Exceptions stopped working

David Luca said:

as I understand profiling doesn't work on C6748

It should.  Is there a SDSCM number for this issue?  Where did you find documentation --gen_profile_info doesn't work on C6748?

Hi Archaeologist,

I found some resources a while ago and forum replies that C6748 hardware doesn't support because is cache based. I think profiling should be supported, for example in our code cache is disabled just to test optimization of functions and real time spent. And until we disabled gen_profile_info, exceptions stopped working, I don't know what "SDSCM" means.

Please see: http://processors.wiki.ti.com/index.php/FAQ_-_CCSv4#Q:_Why_is_function_profiling_not_supported_on_C6000_hardware.3F

and http://e2e.ti.com/support/development_tools/code_composer_studio/f/81/t/164978.aspx

Best regards,

David.

Our defect reports have numbers that look like SDSCM00012345

The --gen_profile_info option is not directly related to function profiling described in that wiki page or event counting in that forum thread.

That wiki page is not very clear about why the cache has anything to do with profiling.  I think the answer is that when the cache is enabled, the cycle counts will be greatly distorted by the profling itself.  I think profiling would work just fine with cache disabled, but then the results might not be useful.

Do you know any resources regarding profiling on C6748 that I can start reading more information?

Best regards,

David.

Unfortunately I'm not an expert on the various methods of profiling, so I don't really know where to look.

I lack expertise in profiling methods, but I can shed some light regarding --gen_profile_info.

The information gathered by --gen_profile_info is not cycle counts.  It counts how many times each C statement executes.  You don't know how long anything takes, but you can make relative comparisons.  (I'm simplifying things a bit here, but this is close enough for this discussion.)  That said, I don't know why using --gen_profile_info would cause exceptions to not work.  I suspect a compiler bug here, and I'd appreciate your help in finding it.

Thanks and regards,

-George

A sanity check first: you're not attempting to catch a C++ exception thrown in a different thread, are you?

That would be impossible, isn't it? It was a simple test:

try{ throw "exception";}

catch(const char *s){printf(s);}

catch(...){printf("unknown");}

Whenever the exception was thrown, the code was going into C$$_exit() as seen on debugger, would never enter any of the catch blocks. The .cproject was stored on SVN and we've seen that only this option was changed, removing it the above code would work fine.

Analysis shows that --gen_profile_info completely removes the exception handling bookkeeping, so at this time --gen_profile_info is completely incompatible with --exceptions.  This is a bug.