Annotation of OpenXM_contrib/gmp/mpn/alpha/rshift.s, Revision 1.1.1.1
1.1 maekawa 1: # Alpha 21064 __mpn_rshift --
2:
3: # Copyright (C) 1994, 1995 Free Software Foundation, Inc.
4:
5: # This file is part of the GNU MP Library.
6:
7: # The GNU MP Library is free software; you can redistribute it and/or modify
8: # it under the terms of the GNU Library General Public License as published by
9: # the Free Software Foundation; either version 2 of the License, or (at your
10: # option) any later version.
11:
12: # The GNU MP Library is distributed in the hope that it will be useful, but
13: # WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
14: # or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
15: # License for more details.
16:
17: # You should have received a copy of the GNU Library General Public License
18: # along with the GNU MP Library; see the file COPYING.LIB. If not, write to
19: # the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
20: # MA 02111-1307, USA.
21:
22:
23: # INPUT PARAMETERS
24: # res_ptr r16
25: # s1_ptr r17
26: # size r18
27: # cnt r19
28:
29: # This code runs at 4.8 cycles/limb on the 21064. With infinite unrolling,
30: # it would take 4 cycles/limb. It should be possible to get down to 3
31: # cycles/limb since both ldq and stq can be paired with the other used
32: # instructions. But there are many restrictions in the 21064 pipeline that
33: # makes it hard, if not impossible, to get down to 3 cycles/limb:
34:
35: # 1. ldq has a 3 cycle delay, srl and sll have a 2 cycle delay.
36: # 2. Only aligned instruction pairs can be paired.
37: # 3. The store buffer or silo might not be able to deal with the bandwidth.
38:
39: .set noreorder
40: .set noat
41: .text
42: .align 3
43: .globl __mpn_rshift
44: .ent __mpn_rshift
45: __mpn_rshift:
46: .frame $30,0,$26,0
47:
48: ldq $4,0($17) # load first limb
49: addq $17,8,$17
50: subq $31,$19,$7
51: subq $18,1,$18
52: and $18,4-1,$20 # number of limbs in first loop
53: sll $4,$7,$0 # compute function result
54:
55: beq $20,.L0
56: subq $18,$20,$18
57:
58: .align 3
59: .Loop0:
60: ldq $3,0($17)
61: addq $16,8,$16
62: addq $17,8,$17
63: subq $20,1,$20
64: srl $4,$19,$5
65: sll $3,$7,$6
66: bis $3,$3,$4
67: bis $5,$6,$8
68: stq $8,-8($16)
69: bne $20,.Loop0
70:
71: .L0: beq $18,.Lend
72:
73: .align 3
74: .Loop: ldq $3,0($17)
75: addq $16,32,$16
76: subq $18,4,$18
77: srl $4,$19,$5
78: sll $3,$7,$6
79:
80: ldq $4,8($17)
81: srl $3,$19,$1
82: bis $5,$6,$8
83: stq $8,-32($16)
84: sll $4,$7,$2
85:
86: ldq $3,16($17)
87: srl $4,$19,$5
88: bis $1,$2,$8
89: stq $8,-24($16)
90: sll $3,$7,$6
91:
92: ldq $4,24($17)
93: srl $3,$19,$1
94: bis $5,$6,$8
95: stq $8,-16($16)
96: sll $4,$7,$2
97:
98: addq $17,32,$17
99: bis $1,$2,$8
100: stq $8,-8($16)
101:
102: bgt $18,.Loop
103:
104: .Lend: srl $4,$19,$8
105: stq $8,0($16)
106: ret $31,($26),1
107: .end __mpn_rshift
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>