Diving problems with ifort and totalview

[emrys56]

We have installed an evaluation version of ifort (version 11) on a machine running Mandriva Linux release 2008.1 (two processor Intel Xeon CPU 1400MHz), and are having difficulty using it in conjunction with the totalview debugger. The program we are testing on is written to be f90 compatible and has been compiled with the -g option. The lines of code appear nicely in the TV gui window and breakpoints can be set. However, when we attempt to stop at a breakpoint and dive on an array variable, a pop-up box is obtained stating that the variable cannot be found and suggests some alternatives, none of which are correct. The ifort installation was a standard rpm installation, using all defaults. Is there something that we should be doing in addition, to cause an ifort-compiled code to work with totalview? Eventually we would like to use totalview to debug ifort/MPI compilations.

-- Rich Naff

[clabra]

Quoting - emrys56

We have installed an evaluation version of ifort (version 11) on a machine running Mandriva Linux release 2008.1 (two processor Intel Xeon CPU 1400MHz), and are having difficulty using it in conjunction with the totalview debugger. The program we are testing on is written to be f90 compatible and has been compiled with the -g option. The lines of code appear nicely in the TV gui window and breakpoints can be set. However, when we attempt to stop at a breakpoint and dive on an array variable, a pop-up box is obtained stating that the variable cannot be found and suggests some alternatives, none of which are correct. The ifort installation was a standard rpm installation, using all defaults. Is there something that we should be doing in addition, to cause an ifort-compiled code to work with totalview? Eventually we would like to use totalview to debug ifort/MPI compilations.

-- Rich Naff

It is not a problem of ifort. Sometimes totalview not understand the "symbol" required to access the variable when it is part of a module, because the different decorations of the modules. Try using the full name of the variables. If you want to watch the variable ARRAY, who is part of the module MOD, you should ask for mod_mp_array_.

clabra

[emrys56]

Clabra: Your suggestion, while clumsy, does work. There must be an incompatability between the ifort -g compilation and totalview, as I have not had this problem with other compilers. I haven't tried the idb debugger that apparently comes with the ifort release for linux OS; can it be used with MPI-fortran codes where multiple processors are employed?

[clabra1]

Quoting - emrys56

Clabra: Your suggestion, while clumsy, does work. There must be an incompatability between the ifort -g compilation and totalview, as I have not had this problem with other compilers. I haven't tried the idb debugger that apparently comes with the ifort release for linux OS; can it be used with MPI-fortran codes where multiple processors are employed?

Yes, you can. I never try it before, but you can see the idb documentation or in the forum for intel cluster and HPC.

[chris-gottbrath]

Quoting - emrys56

We have installed an evaluation version of ifort (version 11) on a machine running Mandriva Linux release 2008.1 (two processor Intel Xeon CPU 1400MHz), and are having difficulty using it in conjunction with the totalview debugger. The program we are testing on is written to be f90 compatible and has been compiled with the -g option. The lines of code appear nicely in the TV gui window and breakpoints can be set. However, when we attempt to stop at a breakpoint and dive on an array variable, a pop-up box is obtained stating that the variable cannot be found and suggests some alternatives, none of which are correct. The ifort installation was a standard rpm installation, using all defaults. Is there something that we should be doing in addition, to cause an ifort-compiled code to work with totalview? Eventually we would like to use totalview to debug ifort/MPI compilations.

-- Rich Naff

What version of TotalView are you using?

There have been problems, occasionally, with the debug information emitted by the fortran compiler. I don't want to point fingers prematurely though. ;)

We're always happy to look at these kinds of issues from the TotalView side of things. In particular we can examine the debug information that is being generated and provide input as to whether it gives enough info for the debugger to be able to locate the variable. We have an on-going dialog with the compiler developers at Intel and can usually hash these things out pretty straightforwardly.

Chris Gottbrath
Product Manager for TotalView
Chris(dot)Gottbrath(at)TotalViewTech(dot)com

[emrys56]

Quoting - chris-gottbrath

What version of TotalView are you using?

There have been problems, occasionally, with the debug information emitted by the fortran compiler. I don't want to point fingers prematurely though. ;)

We're always happy to look at these kinds of issues from the TotalView side of things. In particular we can examine the debug information that is being generated and provide input as to whether it gives enough info for the debugger to be able to locate the variable. We have an on-going dialog with the compiler developers at Intel and can usually hash these things out pretty straightforwardly.

Chris Gottbrath
Product Manager for TotalView
Chris(dot)Gottbrath(at)TotalViewTech(dot)com

We are using version 8.3.0-1; I should have posted that information with the original message. We have experienced similar problems with the ifc version of the Intel fortran compiler. I am now in contact with Rob Solomon at TV now, so we will see if the problem can be resolved. To be honest, it seems like a basic sort of problem that would have been discovered by now, so I thought perhaps it was a bad insallation.

-- Rich Naff