Using Totalview

Starting Totalview

• Running totalview on hpc (Strategy Number One)
  • Run totalview cm64-debug (Make sure you are using a debugged version)
  • Go to Process->Startup Parameters and type your comfile as a Command-line argument
• Running totalview on hpc (Strategy Number Two)
  • Run totalview cm64-debug -a myfile.com

Overview of Important Commands Used in Totalview

• Type f in totalview window allows you to search for file. For example, type march8_ will bring up MARCH8.f which you can then place a breakpoint in.
• To set a breakpoint, e.g. line 129 of MARCH8.f, just click on line 129 on left hand side of page and a red STOP will cover the line number.
• Click on Go to get the programming running. You can Halt at anytime to stop at a point in the program. Note that this does not end your program, just temporarily suspend computation.
• Click Next to step to the next active line in the code without stepping into any subroutines. (Next simply steps over all subroutines.) You can also simply type n to go to the next line.
• Click Step to step to the next active line, or into a subroutine. You can also click s to go to the next line or into a subroutine.
• Click Out to go out of the current subroutine and into the next higher subroutine on the tree.

Viewing Contents of Arrays and Variables

• First option (one I never use) is to look at the Stack Frame window. Gives all variables on the stack. (Note that a subroutine places all local variables on the stack versus global variables which are placed on the heap. Google stack and heap and learn more information about them.)
• Common means of viewing arrays are variables is to double click on them. A window will pop-up with the array values in the window. Use that to search for values or to only look at a slice. (Use Slice window.)
• You can view a variable value (not an array) by simply placing the mouse pointer over that variable.

Creating Breakpoint based on Variable Values

• To set a condition breakpoint edit the properties of the breakpoint and set the Evaluation Expression to be something of the form: "if(np.eq.840) $stop"
• To set a breakpoint for writing to a variable, highlight the variable in the Expression View and create a Watchpoint (under Tools)

Tracing segmentation faults with a coredump file and totalview

• First change your limits to allow a coredump limit coredumpsize unlimited
• Then start from the command line your comfile, e.g. /hpc/cmiss/bin/rs6000-aix/cm64-debug -batch "comfile" >& "outputfile" </dev/null (I would recommend to use the full path for cm64-debug instead of aliases). The option -batch supresses any further input from stdin.
• Start totalview: 'totalview':
  • In the start-up window, change the box Start a new process to Open a core file
  • Choose in the field Program the cm-debug executable which produced the coredump file (with its full path, e.g. /hpc/cmiss/bin/rs6000-aix/cm64-debug )
  • Then browse for the coredump file core produced by running cm64-debug
  • Hit the OK button.
• Totalview pops-up with a window which displays (just as in normal debugging mode) the source code. An arrow points to the position at which the segmentation fault occured.
• Now, you can look at the variables, trace back subroutines, ... You can now use totalview just as you would be in normal debug-mode (of course "execute the next command"-button is not an option).
• After debugging, make sure that you remove your cordump file again. Coredump files are extremely large (can easily be >10G).
• Set your limits back such that no coredump file gets created any more: limit coredumpsize 0