Are Matrix working fine?

[MLobatoR]

Hi everyone,

I'm just making a real small example to try and understand how your object Matrix works. I came across something weird...

Here's my example:

Matrix2DInt matrix = new Matrix2DInt(3, 2);
matrix.set(0, 0, 1);
matrix.set(0, 1, 2);
matrix.set(0, 2, 3);
		
matrix.set(1, 0, 4);
matrix.set(1, 1, 5);
matrix.set(1, 2, 6);
		
System.out.println(matrix.toString());

I don't know how but the output is wrong...

MatrixInt <2 x 3>
1 2 4 
4 5 6

Actually I do know.. I investigated a little more your implementation and if I got it right you are transforming this matrix into a StorageFormat, an array. But when trying to find out the position of each number the math seems wrong. Here is your method (I'm just copying and pasting it here):

public void set(int i, int j, int value) {
     storage[StorageFormats.toRowMajor(i, j, M)] = value;
}

Checking this method on class StoreFormat:

public static int toRowMajor(int i, int j, int yMax) {
     return (i * yMax) + j;
}

So when I'm trying to set line 0 column 2 the position in thos StorageFormat is 0 * 2+2 = 2 and when I'm trying to set line 1 column 0 1 * 2 + 0 = 2. Same index...

If I'm seeing this the wrong way can you explain how Matrix work. How should I initialize it? And how to set values please?

Thank you in advance,
ML

[jjfumero]

It seems there is a bug in Tornado related to the compiler replacement phase for the matrix API. We will analyze this case. Which backend are you using? Juan

…

On Sat, 29 Oct 2022, 17:29 MLobatoR, ***@***.***> wrote: Hi everyone, I'm just making a real small example to try and understand how your object Matrix works. I came across something weird... Here's my example: Matrix2DInt matrix = new Matrix2DInt(3, 2); matrix.set(0, 0, 1); matrix.set(0, 1, 2); matrix.set(0, 2, 3); matrix.set(1, 0, 4); matrix.set(1, 1, 5); matrix.set(1, 2, 6); System.out.println(matrix.toString()); I don't know how but the output is wrong... MatrixInt <2 x 3> 1 2 4 4 5 6 Actually I do know.. I investigated a little more your implementation and if I got it right you are transforming this matrix into a StorageFormat, an array. But when trying to find out the position of each number the math seems wrong. Here is your method (I'm just copying and pasting it here): public void set(int i, int j, int value) { storage[StorageFormats.toRowMajor(i, j, M)] = value; } Checking this method on class StoreFormat: public static int toRowMajor(int i, int j, int yMax) { return (i * yMax) + j; } So when I'm trying to set line 0 column 2 the position in thos StorageFormat is 0 * 2+2 = 2 and when I'm trying to set line 1 column 0 1 * 2 + 0 = 2. Same index... If I'm seeing this the wrong way can you explain how Matrix work. How should I initialize it? And how to set values please? Thank you in advance, ML — Reply to this email directly, view it on GitHub <#213>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABKX2BPYZBXGTEHJ5SZBBU3WFU7FBANCNFSM6AAAAAARR2XSSQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[stratika]

hi @MLobatoR, if you build the latest develop branch and you run tornado --version, it will display the installed backend.

For example:

version=0.15-dev
branch=X
commit=Y

Backends installed: 
	 - opencl

I suggest to try rerun with the latest develop if possible, since I see you are running with JDK8 which is deprecated from TornadoVM. Also, I believe it makes sense to open an issue from this discussion so that you fill the template of the issues and for us to be able to use your filled information to reproduce the problem and provide a fix.

[jjfumero]

@MLobatoR , we got the bug in Tornado. It is now under code review. As @stratika mentioned. JDK8 is no longer supported in TornadoVM. Also, the Graal version we support is 22.2.0 for JDK 11 and 17.

[MLobatoR]

Ok, I will upgrade JDK and wait for the new release from tornado with the bug fixed. Can you tell me when you release the fix please, so I can try again?

Thank you in advance,
ML

[jjfumero]

@MLobatoR , the fix is already merged.

Please, notice that we are transitioning to a new API in TornadoVM, so the latest commit (7cbffcf) includes a first phase of this new transition.

Some changes:

• TaskSchedule -> TaskGraph
• streamIn -> transferToDevice(mode, data, ... ); This call is mandatory
• streamOut -> transferToHost(data...)

See an example: https://github.com/beehive-lab/TornadoVM/blob/master/examples/src/main/java/uk/ac/manchester/tornado/examples/compute/MatrixMultiplication2D.java#L78-L82

[MLobatoR]

Hi @jjfumero and @stratika,

Matrixes are working fine in this version.

Thank you,
ML

[jjfumero]

This has been solved. This message is just to mark the solution.