IPC - MultiProc_setBaseIdOfCluster()

Ralf Goebel

Hello,

I'm using hardware with two C6678 devices. I want to use IPC with SRIO using a single executable. The IPC Users Guide explains how the processor ID can be changed during runtime by using a startup function. Because my system now has two clusters, I additionally tried to call MultiProc_setBaseIdOfCluster() in my startup function.

First, there seems to be a bug in the implementation of the function:

    Assert_isTrue(((baseId + ti_sdo_utils_MultiProc_numProcsInCluster)
            < ti_sdo_utils_MultiProc_numProcessors), 
            ti_sdo_utils_MultiProc_A_invalidMultiProcId);

baseId is 8 for the second chip, ti_sdo_utils_MultiProc_numProcsInCluster is 8 and ti_sdo_utils_MultiProc_numProcessors is 16. Therefore the condition will fail.

Second, it doesn't seem to be possible to change the clusters base ID during runtime. I think it has to be valid in the RTSC configuration, before calling MultiProc.setConfig. But what would be a use case for MultiProc_setBaseIdOfCluster()?

Is there any chance to use a single binary for both processors? It would be annoying to have to build two binaries with the only difference in the clusters base ID.

Thanks,
Ralf

over 12 years ago

0 Ramsey over 12 years ago

TI__Genius 12025 points

Ralf,

Looks like a bug in the assert. I will report this.

The call to MultiProc_setBaseIdOfCluster() must be called early. Try calling it from a reset hook function. See the following link for documentation.

http://rtsc.eclipse.org/cdoc-tip/xdc/runtime/Reset.html

Unfortunately, you cannot build one executable for both devices because the MultiProc ids must always start at zero on each device.

~Ramsey

0 judahvang over 12 years ago in reply to Ramsey

TI__Mastermind 32475 points

Ralf,

Just a slight correction. The MultiProc ids do not have to always start at zero, however, I do recall that there are other reasons why you cannot build a single executable for both C6678. I'm pretty sure that you have to build at least 2 executables but you can give it a try if you want to experiment.

Judah

0 Ralf Goebel over 12 years ago in reply to judahvang

Genius 3715 points

Another reason could be a different ownerProcId value for the SharedRegion. But maybe it could be changed with SharedRegion_setEntry().

But I still don't see any use for MultiProc_setBaseIdOfCluster(). I think MultiProc.baseIdOfCluster must be set before calling MultiProc.setConfig() within the RTSC configuration.

I will work with two executables for now.

Thanks,
Ralf

0 Andrey Lisnevich over 12 years ago in reply to judahvang

Genius 3305 points

Hi Judah,

Personally I see that baseIdOfCluster is not used by ti.transport.ipc.srio.transports module. Everything depends on SRIO configuration done outside of this module. So it should be possible to use SRIO Transport with one executable (that has the same baseIdOfCluster) that runs on different chips.

Regards,

Andrey Lisnevich

0 Ramsey over 12 years ago in reply to Andrey Lisnevich

TI__Genius 12025 points

We have revisited the MultiProc_setBaseIdOfCluster() API and have concluded that this function does not work. As Ralph pointed out above, the config param MultiProc.baseIdOfCluster must be set correctly at config time. It is not possible to change the base ID of a cluster at runtime. Therefore, we are removing this API from the MulitProc module.

Thank you for bringing this to our attention.

~Ramsey

0 Ralf Goebel over 12 years ago in reply to Ramsey

Genius 3715 points

Hi Ramsey,

could you also take a look at the following issue?
http://e2e.ti.com/support/dsp/c6000_multi-core_dsps/f/639/p/284083/996887.aspx

I'm not sure if this is a bug in IPC_attach() or in the SRIO transport driver.

Thanks,
Ralf

0 Ramsey over 12 years ago in reply to Ralf Goebel

TI__Genius 12025 points

Ralf,

I've notified the team working on SRIO to help out on the thread.

~Ramsey

0 Ralf Goebel over 12 years ago in reply to Ramsey

Genius 3715 points

OK thanks,

Ralf

Processors

Processors forum

IPC - MultiProc_setBaseIdOfCluster()