Prepared by Oussama Jebbar Threads COEN 346 Prepared by Oussama Jebbar

Outline General Context Threads Creation: Runnables Threads Creation: Callables Synchronization

Threads The smallest execution unit Shared Memory They can Challenges Execute a process (sometimes even periodically) Return values Challenges Communication (synchronous Vs asynchronous) Scheduling Control access to shared data

Threads Creation: Runnables JAVA

Threads Creation: Callables JAVA

Threads Creation Python import threading #define a class for your thread class myThread (threading.Thread): #constructor def __init__(self): #put constructor logic here #run method that contains the thread routine def run(self): #put the logic of your routine here # Create new threads thread1 = myThread() thread2 = myThread() # Start new Threads thread1.start() thread2.start()

Mutexes in C++ Used to control access to shared data To prevent data from being altered simultaneously To make sure all threads access the right data (variable holding the right value) Include <mutex> use methods such as: lock or try_lock to make sure only one thread accesses the block of code Use the method unlock to release it after the thread is finished

Mutex Example

Common Threads Control Utilities join(): a method used to make a thread wait for another one to finish (Rendez-vous) detach(): used in C++ to tell a thread that it does not have to wait for another one Some frameworks and programming languages include suspend/resume utilities

Exercise Fork-join model is a multithreaded implementation of a divide and conquer algorithm in which the sub-problems can be solved independently In this exercise we will have to implement the Strassen algorithm for matrix multiplication in a fork-join model Strassen algorithm for matrix multiplication Let A and B two square matrices, and C is their product If A and B are not 2n * 2n matrices, fill the missing rows and columns with zeros to make them so Partition A and B into equally sized blocks of size 2n-1 * 2n-1 Calculate the following matrices M1 = (A11 + A22)*(B11 + B22) M2 = (A21 + A22)*B11 M3 = A11*(B12-B22) M4 = A22*(B21-B11) M5 = (A11+A12)*B22 M6 = (A21-A11)*(B11+B12) M7 = (A12-A22)*(B21+B22) Calculate the four blocks that compose the product C as follows C11 = M1 + M4 – M5 + M7 C12 = M3 + M5 C21 = M2 + M4 C22 = M1 – M2 + M3 + M6 Adjust the size of the resulting matrix (in case zeros were added in the first step of the algorithm) Q1: Propose a fork-join model to implement this algorithm Q2: What is the number of threads that the fork-join model you proposed use to calculate the product of two matrices of size n? Q3: Knowing that the machines you have in the lab can only run up to ten threads concurrently, can you propose a solution to make your implementation work with inputs of a big size

Hint! Decoupling function/method invocation from its execution Allows for a better use of resources and reuse of threads (thread pooling) Active Object pattern