mt/bc_vvadd/bc_vvadd.c

   1 //**************************************************************************
   2 // Vector-vector add benchmark
   3 //--------------------------------------------------------------------------
   4 // Author  : Andrew Waterman
   5 // TA      : Christopher Celio
   6 // Student :
   7 //
   8 // This benchmark adds two vectors and writes the results to a
   9 // third vector. The input data (and reference data) should be
  10 // generated using the vvadd_gendata.pl perl script and dumped
  11 // to a file named dataset.h
  12
  13 // to print out arrays, etc.
  14 //#define DEBUG
  15
  16 //--------------------------------------------------------------------------
  17 // Includes
  18
  19 #include <string.h>
  20 #include <stdlib.h>
  21 #include <stdio.h>
  22
  23
  24 //--------------------------------------------------------------------------
  25 // Input/Reference Data
  26
  27 typedef float data_t;
  28 #include "dataset.h"
  29
  30 #define MIN(X,Y) (X < Y ? X : Y)
  31 #define MAX(X,Y) (X > Y ? X : Y)
  32
  33 //--------------------------------------------------------------------------
  34 // Basic Utilities and Multi-thread Support
  35
  36 __thread unsigned long coreid;
  37 unsigned long ncores;
  38
  39 #include "util.h"
  40
  41 #define stringify_1(s) #s
  42 #define stringify(s) stringify_1(s)
  43 #define stats(code) do { \
  44     unsigned long _c = -rdcycle(), _i = -rdinstret(); \
  45     code; \
  46     _c += rdcycle(), _i += rdinstret(); \
  47     if (coreid == 0) \
  48       printf("%s: %ld cycles, %ld.%ld cycles/iter, %ld.%ld CPI\n", \
  49              stringify(code), _c, _c/DATA_SIZE, 10*_c/DATA_SIZE%10, _c/_i, 10*_c/_i%10); \
  50   } while(0)
  51
  52
  53 //--------------------------------------------------------------------------
  54 // Helper functions
  55
  56 void printArray( char name[], int n, data_t arr[] )
  57 {
  58   int i;
  59   if (coreid != 0)
  60      return;
  61
  62   printf( " %10s :", name );
  63   for ( i = 0; i < n; i++ )
  64     printf( " %4ld ", (long) arr[i] );
  65   printf( "\n" );
  66 }
  67
  68 void __attribute__((noinline)) verify(size_t n, const data_t* test, const data_t* correct)
  69 {
  70    if (coreid != 0)
  71       return;
  72
  73    size_t i;
  74    for (i = 0; i < n; i++)
  75    {
  76       if (test[i] != correct[i])
  77       {
  78          printf("FAILED test[%d]= %4ld, correct[%d]= %4ld\n",
  79             i, (long) test[i], i, (long)correct[i]);
  80          exit(-1);
  81       }
  82    }
  83
  84    return;
  85 }
  86
  87 //--------------------------------------------------------------------------
  88 // vvadd function
  89
  90 //perform in-place vvadd
  91 void __attribute__((noinline)) vvadd(size_t n, data_t* __restrict__ x, const data_t* __restrict__ y)
  92 {
  93    size_t i;
  94
  95    // interleave accesses
  96    for (i = coreid; i < n; i+=ncores)
  97    {
  98       x[i] = x[i] + y[i];
  99    }
 100 }
 101
 102 void __attribute__((noinline)) vvadd_opt(size_t n, data_t* __restrict__ x, const data_t* __restrict__ y)
 103 {
 104    // ***************************** //
 105    // **** ADD YOUR CODE HERE ***** //
 106    // ***************************** //
 107   size_t i, start = coreid * (n / ncores), end = (coreid == ncores - 1 ? n : (coreid + 1) * (n / ncores));
 108
 109   for (i = start; i < end; i++) {
 110     x[i] += y[i];
 111   }
 112 }
 113
 114 //--------------------------------------------------------------------------
 115 // Main
 116 //
 117 // all threads start executing thread_entry(). Use their "coreid" to
 118 // differentiate between threads (each thread is running on a separate core).
 119
 120 void thread_entry(int cid, int nc)
 121 {
 122    coreid = cid;
 123    ncores = nc;
 124
 125    // static allocates data in the binary, which is visible to both threads
 126    static data_t results_data[DATA_SIZE];
 127
 128    // because we're going to perform an in-place vvadd (and we're going to run
 129    // it a couple of times) let's copy the input data to a temporary results
 130    // array
 131
 132    size_t i;
 133    if (coreid == 0)
 134    {
 135       for (i = 0; i < DATA_SIZE; i++)
 136          results_data[i] = input1_data[i];
 137    }
 138
 139
 140    // Execute the provided, terrible vvadd
 141    barrier();
 142    stats(vvadd(DATA_SIZE, results_data, input2_data); barrier());
 143
 144
 145    // verify
 146    verify(DATA_SIZE, results_data, verify_data);
 147
 148    // reset results from the first trial
 149    if (coreid == 0)
 150    {
 151       for (i=0; i < DATA_SIZE; i++)
 152          results_data[i] = input1_data[i];
 153    }
 154    barrier();
 155
 156
 157    // Execute your faster vvadd
 158    barrier();
 159    stats(vvadd_opt(DATA_SIZE, results_data, input2_data); barrier());
 160
 161 #ifdef DEBUG
 162    printArray("results: ", DATA_SIZE, results_data);
 163    printArray("verify : ", DATA_SIZE, verify_data);
 164 #endif
 165
 166    // verify
 167    verify(DATA_SIZE, results_data, verify_data);
 168    barrier();
 169
 170    exit(0);
 171 }
 172