Scribd
Upload
472 views2 pages

SIMD and MIMD

This document provides examples of parallel matrix multiplication algorithms for SIMD and MIMD machines. For SIMD machines, it shows a pseudocode that assigns each processing element to calculate one column of the product matrix in parallel. For MIMD machines, it shows a pseudocode that forks multiple processes, with each process calculating one column of the product matrix independently and in parallel. On both architectures, the algorithms take advantage of parallelism to calculate all elements of the product matrix simultaneously.

Uploaded by

BravoYusuf
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
472 views2 pages

SIMD and MIMD

This document provides examples of parallel matrix multiplication algorithms for SIMD and MIMD machines. For SIMD machines, it shows a pseudocode that assigns each processing element to calculate one column of the product matrix in parallel. For MIMD machines, it shows a pseudocode that forks multiple processes, with each process calculating one column of the product matrix independently and in parallel. On both architectures, the algorithms take advantage of parallelism to calculate all elements of the product matrix simultaneously.

Uploaded by

BravoYusuf
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

CS-421 Parallel Processing BE (CIS) Batch 2004-05

Handout_11

Programming Examples for SIMD and MIMD Machines

1. SIMD Programming
Below is presented a matrix multiply pseudocode for an SIMD computer. This assumes that
calculation of one column of product matrix is assigned to one PE. That is, for N x N product
matrix, we require N PEs.
N
/* Matrix multiply, C: = A.B. Compute elements of C by cij = ∑ aik bkj */
k =1

for i := 1 step 1 until N

begin /*Compute one row of C. */


/* Initialize the sums for each element of a row of C */
C[i, j] := 0, (1 ≤ j ≤ N);
/* Loop over the terms of the inner product */
for k := 1 step 1 until N

/* Add the kth inner product term across columns in parallel. */


C[i, j] := C[i, j] + A[i, k] * B[k, j], (1 ≤ j ≤ N);
/* End of product term loop */

end /* of all rows */

For parallel algorithms, it’s also important to consider the data access requirements i.e. the data
layout in memory of the machine. For instance, this SIMD algorithm requires that PEk must have
access to kth-column of matrix B while access of whole matrix A is required by every PE.

2. MIMD Programming
Below is presented a matrix multiply pseudocode for a MIMD computer (Multiprocessor).
private i, j, k;
shared A[N,N], B[N,N], C[N,N], N;

/* Start N – 1 new processes, each for a different column of C. */


for j := 1 step 1 until N-1 fork DOCOL;

/* The original process reaches this point and does the processing for Nth column */
j := N;

DOCOL: /* Executed by N processes, each doing one column of C. */


for i := 1 step 1 until N

begin /*Compute one row of C. */

/* Initialize the sums for each element of a row of C */


C[i, j] := 0,
Page - 1 - of 2
CS-421 Parallel Processing BE (CIS) Batch 2004-05
Handout_11
/* Loop over the terms of the inner product */
for k := 1 step 1 until N

/* Add the kth inner product term */


C[i, j] := C[i, j] + A[i, k] * B[k, j];

/* End of product term loop */


end /* of all rows */
join N;

Of particular importance is the join command here. It has an implicit counter associated with it
that starts off at a count = 0. This counter is shared by all processes (processors) executing the
code. A process executing join increments this counter and compares it to N, the argument to
join. If the value of count is not N, then this process kills itself. The process finding this count
to be N, continues execution beyond join command. This way only one process continues
beyond join and all other processes get terminated.

*****

Page - 2 - of 2

You might also like