TI_start/stop_pprof_collection() early exit problem

zhongfan yang said:

 Internal variables  _TI_pprof_out_hndl =  -1  and _TI_prof_data_start = -1 etc..

I cannot reproduce this result.  I'd appreciate if you would submit your project so I can try to understand how that happened.

Thanks and regards,

-George

My apology for not making clear about the context of the project. My project is used with SYSBIOS, and is written in C++. I connect the CCS to target board via emulator, and run under RELEASE mode. 

When I switch on the - -gen_profile_info, compile and run the test, the program is stuck in  TI_stop_pprof_collect() and never exit (like in a dead cycle). I thought the the compiler warning: "creating output section ".ppdata" without a SECTION" is to blame, so I added the line "Program.sectMap[".ppdata"] = "DDR" into the *.cfg file. Now it gives no warning anymore, but TI_start /stop_pprof_collect() will exit early due to internal error.

I don't know where I take the wrong step.This is the first time I use feedback optimization and I thought the process should be straightforward. Enclosed is the exported project (Note: the IO file test is OK with my system.) Any advice is appreciated!

Regards

ZFYoung

 3568.Test.zip

I still cannot reproduce the problem.

I cannot run the code.  I lack the hardware.  But I can build it.  When I look for the pprof variables in the map file, this is what I see ...

c0813b80   __TI_start_pprof_collection
c0812d40   __TI_stop_pprof_collection
c0815d40   ___TI_pprof_out_hndl

Nothing is -1.  So I still cannot tell you why you see -1 for some or all of the symbols.

Thanks and regards,

-George

Thanks for your timely response! Did you notice the line:"//Program.sectMap[".ppdata"] = "DDR";" that I've commented out in Test.cfg? If that line is in effect, It will cause the early exit problem. If it is not included in compile process as it is now, program will stuck in  TI_stop_pprof_collection() and never return. Neither generates *.pdat file successfully. 

Regards

ZFYoung

zhongfan yang said:

 Did you notice the line:"//Program.sectMap[".ppdata"] = "DDR";" that I've commented out in Test.cfg? If that line is in effect, It will cause the early exit problem.

When I uncomment that line, I still see all the pprof symbols have sensible addresses, i.e. not -1.  So I still cannot reproduce that part of your problem.

Thanks and regards,

-George

Well...Seems like a dead end to me. Anyway, I like to know: Normally when generating feedback file *.pdat,  "-gen_profile_info" switch is enough or other changes in the project config is still needed? ie. should I ignore the linker waring: "creating output section ".ppdata" without a SECTION " or not ?

Regards

ZFYoung

zhongfan yang said:

Normally when generating feedback file *.pdat,  "-gen_profile_info" switch is enough or other changes in the project config is still needed?

Normally --gen_profile_info is all you need.  However ...

zhongfan yang said:

should I ignore the linker warning: "creating output section ".ppdata" without a SECTION "

You should never ignore this warning, for any section.  The linker is saying that it allocates .ppdata in whatever memory range it finds that can hold it.  This is almost certainly the wrong one.  You want to place all of your output sections explicitly.

Thanks and regards,

-George

Hi, George, Since you can not repeat my observation, I've update the project to reflect the problem in my system(Nothing special, just some section reassignment directive in test.cfg file) . Hope you can take a look, Thanks !

Regards

ZFYoung 

1220.Test.zip

Thank you, but no luck once again.

I do have to make a few changes to the project so it will build.  I don't think these changes make a difference, but they might.  So I will describe them.

The first is that I remove the reference to D:\packages in the list of package repositories used by the RTSC option --xdcpath.  Then I change the RTSC platform from ti.platforms.LCSDK6748 (or something like that) to ti.platforms.evm6748.  Then I add the compiler option --gen_profile_info.  After those changes, I check the values of the pprof related symbols in the map file.  I hope to see one of them set to -1.  But I never do.

Thanks and regards,

-George