int main(int argc, char **argv)
{
int my_rank=0; /* My process rank */
int p; /* The number of processes */
//clock time recording variables
double start_time,end_time,comm_time_start,comm_time_end,total_comm_time=0.0;
/* Let the system do what it needs to start up MPI */
MPI_Init(&argc, &argv);
/* Get my process rank */
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
/* Find out how many processes are being used */
MPI_Comm_size(MPI_COMM_WORLD, &p);
float dataA[N][N], dataB[N][N];
complex A[N][N], B[N][N];
//create MPI type for complex datatype
MPI_Aint displ[2];
displ[0] = 0;
displ[1] = sizeof(float);
MPI_Datatype mpi_complex,types[2];
types[0] = MPI_FLOAT;
types[1] = MPI_FLOAT;
int block_len[2];
block_len[0]= 1;
block_len[1]= 1;
MPI_Type_struct(2, block_len, displ, types,&mpi_complex);
MPI_Type_commit(&mpi_complex);
//create 4 communicators by splitting the default one
MPI_Comm new_comm;
/* Find out how many processes are being used */
int color = my_rank / 2;
MPI_Comm_split(MPI_COMM_WORLD, color, my_rank, &new_comm);
int row_rank, row_size;
MPI_Comm_rank(new_comm, &row_rank);
MPI_Comm_size(new_comm, &row_size);
//we will read the files by corresponding processes
if(my_rank==0)
{
readFromFile("data/1_im1", dataA);
convertToComplex(A, dataA);
}
else if(my_rank==2)
{
readFromFile("data/1_im2", dataB);
convertToComplex(B, dataB);
}
if(my_rank==6)
{
//Start clock and record the start time.
start_time = MPI_Wtime();
}
int local_n = N/row_size;
int rows_per_process = N/row_size;
complex local_A[N/row_size][N];
complex local_B[N/row_size][N];
//broadcast the number of ros each process will work on
MPI_Bcast(&rows_per_process , 1, MPI_INT, 0,MPI_COMM_WORLD);
//process 0 and 1 work on Matrix A
if(my_rank<2)
{
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*rows_per_process, mpi_complex, 0,
new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
c_rowwise_fft2d(&local_A[0][0],rows_per_process);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Gather(&local_A, N*rows_per_process, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
if(row_rank==0)
transpose(A);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*rows_per_process, mpi_complex, 0,
new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
c_rowwise_fft2d(&local_A[0][0],rows_per_process);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Gather(&local_A, N*rows_per_process, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
MPI_Barrier(new_comm);
if(row_rank==0)
{
transpose(A);
comm_time_start=MPI_Wtime();
//send the intermediate matrix to process 4
MPI_Send(A, N*N, mpi_complex, 4,11,MPI_COMM_WORLD);
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
}
if(my_rank>=2 && my_rank<4) //process 2 and 3 work on Matrix B
{
//printf(" %d ",rows_per_process);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Scatter(&B[0][0], N*rows_per_process, mpi_complex,
&local_B, N*rows_per_process, mpi_complex, 0,
new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
c_rowwise_fft2d(&local_B[0][0],rows_per_process);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Gather(&local_B, N*local_n, mpi_complex,
&B[0][0], N*rows_per_process, mpi_complex,
0, new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
if(row_rank==0)
transpose(B);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Scatter(&B[0][0], N*rows_per_process, mpi_complex,
&local_B, N*rows_per_process, mpi_complex, 0,
new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
c_rowwise_fft2d(&local_B[0][0],rows_per_process);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Gather(&local_B, N*local_n, mpi_complex,
&B[0][0], N*rows_per_process, mpi_complex,
0, new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
MPI_Barrier(new_comm);
if(row_rank==0)
{
transpose(B);
//lets send intermediate matrix B to process 4
MPI_Send(B, N*N, mpi_complex, 4, 12,MPI_COMM_WORLD);
}
}
if(my_rank==4) //process 4 multiplies the two matrices
{
MPI_Status status;
comm_time_start=MPI_Wtime();
MPI_Recv(A, N*N, mpi_complex, 0, 11,
MPI_COMM_WORLD, &status);
MPI_Recv(B, N*N, mpi_complex, 2, 12,
MPI_COMM_WORLD, &status);
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
mmul_point(A,B);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
MPI_Send(A, N*N, mpi_complex, 6, 13,MPI_COMM_WORLD);
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
}
if(my_rank>=6 && my_rank<8) //process 6 and 7 does inverse transform on result
{
MPI_Status status;
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
MPI_Recv(A, N*N, mpi_complex, 4, 13,
MPI_COMM_WORLD, &status);
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*rows_per_process, mpi_complex, 0,
new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
c_rowwise_inv_fft2d(&local_A[0][0],rows_per_process);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Gather(&local_A, N*rows_per_process, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
if(row_rank==0)
transpose(A);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*rows_per_process, mpi_complex, 0,
new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
c_rowwise_inv_fft2d(&local_A[0][0],rows_per_process);
if(row_rank==0)
{
comm_time_start=MPI_Wtime();
}
MPI_Gather(&local_A, N*rows_per_process, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, new_comm);
if(row_rank==0)
{
comm_time_end=MPI_Wtime();
total_comm_time+=comm_time_end-comm_time_start;
}
if(row_rank==0)
{ transpose(A);
//Stop clock as operation is finished.
end_time = MPI_Wtime();
//print the execution time for performance analysis purpose.
printf("\n\nThe total execution time as recorded on process 0 = %f seconds!!\n!",end_time-start_time);
convertToReal(dataA,A);
writeToFile("final_output.txt", dataA);
}
}
MPI_Barrier(MPI_COMM_WORLD);
//get a total of communication cost recorded by all processes
MPI_Reduce(&total_comm_time, &total_comm_time, 1, MPI_FLOAT,
MPI_SUM, 0, MPI_COMM_WORLD);
if(my_rank ==0)
{
printf("\n\nThe total Communication time = %f seconds!!\n!",total_comm_time);
}
MPI_Comm_free(&new_comm);
MPI_Finalize();
return 0;
}
int main(int argc, char **argv)
{
int my_rank=0; /* My process rank */
int p; /* The number of processes */
//clock time recording variables
double start_time,end_time,comm_time_start,comm_time_end,total_comm_time=0.0;
/* Let the system do what it needs to start up MPI */
MPI_Init(&argc, &argv);
/* Get my process rank */
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
/* Find out how many processes are being used */
MPI_Comm_size(MPI_COMM_WORLD, &p);
float dataA[N][N], dataB[N][N];
complex A[N][N], B[N][N];
MPI_Aint displ[2];
displ[0] = 0;
displ[1] = sizeof(float);
MPI_Datatype mpi_complex,types[2];
types[0] = MPI_FLOAT;
types[1] = MPI_FLOAT;
int block_len[2];
block_len[0]= 1;
block_len[1]= 1;
MPI_Type_struct(2, block_len, displ, types,&mpi_complex);
MPI_Type_commit(&mpi_complex);
if(my_rank==0)
{
readFromFile("data/1_im1", dataA);
readFromFile("data/1_im2", dataB);
convertToComplex(A, dataA);
convertToComplex(B, dataB);
//Start clock and record the start time.
start_time = MPI_Wtime();
}
int local_n = N/p;
int rows_per_process = N/p;
complex local_A[N/p][N];
MPI_Bcast(&rows_per_process , 1, MPI_INT, 0,MPI_COMM_WORLD);
//scatter A matrix aross all processes
startCommTimer(my_rank,&comm_time_start);
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*local_n, mpi_complex, 0,
MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
c_rowwise_fft2d(&local_A[0][0],local_n);
startCommTimer(my_rank,&comm_time_start);
MPI_Gather(&local_A, N*local_n, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
complex local_B[N/p][N];
//scatter A matrix aross all processes
startCommTimer(my_rank,&comm_time_start);
MPI_Scatter(&B[0][0], N*rows_per_process, mpi_complex,
&local_B, N*local_n, mpi_complex, 0,
MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
c_rowwise_fft2d(&local_B[0][0],local_n);
startCommTimer(my_rank,&comm_time_start);
MPI_Gather(&local_B, N*local_n, mpi_complex,
&B[0][0], N*rows_per_process, mpi_complex,
0, MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
if(my_rank==0)
{
transpose(A);
transpose(B);
}
//scatter A matrix aross all processes for inverse transform
startCommTimer(my_rank,&comm_time_start);
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*local_n, mpi_complex, 0,
MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
c_rowwise_fft2d(&local_A[0][0],local_n);
startCommTimer(my_rank,&comm_time_start);
MPI_Gather(&local_A, N*local_n, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
startCommTimer(my_rank,&comm_time_start);
MPI_Scatter(&B[0][0], N*rows_per_process, mpi_complex,
&local_B, N*local_n, mpi_complex, 0,
MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
c_rowwise_fft2d(&local_B[0][0],local_n);
startCommTimer(my_rank,&comm_time_start);
MPI_Gather(&local_B, N*local_n, mpi_complex,
&B[0][0], N*rows_per_process, mpi_complex,
0, MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
if(my_rank==0)
{
transpose(A);
transpose(B);
mmul_point(A,B);
}
startCommTimer(my_rank,&comm_time_start);
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*local_n, mpi_complex, 0,
MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
c_rowwise_inv_fft2d(&local_A[0][0],local_n);
startCommTimer(my_rank,&comm_time_start);
MPI_Gather(&local_A, N*local_n, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
if(my_rank==0)
{
transpose(A);
}
startCommTimer(my_rank,&comm_time_start);
MPI_Scatter(&A[0][0], N*rows_per_process, mpi_complex,
&local_A, N*local_n, mpi_complex, 0,
MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
c_rowwise_inv_fft2d(&local_A[0][0],local_n);
startCommTimer(my_rank,&comm_time_start);
MPI_Gather(&local_A, N*local_n, mpi_complex,
&A[0][0], N*rows_per_process, mpi_complex,
0, MPI_COMM_WORLD);
stopCommTimer(my_rank,&total_comm_time,comm_time_start);
if(my_rank==0)
{
transpose(A);
//Stop clock as operation is finished.
end_time = MPI_Wtime();
//print the execution time for performance analysis purpose.
printf("\n\nThe total execution time as recorded on process 0 = %f seconds!!\n!",end_time-start_time);
printf("\n\nThe total communication time = %f seconds!!\n!",total_comm_time);
convertToReal(dataA,A);
}
writeToFile("final_output.txt", dataA);
MPI_Finalize();
return 0;
}
void CMLcdPixDefault_Noai::updatePixel(int start_x, int start_y, enum UpdateRegion updateRegion)
{
int x, y;
int end_y, end_x;
int place_x, place_y;
start_x = convertToReal( start_x );
start_y = convertToReal( start_y );
if(updateRegion == PART)
{
end_x = getRealWidth102();
end_y = getRealHeight64();
}
else
{
end_y = getRealHeight240();
end_x = getRealWidth320();
}
for( y = 0; y < end_y; y++ )
{
place_y = start_y + y;
for( x = 0; x < end_x; x++ )
{
place_x = start_x + x;
switch(LcdBlRvMgr::getPixelStatus( convertFromReal( place_x ), convertFromReal( place_y )))
{
case ON:
fgPixelData[y][x] = pixelData[place_y][place_x];
break;
case OFF:
fgPixelData[y][x] = 0;
break;
case REVERSE:
fgPixelData[y][x] = 255 - pixelData[place_y][place_x];
break;
}
}
}
#if 0
int x, y;
int end_y, end_x;
if(updateRegion == PART)
{
start_x = convertToReal( start_x );
start_y = convertToReal( start_y );
end_x = start_x + getRealWidth102();
end_y = start_y + getRealHeight64();
}
else
{
end_y = getRealHeight240();
end_x = getRealWidth320();
}
for( y = start_y; y < end_y; y++ )
{
for( x = start_x; x < end_x; x++ )
{
switch(LcdBlRvMgr::getPixelStatus( convertFromReal( x ), convertFromReal( y )))
{
case ON:
fgPixelData[y - start_y ][x - start_x ] = pixelData[y][x];
break;
case OFF:
fgPixelData[y - start_y ][x - start_x ] = 0;
break;
case REVERSE:
fgPixelData[y - start_y ][x - start_x ] = 255 - (int)pixelData[y][x];
break;
}
}
}
#endif
}
