-
Notifications
You must be signed in to change notification settings - Fork 56
/
ssebfly27.s
205 lines (174 loc) · 4.72 KB
/
ssebfly27.s
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
/* Intel SIMD (SSE) implementation of Viterbi ACS butterflies
for 64-state (k=7) convolutional code
Copyright 2001 Phil Karn, KA9Q
This code may be used under the terms of the GNU Lesser General Public License (LGPL)
int update_viterbi27_blk_sse(struct v27 *vp,unsigned char syms[],int nbits) ;
*/
# SSE (64-bit integer SIMD) version
# Requires Pentium III or better
# These are offsets into struct v27, defined in viterbi27.h
.set DP,128
.set OLDMETRICS,132
.set NEWMETRICS,136
.text
.global update_viterbi27_blk_sse,Branchtab27_sse
.type update_viterbi27_blk_sse,@function
.align 16
update_viterbi27_blk_sse:
pushl %ebp
movl %esp,%ebp
pushl %esi
pushl %edi
pushl %edx
pushl %ebx
movl 8(%ebp),%edx # edx = vp
testl %edx,%edx
jnz 0f
movl -1,%eax
jmp err
0: movl OLDMETRICS(%edx),%esi # esi -> old metrics
movl NEWMETRICS(%edx),%edi # edi -> new metrics
movl DP(%edx),%edx # edx -> decisions
1: movl 16(%ebp),%eax # eax = nbits
decl %eax
jl 2f # passed zero, we're done
movl %eax,16(%ebp)
xorl %eax,%eax
movl 12(%ebp),%ebx # %ebx = syms
movb (%ebx),%al
movd %eax,%mm6 # mm6[0] = first symbol
movb 1(%ebx),%al
movd %eax,%mm5 # mm5[0] = second symbol
addl $2,%ebx
movl %ebx,12(%ebp)
punpcklbw %mm6,%mm6 # mm6[1] = mm6[0]
punpcklbw %mm5,%mm5
movq thirtyones,%mm7
pshufw $0,%mm6,%mm6 # copy low word to upper 3
pshufw $0,%mm5,%mm5
# mm6 now contains first symbol in each byte, mm5 the second
# each invocation of this macro does 8 butterflies in parallel
.MACRO butterfly GROUP
# compute branch metrics
movq Branchtab27_sse+(8*\GROUP),%mm4
movq Branchtab27_sse+32+(8*\GROUP),%mm3
pxor %mm6,%mm4
pxor %mm5,%mm3
pavgb %mm3,%mm4 # mm4 contains branch metrics
psrlw $3,%mm4
pand %mm7,%mm4
movq (8*\GROUP)(%esi),%mm0 # Incoming path metric, high bit = 0
movq ((8*\GROUP)+32)(%esi),%mm3 # Incoming path metric, high bit = 1
movq %mm0,%mm2
movq %mm3,%mm1
paddusb %mm4,%mm0
paddusb %mm4,%mm3
# invert branch metrics. This works only because they're 5 bits
pxor %mm7,%mm4
paddusb %mm4,%mm1
paddusb %mm4,%mm2
# Find survivors, leave in mm0,2
pminub %mm1,%mm0
pminub %mm3,%mm2
# get decisions, leave in mm1,3
pcmpeqb %mm0,%mm1
pcmpeqb %mm2,%mm3
# interleave and store new branch metrics in mm0,2
movq %mm0,%mm4
punpckhbw %mm2,%mm0 # interleave second 8 new metrics
punpcklbw %mm2,%mm4 # interleave first 8 new metrics
movq %mm0,(16*\GROUP+8)(%edi)
movq %mm4,(16*\GROUP)(%edi)
# interleave decisions, accumulate into %ebx
movq %mm1,%mm4
punpckhbw %mm3,%mm1
punpcklbw %mm3,%mm4
# Due to an error in the Intel instruction set ref (the register
# fields are swapped), gas assembles pmovmskb incorrectly
# See http://mail.gnu.org/pipermail/bug-gnu-utils/2000-August/002341.html
.byte 0x0f,0xd7,0xc1 # pmovmskb %mm1,%eax
shll $((16*\GROUP+8)&31),%eax
orl %eax,%ebx
.byte 0x0f,0xd7,0xc4 # pmovmskb %mm4,%eax
shll $((16*\GROUP)&31),%eax
orl %eax,%ebx
.endm
# invoke macro 4 times for a total of 32 butterflies
xorl %ebx,%ebx # clear decisions
butterfly GROUP=0
butterfly GROUP=1
movl %ebx,(%edx) # stash first 32 decisions
xorl %ebx,%ebx
butterfly GROUP=2
butterfly GROUP=3
movl %ebx,4(%edx) # stash second 32 decisions
addl $8,%edx # bump decision pointer
# see if we have to normalize
movl (%edi),%eax # extract first output metric
andl $255,%eax
cmpl $150,%eax # is it greater than 150?
movl $0,%eax
jle done # No, no need to normalize
# Normalize by finding smallest metric and subtracting it
# from all metrics
movq (%edi),%mm0
pminub 8(%edi),%mm0
pminub 16(%edi),%mm0
pminub 24(%edi),%mm0
pminub 32(%edi),%mm0
pminub 40(%edi),%mm0
pminub 48(%edi),%mm0
pminub 56(%edi),%mm0
# mm0 contains 8 smallest metrics
# crunch down to single lowest metric
movq %mm0,%mm1
psrlq $32,%mm0
pminub %mm1,%mm0
movq %mm0,%mm1
psrlq $16,%mm0
pminub %mm1,%mm0
movq %mm0,%mm1
psrlq $8,%mm0
pminub %mm1,%mm0
punpcklbw %mm0,%mm0 # expand to all 8 bytes
pshufw $0,%mm0,%mm0
# mm0 now contains lowest metric in all 8 bytes
# subtract it from every output metric
# Trashes %mm7
.macro PSUBUSBM REG,MEM
movq \MEM,%mm7
psubusb \REG,%mm7
movq %mm7,\MEM
.endm
PSUBUSBM %mm0,(%edi)
PSUBUSBM %mm0,8(%edi)
PSUBUSBM %mm0,16(%edi)
PSUBUSBM %mm0,24(%edi)
PSUBUSBM %mm0,32(%edi)
PSUBUSBM %mm0,40(%edi)
PSUBUSBM %mm0,48(%edi)
PSUBUSBM %mm0,56(%edi)
movd %mm0,%eax
and $0xff,%eax
done: # swap metrics
movl %esi,%eax
movl %edi,%esi
movl %eax,%edi
jmp 1b
2: emms
movl 8(%ebp),%ebx # ebx = vp
# stash metric pointers
movl %esi,OLDMETRICS(%ebx)
movl %edi,NEWMETRICS(%ebx)
movl %edx,DP(%ebx) # stash incremented value of vp->dp
xorl %eax,%eax
err: popl %ebx
popl %edx
popl %edi
popl %esi
popl %ebp
ret
.data
.align 16
thirtyones:
.byte 31,31,31,31,31,31,31,31