Finish Lab 3

Lab3/examples/1-no_join-fixed.c

@@ -0,0 +1,17 @@
+/**To compile, don't forget to add -lpthread. Might not work without that */
+#include<pthread.h>
+#include<stdio.h>
+
+void* run (void* arg){
+    printf("Hello from run\n");
+    return NULL;
+}
+
+int main(){
+    pthread_t thread; // variable to store the reference to the thread
+    pthread_create(&thread, NULL, &run, NULL);
+    printf("In main"); // This section will be executed in parallel
+    pthread_join(thread,NULL); // necessary for waiting for the thread to finish
+    printf("In main 2");
+    return 0;
+}

Lab3/examples/2-sleep no_join-fixed.c

@@ -0,0 +1,23 @@
+#include<pthread.h>
+#include<stdio.h>
+#include<stdlib.h>
+#include<unistd.h>
+#include<time.h>
+#include<sys/time.h>
+
+void* run (void* arg){
+    printf("Hello from run\n");
+    return NULL;
+}
+
+int main(){
+    pthread_t thread;
+    pthread_create(&thread, NULL, &run, NULL);
+    // While the sleep() function will appear to provide synchronization,
+    // it is incorrect, and if used for this purposes instead of join
+    // or synchronization mechanism - the grade for the assignment will be
+    // reduced
+    sleep(2);
+    printf("Back in main");
+    return 0;
+}

Lab3/examples/3-losing_track_of_threads-fixed.c

@@ -0,0 +1,24 @@
+#include<pthread.h>
+#include<stdio.h>
+#include<stdlib.h>
+#include<unistd.h>
+#include<time.h>
+#include<sys/time.h>
+
+void* run (void* arg){
+    sleep(10-(int)arg);
+    printf("Hello from run\n");
+}
+
+int main(){
+    pthread_t thread[10];
+    int i = 0;
+    for (i = 0 ; i < 10; ++i)
+        pthread_create(&thread[i], NULL, &run,(void*)i);
+
+    // If you plan to use the results of all the threads, consider
+    // using join for all of the threads
+    for (i = 0 ; i < 10; ++i)
+        pthread_join(thread[i],NULL);
+    printf("In main");
+}

Lab3/examples/4-passing_data_wrongly.c

@@ -0,0 +1,21 @@
+#include<pthread.h>
+#include<stdio.h>
+
+void* run (void* arg){
+    int i = (int) arg;
+    printf("Hello from run, arg is %i\n",i);
+}
+
+int main(){
+    pthread_t thread[10];
+    int i = 0;
+    for (i = 0 ; i < 10; ++i)
+        pthread_create(&thread, NULL, &run,(void*)i);
+    // Notice how warnings are generated. This can be resolved by properly
+    // allocating space for the thread parameters.
+    // Note, that if you are using stack variable, the values might be corrupted
+    for (i = 0 ; i < 10; i++){
+        pthread_join(thread[i],NULL);
+    }
+    printf("In main");
+}

Lab3/examples/4-passing_data_wrongly_fixed.c

@@ -0,0 +1,31 @@
+#include<pthread.h>
+#include<stdio.h>
+#include <stdlib.h>
+
+typedef struct thread_args{
+    char letter;
+    int id;
+}Args;
+void* run (void* arg){
+    Args* argg = (Args*) arg;
+    int i = argg->id;
+    printf("Hello from run %d\n", i);
+    return NULL;
+}
+
+int main(int argc, char** argv){
+    pthread_t thread[10];
+    int i = 0;
+    for (i = 0 ; i < 10; i++){
+        Args* argg = (Args*) (malloc(sizeof(Args)));
+        argg->letter = 'q';
+        argg->id = i;
+        pthread_create(&(thread[i]), NULL, &run,(void*)argg);
+    }
+    for (i = 0 ; i < 10; i++){
+        pthread_join(thread[i],NULL);
+    }
+    // Free the resources somewhere, for long running program. Or let the OS handle that
+    printf("In main");
+    return 0;
+}

Lab3/examples/5-pthread_detach.c

@@ -0,0 +1,19 @@
+#include<pthread.h>
+#include<stdio.h>
+
+void* run (void* arg){
+    printf("Hello from run\n");
+    return NULL;
+}
+
+int main(){
+    pthread_t thread;
+    pthread_create(&thread, NULL, &run, NULL);
+    pthread_detach(thread);
+    // This command will make thread detached. This means the resources will be released 
+    // upon the thread's completion - calling return
+    // However, detached threads cannot be joined, which means if you care about the result
+    // of the thread execution - you should not be using pthread_detach
+    printf("In main");
+    return 0;
+}

Lab3/main.c

@@ -1,8 +1,42 @@
-#include<stdio.h>
+/**
-#include<sys/time.h>
+ * Question 1: Run the program with the problem size of 1000 and 10 threads, what is the approximate speedup you are achieving?
-#include<stdlib.h>
+ *
-#include<unistd.h>
+ * Answer 1: On my home machine single threading took 0.640239 seconds and multithreading took 0.194644 seconds.
-#include<string.h>
+ * On the lab machine single threading took 0.697472 seconds and multithreading took 0.114415 seconds.
+ * This means that the my home machine is 3.29 times faster and the lab machine is 6.10 times faster.
+ *
+ * Question 2: Is there a problem size / number of threads combination that slows down the computation process? Why do you think it is happening?
+ *
+ * Answer 2: There is a problem with having too many threads for the number of hardware threads available. Increasing past this only
+ * increases the overhead of creating and managing the threads. This is because the threads are not running in parallel and are instead
+ * being switched between by the OS. At lower matrix sizes the cost of creating threads and managing them is greater than the cost of
+ * just doing the computation in a single thread, so any combination where threads > the number of hardware threads will be slower, and
+ * as the matrix size approaches 1, than the greater effect thread creation and management will have on the speed of the program.
+ *
+ * Question 3: What is the minimum size of the problem that benefits from creating an extra thread?
+ *
+ * Answer 3: The lowest size on my home machine that consistently benefited from an extra thread was 150, but this can change depending
+ * on the specifications of the machine.
+ *
+ * Question 4: Does using the threads always improve execution duration?
+ *
+ * Answer 4: No, as the number of threads increases past the number of hardware threads available the execution duration increases
+ * due to managing the threads, as well as low size matrices where the cost of creating and managing threads is greater than the
+ * cost of just doing the computation in a single thread.
+ *
+ * Question 5: Guesstimate and comment on the nature of growth of the speedup with the number of threads – is it linear, exponential, are there any limits?
+ *
+ * Answer 5: The speedup is linear up to the number of hardware threads available, given a large enough matrix size.
+ * Given a matrix of size x, the speedup of using y threads is approximately x/y, up to the number of hardware threads available.
+ * After this the speedup will decrease as the number of threads increases, due to the overhead of creating and managing threads.
+ **/
+
+#include <stdio.h>
+#include <sys/time.h>
+#include <time.h>
+#include <stdlib.h>
+#include <string.h>
+#include <pthread.h>
 
 #define MAXN 5
 
@@ -82,6 +116,23 @@ void *multiply_matrices_threaded(void *threadParams) {
      * threading capacity and parallelize the computation in such a
      * way that a thread computes result per one or more rows
      */
+    ThreadParams *t = (ThreadParams *) threadParams;
+    int N = t->size;
+    int row = t->row_index;
+    int column = 0;
+    int temp_result = 0;
+    while (row < N) {
+        column = 0;
+        while (column < N) {
+            temp_result = 0;
+            for (int i = 0; i < t->size; i++) {
+                temp_result = temp_result + t->first_array[row][i] * t->second_array[i][column];
+            }
+            t->result[row][column] = temp_result;
+            column = column + 1;
+        }
+        row += t->max_threads;
+    }
 }
 
 int main(int argc, char **argv) {
@@ -121,6 +172,22 @@ int main(int argc, char **argv) {
      * Write your code to create and use max_threads here, such that the threaded_result 
      * is populated with the result of the computation.
      */
+    thr->result = threaded_result;
+    pthread_t threads[max_threads];
+    for (int i = 0; i < max_threads; i++) {
+        ThreadParams *params = (ThreadParams *) malloc(sizeof(ThreadParams));
+        params->first_array = array1;
+        params->second_array = array2;
+        params->result = threaded_result;
+        params->row_index = i;
+        params->size = size;
+        params->max_threads = max_threads;
+        pthread_create(&threads[i], NULL, &multiply_matrices_threaded, (void *) params);
+    }
+    for (int i = 0; i < max_threads; i++) {
+        pthread_join(threads[i], NULL);
+    }
+
     gettimeofday(&end, NULL);
     //The next line is inspired by https://linuxhint.com/gettimeofday_c_language/
     microseconds = (end.tv_sec * 1000000 + end.tv_usec) - (begin.tv_sec * 1000000 + begin.tv_usec);
@@ -129,7 +196,7 @@ int main(int argc, char **argv) {
 
     if (check_if_matrices_differ(result, threaded_result, size) != 0) {
         printf("Threaded result differ from single core computation, error\n");
-        exit(0);
+        exit(1);
     }
     return 0;
 }

Lab3/try-for-error.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+cond=0
+while [ $cond -eq 0 ]
+do
+    ./build/Lab3 10 10 &> output.txt
+    cond=$?
+    echo $cond
+done