计算机代写 SOFT3410 Tutorial 9 OpenMP – cscodehelp代写
SOFT3410 Tutorial 9 OpenMP
Introduction to OpenMP and how it can be used to transform the performance of our applications
Question 1: Hello OpenMP
Ensure that you have the OpenMP libraries installed on your system. You can check this by construct- ing the following OpenMP HelloWorld program.
#include
#include
int main(void) {
#pragma omp parallel printf(“Hello, world.
”); return 0;
}
You should observe your application outputting “Hello, world.” to the terminal based on the
number of cores your device has.
Question 2: Parallel Sum
By now, you should be completely capable to write a parallel sum program using pthreads. In this exercise, we want you to use the code below and the correct OpenMP preprocessor directives to transform this single threaded program into a threaded version.
int main() {
//data here
int64_t sum = 0;
for(size_t i = 0; i < end; i++) {
sum += data[i];
}
printf("%ld
", sum); return 0;
}
1
1 2 3 4 5
intersect(box1, box2):
return (box1.x < (box2.x + box2.width)) &&
((box1.x + box1.width) > box2.x && ((box1.y < (box2.y + box2.height) && ((box1.y + box1.height) > box2.y)
Construct a SDL application that will recolour moving boxes to show if they have collided.
SOFT3410 OpenMP
• Try to observe what OpenMP is doing by running the preprocessor only, what do you observe?
• What are the advantages and disadvantages to this approach? When would we prefer this ap- proach over pthreads?
Question 3: Parallel Intersection
Often in computer simulations and video games we want to check if two objects collide into each other. We are able to simply achieve this by constructing axis-aligned bounding boxes and comparing their intersections with each other.
Concurrency Page 2 of 3
Question 4: Translating matrix multiplication – Homework
Time to revisit matrix multiplication again! This time, you will need to modify the following code to utilise OpenMP. Consider how you could refactor this code to make it work with OpenMP.
void multiply(const float* mata, size_t mata_width, size_t mata_height, const float* matb, size_t matb_width, size_t matb_height,
float** result_mat, size_t* res_width, size_t* res_height) {
if(result_mat) {
*res_width = matb_width;
*res_height = mata_height);
*result_mat = calloc(mata_height * matb_width
* sizeof(float));
for(size_t y = 0; y < mata_height; y++) { for(size_t x = 0; x < matb_width; x++) {
for(size_t k = 0; k < mata_width; k++) { (*result_mat)[(y * matb_width) + x] +=
(mata[(y * mata_width) + k] *
matb[(k * matb_width) + x]);
}
} }
} }
• Refactor the code to use multiple threads with OpenMP
• Construct a set of test cases of different sizes (try 32x32, 64x64, 128x128 ... 2048x2048 and larger)
• Record these results and use gnuplot to graph the data
• Write a conclusion and explain your observations
For your submission, your code submission must be uploaded onto USYD Github under the repo name soft3410hw9. We will only consider the last submission before 11:59pm, 9th November, 2020.
For your submission, your report must be submitted to canvas via TurnitIn. You must submit your report to canvas portal by 11:59pm, 9th November, 2020.
SOFT3410 OpenMP
Concurrency Page 3 of 3
