#define TS 32

#define WPT 8

#define RTS (TS/WPT)

  int AA=-1;
 int BB=-1;
 int CC=-1;
 int DD=-1;

__kernel void sgemm(__global float *A, __global float *B, __global float *C, int M, int N, int K) {

    // Thread identifiers

    const int row = get_local_id(0); // Local row ID (max: TS)

    const int col = get_local_id(1); // Local col ID (max: TS/WPT == RTS)

    const int globalRow = TS*get_group_id(0) + row; // Row ID of C (0..M)

    const int globalCol = TS*get_group_id(1) + col; // Col ID of C (0..N)

 

    // Local memory to fit a tile of TS*TS elements of A and B

    __local float Asub[TS][TS];

    __local float Bsub[TS][TS];

 

    // Initialise the accumulation registers

    float acc[WPT];

    for (int w=0; w<WPT; w++) {

        acc[w] = 0.0f;

    }

 

    // Loop over all tiles   

    const int numTiles = (K + TS - 1)/TS;

 

    for (int t=0; t<numTiles; t++) {

        // Load one tile of A and B into local memory

        for (int w=0; w<WPT; w++) {

            const int tiledRow = TS*t + row;

            const int tiledCol = TS*t + col;

            if((globalRow + w*RTS) < M && (tiledCol<K))

            {

                Asub[row + w*RTS][col] = A[(globalRow + w*RTS)*K + tiledCol];

            }

            else{

                Asub[row + w*RTS][col] = 0.0f;

            }

            

            if((tiledRow + w*RTS)<K  && globalCol < N)

            {

                Bsub[row + w*RTS][col] = B[(tiledRow + w*RTS)*N + globalCol];

            }

            else{

                Bsub[row + w*RTS][col] = 0.0f;

            }

        }

#if 0        
if(get_group_id(0)==AA && get_group_id(1)==BB && row ==CC && col==DD)
//if(get_group_id(0)==1 && get_group_id(1)==0)
{ 
printf("============A===========\n");
          for(int i=0;i<M;i++)
          {
            for(int j=0;j<K;j++)
            {
              printf("%+.3f ",A[i*K+j]);
            }
            printf("\n");
          }
printf("============B===========\n");          
          for(int i=0;i<K;i++)
          {
            for(int j=0;j<N;j++)
            {
              printf("%+.3f ",B[i*N+j]);
            }
            printf("\n");
          }
printf("============ASUB===========\n");

          for(int i=0;i<TS;i++)
          {
            for(int j=0;j<TS;j++)
            {
              printf("%+.3f ",Asub[i][j]);
            }
            printf("\n");
          }

printf("============BSUB===========\n");
          for(int i=0;i<TS;i++)
          {
            for(int j=0;j<TS;j++)
            {
              printf("%+.3f ",Bsub[i][j]);
            }
            printf("\n");
          }
}
#endif
 

        // Synchronise to make sure the tile is loaded

        barrier(CLK_LOCAL_MEM_FENCE);

 

        // Perform the computation for a single tile

        for (int k=0; k<TS; k++) {

            for (int w=0; w<WPT; w++) {

                acc[w] += Asub[row + w*RTS][k] * Bsub[k][col];

            }

        }

 

        // Synchronise before loading the next tile

        barrier(CLK_LOCAL_MEM_FENCE);

    }

    // Store the final results in C

    for (int w=0; w<WPT; w++) {

        if(globalRow < M && globalCol < N) 

            C[(globalRow + w*RTS)*N + globalCol] = acc[w];

    }

}