CSC/ECE 506 Spring 2010/ch 2 maf

Supplement to Chapter 2: The Data Parallel Programming Model

This chapter is a supplement to Chapter 2 of the Solihin textbook. The textbook covers the shared memory and message passing parallel programming models. However, it does not address the data parallel model

Comparing the Data Parallel Model with the Shared Memory and Message Passing Models

Comparison between shared memory, message passing, and data parallel programming models (adapted from Solihin 2008, page 22).

Aspects	Shared Memory	Message Passing	Data Parallel
Communication	implicit (via loads/stores)	explicit messages	implicit
Synchronization	explicit	implicit (via messages)	implicit for SIMD; explicit for SPMD
Hardware support	typically required	none
Development effort	lower	higher	higher
Tuning effort	higher	lower

A Code Example

// Simple sequential code from Solihin 2008, page 25.

for (i = 0; i < 8; i++)
    a[i] = b[i] + c[i];
sum = 0;
for (i = 0; i < 8; i++)
    if (a[i] > 0)
        sum = sum + a[i];
Print sum;

// Data parallel implementation in C++ with OpenMP.

int main(void)
{
    double a[8], b[8], c[8], localSum[2];

    #pragma omp parallel for
    for (int id = 0; id < 2; id++)
    {
        int local_iter = 4;
        int start_iter = id * local_iter;
        int end_iter = start_iter + local_iter;

        for (int i = start_iter; i < end_iter; i++)
            a[i] = b[i] + c[i];
   
        local_sum[id] = 0;
        for (int i = start_iter; i < end_iter; i++)
            if (a[i] > 0)
                localSum[id] = localSum[id] + a[i];
    }

    double sum = localSum[0] + localSum[1];
    cout << sum;
}

// Data parallel implementation in C for CUDA.

__global__ void kernel(
    double* a,
    double* b,
    double* c,
    double* localSum)
{
    int id = threadIdx.x;
    int local_iter = 4;
    int start_iter = id * local_iter;
    int end_iter = start_iter + local_iter;
   
    for (int i = start_iter; i < end_iter; i++)
        a[i] = b[i] + c[i];
   
    local_sum[id] = 0;
    for (int i = start_iter; i < end_iter; i++)
        if (a[i] > 0)
            localSum[id] = localSum[id] + a[i];
}
 
int main()
{
    double a[8], b[8], c[8], localSum[2];
    kernel<<<1, 2>>>(a, b, c, localSum);
    double sum = localSum[0] + localSum[1];
    cout << sum;
}

C DATA PARALLEL IMPLEMENTATION IN FORTRAN

REAL A(8), B(8), C(8), LOCAL_SUM(2), SUM

FORALL ID = 1:2
    LOCAL_ITER = 4
    START_ITER = (ID - 1) * LOCAL_ITER + 1
    END_ITER = START_ITER + LOCAL_ITER - 1

    DO I = START_ITER:END_ITER
        A[I] = B[I] + C[I]
    END DO
 
    LOCAL_SUM[ID] = 0;
    DO I = START_ITER:END_ITER
        IF A[I] > 0 THEN
            LOCAL_SUM[ID] = LOCAL_SUM[ID] + A[I]
        END IF
    END DO
END FORALL

SUM = LOCAL_SUM[0] + LOCAL_SUM[1]
WRITE(*,*) SUM

References

• David E. Culler, Jaswinder Pal Singh, and Anoop Gupta, Parallel Computer Architecture: A Hardware/Software Approach, Morgan-Kauffman, 1999.
• Magne Haveraaen, "Machine and collection abstractions for user-implemented data-parallel programming," Scientific Programming, 8(4):231-246, 2000.
• W. Daniel Hillis and Guy L. Steele, Jr., "Data parallel algorithms," Communications of the ACM, 29(12):1170-1183, December 1986.
• Yan Solihin, Fundamentals of Parallel Computer Architecture: Multichip and Multicore Systems, Solihin Books, 2008.