CCSv4.2 - RTSC profile "whole_program" outputs error

Hi Eddie,

Could you try updating your Code Gen Tools?  Your version of BIOS/IPC requires 7.0.2 or later.

Regards,

Shreyas

Hi

Before I switch to CGT 7, could you tell me if there will be any more CGT 6 releases that could support the above BIOS/IPC packages? Or are we being forced to go to CGT 7.0 if we wish to get the latest compiler fixes.

The documentation says the COEFF format will be expired in 7.0 and changed to ELF. This is pretty significant. Does 7.0.4 output ELF ?

C:\Program Files\Texas Instruments\ccsv4\tools\compiler\C6000 Code Generation Tools 7.0.4\doc\sprab90.pdf

There is also some 40 bit int changes we'd have to lookout for.

Cheers

7.0.x supports COFF.   7.0.x should be able to link existing COFF libraries built with 6.1.x.

While TI has added ELF support, COFF will be supported for the foreseeable future. 

More info here:

http://processors.wiki.ti.com/index.php/C6000_EABI_Migration

Hi

Changing to CGT 7.0.4 got rid of the error. Unfortunately, my code only works with RTSC profile "debug" or "release". For some unknown reason, "whole_program" does not work (I don't get any EDMA interrupts from SRIO transfers).

But this also was a problem with CGT 6.x.

Whats happening with the CGT 6.x releases? Are customers being asked to switch to 7.x since 6.x bug updates are being phased out?

Cheers

Normal 0 false false false EN-US X-NONE X-NONE MicrosoftInternetExplorer4

Normal 0 false false false EN-US X-NONE X-NONE MicrosoftInternetExplorer4

Well now we are between a rock and a hard place. We get stack overflow in some IPC module with “debug” profile and SRIO doesn’t work with “whole_program”.

The good news (in a sad way) is that our old configuration does the same thing

XDCtools version: 3.16.3.36
Code Gen Tools: 6.1.18
IPC: 1.20.00.23
BIOS: 6.21.03.21
XDAIS: 6.25.01.08

I wonder if SRIO doesn't work in "whole_program" because the SRIO library wasn't built with "whole_program"?

Cheers

Do you know which stack is overflowing?   Can you change the stack size for that task?   Are you using the ROV tools to check the stack usage?

Can you give more info on "SRIO doesn't work"?  Does it build but not run correctly?  Or are you having build problems?  

Hi

Karl Wechsler said:

Do you know which stack is overflowing? 

When running under debug, the console outputs:

ti.sysbios.knl.Task: line 330: E_stackOverflow: Task 0x86ec28 stack overflow.

xdc.runtime.Error.raise: terminating execution

I'm no expert at CCSv4, so I'm pasting below what I see in the ROV window. From what I saw in ROV, I suspect its the Core0 ti.sysbios.knl.Task.IdleTask but I can't tell since the "Basic" tab has a Java error (see below). The ROV module tab says my max was about 668 which is far from the available 4096.

I went to ROV -> ti -> sysbios -> knl -> Task for each core

Core 0 -(This is the one who's state in the debug window is "abort() at exit.c: 67 0x8526c0)

Tab "Basic" says:

Received exception from ROB server: java.lang.Exception: Target memory read failed at address 0xfffffffb, length 8. The read is at an INVALID address according to the applications section map. The application is likely either unititialized of corrupt.

From "module" tab (it says the peak is 292):

,0x00868468,Locked,0x8001,false,5,0x00868460,292,4096,0x87ca60

Core 1 to Core 5

From Base tab

,0x008683a0,ti.sysbios.knl.Task.IdleTask,0,Running,ti_sysbios_knl_Idle_loop__E,0x00000000,0x00000000,2048,0x87da60

,0x0086e118,,15,Blocked,WorkerTask_task,0x00000000,0x00000000,8192,0x86e168

From the module tab (the max peak was 668)

,0x00868468,Unlocked,0x1,false,2,0x008683a0,668,4096,0x87ca60

Karl Wechsler said:

Can you give more info on "SRIO doesn't work"?

I don't get any EDMA interrupts indicating a streaming transfer has completed. Any transfers under 4096 bytes work as they use the IDMA and not the EDMA. I have yet to check the EDMA params registers to figure out if the transfer actually happened. It could be that the RTOS under "whole_program" is not linking the EDMA interrupt somehow.

Cheers

Hi

I think I have found the problem about why the SRIO didn't work under RTSC profile "whole_program". I tried running a "whole_program" on the SRIO test jig I had and I got a "redefined" error (same as the post below)

http://e2e.ti.com/support/embedded/f/355/p/52404/187761.aspx#187761

So I added this line to my .cfg

var ti_sysbios_family_c64p_Hwi = xdc.useModule('ti.sysbios.family.c64p.Hwi');

and the redefined error went away. The SRIO also worked under "whole_program" in my test jig.

So next I added it to our production code and guess what, the SRIO started working!

So I wonder why this line of code isn't needed when I do "debug" or "program", but needed when I use "whole_program".

Seems like the compiler should be outputting an error when on doesn't have 'ti.sysbios.family.c64p.Hwi' in the .cfg file. How else is it supposed to know about connecting up hardware interrupts?

Unresolved is the stack overflow issue under RTSC profile "debug".......

Cheers

Thanks Karl