interpretor/preprocessor/faust-0.9.47mr3/architecture/bench.cpp

   1 /************************************************************************
   2
   3         IMPORTANT NOTE : this file contains two clearly delimited sections :
   4         the ARCHITECTURE section (in two parts) and the USER section. Each section
   5         is governed by its own copyright and license. Please check individually
   6         each section for license and copyright information.
   7 *************************************************************************/
   8
   9 /*******************BEGIN ARCHITECTURE SECTION (part 1/2)****************/
  10
  11 /************************************************************************
  12     FAUST Architecture File
  13         Copyright (C) 2003-2011 GRAME, Centre National de Creation Musicale
  14     ---------------------------------------------------------------------
  15     This Architecture section is free software; you can redistribute it
  16     and/or modify it under the terms of the GNU General Public License
  17         as published by the Free Software Foundation; either version 3 of
  18         the License, or (at your option) any later version.
  19
  20     This program is distributed in the hope that it will be useful,
  21     but WITHOUT ANY WARRANTY; without even the implied warranty of
  22     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  23     GNU General Public License for more details.
  24
  25     You should have received a copy of the GNU General Public License
  26         along with this program; If not, see <http://www.gnu.org/licenses/>.
  27
  28         EXCEPTION : As a special exception, you may create a larger work
  29         that contains this FAUST architecture section and distribute
  30         that work under terms of your choice, so long as this FAUST
  31         architecture section is not modified.
  32
  33
  34  ************************************************************************
  35  ************************************************************************/
  36
  37
  38 /* link with  */
  39 /* link with : "" */
  40 #include <stdlib.h>
  41 #include <stdlib.h>
  42 #include <stdio.h>
  43 #include <string.h>
  44 #include <limits.h>
  45 #include <math.h>
  46 #include <errno.h>
  47 #include <time.h>
  48 #include <sys/ioctl.h>
  49 #include <unistd.h>
  50 #include <fcntl.h>
  51 #include <pwd.h>
  52 #include <sys/types.h>
  53 #include <assert.h>
  54 #include <pthread.h>
  55 #include <sys/wait.h>
  56 #include <list>
  57
  58 #include <iostream>
  59
  60
  61 using namespace std;
  62
  63 #include <map>
  64
  65 using namespace std;
  66
  67 // On Intel set FZ (Flush to Zero) and DAZ (Denormals Are Zero)
  68 // flags to avoid costly denormals
  69 #ifdef __SSE__
  70     #include <xmmintrin.h>
  71     #ifdef __SSE2__
  72         #define AVOIDDENORMALS _mm_setcsr(_mm_getcsr() | 0x8040)
  73     #else
  74         #define AVOIDDENORMALS _mm_setcsr(_mm_getcsr() | 0x8000)
  75     #endif
  76 #else
  77     #define AVOIDDENORMALS
  78 #endif
  79
  80 struct Meta : map<const char*, const char*>
  81 {
  82     void declare (const char* key, const char* value) { (*this)[key]=value; }
  83 };
  84
  85
  86
  87 float*          gBuffer = 0;        // a buffer of NV*VSize samples
  88
  89 unsigned int    COUNT   = 2000;     // number of measures
  90 unsigned int    NV      = 4096;     // number of vectors in BIG buffer (should exceed cache)
  91 unsigned int    ITER    = 10;       // number of iterations per measure
  92 unsigned int    VSIZE   = 4096;     // size of a vector in samples
  93 unsigned int    IDX     = 0;        // current vector number (0 <= VIdx < NV)
  94
  95
  96 //inline void *aligned_calloc(size_t nmemb, size_t size) { return (void*)((unsigned)(calloc((nmemb*size)+15,sizeof(char)))+15 & 0xfffffff0); }
  97 //inline void *aligned_calloc(size_t nmemb, size_t size) { return (void*)((size_t)(calloc((nmemb*size)+15,sizeof(char)))+15 & ~15); }
  98
  99
 100
 101 // g++ -Wall -O3 -lm -lpthread -lasound `gtk-config --cflags --libs` test.cpp -o test
 102
 103
 104 //-------------------------------------------------------------------
 105 // Generic min and max using c++ inline
 106 //-------------------------------------------------------------------
 107
 108 inline int              max (unsigned int a, unsigned int b) { return (a>b) ? a : b; }
 109 inline int              max (int a, int b)                      { return (a>b) ? a : b; }
 110
 111 inline long     max (long a, long b)            { return (a>b) ? a : b; }
 112 inline long     max (int a, long b)             { return (a>b) ? a : b; }
 113 inline long     max (long a, int b)             { return (a>b) ? a : b; }
 114
 115 inline float    max (float a, float b)          { return (a>b) ? a : b; }
 116 inline float    max (int a, float b)            { return (a>b) ? a : b; }
 117 inline float    max (float a, int b)            { return (a>b) ? a : b; }
 118 inline float    max (long a, float b)           { return (a>b) ? a : b; }
 119 inline float    max (float a, long b)           { return (a>b) ? a : b; }
 120
 121 inline double   max (double a, double b)        { return (a>b) ? a : b; }
 122 inline double   max (int a, double b)           { return (a>b) ? a : b; }
 123 inline double   max (double a, int b)           { return (a>b) ? a : b; }
 124 inline double   max (long a, double b)          { return (a>b) ? a : b; }
 125 inline double   max (double a, long b)          { return (a>b) ? a : b; }
 126 inline double   max (float a, double b)         { return (a>b) ? a : b; }
 127 inline double   max (double a, float b)         { return (a>b) ? a : b; }
 128
 129
 130 inline int              min (int a, int b)                      { return (a<b) ? a : b; }
 131
 132 inline long     min (long a, long b)            { return (a<b) ? a : b; }
 133 inline long     min (int a, long b)             { return (a<b) ? a : b; }
 134 inline long     min (long a, int b)             { return (a<b) ? a : b; }
 135
 136 inline float    min (float a, float b)          { return (a<b) ? a : b; }
 137 inline float    min (int a, float b)            { return (a<b) ? a : b; }
 138 inline float    min (float a, int b)            { return (a<b) ? a : b; }
 139 inline float    min (long a, float b)           { return (a<b) ? a : b; }
 140 inline float    min (float a, long b)           { return (a<b) ? a : b; }
 141
 142 inline double   min (double a, double b)        { return (a<b) ? a : b; }
 143 inline double   min (int a, double b)           { return (a<b) ? a : b; }
 144 inline double   min (double a, int b)           { return (a<b) ? a : b; }
 145 inline double   min (long a, double b)          { return (a<b) ? a : b; }
 146 inline double   min (double a, long b)          { return (a<b) ? a : b; }
 147 inline double   min (float a, double b)         { return (a<b) ? a : b; }
 148 inline double   min (double a, float b)         { return (a<b) ? a : b; }
 149
 150
 151 inline int              lsr (int x, int n)                      { return int(((unsigned int)x) >> n); }
 152
 153
 154 bool setRealtimePriority ()
 155 {
 156     struct passwd *         pw;
 157     int                     err;
 158     uid_t                   uid;
 159     int                     policy;
 160     struct sched_param      param;
 161
 162     uid = getuid ();
 163     pw = getpwnam ("root");
 164     setuid (pw->pw_uid);
 165
 166     pthread_getschedparam(pthread_self(), &policy, &param);
 167     policy = SCHED_RR;
 168     param.sched_priority = 50;
 169     err = pthread_setschedparam(pthread_self(), policy, &param);
 170
 171     setuid (uid);
 172     return (err != -1);
 173 }
 174
 175 #include <sys/time.h>
 176
 177 double mysecond()
 178 {
 179         struct timeval tp;
 180         struct timezone tzp;
 181         int i;
 182
 183         i = gettimeofday(&tp,&tzp);
 184         return ( (double) tp.tv_sec + (double) tp.tv_usec * 1.e-6 );
 185 }
 186
 187 /******************************************************************************
 188 *******************************************************************************
 189
 190                                                                VECTOR INTRINSICS
 191
 192 *******************************************************************************
 193 *******************************************************************************/
 194
 195
 196
 197 <<includeIntrinsic>>
 198
 199
 200
 201 /******************************************************************************
 202 *******************************************************************************
 203
 204                         ABSTRACT USER INTERFACE
 205
 206 *******************************************************************************
 207 *******************************************************************************/
 208
 209 class UI
 210 {
 211         bool    fStopped;
 212 public:
 213
 214         UI() : fStopped(false) {}
 215         virtual ~UI() {}
 216
 217         virtual void addButton(const char* label, float* zone) = 0;
 218         virtual void addToggleButton(const char* label, float* zone) = 0;
 219         virtual void addCheckButton(const char* label, float* zone) = 0;
 220         virtual void addVerticalSlider(const char* label, float* zone, float init, float min, float max, float step) = 0;
 221         virtual void addHorizontalSlider(const char* label, float* zone, float init, float min, float max, float step) = 0;
 222         virtual void addNumEntry(const char* label, float* zone, float init, float min, float max, float step) = 0;
 223
 224         // -- passive widgets
 225
 226         virtual void addNumDisplay(const char* label, float* zone, int precision) = 0;
 227         virtual void addTextDisplay(const char* label, float* zone, char* names[], float min, float max) = 0;
 228         virtual void addHorizontalBargraph(const char* label, float* zone, float min, float max) = 0;
 229         virtual void addVerticalBargraph(const char* label, float* zone, float min, float max) = 0;
 230
 231         virtual void openFrameBox(const char* label) = 0;
 232         virtual void openTabBox(const char* label) = 0;
 233         virtual void openHorizontalBox(const char* label) = 0;
 234         virtual void openVerticalBox(const char* label) = 0;
 235
 236         virtual void closeBox() = 0;
 237
 238         virtual void run() = 0;
 239
 240         void stop()     { fStopped = true; }
 241         bool stopped()  { return fStopped; }
 242
 243     virtual void declare(float* zone, const char* key, const char* value) {}
 244 };
 245
 246
 247
 248
 249 /******************************************************************************
 250 *******************************************************************************
 251
 252                                                                 DSP
 253
 254 *******************************************************************************
 255 *******************************************************************************/
 256
 257
 258
 259 //----------------------------------------------------------------
 260 //  definition du processeur de signal
 261 //----------------------------------------------------------------
 262
 263 class dsp {
 264  protected:
 265         int fSamplingFreq;
 266  public:
 267         dsp() {}
 268         virtual ~dsp() {}
 269         virtual int getNumInputs()                                                                              = 0;
 270         virtual int getNumOutputs()                                                                     = 0;
 271         virtual void buildUserInterface(UI* interface)                                  = 0;
 272         virtual void init(int samplingRate)                                                     = 0;
 273         virtual void compute(int len, float** inputs, float** outputs)  = 0;
 274 };
 275
 276 /********************END ARCHITECTURE SECTION (part 1/2)****************/
 277
 278 /**************************BEGIN USER SECTION **************************/
 279
 280 <<includeclass>>
 281
 282 /***************************END USER SECTION ***************************/
 283
 284 /*******************BEGIN ARCHITECTURE SECTION (part 2/2)***************/
 285
 286 mydsp   DSP;
 287
 288
 289 #if 0
 290
 291 static __inline__ unsigned long long int rdtsc(void)
 292 {
 293   unsigned long long int x;
 294      __asm__ volatile (".byte 0x0f, 0x31" : "=A" (x));
 295      return x;
 296 }
 297
 298 #endif
 299
 300 /**
 301  * Bench by calling COUNT times the compute() method for
 302  * the computation of vsize samples
 303  */
 304
 305 #define MEGABYTE 1048576.0
 306
 307 void statistic(const char* name, double* timing)
 308 {
 309     double lo, hi, tot;
 310     double mega =  double(VSIZE*ITER)/MEGABYTE; // mega samples
 311     // skip first 10 values to avoid cache bias ???
 312     lo = hi = tot = mega/(timing[11] - timing[10]);
 313     for (int i = 11; i<COUNT; i++) {
 314         double delta = mega/(timing[i] - timing[i-1]);
 315         if (delta < lo) {
 316             lo = delta;
 317         } else if (delta > hi) {
 318             hi = delta;
 319         }
 320         tot += delta;
 321     }
 322
 323     cout << '\t' << hi
 324          << '\t' << hi*4*DSP.getNumInputs() << '\t' << "MB/s inputs"
 325          << '\t' << hi*4*DSP.getNumOutputs() << '\t' << "MB/s outputs"
 326          << '\t' << tot/(COUNT-11)
 327          << '\t' << lo
 328          << endl;
 329 }
 330
 331 void allocBuffer()
 332 {
 333     unsigned int BSIZE = NV * VSIZE;
 334     gBuffer = (float*) calloc (BSIZE, sizeof(float));
 335
 336     int R0_0 = 0;
 337     for (int j = 0; j < BSIZE; j++) {
 338         int R0temp0 = (12345 + (1103515245 * R0_0));
 339         gBuffer[j] = 4.656613e-10f*R0temp0;
 340         R0_0 = R0temp0;
 341     }
 342 }
 343
 344 float* nextVect()
 345 {
 346     IDX = (1+IDX)%NV;
 347     return &gBuffer[IDX*VSIZE];
 348 }
 349
 350 void bench(const char* name)
 351 {
 352
 353     int numInChan = DSP.getNumInputs();
 354     int numOutChan = DSP.getNumOutputs();
 355
 356     assert (numInChan < 256);
 357     assert (numOutChan < 256);
 358
 359     float*  inChannel[256];
 360     float*  outChannel[256];
 361
 362     // allocate input buffers (initialized with white noise)
 363     allocBuffer();
 364
 365     // allocate output channels (not initialized)
 366     for (int i = 0; i < numOutChan; i++) outChannel[i] = (float*) calloc (VSIZE, sizeof(float));
 367
 368     // init the dsp with a resoneable sampling rate)
 369     DSP.init(48000);
 370     double* timing = (double*) calloc (COUNT, sizeof(double));
 371
 372     for (int i = 0; i<COUNT; i++) {
 373         timing[i] = mysecond();
 374         for (int k = 0; k<ITER; k++) {
 375             // allocate new input buffers to avoid L2 cache
 376             for (int c=0; c<numInChan; c++) { inChannel[c] = nextVect(); }
 377             DSP.compute(VSIZE,inChannel,outChannel);
 378         }
 379     }
 380
 381     statistic(name, timing);
 382 }
 383
 384 //-------------------------------------------------------------------------
 385 //                                                                      MAIN
 386 //-------------------------------------------------------------------------
 387
 388 // lopt : Scan Command Line long int Arguments
 389
 390 long lopt (int argc, char *argv[], const char* longname, const char* shortname, long def)
 391 {
 392         for (int i=2; i<argc; i++)
 393                 if ( strcmp(argv[i-1], shortname) == 0 || strcmp(argv[i-1], longname) == 0 )
 394                         return atoi(argv[i]);
 395         return def;
 396 }
 397
 398
 399 int main(int argc, char *argv[] )
 400 {
 401     AVOIDDENORMALS;
 402     VSIZE = lopt(argc, argv, "--vector-size", "-vec", 4096);
 403     NV = lopt(argc, argv, "--num-vector", "-n", 20000);
 404     COUNT = lopt(argc, argv, "--count", "-c", 1000);
 405     ITER = lopt(argc, argv, "--iteration", "-i", 10);
 406     //setRealtimePriority();
 407         bench(argv[0]);
 408         return 0;
 409 }
 410
 411
 412 /********************END ARCHITECTURE SECTION (part 2/2)****************/
 413
 414
 415