]> sourceware.org Git - glibc.git/blob - sysdeps/ia64/memccpy.S
* sysdeps/mach/hurd/bits/libc-lock.h
[glibc.git] / sysdeps / ia64 / memccpy.S
1 /* Optimized version of the memccpy() function.
2 This file is part of the GNU C Library.
3 Copyright (C) 2000, 2001, 2003 Free Software Foundation, Inc.
4 Contributed by Dan Pop <Dan.Pop@cern.ch>.
5
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
10
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
19 02111-1307 USA. */
20
21 /* Return: a pointer to the next byte after char in dest or NULL
22
23 Inputs:
24 in0: dest
25 in1: src
26 in2: char
27 in3: byte count
28
29 This implementation assumes little endian mode (UM.be = 0).
30
31 This implementation assumes that it is safe to do read ahead
32 in the src block, without getting beyond its limit. */
33
34 #include <sysdep.h>
35 #undef ret
36
37 #define OP_T_THRES 16
38 #define OPSIZ 8
39
40 #define saved_pr r17
41 #define saved_lc r18
42 #define dest r19
43 #define src r20
44 #define len r21
45 #define asrc r22
46 #define tmp r23
47 #define char r24
48 #define charx8 r25
49 #define saved_ec r26
50 #define sh2 r28
51 #define sh1 r29
52 #define loopcnt r30
53 #define value r31
54
55 ENTRY(memccpy)
56 .prologue
57 alloc r2 = ar.pfs, 4, 40 - 4, 0, 40
58
59 #include "softpipe.h"
60 .rotr r[MEMLAT + 7], tmp1[4], tmp2[4], val[4], tmp3[2], pos0[2]
61 .rotp p[MEMLAT + 6 + 1]
62
63 mov ret0 = r0 // return NULL if no match
64 .save pr, saved_pr
65 mov saved_pr = pr // save the predicate registers
66 mov dest = in0 // dest
67 .save ar.lc, saved_lc
68 mov saved_lc = ar.lc // save the loop counter
69 mov saved_ec = ar.ec // save the loop counter
70 .body
71 mov src = in1 // src
72 extr.u char = in2, 0, 8 // char
73 mov len = in3 // len
74 sub tmp = r0, in0 // tmp = -dest
75 cmp.ne p7, p0 = r0, r0 // clear p7
76 ;;
77 and loopcnt = 7, tmp // loopcnt = -dest % 8
78 cmp.ge p6, p0 = OP_T_THRES, len // is len <= OP_T_THRES
79 mov ar.ec = 0 // ec not guaranteed zero on entry
80 (p6) br.cond.spnt .cpyfew // copy byte by byte
81 ;;
82 cmp.eq p6, p0 = loopcnt, r0
83 mux1 charx8 = char, @brcst
84 (p6) br.cond.sptk .dest_aligned
85 sub len = len, loopcnt // len -= -dest % 8
86 adds loopcnt = -1, loopcnt // --loopcnt
87 ;;
88 mov ar.lc = loopcnt
89 .l1: // copy -dest % 8 bytes
90 ld1 value = [src], 1 // value = *src++
91 ;;
92 st1 [dest] = value, 1 // *dest++ = value
93 cmp.eq p6, p0 = value, char
94 (p6) br.cond.spnt .foundit
95 br.cloop.dptk .l1
96 .dest_aligned:
97 and sh1 = 7, src // sh1 = src % 8
98 and tmp = -8, len // tmp = len & -OPSIZ
99 and asrc = -8, src // asrc = src & -OPSIZ -- align src
100 shr.u loopcnt = len, 3 // loopcnt = len / 8
101 and len = 7, len ;; // len = len % 8
102 shl sh1 = sh1, 3 // sh1 = 8 * (src % 8)
103 adds loopcnt = -1, loopcnt // --loopcnt
104 mov pr.rot = 1 << 16 ;; // set rotating predicates
105 sub sh2 = 64, sh1 // sh2 = 64 - sh1
106 mov ar.lc = loopcnt // set LC
107 cmp.eq p6, p0 = sh1, r0 // is the src aligned?
108 (p6) br.cond.sptk .src_aligned ;;
109 add src = src, tmp // src += len & -OPSIZ
110 mov ar.ec = MEMLAT + 6 + 1 // six more passes needed
111 ld8 r[1] = [asrc], 8 // r[1] = w0
112 cmp.ne p6, p0 = r0, r0 ;; // clear p6
113 .align 32
114 .l2:
115 (p[0]) ld8.s r[0] = [asrc], 8 // r[0] = w1
116 (p[MEMLAT]) shr.u tmp1[0] = r[1 + MEMLAT], sh1 // tmp1 = w0 >> sh1
117 (p[MEMLAT]) shl tmp2[0] = r[0 + MEMLAT], sh2 // tmp2 = w1 << sh2
118 (p[MEMLAT+4]) xor tmp3[0] = val[1], charx8
119 (p[MEMLAT+5]) czx1.r pos0[0] = tmp3[1]
120 (p[MEMLAT+6]) chk.s r[6 + MEMLAT], .recovery1 // our data isn't
121 // valid - rollback!
122 (p[MEMLAT+6]) cmp.ne p6, p0 = 8, pos0[1]
123 (p6) br.cond.spnt .gotit
124 (p[MEMLAT+6]) st8 [dest] = val[3], 8 // store val to dest
125 (p[MEMLAT+3]) or val[0] = tmp1[3], tmp2[3] // val = tmp1 | tmp2
126 br.ctop.sptk .l2
127 br.cond.sptk .cpyfew
128
129 .src_aligned:
130 cmp.ne p6, p0 = r0, r0 // clear p6
131 mov ar.ec = MEMLAT + 2 + 1 ;; // set EC
132 .l3:
133 (p[0]) ld8.s r[0] = [src], 8
134 (p[MEMLAT]) xor tmp3[0] = r[MEMLAT], charx8
135 (p[MEMLAT+1]) czx1.r pos0[0] = tmp3[1]
136 (p[MEMLAT+2]) cmp.ne p7, p0 = 8, pos0[1]
137 (p[MEMLAT+2]) chk.s r[MEMLAT+2], .recovery2
138 (p7) br.cond.spnt .gotit
139 .back2:
140 (p[MEMLAT+2]) st8 [dest] = r[MEMLAT+2], 8
141 br.ctop.dptk .l3
142 .cpyfew:
143 cmp.eq p6, p0 = len, r0 // is len == 0 ?
144 adds len = -1, len // --len;
145 (p6) br.cond.spnt .restore_and_exit ;;
146 mov ar.lc = len
147 .l4:
148 ld1 value = [src], 1
149 ;;
150 st1 [dest] = value, 1
151 cmp.eq p6, p0 = value, char
152 (p6) br.cond.spnt .foundit
153 br.cloop.dptk .l4 ;;
154 .foundit:
155 (p6) mov ret0 = dest
156 .restore_and_exit:
157 mov pr = saved_pr, -1 // restore the predicate registers
158 mov ar.lc = saved_lc // restore the loop counter
159 mov ar.ec = saved_ec ;; // restore the epilog counter
160 br.ret.sptk.many b0
161 .gotit:
162 .pred.rel "mutex" p6, p7
163 (p6) mov value = val[3] // if coming from l2
164 (p7) mov value = r[MEMLAT+2] // if coming from l3
165 mov ar.lc = pos0[1] ;;
166 .l5:
167 extr.u tmp = value, 0, 8 ;;
168 st1 [dest] = tmp, 1
169 shr.u value = value, 8
170 br.cloop.sptk .l5 ;;
171 mov ret0 = dest
172 mov pr = saved_pr, -1
173 mov ar.lc = saved_lc
174 br.ret.sptk.many b0
175
176 .recovery1:
177 adds src = -(MEMLAT + 6 + 1) * 8, asrc
178 mov loopcnt = ar.lc
179 mov tmp = ar.ec ;;
180 sub sh1 = (MEMLAT + 6 + 1), tmp
181 shr.u sh2 = sh2, 3
182 ;;
183 shl loopcnt = loopcnt, 3
184 sub src = src, sh2
185 shl sh1 = sh1, 3
186 shl tmp = tmp, 3
187 ;;
188 add len = len, loopcnt
189 add src = sh1, src ;;
190 add len = tmp, len
191 .back1:
192 br.cond.sptk .cpyfew
193
194 .recovery2:
195 add tmp = -(MEMLAT + 3) * 8, src
196 (p7) br.cond.spnt .gotit
197 ;;
198 ld8 r[MEMLAT+2] = [tmp] ;;
199 xor pos0[1] = r[MEMLAT+2], charx8 ;;
200 czx1.r pos0[1] = pos0[1] ;;
201 cmp.ne p7, p6 = 8, pos0[1]
202 (p7) br.cond.spnt .gotit
203 br.cond.sptk .back2
204 END(memccpy)
This page took 0.046676 seconds and 5 git commands to generate.