1 # Optimized strlen implementation for PowerPC.
2 # Copyright (C) 1997 Free Software Foundation, Inc.
3 # This file is part of the GNU C Library.
5 # The GNU C Library is free software; you can redistribute it and/or
6 # modify it under the terms of the GNU Library General Public License as
7 # published by the Free Software Foundation; either version 2 of the
8 # License, or (at your option) any later version.
10 # The GNU C Library is distributed in the hope that it will be useful,
11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 # Library General Public License for more details.
15 # You should have received a copy of the GNU Library General Public
16 # License along with the GNU C Library; see the file COPYING.LIB. If not,
17 # write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
18 # Boston, MA 02111-1307, USA.
20 # The algorithm here uses the following techniques:
22 # 1) Given a word 'x', we can test to see if it contains any 0 bytes
23 # by subtracting 0x01010101, and seeing if any of the high bits of each
24 # byte changed from 0 to 1. This works because the least significant
25 # 0 byte must have had no incoming carry (otherwise it's not the least
26 # significant), so it is 0x00 - 0x01 == 0xff. For all other
27 # byte values, either they have the high bit set initially, or when
28 # 1 is subtracted you get a value in the range 0x00-0x7f, none of which
29 # have their high bit set. The expression here is
30 # (x + 0xfefefeff) & ~(x | 0x7f7f7f7f), which gives 0x00000000 when
31 # there were no 0x00 bytes in the word.
33 # 2) Given a word 'x', we can test to see _which_ byte was zero by
34 # calculating ~(((x & 0x7f7f7f7f) + 0x7f7f7f7f) | x | 0x7f7f7f7f).
35 # This produces 0x80 in each byte that was zero, and 0x00 in all
36 # the other bytes. The '| 0x7f7f7f7f' clears the low 7 bits in each
37 # byte, and the '| x' part ensures that bytes with the high bit set
38 # produce 0x00. The addition will carry into the high bit of each byte
39 # iff that byte had one of its low 7 bits set. We can then just see
40 # which was the most significant bit set and divide by 8 to find how
41 # many to add to the index.
42 # This is from the book 'The PowerPC Compiler Writer's Guide',
43 # by Steve Hoxey, Faraydon Karim, Bill Hay and Hank Warren.
45 # We deal with strings not aligned to a word boundary by taking the
46 # first word and ensuring that bytes not part of the string
47 # are treated as nonzero. To allow for memory latency, we unroll the
48 # loop a few times, being careful to ensure that we do not read ahead
49 # across cache line boundaries.
51 # Questions to answer:
52 # 1) How long are strings passed to strlen? If they're often really long,
53 # we should probably use cache management instructions and/or unroll the
54 # loop more. If they're often quite short, it might be better to use
55 # fact (2) in the inner loop than have to recalculate it.
56 # 2) How popular are bytes with the high bit set? If they are very rare,
57 # on some processors it might be useful to use the simpler expression
58 # ~((x - 0x01010101) | 0x7f7f7f7f) (that is, on processors with only one
59 # ALU), but this fails when any character has its high bit set.
61 # Some notes on register usage: Under the SVR4 ABI, we can use registers
62 # 0 and 3 through 12 (so long as we don't call any procedures) without
63 # saving them. We can also use registers 14 through 31 if we save them.
64 # We can't use r1 (it's the stack pointer), r2 nor r13 because the user
65 # program may expect them to hold their usual value if we get sent
66 # a signal. Integer parameters are passed in r3 through r10.
67 # We can use condition registers cr0, cr1, cr5, cr6, and cr7 without saving
68 # them, the others we must save.
73 .type strlen,@function
75 # On entry, r3 points to the string, and it's left that way.
76 # We use r6 to store 0xfefefeff, and r7 to store 0x7f7f7f7f.
77 # r4 is used to keep the current index into the string; r5 holds
78 # the number of padding bits we prepend to the string to make it
79 # start at a word boundary. r8 holds the 'current' word.
80 # r9-12 are temporaries. r0 is used as a temporary and for discarded
84 rlwinm %r5,%r3,3,27,28
88 # That's the setup done, now do the first pair of words.
89 # We make an exception and use method (2) on the first two words, to reduce
101 # Are we now aligned to a doubleword boundary?
104 # Handle second word of pair.
130 done1: and %r0,%r7,%r8
135 # When we get to here, r4 points to the first word in the string that
136 # contains a zero byte, and the most significant set bit in r8 is in that
138 done0: cntlzw %r11,%r8
144 .size strlen,0b-strlen