OpenCores
URL https://opencores.org/ocsvn/or1k/or1k/trunk

Subversion Repositories or1k

[/] [or1k/] [trunk/] [linux/] [linux-2.4/] [arch/] [alpha/] [lib/] [memchr.S] - Blame information for rev 1765

Details | Compare with Previous | View Log

Line No. Rev Author Line
1 1275 phoenix
/* Copyright (C) 1996 Free Software Foundation, Inc.
2
   This file is part of the GNU C Library.
3
   Contributed by David Mosberger (davidm@cs.arizona.edu).
4
 
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.
9
 
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.
14
 
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.  */
19
 
20
/* Finds characters in a memory area.  Optimized for the Alpha:
21
 
22
      - memory accessed as aligned quadwords only
23
      - uses cmpbge to compare 8 bytes in parallel
24
      - does binary search to find 0 byte in last
25
        quadword (HAKMEM needed 12 instructions to
26
        do this instead of the 9 instructions that
27
        binary search needs).
28
 
29
For correctness consider that:
30
 
31
      - only minimum number of quadwords may be accessed
32
      - the third argument is an unsigned long
33
*/
34
 
35
        .set noreorder
36
        .set noat
37
 
38
        .globl memchr
39
        .ent memchr
40
memchr:
41
        .frame $30,0,$26,0
42
        .prologue 0
43
 
44
        # Hack -- if someone passes in (size_t)-1, hoping to just
45
        # search til the end of the address space, we will overflow
46
        # below when we find the address of the last byte.  Given
47
        # that we will never have a 56-bit address space, cropping
48
        # the length is the easiest way to avoid trouble.
49
        zap     $18, 0x80, $5   #-e0    :
50
 
51
        beq     $18, $not_found # .. e1 :
52
        ldq_u   $1, 0($16)      # e1    : load first quadword
53
        insbl   $17, 1, $2      # .. e0 : $2 = 000000000000ch00
54
        and     $17, 0xff, $17  #-e0    : $17 = 00000000000000ch
55
        cmpult  $18, 9, $4      # .. e1 :
56
        or      $2, $17, $17    # e0    : $17 = 000000000000chch
57
        lda     $3, -1($31)     # .. e1 :
58
        sll     $17, 16, $2     #-e0    : $2 = 00000000chch0000
59
        addq    $16, $5, $5     # .. e1 :
60
        or      $2, $17, $17    # e1    : $17 = 00000000chchchch
61
        unop                    #       :
62
        sll     $17, 32, $2     #-e0    : $2 = chchchch00000000
63
        or      $2, $17, $17    # e1    : $17 = chchchchchchchch
64
        extql   $1, $16, $7     # e0    :
65
        beq     $4, $first_quad # .. e1 :
66
 
67
        ldq_u   $6, -1($5)      #-e1    : eight or less bytes to search
68
        extqh   $6, $16, $6     # .. e0 :
69
        mov     $16, $0         # e0    :
70
        or      $7, $6, $1      # .. e1 : $1 = quadword starting at $16
71
 
72
        # Deal with the case where at most 8 bytes remain to be searched
73
        # in $1.  E.g.:
74
        #       $18 = 6
75
        #       $1 = ????c6c5c4c3c2c1
76
$last_quad:
77
        negq    $18, $6         #-e0    :
78
        xor     $17, $1, $1     # .. e1 :
79
        srl     $3, $6, $6      # e0    : $6 = mask of $18 bits set
80
        cmpbge  $31, $1, $2     # .. e1 :
81
        and     $2, $6, $2      #-e0    :
82
        beq     $2, $not_found  # .. e1 :
83
 
84
$found_it:
85
        # Now, determine which byte matched:
86
        negq    $2, $3          # e0    :
87
        and     $2, $3, $2      # e1    :
88
 
89
        and     $2, 0x0f, $1    #-e0    :
90
        addq    $0, 4, $3       # .. e1 :
91
        cmoveq  $1, $3, $0      # e0    :
92
 
93
        addq    $0, 2, $3       # .. e1 :
94
        and     $2, 0x33, $1    #-e0    :
95
        cmoveq  $1, $3, $0      # .. e1 :
96
 
97
        and     $2, 0x55, $1    # e0    :
98
        addq    $0, 1, $3       # .. e1 :
99
        cmoveq  $1, $3, $0      #-e0    :
100
 
101
$done:  ret                     # .. e1 :
102
 
103
        # Deal with the case where $18 > 8 bytes remain to be
104
        # searched.  $16 may not be aligned.
105
        .align 4
106
$first_quad:
107
        andnot  $16, 0x7, $0    #-e1    :
108
        insqh   $3, $16, $2     # .. e0 : $2 = 0000ffffffffffff ($16<0:2> ff)
109
        xor     $1, $17, $1     # e0    :
110
        or      $1, $2, $1      # e1    : $1 = ====ffffffffffff
111
        cmpbge  $31, $1, $2     #-e0    :
112
        bne     $2, $found_it   # .. e1 :
113
 
114
        # At least one byte left to process.
115
 
116
        ldq     $1, 8($0)       # e0    :
117
        subq    $5, 1, $18      # .. e1 :
118
        addq    $0, 8, $0       #-e0    :
119
 
120
        # Make $18 point to last quad to be accessed (the
121
        # last quad may or may not be partial).
122
 
123
        andnot  $18, 0x7, $18   # .. e1 :
124
        cmpult  $0, $18, $2     # e0    :
125
        beq     $2, $final      # .. e1 :
126
 
127
        # At least two quads remain to be accessed.
128
 
129
        subq    $18, $0, $4     #-e0    : $4 <- nr quads to be processed
130
        and     $4, 8, $4       # e1    : odd number of quads?
131
        bne     $4, $odd_quad_count # e1 :
132
 
133
        # At least three quads remain to be accessed
134
 
135
        mov     $1, $4          # e0    : move prefetched value to correct reg
136
 
137
        .align  4
138
$unrolled_loop:
139
        ldq     $1, 8($0)       #-e0    : prefetch $1
140
        xor     $17, $4, $2     # .. e1 :
141
        cmpbge  $31, $2, $2     # e0    :
142
        bne     $2, $found_it   # .. e1 :
143
 
144
        addq    $0, 8, $0       #-e0    :
145
$odd_quad_count:
146
        xor     $17, $1, $2     # .. e1 :
147
        ldq     $4, 8($0)       # e0    : prefetch $4
148
        cmpbge  $31, $2, $2     # .. e1 :
149
        addq    $0, 8, $6       #-e0    :
150
        bne     $2, $found_it   # .. e1 :
151
 
152
        cmpult  $6, $18, $6     # e0    :
153
        addq    $0, 8, $0       # .. e1 :
154
        bne     $6, $unrolled_loop #-e1 :
155
 
156
        mov     $4, $1          # e0    : move prefetched value into $1
157
$final: subq    $5, $0, $18     # .. e1 : $18 <- number of bytes left to do
158
        bne     $18, $last_quad # e1    :
159
 
160
$not_found:
161
        mov     $31, $0         #-e0    :
162
        ret                     # .. e1 :
163
 
164
        .end memchr

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.