![[BACK]](/icons/cvsweb/back.gif){kind=icon}
Return to copyd.asm CVS log ![[TXT]](/icons/cvsweb/text.gif){kind=icon}
![[DIR]](/icons/cvsweb/dir.gif){kind=icon}
Up to [local] / OpenXM_contrib / gmp / mpn / ia64|
Return to copyd.asm CVS log | Up to [local] / OpenXM_contrib / gmp / mpn / ia64 |
1.1 ohara 1: dnl IA-64 mpn_copyd -- copy limb vector, decrementing.
2:
3: dnl Copyright 2001, 2002 Free Software Foundation, Inc.
4:
5: dnl This file is part of the GNU MP Library.
6:
7: dnl The GNU MP Library is free software; you can redistribute it and/or modify
8: dnl it under the terms of the GNU Lesser General Public License as published
9: dnl by the Free Software Foundation; either version 2.1 of the License, or (at
10: dnl your option) any later version.
11:
12: dnl The GNU MP Library is distributed in the hope that it will be useful, but
13: dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
14: dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
15: dnl License for more details.
16:
17: dnl You should have received a copy of the GNU Lesser General Public License
18: dnl along with the GNU MP Library; see the file COPYING.LIB. If not, write to
19: dnl the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
20: dnl MA 02111-1307, USA.
21:
22: include(`../config.m4')
23:
24: C INPUT PARAMETERS
25: C rp = r32
26: C sp = r33
27: C n = r34
28:
29: ASM_START()
30: PROLOGUE(mpn_copyd)
31: .prologue
32: .save ar.lc, r2
33: mov r2 = ar.lc
34: .body
35: ifdef(`HAVE_ABI_32',
36: ` addp4 r32 = 0, r32
37: addp4 r33 = 0, r33
38: sxt4 r34 = r34
39: ;;
40: ')
41: and r14 = 3, r34
42: cmp.ge p14, p15 = 3, r34
43: add r34 = -1, r34
44: ;;
45: cmp.eq p8, p9 = 1, r14
46: shladd r32 = r34, 3, r32
47: shladd r33 = r34, 3, r33
48: add r34 = -3, r34
49: cmp.eq p10, p11 = 2, r14
50: cmp.eq p12, p13 = 3, r14
51: ;;
52: (p8) br.dptk .Lb01
53: (p10) br.dptk .Lb10
54: (p12) br.dptk .Lb11
55:
56: .Lb00: C n = 4, 8, 12, ...
57: (p14) br.dptk .Ls00
58: ;;
59: ld8 r16 = [r33], -8
60: shr r15 = r34, 2
61: ;;
62: ld8 r17 = [r33], -8
63: mov ar.lc = r15
64: ;;
65: ld8 r18 = [r33], -8
66: ;;
67: ld8 r19 = [r33], -8
68: br.cloop.dptk .Loop
69: ;;
70: br.sptk .Lend
71: ;;
72:
73: .Lb01: C n = 1, 5, 9, 13, ...
74: ld8 r19 = [r33], -8
75: shr r15 = r34, 2
76: (p14) br.dptk .Ls01
77: ;;
78: ld8 r16 = [r33], -8
79: mov ar.lc = r15
80: ;;
81: ld8 r17 = [r33], -8
82: ;;
83: ld8 r18 = [r33], -8
84: br.sptk .Li01
85: ;;
86:
87: .Lb10: C n = 2,6, 10, 14, ...
88: ld8 r18 = [r33], -8
89: shr r15 = r34, 2
90: ;;
91: ld8 r19 = [r33], -8
92: mov ar.lc = r15
93: (p14) br.dptk .Ls10
94: ;;
95: ld8 r16 = [r33], -8
96: ;;
97: ld8 r17 = [r33], -8
98: br.sptk .Li10
99: ;;
100:
101: .Lb11: C n = 3, 7, 11, 15, ...
102: ld8 r17 = [r33], -8
103: shr r15 = r34, 2
104: ;;
105: ld8 r18 = [r33], -8
106: mov ar.lc = r15
107: ;;
108: ld8 r19 = [r33], -8
109: (p14) br.dptk .Ls11
110: ;;
111: ld8 r16 = [r33], -8
112: br.sptk .Li11
113: ;;
114:
115: .Loop:
116: .Li00:
117: { .mmb; st8 [r32] = r16, -8
118: ld8 r16 = [r33], -8
119: ;;
120: }
121: .Li11:
122: { .mmb; st8 [r32] = r17, -8
123: ld8 r17 = [r33], -8
124: ;;
125: }
126: .Li10:
127: { .mmb; st8 [r32] = r18, -8
128: ld8 r18 = [r33], -8
129: ;;
130: }
131: .Li01:
132: { .mmb; st8 [r32] = r19, -8
133: ld8 r19 = [r33], -8
134: br.cloop.dptk .Loop
135: ;;
136: }
137: .Lend: st8 [r32] = r16, -8
138: ;;
139: .Ls11: st8 [r32] = r17, -8
140: ;;
141: .Ls10: st8 [r32] = r18, -8
142: ;;
143: .Ls01: st8 [r32] = r19, -8
144: .Ls00:
145: mov ar.lc = r2
146: br.ret.sptk.many rp
147: EPILOGUE(mpn_copyd)
148: ASM_END()