: 
yann@1365: -= END original header =-
yann@1365: 
yann@1365: diff -durN glibc-2_9.orig/string/strlen.c glibc-2_9/string/strlen.c
yann@1365: --- glibc-2_9.orig/string/strlen.c	2005-12-14 12:09:07.000000000 +0100
yann@1365: +++ glibc-2_9/string/strlen.c	2009-02-02 22:00:51.000000000 +0100
yann@1365: @@ -32,7 +32,7 @@
yann@1365:  {
yann@1365:    const char *char_ptr;
yann@1365:    const unsigned long int *longword_ptr;
yann@1365: -  unsigned long int longword, magic_bits, himagic, lomagic;
yann@1365: +  unsigned long int longword, himagic, lomagic;
yann@1365:  
yann@1365:    /* Handle the first few characters by reading one character at a time.
yann@1365:       Do this until CHAR_PTR is aligned on a longword boundary.  */
yann@1365: @@ -42,28 +42,14 @@
yann@1365:      if (*char_ptr == '\0')
yann@1365:        return char_ptr - str;
yann@1365:  
yann@1365: -  /* All these elucidatory comments refer to 4-byte longwords,
yann@1365: -     but the theory applies equally well to 8-byte longwords.  */
yann@1365: -
yann@1365:    longword_ptr = (unsigned long int *) char_ptr;
yann@1365:  
yann@1365: -  /* Bits 31, 24, 16, and 8 of this number are zero.  Call these bits
yann@1365: -     the "holes."  Note that there is a hole just to the left of
yann@1365: -     each byte, with an extra at the end:
yann@1365: -
yann@1365: -     bits:  01111110 11111110 11111110 11111111
yann@1365: -     bytes: AAAAAAAA BBBBBBBB CCCCCCCC DDDDDDDD
yann@1365: -
yann@1365: -     The 1-bits make sure that carries propagate to the next 0-bit.
yann@1365: -     The 0-bits provide holes for carries to fall into.  */
yann@1365: -  magic_bits = 0x7efefeffL;
yann@1365:    himagic = 0x80808080L;
yann@1365:    lomagic = 0x01010101L;
yann@1365:    if (sizeof (longword) > 4)
yann@1365:      {
yann@1365:        /* 64-bit version of the magic.  */
yann@1365:        /* Do the shift in two steps to avoid a warning if long has 32 bits.  */
yann@1365: -      magic_bits = ((0x7efefefeL << 16) << 16) | 0xfefefeffL;
yann@1365:        himagic = ((himagic << 16) << 16) | himagic;
yann@1365:        lomagic = ((lomagic << 16) << 16) | lomagic;
yann@1365:      }
yann@1365: @@ -75,56 +61,12 @@
yann@1365:       if *any of the four* bytes in the longword in question are zero.  */
yann@1365:    for (;;)
yann@1365:      {
yann@1365: -      /* We tentatively exit the loop if adding MAGIC_BITS to
yann@1365: -	 LONGWORD fails to change any of the hole bits of LONGWORD.
yann@1365: -
yann@1365: -	 1) Is this safe?  Will it catch all the zero bytes?
yann@1365: -	 Suppose there is a byte with all zeros.  Any carry bits
yann@1365: -	 propagating from its left will fall into the hole at its
yann@1365: -	 least significant bit and stop.  Since there will be no
yann@1365: -	 carry from its most significant bit, the LSB of the
yann@1365: -	 byte to the left will be unchanged, and the zero will be
yann@1365: -	 detected.
yann@1365: -
yann@1365: -	 2) Is this worthwhile?  Will it ignore everything except
yann@1365: -	 zero bytes?  Suppose every byte of LONGWORD has a bit set
yann@1365: -	 somewhere.  There will be a carry into bit 8.  If bit 8
yann@1365: -	 is set, this will carry into bit 16.  If bit 8 is clear,
yann@1365: -	 one of bits 9-15 must be set, so there will be a carry
yann@1365: -	 into bit 16.  Similarly, there will be a carry into bit
yann@1365: -	 24.  If one of bits 24-30 is set, there will be a carry
yann@1365: -	 into bit 31, so all of the hole bits will be changed.
yann@1365: -
yann@1365: -	 The one misfire occurs when bits 24-30 are clear and bit
yann@1365: -	 31 is set; in this case, the hole at bit 31 is not
yann@1365: -	 changed.  If we had access to the processor carry flag,
yann@1365: -	 we could close this loophole by putting the fourth hole
yann@1365: -	 at bit 32!
yann@1365: -
yann@1365: -	 So it ignores everything except 128's, when they're aligned
yann@1365: -	 properly.  */
yann@1365: -
yann@1365:        longword = *longword_ptr++;
yann@1365:  
yann@1365: -      if (
yann@1365: -#if 0
yann@1365: -	  /* Add MAGIC_BITS to LONGWORD.  */
yann@1365: -	  (((longword + magic_bits)
yann@1365: -
yann@1365: -	    /* Set those bits that were unchanged by the addition.  */
yann@1365: -	    ^ ~longword)
yann@1365: -
yann@1365: -	   /* Look at only the hole bits.  If any of the hole bits
yann@1365: -	      are unchanged, most likely one of the bytes was a
yann@1365: -	      zero.  */
yann@1365: -	   & ~magic_bits)
yann@1365: -#else
yann@1365: -	  ((longword - lomagic) & himagic)
yann@1365: -#endif
yann@1365: -	  != 0)
yann@1365: +      /* This hack taken from Alan Mycroft's HAKMEMC postings.
yann@1365: +         See: http://www.cl.cam.ac.uk/~am21/progtricks.html */
yann@1365: +      if (((longword - lomagic) & ~longword & himagic) != 0)
yann@1365:  	{
yann@1365: -	  /* Which of the bytes was the zero?  If none of them were, it was
yann@1365: -	     a misfire; continue the search.  */
yann@1365:  
yann@1365:  	  const char *cp = (const char *) (longword_ptr - 1);
yann@1365: