1 /* Optimized version of the memccpy() function.
2 This file is part of the GNU C Library.
3 Copyright (C) 2000,2001,2003,2006 Free Software Foundation, Inc.
4 Contributed by Dan Pop <Dan.Pop@cern.ch>.
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.
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.
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
21 /* Return: a pointer to the next byte after char in dest or NULL
29 This implementation assumes little endian mode (UM.be = 0).
31 This implementation assumes that it is safe to do read ahead
32 in the src block, without getting beyond its limit. */
55 #ifdef GAS_ALIGN_BREAKS_UNWIND_INFO
56 /* Manually force proper loop-alignment. Note: be sure to
57 double-check the code-layout after making any changes to
59 # define ALIGN(n) { nop 0 }
61 # define ALIGN(n) .align n
66 alloc r2 = ar.pfs, 4, 40 - 4, 0, 40
69 .rotr r[MEMLAT + 7], tmp1[4], tmp2[4], val[4], tmp3[2], pos0[2]
70 .rotp p[MEMLAT + 6 + 1]
72 mov ret0 = r0 // return NULL if no match
74 mov saved_pr = pr // save the predicate registers
75 mov dest = in0 // dest
77 mov saved_lc = ar.lc // save the loop counter
78 mov saved_ec = ar.ec // save the loop counter
81 extr.u char = in2, 0, 8 // char
83 sub tmp = r0, in0 // tmp = -dest
84 cmp.ne p7, p0 = r0, r0 // clear p7
86 and loopcnt = 7, tmp // loopcnt = -dest % 8
87 cmp.ge p6, p0 = OP_T_THRES, len // is len <= OP_T_THRES
88 mov ar.ec = 0 // ec not guaranteed zero on entry
89 (p6) br.cond.spnt .cpyfew // copy byte by byte
91 cmp.eq p6, p0 = loopcnt, r0
92 mux1 charx8 = char, @brcst
93 (p6) br.cond.sptk .dest_aligned
94 sub len = len, loopcnt // len -= -dest % 8
95 adds loopcnt = -1, loopcnt // --loopcnt
98 .l1: // copy -dest % 8 bytes
99 ld1 value = [src], 1 // value = *src++
101 st1 [dest] = value, 1 // *dest++ = value
102 cmp.eq p6, p0 = value, char
103 (p6) br.cond.spnt .foundit
106 and sh1 = 7, src // sh1 = src % 8
107 and tmp = -8, len // tmp = len & -OPSIZ
108 and asrc = -8, src // asrc = src & -OPSIZ -- align src
109 shr.u loopcnt = len, 3 // loopcnt = len / 8
110 and len = 7, len ;; // len = len % 8
111 shl sh1 = sh1, 3 // sh1 = 8 * (src % 8)
112 adds loopcnt = -1, loopcnt // --loopcnt
113 mov pr.rot = 1 << 16 ;; // set rotating predicates
114 sub sh2 = 64, sh1 // sh2 = 64 - sh1
115 mov ar.lc = loopcnt // set LC
116 cmp.eq p6, p0 = sh1, r0 // is the src aligned?
117 (p6) br.cond.sptk .src_aligned ;;
118 add src = src, tmp // src += len & -OPSIZ
119 mov ar.ec = MEMLAT + 6 + 1 // six more passes needed
120 ld8 r[1] = [asrc], 8 // r[1] = w0
121 cmp.ne p6, p0 = r0, r0 ;; // clear p6
124 (p[0]) ld8.s r[0] = [asrc], 8 // r[0] = w1
125 (p[MEMLAT]) shr.u tmp1[0] = r[1 + MEMLAT], sh1 // tmp1 = w0 >> sh1
126 (p[MEMLAT]) shl tmp2[0] = r[0 + MEMLAT], sh2 // tmp2 = w1 << sh2
127 (p[MEMLAT+4]) xor tmp3[0] = val[1], charx8
128 (p[MEMLAT+5]) czx1.r pos0[0] = tmp3[1]
129 (p[MEMLAT+6]) chk.s r[6 + MEMLAT], .recovery1 // our data isn't
131 (p[MEMLAT+6]) cmp.ne p6, p0 = 8, pos0[1]
132 (p6) br.cond.spnt .gotit
133 (p[MEMLAT+6]) st8 [dest] = val[3], 8 // store val to dest
134 (p[MEMLAT+3]) or val[0] = tmp1[3], tmp2[3] // val = tmp1 | tmp2
139 cmp.ne p6, p0 = r0, r0 // clear p6
140 mov ar.ec = MEMLAT + 2 + 1 ;; // set EC
142 (p[0]) ld8.s r[0] = [src], 8
143 (p[MEMLAT]) xor tmp3[0] = r[MEMLAT], charx8
144 (p[MEMLAT+1]) czx1.r pos0[0] = tmp3[1]
145 (p[MEMLAT+2]) cmp.ne p7, p0 = 8, pos0[1]
146 (p[MEMLAT+2]) chk.s r[MEMLAT+2], .recovery2
147 (p7) br.cond.spnt .gotit
149 (p[MEMLAT+2]) st8 [dest] = r[MEMLAT+2], 8
152 cmp.eq p6, p0 = len, r0 // is len == 0 ?
153 adds len = -1, len // --len;
154 (p6) br.cond.spnt .restore_and_exit ;;
159 st1 [dest] = value, 1
160 cmp.eq p6, p0 = value, char
161 (p6) br.cond.spnt .foundit
166 mov pr = saved_pr, -1 // restore the predicate registers
167 mov ar.lc = saved_lc // restore the loop counter
168 mov ar.ec = saved_ec ;; // restore the epilog counter
171 .pred.rel "mutex" p6, p7
172 (p6) mov value = val[3] // if coming from l2
173 (p7) mov value = r[MEMLAT+2] // if coming from l3
174 mov ar.lc = pos0[1] ;;
176 extr.u tmp = value, 0, 8 ;;
178 shr.u value = value, 8
181 mov pr = saved_pr, -1
187 # error "MEMLAT must be 6!"
193 (p[0]) adds src = -8, src
195 (p[1]) adds src = -8, src
196 sub sh1 = (MEMLAT + 6 + 1), tmp
198 (p[2]) adds src = -8, src
200 (p[3]) adds src = -8, src
201 shl loopcnt = loopcnt, 3
203 (p[4]) adds src = -8, src
205 (p[5]) adds src = -8, src
208 (p[6]) adds src = -8, src
210 (p[7]) adds src = -8, src
213 (p[8]) adds src = -8, src
215 (p[9]) adds src = -8, src
218 (p[10]) adds src = -8, src
220 (p[11]) adds src = -8, src
221 add len = len, loopcnt
231 # error "MEMLAT must be 6!"
234 (p7) br.cond.spnt .gotit
236 (p[0]) add tmp = -8, tmp ;;
237 (p[1]) add tmp = -8, tmp ;;
238 (p[2]) add tmp = -8, tmp ;;
239 (p[3]) add tmp = -8, tmp ;;
240 (p[4]) add tmp = -8, tmp ;;
241 (p[5]) add tmp = -8, tmp ;;
242 (p[6]) add tmp = -8, tmp ;;
243 (p[7]) add tmp = -8, tmp ;;
244 ld8 r[MEMLAT+2] = [tmp] ;;
245 xor pos0[1] = r[MEMLAT+2], charx8 ;;
246 czx1.r pos0[1] = pos0[1] ;;
247 cmp.ne p7, p6 = 8, pos0[1]
248 (p7) br.cond.spnt .gotit