Scribd
Upload
3 views

Operating System Threads Lab 3

Uploaded by

Prahalad V
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
3 views

Operating System Threads Lab 3

Uploaded by

Prahalad V
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

LAB-3

Prahalad Vijaykumar
2003128

Using mat .c to generate matrices of random size of m*n and n*k and then my_matrix_multiply.c first accepts
three flags -a for matrix a and -b for matrix b and -p for number of threads and then computes the matrix
multiplication of the two.

Using 1 D array to represent a 2d array and then using a thread to execute each row if the threads are greater
than or equal to number of rows or else threads repeatedly execute the rows.

And the issue faced is when the multiplication between threads are ununiform this is because of access of all
threads at the same time which will lead to skipping of the iteration.

Eg 0,1,1,3…this has happened because two threads have accessed 1 at nearly the same time the iter
variable(reference of iter variable in the code.). so to avoid this there are two cases compare and swap principle
and mutex lock. Since it involves using atomic variable and mutex lock is easier to implement so I implemented
mutex lock.
Mutex lock works just like a lock in the sense if two threads come towards accessing the same function then it
gives access to one of the thread and the other thread has to wait until the previous thread completes its
execution in c mutex_thread_lock() and unlock() are used respectively to implement the same

Observation :
For smaller matrix it is better to work normally than performing multithreading
For larger matrix of size greater than 2500 I found a major difference when comes to execution time . though it
takes a lot of time to executed.

prahaladvijaykumar$ ./my_matrix_multiply -a i -b o -t 1

213255720
123437373

prahaladvijaykumar$ ./my_matrix_multiply -a i -b o -t 100


207723403
56871058
Size of i and o is 2500*2500

prahaladvijaykumar$ ./my_matrix_multiply -a i -b o -t 1

7
549

Size of i and o is 10*10

The first output is the time taken without multithreading and the second is with multithreading
Plot for multiplying 40*40 matrix for random threads from 1-100.

100*100 matrix result on increasing threads from 1-100

Each time is taken by running each number of thread 100 times and then taking an average and reporting in the
graph and plotted it using excel.

512
354
529
359
474
385
517
320
385
558
513
465
354
360
526
496
508
492
494
502
338
434
531
467
520
514
442
513
490
536
527
527
470
528
569
477
434
586
347
360
515
572
581
544
501
571
521
495
368
373
367
439
442
374
482
493
528
571
498
540
432
386
564
506
454
455
591
469
473
499
593
503
503
517
505
494
505
458
548
464
477
539
436
524
471
532
468
435
457
488
496
510
648
549
563
518
480
462
491
503

If u take a bigger matrix then it can be observed that a decreasing graph will be obtained

You might also like