Multi2MonoToken Tranformation errors about orcc HOT 27 CLOSED

Would you please tell me with which design you had errors wih the C backend? (I left the transformation OK with C backend for MPEG4SP, CBP and HEVC).
Concerning the instreturn I will check this soon. The C backend does not show the return so I could have missed his detail.

from orcc.

I have tested with epfl mpeg 4 sp. Orcc is complaining if this transformation is active. I have null pointer errors in xronos because the fsm is missing states...

from orcc.

I tested the mpeg4sp parallel decoder with the C backend without the multi2mono transformation and it does not work (compilation ok but display is not ok) !! I think there is an error some where in the epfl design. please check that and confirm the source of display error

from orcc.

For me it is working perfectly i have 588fps in my machine and the display is correct, are you in the latest revision? I am...

from orcc.

what is the bitstream to test? Khaled is using hit001.m4v which is correct with all previous MPEG4 SP decoder

from orcc.

I am using foreman_qcif_30.bit, so we are always talking about the RVC mpeg4sp decoder found on the RVC project...

from orcc.

it is not part of a conformance test streams we can try it, but you can also try hit001 which is part of conformance test stream and let us know if it works
__
Mickaël

Le 7 févr. 2013 à 16:34, Endri Bezati [email protected] a écrit :

I am using foreman_qcif_30.bit, so we are always talking about the RVC mpeg4sp decoder found on the RVC project...

—
Reply to this email directly or view it on GitHub.

from orcc.

could you send me the sequence i do not have it !

from orcc.

It is working correctly for me i see the video, 175fps...

from orcc.

There was a confusion between Serial and Parallel version... The serial version is tested on the CI server so it's working fine on many sequences. I've just remove the top network for the parallel version which is broken for a long time to avoid confusion...

But, next time, please give us the full qualified name (i.e. package and xdf name) of the application when you write the issue to avoid this lost of time. Keep in mind that we are not with you behind your computer so you have to be precise !

from orcc.

Endri, you told me yesterday that it was the epfl design (so certainly in the research project). for RVC project we already know that all designs are ok (mpeg4sp, cbp and hevc)

from orcc.

Ok for the design serialize in research epfl here are errors with the multi 2 mono, I ry to fix it as soon as possible

from orcc.

FYI in the second message I posted "I have tested with epfl mpeg 4 sp.". I do not understand you comment Herve...

from orcc.

There was a second qualified name in the package Research/src/ch/epfl/mpeg4/part2/Top_Parallel.xdf
__
Mickaël

Le 7 févr. 2013 à 17:19, Endri Bezati [email protected] a écrit :

FYI in the second message I posted "I have tested with epfl mpeg 4 sp.". I do not understand you comment Herve...

—
Reply to this email directly or view it on GitHub.

from orcc.

orcc/orc-apps@0afdc07

from orcc.

It should work with the new commit

from orcc.

may be we should also fix a way to put tickets. with such case a ticket would contain:

1. the nature of the error (code generation, C compilation or video display)
2. the exact name of he design
3. what ever the souce of the error a copy paste of the error message (don't worry if it is a long message we don't pay for that lol)
4. any other precision on the used version of orcc or orc-apps would be appreciated.

Such policy of making tickets has to be added to avoid all the time we lost (for example today) . what d you think/suggest?

from orcc.

Running the transformation with DEBUG enabled it gives some warnings on the transformation. In class ModifyProcessActionStore, in the caseInstLoad, there is a missing Load.

    public Object caseInstLoad(InstLoad load) {
        Var varSource = load.getSource().getVariable();
        Pattern pattern = EcoreHelper.getContainerOfType(varSource,
                Pattern.class);
        if (pattern != null) {
            Port testPort = pattern.getPort(varSource);
            if (port.equals(testPort)) {
                // change tab Name
                load.getSource().setVariable(tab);
                Expression indexInit = load.getIndexes().get(0);
                // FIXME: missing load here
                Expression indexFinal = irFactory.createExprBinary(
                        indexInit, OpBinary.PLUS,
                        irFactory.createExprVar(writeIndex),
                        irFactory.createTypeInt(32));
                Expression exprMask = irFactory
                        .createExprInt(bufferSize - 1);
                Expression maskValue = irFactory.createExprBinary(
                        indexFinal, OpBinary.BITAND, exprMask,
                        irFactory.createTypeInt(32));

                load.getIndexes().add(maskValue);
            }
        }
        return null;
    }

The writeIndex is a global variable and it should be loaded firstly. I think the testing only with the C backend is not enought.

from orcc.

True, we could test it as well on the LLVM backend which is not so permissive..

from orcc.

And another issue that i have toot is that it is not adding the transition to the fsm
EMF gives that error after the transfornation
Caused by: org.eclipse.emf.ecore.xmi.DanglingHREFException: The object 'newStateWriteuntagged_0 (untagged_0newWriteDone) --> init' is not contained in a resource.

this is the test actor

package hdl;

actor Gen()
==>
int Out:

action ==> Out:[[i: for int i in 0 .. 63]] repeat 64
end

end

from orcc.

transitions and states are now put in their containers. try now and tell me if the problem of transitions is resolved. I continue cleaning uses and def containment warnings ...

from orcc.

I cannot compile even in C now,

An internal error occurred during: "Launching hdl.test - C".
net.sf.orcc.df.FSM.addState(Lnet/sf/orcc/df/State;)V

from orcc.

I changed also the fsm and fsmImpl in the core did you update them ?

from orcc.

Ok, i clean up everything, it is working with C, but it does not work with Xronos yet, do not forget to add the missing load as i posted previously...

from orcc.

Is it fixed now ? Can you close the ticket ?

from orcc.

I think it's ok ! But we wait for Endri to confirm that before closing the ticket.

from orcc.

The transformation is working correctly after my quick fix 59816ed... I am closing the issue.

from orcc.