MatrixMultiplication | little playground , to test and try the benefits | GPU library

Turns out no matter how much I searched, I couldn't find the answer because I didn't know the word "overload." Thank you Jimi for providing the answer in the comments. I overloaded the function and it works! Here's my code, in case anyone's wondering. Thanks again.

In MatrixMultiplication function, the array q is allocated on the stack. So at function exit, its memory is freed and matrix datas may be replaced with other values. The solution is to allocate memory with malloc so it is kept even outside of the function.

From your code, I just removed the variable q and replace ptr with a memory allocation :

In order for this function to work:

This line isn't doing what you need:

There are multiple possible causes for this: Either you did not install OpenMP with the compiler or you made a mistake in the Code::Blocks configuration. Anyways go through the following steps and you should be able to fix it. You seem to be working on Windows but I also added remarks on how to do it on Linux.

1) You will need a compiler that comes with OpenMP. For Windows download TDM-GCC preferably the 64-bit executable (second file) and install it. Make sure you select OpenMP in the component tab: Components > gcc (TDM current: ....) > OpenMP (the last entry). Linux already comes with GCC so just open the command line and get yourself OpenMP by typing sudo apt-get install libomp-dev in the terminal.

2) You will have to configure the Compiler in Code::Blocks now: Go to Settings > Compiler under Selected Compiler select GNU GCC Compiler and click Copy and type in a convenient name for your new compiler such as TDM-GCC Compiler. Go to Toolchain executables and browse the directory for all the Program files (C compiler, C++ compiler, Linkers, Debugger, Resource compiler, Make) you should be able to find them in C:\TDM-GCC-64\ ...\bin on a Windows machine if you chose the default installation. I can't tell you the precise sub-directory as I working on a Linux machine but you should be able to find it pretty easily. Depending on your installation there might be two folders for 32- and 64-bit. For Linux this step is not necessary.

4) Then you need to set the Linker settings. This can be done for all projects (which I would not recommend) by doing the following steps in the aforementioned menu or for your current project by clicking Project > Build options. Go to Linker settings of the corresponding configuration (Debug or Release) and click Add under Link libraries. On a Windows machine you will need to browse a file called libgomp-1.dll (32-bit) or libgomp_64-1.dll (64-bit) which should be located in the same folder as the aforementioned Program files. Under Linux instead choose -lgomp under Other linker options.

3) Set the compiler flag -fopenmp (for all the projects or only the current one) by going to Compiler settings > Other compiler options and typing in there -fopenmp.

4) Test it with a program like the OpenMP "Hello World".

Isn't this method a class method

No, Objective-C initializers are instance methods. You have to create an instance with alloc. So

The plan should be printed if you remove the print() calls.

print() also triggers an execution.

The "difference" (or really a similarity, depending on how you think about it) between an array of int and an array of BigInteger is that when you make a new array of integers, all the integers exist (and are zero) while a new array of BigInteger is filled with null. Of course add cannot be called on null.

There are different fixes, for example you could fill the matrix C with zeros, or you could modify the multiplication so that it does not read from C at all but instead sums into a local variable which you initialize to zero.

I think you are doing it wrong with the indexing. the *offsetM3* should be equal to *i\*COLS_M3+j* , the *offsetM1* should be equal to *i\*COLS_M1+k* , and *offsetM2* to *k\*COLS_M2+j*.

Write the matrices on a paper and do the maths, and then write the matrices in an array like there are in memory , and then multiply them, then you will see the indexing pattern. Remember, every thread(work-item) is for one element of the new matrix. If you change the index of the new matrix through the for loop, you are not following the logic one work item for one matrix element, and you should consider another logic if you want it that way. Hope this helps