1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
|
/* Copyright (C) 2008-2013 Free Software Foundation, Inc.
Contributor: Joern Rennecke <joern.rennecke@embecosm.com>
on behalf of Synopsys Inc.
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 3, or (at your option) any later
version.
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
Under Section 7 of GPL version 3, you are granted additional
permissions described in the GCC Runtime Library Exception, version
3.1, as published by the Free Software Foundation.
You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
<http://www.gnu.org/licenses/>. */
/* XMAC schedule: directly back-to-back multiplies stall; the third
instruction after a multiply stalls unless it is also a multiply. */
#include "arc-ieee-754.h"
#if 0 /* DEBUG */
.global __muldf3
.balign 4
__muldf3:
push_s blink
push_s r2
push_s r3
push_s r0
bl.d __muldf3_c
push_s r1
ld_s r2,[sp,12]
ld_s r3,[sp,8]
st_s r0,[sp,12]
st_s r1,[sp,8]
pop_s r1
bl.d __muldf3_asm
pop_s r0
pop_s r3
pop_s r2
pop_s blink
cmp r0,r2
cmp.eq r1,r3
jeq_s [blink]
b abort
#define __muldf3 __muldf3_asm
#endif /* DEBUG */
/* N.B. This is optimized for ARC700.
ARC600 has very different scheduling / instruction selection criteria. */
/* For the standard multiplier, instead of mpyu rx,DBL0L,DBL1L; tst rx,rx ,
we can do:
sub rx,DBL0L,1; bic rx,DBL0L,rx; lsr rx,rx; norm rx,rx; asl.f 0,DBL1L,rx */
__muldf3_support: /* This label makes debugger output saner. */
/* If one number is denormal, subtract some from the exponent of the other
one (if the other exponent is too small, return 0), and normalize the
denormal. Then re-run the computation. */
.balign 4
FUNC(__muldf3)
.Ldenorm_dbl0:
mov_s r12,DBL0L
mov_s DBL0L,DBL1L
mov_s DBL1L,r12
mov_s r12,DBL0H
mov_s DBL0H,DBL1H
mov_s DBL1H,r12
and r11,DBL0H,r9
.Ldenorm_dbl1:
brhs r11,r9,.Linf_nan
brhs 0x3ca00001,r11,.Lret0
sub_s DBL0H,DBL0H,DBL1H
bmsk_s DBL1H,DBL1H,30
add_s DBL0H,DBL0H,DBL1H
breq_s DBL1H,0,.Ldenorm_2
norm r12,DBL1H
sub_s r12,r12,10
asl r5,r12,20
asl_s DBL1H,DBL1H,r12
sub DBL0H,DBL0H,r5
neg r5,r12
lsr r6,DBL1L,r5
asl_s DBL1L,DBL1L,r12
b.d __muldf3
add_s DBL1H,DBL1H,r6
.balign 4
.Linf_nan:
bclr r12,DBL1H,31
xor_s DBL1H,DBL1H,DBL0H
bclr_s DBL0H,DBL0H,31
max r8,DBL0H,r12 ; either NaN -> NaN ; otherwise inf
or.f 0,DBL0H,DBL0L
mov_s DBL0L,0
or.ne.f DBL1L,DBL1L,r12
not_s DBL0H,DBL0L ; inf * 0 -> NaN
mov.ne DBL0H,r8
tst_s DBL1H,DBL1H
j_s.d [blink]
bset.mi DBL0H,DBL0H,31
.Lret0: xor_s DBL0H,DBL0H,DBL1H
bclr DBL1H,DBL0H,31
xor_s DBL0H,DBL0H,DBL1H
j_s.d [blink]
mov_l DBL0L,0
.balign 4
.Ldenorm_2:
breq_s DBL1L,0,.Lret0 ; 0 input -> 0 output
norm.f r12,DBL1L
mov.mi r12,21
add.pl r12,r12,22
neg r11,r12
asl_s r12,r12,20
lsr.f DBL1H,DBL1L,r11
ror DBL1L,DBL1L,r11
sub_s DBL0H,DBL0H,r12
mov.eq DBL1H,DBL1L
sub_s DBL1L,DBL1L,DBL1H
/* Fall through. */
.global __muldf3
.balign 4
__muldf3:
ld.as r9,[pcl,0x4b] ; ((.L7ff00000-.+2)/4)]
mpyhu r4,DBL0L,DBL1L
bmsk r6,DBL0H,19
bset r6,r6,20
mpyu r7,r6,DBL1L
and r11,DBL0H,r9
breq r11,0,.Ldenorm_dbl0
mpyhu r8,r6,DBL1L
bmsk r10,DBL1H,19
bset r10,r10,20
mpyhu r5,r10,DBL0L
add.f r4,r4,r7
and r12,DBL1H,r9
mpyhu r7,r6,r10
breq r12,0,.Ldenorm_dbl1
adc.f r5,r5,r8
mpyu r8,r10,DBL0L
breq r11,r9,.Linf_nan
breq r12,r9,.Linf_nan
mpyu r6,r6,r10
add.cs r7,r7,1
add.f r4,r4,r8
mpyu r10,DBL1L,DBL0L
bclr r8,r9,30 ; 0x3ff00000
adc.f r5,r5,r6
; XMAC write-back stall / std. mult stall is one cycle later
bclr r6,r9,20 ; 0x7fe00000
add.cs r7,r7,1 ; fraction product in r7:r5:r4
tst r10,r10
bset.ne r4,r4,0 ; put least significant word into sticky bit
lsr.f r10,r7,9
add_l r12,r12,r11 ; add exponents
rsub.eq r8,r8,r9 ; 0x40000000
sub r12,r12,r8 ; subtract bias + implicit 1
brhs.d r12,r6,.Linf_denorm
rsub r10,r10,12
.Lshift_frac:
neg r8,r10
asl r6,r4,r10
lsr DBL0L,r4,r8
add.f 0,r6,r6
btst.eq DBL0L,0
cmp.eq r4,r4 ; round to nearest / round to even
asl r4,r5,r10
lsr r5,r5,r8
adc.f DBL0L,DBL0L,r4
xor.f 0,DBL0H,DBL1H
asl r7,r7,r10
add_s r12,r12,r5
adc DBL0H,r12,r7
j_s.d [blink]
bset.mi DBL0H,DBL0H,31
/* We have checked for infinity / NaN input before, and transformed
denormalized inputs into normalized inputs. Thus, the worst case
exponent overflows are:
1 + 1 - 0x400 == 0xc02 : maximum underflow
0x7fe + 0x7fe - 0x3ff == 0xbfd ; maximum overflow
N.B. 0x7e and 0x7f are also values for overflow.
If (r12 <= -54), we have an underflow to zero. */
.balign 4
.Linf_denorm:
brlo r12,0xc0000000,.Linf
asr r6,r12,20
mov_s r12,0
add.f r10,r10,r6
brgt r10,0,.Lshift_frac
beq_s .Lround_frac
add.f r10,r10,32
.Lshift32_frac:
tst r4,r4
mov r4,r5
bset.ne r4,r4,1
mov r5,r7
mov r7,0
brge r10,1,.Lshift_frac
breq r10,0,.Lround_frac
add.f r10,r10,32
brgt r10,21,.Lshift32_frac
b_s .Lret0
.Lround_frac:
add.f 0,r4,r4
btst.eq r5,0
mov_s DBL0L,r5
mov_s DBL0H,r7
adc.eq.f DBL0L,DBL0L,0
j_s.d [blink]
adc.eq DBL0H,DBL0H,0
.Linf: xor.f DBL1H,DBL1H,DBL0H
mov_s DBL0L,0
mov_s DBL0H,r9
j_s.d [blink]
bset.mi DBL0H,DBL0H,31
ENDFUNC(__muldf3)
.balign 4
.L7ff00000:
.long 0x7ff00000
|
