Description
This assignment is to practice programming in the new parallel programming language, Chapel. You are
going to continue with Lab 6’s work. Specifically, you are going to implement a matrix-multiplication program
and a couple of producer-consumer programs. CS515 students need to do an extra part (see below). This
assignment carries a total of 10 points.
1. Matrix Multiplication
The file mmul.c contains a matrix multiplication program in C. Create a Chapel version of this program,
mmul.chpl. The specific requirements are:
• Represent the array dimension size parameter, N, by a configurable constant in the Chapel program.
Set its default value to 8.
• Define a two-dimensional domain D to represent the array index set, {0..N-1} × {0..N-1}. Declare
the three arrays, a, b, and c, over this domain.
• Parallelize all loops in mmul.c. Convert them to array operations, reduction operations, and/or parallel loops. The resulting Chapel program should have just one loop, a forall parallel loop for the
multiplication section.
• Use the script command to create a run script of your program running with N=8 (default), N=16,
N=32, and N=64.
linux> script mmul-script.txt
Script started, file is mmul-script.txt
linux> ./mmul -nl 1
total = 3584 (should be 3584)
linux> ./mmul –N=16 -nl 1
total = 61440 (should be 3584)
…
linux> exit
Script done, file is mmul-script.txt
2. Producer-Consumer
The file circQueue.chpl contains a representation of circular queue data structure. The queue items are
stored in a buffer array; when the end of the buffer is reached, it continues back from the beginning.
Read and understand this program. Pay special attention to the sync variable declarations, especially the
buffer array, which means every array element is a self-sync item, allowing only alternating reads and writes.
Your task is to write several produce-consumer programs. Note: You should not modify the provided
circQueue.chpl program.
Version 1: prodcons1.chpl
In this version you are to write a producer routine to add items to a queue, and a consumer routine to
remove items from the same queue. The specific requirements are:
1
• The total number of items to add and remove is represented by a configurable constant numItems, with
a default value of 32.
• Both routines print out a line for each item added or removed, in the following form:
Producer added item 28 to buf[7]
consumer removed item 21 to buf[0]
• The producer and the consumer routines must run concurrently. (When you run the program, you
should see producer messages and consumer messages interleave.)
Version 2: prodcons2.chpl
In this version you are to modify the previous version to allow multiple copies of both producer and consumer
routines to be created and run concurrently. Additional requirements are:
• Both the producer and consumer functions take an integer parameter, serving as an ID to allow
differentiation among multiple copies of the same function. So messages from this program will look
like
Producer 2 added item 28 to buf[7]
consumer 4 removed item 21 to buf[0]
• The numbers of producers and consumers are represented by configurable constants, numProds and
numCons, respectively; both have a default value of 2.
• The workload of adding numItems items to the queue is evenly partitioned among the numProds
producers; each handles numItems/numProds items.
• Similarly, the workload of removing numItems items from the queue is evenly partitioned among the
numCons consumers; each handles numItems/numCons items.
• Your program should verify that both numProds and numCons evenly divide numItems, and reject
non-conforming values.
• All producers and consumers should run concurrently.
Version 3: prodcons3.chpl [CS515 Students] (3 extra points for CS415 students)
In this version you are going to further enhance the program by using dynamic workload distribution.
Specifically,
• There is no workload partition. All copies of the produce function will compete to add as many items
to the queue as possible; each is capable of adding all numItems items.
• The same is true with the consumers.
• As such, there is no need to require that both numProds and numCons evenly divide numItems.
• However, there is a new challenge that you need to handle: each producer and consumer needs to know
when to terminate.
Test your programs with different parameter values. Create a run script for showing some test cases.
Submission
Make a zip file containing your programs and scripts. Use the Dropbox on the D2L site to submit your
assignment file.
2
