mt/dv_matmul.c

   1 #include "stdlib.h"
   2
   3 #include "util.h"
   4
   5 #include "dataset.h"
   6 void __attribute__((noinline)) matmul(const int coreid, const int ncores, const int lda,  const data_t A[], const data_t B[], data_t C[] )
   7 {
   8
   9    // ***************************** //
  10    // **** ADD YOUR CODE HERE ***** //
  11    // ***************************** //
  12    //
  13    // feel free to make a separate function for MI and MSI versions.
  14    int i, j, k, ii, jj, kk;
  15    if(coreid > 1) return;
  16   if (coreid || ncores == 1) {
  17 // for ( ii = 0; ii < 32; ii+=IC )
  18          for ( kk = 0; kk < 32; kk+=16 )
  19    for ( j = 0; j < 16; j++ )
  20 // for ( j = 0; j < 16; j++ )
  21    {
  22       for ( i =  0; i < 32; i+=8 )
  23 //    for ( i = ii; i < ii + IC && i < 32; i+=8 )
  24       {
  25          data_t temp0 = C[i+j*32];
  26          data_t temp1 = C[i+j*32+1];
  27          data_t temp2 = C[i+j*32+2];
  28          data_t temp3 = C[i+j*32+3];
  29          data_t temp4 = C[i+j*32+4];
  30          data_t temp5 = C[i+j*32+5];
  31          data_t temp6 = C[i+j*32+6];
  32          data_t temp7 = C[i+j*32+7];
  33          for ( k = kk; k < kk+16 && k < 32; k++ )
  34 //       for ( k = 0; k < 32; k++ )
  35          {
  36             data_t tempA = A[j*32+k];
  37             temp0 += tempA * B[k*32 + i];
  38             temp1 += tempA * B[k*32 + i+1];
  39             temp2 += tempA * B[k*32 + i+2];
  40             temp3 += tempA * B[k*32 + i+3];
  41             temp4 += tempA * B[k*32 + i+4];
  42             temp5 += tempA * B[k*32 + i+5];
  43             temp6 += tempA * B[k*32 + i+6];
  44             temp7 += tempA * B[k*32 + i+7];
  45          }
  46          C[i+j*32] = temp0;
  47          C[i+j*32+1] = temp1;
  48          C[i+j*32+2] = temp2;
  49          C[i+j*32+3] = temp3;
  50          C[i+j*32+4] = temp4;
  51          C[i+j*32+5] = temp5;
  52          C[i+j*32+6] = temp6;
  53          C[i+j*32+7] = temp7;
  54       }
  55    }
  56   }
  57   if(coreid == 0){
  58 // for ( ii = 0; ii < 32; ii+=IC )
  59          for ( kk = 0; kk < 32; kk+=16 )
  60    for ( j = 16; j < 32; j++ )
  61 // for ( j = 16; j < 32; j++ )
  62    {
  63       for ( i =   0; i < 32; i+=8 )
  64 //    for ( i = ii; i < ii + IC && i < 32; i+=8 )
  65       {
  66          data_t temp0 = C[i+j*32];
  67          data_t temp1 = C[i+j*32+1];
  68          data_t temp2 = C[i+j*32+2];
  69          data_t temp3 = C[i+j*32+3];
  70          data_t temp4 = C[i+j*32+4];
  71          data_t temp5 = C[i+j*32+5];
  72          data_t temp6 = C[i+j*32+6];
  73          data_t temp7 = C[i+j*32+7];
  74          for ( k = kk; k < kk+16 && k < 32; k++ )
  75          {
  76             data_t tempA = A[j*32+k];
  77             temp0 += tempA * B[k*32 + i];
  78             temp1 += tempA * B[k*32 + i+1];
  79             temp2 += tempA * B[k*32 + i+2];
  80             temp3 += tempA * B[k*32 + i+3];
  81             temp4 += tempA * B[k*32 + i+4];
  82             temp5 += tempA * B[k*32 + i+5];
  83             temp6 += tempA * B[k*32 + i+6];
  84             temp7 += tempA * B[k*32 + i+7];
  85          }
  86          C[i+j*32] = temp0;
  87          C[i+j*32+1] = temp1;
  88          C[i+j*32+2] = temp2;
  89          C[i+j*32+3] = temp3;
  90          C[i+j*32+4] = temp4;
  91          C[i+j*32+5] = temp5;
  92          C[i+j*32+6] = temp6;
  93          C[i+j*32+7] = temp7;
  94       }
  95
  96    }
  97   }
  98 }