Bill Yi | 4e213d5 | 2015-06-23 13:53:11 -0700 | [diff] [blame] | 1 | /* Test viterbi decoder speeds */ |
| 2 | #include "config.h" |
| 3 | #include <stdio.h> |
| 4 | #include <stdlib.h> |
| 5 | #include <unistd.h> |
| 6 | #include <time.h> |
| 7 | #include <math.h> |
| 8 | #include <memory.h> |
| 9 | #include <sys/time.h> |
| 10 | #include <sys/resource.h> |
| 11 | #ifdef HAVE_GETOPT_H |
| 12 | #include <getopt.h> |
| 13 | #endif |
| 14 | #include "fec.h" |
| 15 | |
| 16 | #if HAVE_GETOPT_LONG |
| 17 | struct option Options[] = { |
| 18 | {"frame-length",1,NULL,'l'}, |
| 19 | {"frame-count",1,NULL,'n'}, |
| 20 | {"ebn0",1,NULL,'e'}, |
| 21 | {"gain",1,NULL,'g'}, |
| 22 | {"verbose",0,NULL,'v'}, |
| 23 | {"force-altivec",0,NULL,'a'}, |
| 24 | {"force-port",0,NULL,'p'}, |
| 25 | {"force-mmx",0,NULL,'m'}, |
| 26 | {"force-sse",0,NULL,'s'}, |
| 27 | {"force-sse2",0,NULL,'t'}, |
| 28 | {NULL}, |
| 29 | }; |
| 30 | #endif |
| 31 | |
| 32 | #define RATE (1./2.) |
| 33 | #define MAXBYTES 10000 |
| 34 | |
| 35 | double Gain = 32.0; |
| 36 | int Verbose = 0; |
| 37 | |
| 38 | int main(int argc,char *argv[]){ |
| 39 | int i,d,tr; |
| 40 | int sr=0,trials = 10000,errcnt,framebits=2048; |
| 41 | long long int tot_errs=0; |
| 42 | unsigned char bits[MAXBYTES]; |
| 43 | unsigned char data[MAXBYTES]; |
| 44 | unsigned char xordata[MAXBYTES]; |
| 45 | unsigned char symbols[8*2*(MAXBYTES+6)]; |
| 46 | void *vp; |
| 47 | extern char *optarg; |
| 48 | struct rusage start,finish; |
| 49 | double extime; |
| 50 | double gain,esn0,ebn0; |
| 51 | time_t t; |
| 52 | int badframes=0; |
| 53 | |
| 54 | time(&t); |
| 55 | srandom(t); |
| 56 | ebn0 = -100; |
| 57 | #if HAVE_GETOPT_LONG |
| 58 | while((d = getopt_long(argc,argv,"l:n:te:g:vapmst",Options,NULL)) != EOF){ |
| 59 | #else |
| 60 | while((d = getopt(argc,argv,"l:n:te:g:vapmst")) != EOF){ |
| 61 | #endif |
| 62 | switch(d){ |
| 63 | case 'a': |
| 64 | Cpu_mode = ALTIVEC; |
| 65 | break; |
| 66 | case 'p': |
| 67 | Cpu_mode = PORT; |
| 68 | break; |
| 69 | case 'm': |
| 70 | Cpu_mode = MMX; |
| 71 | break; |
| 72 | case 's': |
| 73 | Cpu_mode = SSE; |
| 74 | break; |
| 75 | case 't': |
| 76 | Cpu_mode = SSE2; |
| 77 | break; |
| 78 | case 'l': |
| 79 | framebits = atoi(optarg); |
| 80 | break; |
| 81 | case 'n': |
| 82 | trials = atoi(optarg); |
| 83 | break; |
| 84 | case 'e': |
| 85 | ebn0 = atof(optarg); |
| 86 | break; |
| 87 | case 'g': |
| 88 | Gain = atof(optarg); |
| 89 | break; |
| 90 | case 'v': |
| 91 | Verbose++; |
| 92 | break; |
| 93 | } |
| 94 | } |
| 95 | if(framebits > 8*MAXBYTES){ |
| 96 | fprintf(stderr,"Frame limited to %d bits\n",MAXBYTES*8); |
| 97 | framebits = MAXBYTES*8; |
| 98 | } |
| 99 | if((vp = create_viterbi27(framebits)) == NULL){ |
| 100 | printf("create_viterbi27 failed\n"); |
| 101 | exit(1); |
| 102 | } |
| 103 | if(ebn0 != -100){ |
| 104 | esn0 = ebn0 + 10*log10((double)RATE); /* Es/No in dB */ |
| 105 | /* Compute noise voltage. The 0.5 factor accounts for BPSK seeing |
| 106 | * only half the noise power, and the sqrt() converts power to |
| 107 | * voltage. |
| 108 | */ |
| 109 | gain = 1./sqrt(0.5/pow(10.,esn0/10.)); |
| 110 | |
| 111 | printf("nframes = %d framesize = %d ebn0 = %.2f dB gain = %g\n",trials,framebits,ebn0,Gain); |
| 112 | |
| 113 | for(tr=0;tr<trials;tr++){ |
| 114 | /* Encode a frame of random data */ |
| 115 | for(i=0;i<framebits+6;i++){ |
| 116 | int bit = (i < framebits) ? (random() & 1) : 0; |
| 117 | |
| 118 | sr = (sr << 1) | bit; |
| 119 | bits[i/8] = sr & 0xff; |
| 120 | symbols[2*i+0] = addnoise(parity(sr & V27POLYA),gain,Gain,127.5,255); |
| 121 | symbols[2*i+1] = addnoise(parity(sr & V27POLYB),gain,Gain,127.5,255); |
| 122 | } |
| 123 | /* Decode it and make sure we get the right answer */ |
| 124 | /* Initialize Viterbi decoder */ |
| 125 | init_viterbi27(vp,0); |
| 126 | |
| 127 | /* Decode block */ |
| 128 | update_viterbi27_blk(vp,symbols,framebits+6); |
| 129 | |
| 130 | /* Do Viterbi chainback */ |
| 131 | chainback_viterbi27(vp,data,framebits,0); |
| 132 | errcnt = 0; |
| 133 | for(i=0;i<framebits/8;i++){ |
| 134 | int e = Bitcnt[xordata[i] = data[i] ^ bits[i]]; |
| 135 | errcnt += e; |
| 136 | tot_errs += e; |
| 137 | } |
| 138 | if(errcnt != 0) |
| 139 | badframes++; |
| 140 | if(Verbose > 1 && errcnt != 0){ |
| 141 | printf("frame %d, %d errors: ",tr,errcnt); |
| 142 | for(i=0;i<framebits/8;i++){ |
| 143 | printf("%02x",xordata[i]); |
| 144 | } |
| 145 | printf("\n"); |
| 146 | } |
| 147 | if(Verbose) |
| 148 | printf("BER %lld/%lld (%10.3g) FER %d/%d (%10.3g)\r", |
| 149 | tot_errs,(long long)framebits*(tr+1),tot_errs/((double)framebits*(tr+1)), |
| 150 | badframes,tr+1,(double)badframes/(tr+1)); |
| 151 | fflush(stdout); |
| 152 | } |
| 153 | if(Verbose > 1) |
| 154 | printf("nframes = %d framesize = %d ebn0 = %.2f dB gain = %g\n",trials,framebits,ebn0,Gain); |
| 155 | else if(Verbose == 0) |
| 156 | printf("BER %lld/%lld (%.3g) FER %d/%d (%.3g)\n", |
| 157 | tot_errs,(long long)framebits*trials,tot_errs/((double)framebits*trials), |
| 158 | badframes,tr+1,(double)badframes/(tr+1)); |
| 159 | else |
| 160 | printf("\n"); |
| 161 | |
| 162 | } else { |
| 163 | /* Do time trials */ |
| 164 | memset(symbols,127,sizeof(symbols)); |
| 165 | printf("Starting time trials\n"); |
| 166 | getrusage(RUSAGE_SELF,&start); |
| 167 | for(tr=0;tr < trials;tr++){ |
| 168 | /* Initialize Viterbi decoder */ |
| 169 | init_viterbi27(vp,0); |
| 170 | |
| 171 | /* Decode block */ |
| 172 | update_viterbi27_blk(vp,symbols,framebits); |
| 173 | |
| 174 | /* Do Viterbi chainback */ |
| 175 | chainback_viterbi27(vp,data,framebits,0); |
| 176 | } |
| 177 | getrusage(RUSAGE_SELF,&finish); |
| 178 | extime = finish.ru_utime.tv_sec - start.ru_utime.tv_sec + 1e-6*(finish.ru_utime.tv_usec - start.ru_utime.tv_usec); |
| 179 | printf("Execution time for %d %d-bit frames: %.2f sec\n",trials, |
| 180 | framebits,extime); |
| 181 | printf("decoder speed: %g bits/s\n",trials*framebits/extime); |
| 182 | } |
| 183 | exit(0); |
| 184 | } |