1 /* K=7 r=1/2 Viterbi decoder for PowerPC G4/G5 Altivec instructions
2  * Feb 2004, Phil Karn, KA9Q
3  */
4 #include <stdio.h>
5 #include <memory.h>
6 #include <stdlib.h>
7 #include "fec.h"
8 
9 typedef union { long long p; unsigned char c[64]; vector bool char v[4]; } decision_t;
10 typedef union { long long p; unsigned char c[64]; vector unsigned char v[4]; } metric_t;
11 
12 static union branchtab27 { unsigned char c[32]; vector unsigned char v[2];} Branchtab27[2];
13 static int Init = 0;
14 
15 /* State info for instance of Viterbi decoder
16  * Don't change this without also changing references in [mmx|sse|sse2]bfly29.s!
17  */
18 struct v27 {
19   metric_t metrics1; /* path metric buffer 1 */
20   metric_t metrics2; /* path metric buffer 2 */
21   decision_t *dp;          /* Pointer to current decision */
22   metric_t *old_metrics,*new_metrics; /* Pointers to path metrics, swapped on every bit */
23   decision_t *decisions;   /* Beginning of decisions for block */
24 };
25 
26 /* Initialize Viterbi decoder for start of new frame */
init_viterbi27_av(void * p,int starting_state)27 int init_viterbi27_av(void *p,int starting_state){
28   struct v27 *vp = p;
29   int i;
30 
31   if(p == NULL)
32     return -1;
33   for(i=0;i<4;i++)
34     vp->metrics1.v[i] = (vector unsigned char)(63);
35   vp->old_metrics = &vp->metrics1;
36   vp->new_metrics = &vp->metrics2;
37   vp->dp = vp->decisions;
38   vp->old_metrics->c[starting_state & 63] = 0; /* Bias known start state */
39   return 0;
40 }
41 
set_viterbi27_polynomial_av(int polys[2])42 void set_viterbi27_polynomial_av(int polys[2]){
43   int state;
44 
45   for(state=0;state < 32;state++){
46     Branchtab27[0].c[state] = (polys[0] < 0) ^ parity((2*state) & abs(polys[0])) ? 255 : 0;
47     Branchtab27[1].c[state] = (polys[1] < 0) ^ parity((2*state) & abs(polys[1])) ? 255 : 0;
48   }
49   Init++;
50 }
51 
52 /* Create a new instance of a Viterbi decoder */
create_viterbi27_av(int len)53 void *create_viterbi27_av(int len){
54   struct v27 *vp;
55 
56   if(!Init){
57     int polys[2] = { V27POLYA,V27POLYB };
58     set_viterbi27_polynomial_av(polys);
59   }
60   if((vp = (struct v27 *)malloc(sizeof(struct v27))) == NULL)
61     return NULL;
62   if((vp->decisions = (decision_t *)malloc((len+6)*sizeof(decision_t))) == NULL){
63     free(vp);
64     return NULL;
65   }
66   init_viterbi27_av(vp,0);
67   return vp;
68 }
69 
70 /* Viterbi chainback */
chainback_viterbi27_av(void * p,unsigned char * data,unsigned int nbits,unsigned int endstate)71 int chainback_viterbi27_av(
72       void *p,
73       unsigned char *data, /* Decoded output data */
74       unsigned int nbits, /* Number of data bits */
75       unsigned int endstate){ /* Terminal encoder state */
76   struct v27 *vp = p;
77   decision_t *d = (decision_t *)vp->decisions;
78 
79   if(p == NULL)
80     return -1;
81 
82   /* Make room beyond the end of the encoder register so we can
83    * accumulate a full byte of decoded data
84    */
85   endstate %= 64;
86   endstate <<= 2;
87 
88   /* The store into data[] only needs to be done every 8 bits.
89    * But this avoids a conditional branch, and the writes will
90    * combine in the cache anyway
91    */
92   d += 6; /* Look past tail */
93   while(nbits-- != 0){
94     int k;
95 
96     k = d[nbits].c[endstate>>2] & 1;
97     data[nbits>>3] = endstate = (endstate >> 1) | (k << 7);
98   }
99   return 0;
100 }
101 
102 /* Delete instance of a Viterbi decoder */
delete_viterbi27_av(void * p)103 void delete_viterbi27_av(void *p){
104   struct v27 *vp = p;
105 
106   if(vp != NULL){
107     free(vp->decisions);
108     free(vp);
109   }
110 }
111 
112 /* Process received symbols */
update_viterbi27_blk_av(void * p,unsigned char * syms,int nbits)113 int update_viterbi27_blk_av(void *p,unsigned char *syms,int nbits){
114   struct v27 *vp = p;
115   decision_t *d;
116 
117   if(p == NULL)
118     return -1;
119   d = (decision_t *)vp->dp;
120   while(nbits--){
121     vector unsigned char survivor0,survivor1,sym0v,sym1v;
122     vector bool char decision0,decision1;
123     vector unsigned char metric,m_metric,m0,m1,m2,m3;
124     void *tmp;
125 
126     /* sym0v.0 = syms[0]; sym0v.1 = syms[1] */
127     sym0v = vec_perm(vec_ld(0,syms),vec_ld(1,syms),vec_lvsl(0,syms));
128 
129     sym1v = vec_splat(sym0v,1); /* Splat syms[1] across sym1v */
130     sym0v = vec_splat(sym0v,0); /* Splat syms[0] across sym0v */
131     syms += 2;
132 
133     /* Do the 32 butterflies as two interleaved groups of 16 each to keep the pipes full */
134 
135     /* Form first set of 16 branch metrics */
136     metric = vec_avg(vec_xor(Branchtab27[0].v[0],sym0v),vec_xor(Branchtab27[1].v[0],sym1v));
137     metric = vec_sr(metric,(vector unsigned char)(3));
138     m_metric = vec_sub((vector unsigned char)(31),metric);
139 
140     /* Form first set of path metrics */
141     m0 = vec_adds(vp->old_metrics->v[0],metric);
142     m3 = vec_adds(vp->old_metrics->v[2],metric);
143     m1 = vec_adds(vp->old_metrics->v[2],m_metric);
144     m2 = vec_adds(vp->old_metrics->v[0],m_metric);
145 
146     /* Form second set of 16 branch metrics */
147     metric = vec_avg(vec_xor(Branchtab27[0].v[1],sym0v),vec_xor(Branchtab27[1].v[1],sym1v));
148     metric = vec_sr(metric,(vector unsigned char)(3));
149     m_metric = vec_sub((vector unsigned char)(31),metric);
150 
151     /* Compare and select first set */
152     decision0 = vec_cmpgt(m0,m1);
153     decision1 = vec_cmpgt(m2,m3);
154     survivor0 = vec_min(m0,m1);
155     survivor1 = vec_min(m2,m3);
156 
157     /* Compute second set of path metrics */
158     m0 = vec_adds(vp->old_metrics->v[1],metric);
159     m3 = vec_adds(vp->old_metrics->v[3],metric);
160     m1 = vec_adds(vp->old_metrics->v[3],m_metric);
161     m2 = vec_adds(vp->old_metrics->v[1],m_metric);
162 
163     /* Interleave and store first decisions and survivors */
164     d->v[0] = vec_mergeh(decision0,decision1);
165     d->v[1] = vec_mergel(decision0,decision1);
166     vp->new_metrics->v[0] = vec_mergeh(survivor0,survivor1);
167     vp->new_metrics->v[1] = vec_mergel(survivor0,survivor1);
168 
169     /* Compare and select second set */
170     decision0 = vec_cmpgt(m0,m1);
171     decision1 = vec_cmpgt(m2,m3);
172     survivor0 = vec_min(m0,m1);
173     survivor1 = vec_min(m2,m3);
174 
175     /* Interleave and store second set of decisions and survivors */
176     d->v[2] = vec_mergeh(decision0,decision1);
177     d->v[3] = vec_mergel(decision0,decision1);
178     vp->new_metrics->v[2] = vec_mergeh(survivor0,survivor1);
179     vp->new_metrics->v[3] = vec_mergel(survivor0,survivor1);
180 
181     /* renormalize if necessary */
182     if(vp->new_metrics->c[0] >= 105){
183       vector unsigned char scale0,scale1;
184 
185       /* Find smallest metric and splat */
186       scale0 = vec_min(vp->new_metrics->v[0],vp->new_metrics->v[1]);
187       scale1 = vec_min(vp->new_metrics->v[2],vp->new_metrics->v[3]);
188       scale0 = vec_min(scale0,scale1);
189       scale0 = vec_min(scale0,vec_sld(scale0,scale0,8));
190       scale0 = vec_min(scale0,vec_sld(scale0,scale0,4));
191       scale0 = vec_min(scale0,vec_sld(scale0,scale0,2));
192       scale0 = vec_min(scale0,vec_sld(scale0,scale0,1));
193 
194       /* Now subtract from all metrics */
195       vp->new_metrics->v[0] = vec_subs(vp->new_metrics->v[0],scale0);
196       vp->new_metrics->v[1] = vec_subs(vp->new_metrics->v[1],scale0);
197       vp->new_metrics->v[2] = vec_subs(vp->new_metrics->v[2],scale0);
198       vp->new_metrics->v[3] = vec_subs(vp->new_metrics->v[3],scale0);
199     }
200     d++;
201     /* Swap pointers to old and new metrics */
202     tmp = vp->old_metrics;
203     vp->old_metrics = vp->new_metrics;
204     vp->new_metrics = tmp;
205   }
206   vp->dp = d;
207 
208   return 0;
209 }
210 
211