benchmarks/vec-vvadd/vec_vvadd_main.c

   1 //**************************************************************************
   2 // Vector-Thread Vector-vector add benchmark
   3 //--------------------------------------------------------------------------
   4 //
   5 // This benchmark uses adds to vectors and writes the results to a third
   6 // vector. The input data (and reference data) should be generated using the
   7 // vvadd_gendata.pl perl script and dumped to a file named dataset.h. The
   8 // riscv-gcc toolchain does not support system calls so printf's can only be
   9 // used on a host system, not on the riscv-v processor simulator itself.
  10 //
  11 // HOWEVER: printstr() and printhex() are provided, for a primitive form of
  12 // printing strings and hexadecimal values to stdout.
  13
  14
  15 // Choose which implementation you wish to test... but leave only one on!
  16 // (only the first one will be executed).
  17 //#define SCALAR_C
  18 //#define SCALAR_ASM
  19 #define VT_ASM
  20
  21 //--------------------------------------------------------------------------
  22 // Macros
  23
  24 // Set HOST_DEBUG to 1 if you are going to compile this for a host
  25 // machine (ie Athena/Linux) for debug purposes and set HOST_DEBUG
  26 // to 0 if you are compiling with the smips-gcc toolchain.
  27
  28 #ifndef HOST_DEBUG
  29 #define HOST_DEBUG 0
  30 #endif
  31
  32 // Set PREALLOCATE to 1 if you want to preallocate the benchmark
  33 // function before starting stats. If you have instruction/data
  34 // caches and you don't want to count the overhead of misses, then
  35 // you will need to use preallocation.
  36
  37 #ifndef PREALLOCATE
  38 #define PREALLOCATE 0
  39 #endif
  40
  41 // Set SET_STATS to 1 if you want to carve out the piece that actually
  42 // does the computation.
  43
  44 #ifndef SET_STATS
  45 #define SET_STATS 0
  46 #endif
  47
  48 //--------------------------------------------------------------------------
  49 // Platform Specific Includes
  50
  51 #if HOST_DEBUG
  52    #include <stdio.h>
  53    #include <stdlib.h>
  54 #else
  55 void printstr(const char*);
  56 void exit();
  57 #endif
  58
  59
  60 //--------------------------------------------------------------------------
  61 // Input/Reference Data
  62
  63 //#include "dataset_test.h"
  64 #include "dataset.h"
  65
  66 //--------------------------------------------------------------------------
  67 // Helper functions
  68
  69 int verify( int n, float test[], float correct[] )
  70 {
  71   int i;
  72   for ( i = 0; i < n; i++ ) {
  73 //    if ( test[i] != correct[i] ) {
  74     if ( test[i] > 1.02*correct[i]
  75        || test[i] < 0.98*correct[i]) {
  76 #if HOST_DEBUG
  77       printf("    test[%d] : %3.2f\n", i, test[i]);
  78       printf("    corr[%d] : %3.2f\n", i, correct[i]);
  79 #endif
  80       // tell us which index fails + 2
  81       // (so that if i==0,i==1 fails, we don't
  82       // think it was a 'not-finished yet' or pass)
  83 //      return i+10;
  84       return 2;
  85     }
  86   }
  87   return 1;
  88 }
  89
  90 #if HOST_DEBUG
  91 void printArray( char name[], int n, float arr[] )
  92 {
  93   int i;
  94   printf( " %10s :", name );
  95   for ( i = 0; i < n; i++ )
  96     printf( " %03.2f ", arr[i] );
  97   printf( "\n" );
  98 }
  99 #endif
 100
 101
 102 void finishTest( int correct, long long num_cycles, long long num_retired )
 103 {
 104    int toHostValue = correct;
 105 #if HOST_DEBUG
 106   if ( toHostValue == 1 )
 107     printf( "*** PASSED ***\n" );
 108   else
 109     printf( "*** FAILED *** (tohost = %d)\n", toHostValue );
 110   exit(0);
 111 #else
 112   // we no longer run in -testrun mode, which means we can't use
 113   // the tohost register to communicate "test is done" and "test results"
 114   // so instead we will communicate through print* functions!
 115   if ( correct == 1 )
 116   {
 117     printstr( "*** PASSED *** (num_cycles = 0x" );
 118     printhex(num_cycles);
 119     printstr( ", num_inst_retired = 0x");
 120     printhex(num_retired);
 121     printstr( ")\n" );
 122   }
 123   else
 124   {
 125     printstr( "*** FAILED *** (num_cycles = 0x");
 126     printhex(num_cycles);
 127     printstr( ", num_inst_retired = 0x");
 128     printhex(num_retired);
 129     printstr( ")\n" );
 130   }
 131   exit();
 132 #endif
 133 }
 134
 135
 136 // deprecated - cr10/stats-enable register no longer exists
 137 void setStats( int enable )
 138 {
 139 #if ( !HOST_DEBUG && SET_STATS )
 140   asm( "mtpcr %0, cr10" : : "r" (enable) );
 141 #endif
 142 }
 143
 144 long long getCycles()
 145 {
 146    long long cycles = 1337;
 147 #if ( !HOST_DEBUG && SET_STATS )
 148   __asm__ __volatile__( "rdcycle %0" : "=r" (cycles) );
 149 #endif
 150   return cycles;
 151 }
 152
 153 long long getInstRetired()
 154 {
 155    long long inst_retired = 1338;
 156 #if ( !HOST_DEBUG && SET_STATS )
 157   __asm__ __volatile__( "rdinstret %0" : "=r" (inst_retired) );
 158 #endif
 159   return inst_retired;
 160 }
 161
 162 //--------------------------------------------------------------------------
 163 // vvadd function
 164
 165 // scalar C implementation
 166 void vvadd( int n, float a[], float b[], float c[] )
 167 {
 168   int i;
 169   for ( i = 0; i < n; i++ )
 170     c[i] = a[i] + b[i];
 171 }
 172
 173 // assembly implementations can be found in *_asm.S
 174
 175 //--------------------------------------------------------------------------
 176 // Main
 177
 178 int main( int argc, char* argv[] )
 179 {
 180   float results_data[DATA_SIZE];
 181   long long start_cycles = 0;
 182   long long stop_cycles = 0;
 183   long long num_cycles;
 184   long long start_retired = 0;
 185   long long stop_retired = 0;
 186   long long num_retired;
 187
 188   // Output the input array
 189
 190 #if HOST_DEBUG
 191   printArray( "input1", DATA_SIZE, input1_data );
 192   printArray( "input2", DATA_SIZE, input2_data );
 193   printArray( "verify", DATA_SIZE, verify_data );
 194 #endif
 195
 196   // --------------------------------------------------
 197   // If needed we preallocate everything in the caches
 198
 199 #if PREALLOCATE
 200
 201 #ifdef SCALAR_C
 202   vvadd( DATA_SIZE, input1_data, input2_data, results_data );
 203 #else
 204 #ifdef SCALAR_ASM
 205   scalar_vvadd_asm( DATA_SIZE, input1_data, input2_data, results_data );
 206 #else
 207 #ifdef VT_ASM
 208   vt_vvadd_asm( DATA_SIZE, input1_data, input2_data, results_data );
 209 #endif
 210 #endif
 211 #endif
 212
 213 #endif
 214
 215   // --------------------------------------------------
 216   // Do the vvadd
 217   start_cycles = getCycles();
 218   start_retired = getInstRetired();
 219
 220 #ifdef SCALAR_C
 221   vvadd( DATA_SIZE, input1_data, input2_data, results_data );
 222 #else
 223 #ifdef SCALAR_ASM
 224   #if HOST_DEBUG==0
 225   scalar_vvadd_asm( DATA_SIZE, input1_data, input2_data, results_data );
 226   #endif
 227 #else
 228 #ifdef VT_ASM
 229   #if HOST_DEBUG==0
 230   vt_vvadd_asm( DATA_SIZE, input1_data, input2_data, results_data );
 231   #endif
 232 #endif
 233 #endif
 234 #endif
 235
 236   stop_cycles  = getCycles();
 237   stop_retired = getInstRetired();
 238   num_cycles = stop_cycles - start_cycles;
 239   num_retired = stop_retired - start_retired;
 240
 241 //  printstr("stop_cycles: "); printhex(stop_cycles); printstr("\n");
 242 //  printstr("star_cycles: "); printhex(start_cycles); printstr("\n");
 243
 244   // --------------------------------------------------
 245   // Print out the results
 246
 247 #if HOST_DEBUG
 248   printArray( "results", DATA_SIZE, results_data );
 249 #endif
 250
 251   // --------------------------------------------------
 252   // Check the results
 253   int correct = verify( DATA_SIZE, results_data, verify_data );
 254   finishTest(correct, num_cycles, num_retired);
 255 }