Using the debug flag in Scarab about scarab HOT 2 CLOSED

Running in debug mode requires 3 steps:

1. Build the debug version of Scarab.
It sounds like from your question you already did this, but I will still put the instructions here for completeness. Compiling the binary in debug mode is also useful if you would like to use gdb.

cd src
make dbg

2. Enable debug mode.

Debug mode prints a lot of information to the stdout, typically too much to dump for an entire run (the log files would be too large). For that reason, it is usually good practice to enable debug mode only for a region of interest. Scarab provides two ways to identify the region of interest: by instruction of by cycle.

To specify a region by instruction, use the --debug_inst_start and the --debug_inst_stop scarab params. By default, debug_inst_start is set to 0 and debug_inst_stop is set to -1, meaning that there is no debug region (and therefore no debug prints). To enable, debug_inst_start must be set to a value >= 1 and debug_inst_stop must be set to a value > debug_inst_start. I would generally try to narrow this range to a few thousand instructions. I usually shoot for 5,000 instructions, no more than 10,000.

To specify a region by cycle, use the --debug_cycle_start and --debug_cycle_stop scarab params. The rules for these parameters are the same as above.

3. Enable debug prints for a particular module.

Finally, you must enable debug prints for whichever module you are interested in debugging. A full list of modules can be found in src/debug/debug.param.def.

Usually the best way to identify which debug module you want to enable is to go to the file that is giving you trouble and look for a DEBUG macro defined at the top of the file. For example, if you having trouble with the decode stage, you could look src/decode_stage.c and search for the word DEBUG. There you will find a macro definition that includes the variable DEBUG_DECODE_STAGE. This is the name of the scarab parameter required to enable debug mode for this file. So if I wanted to enable debug, I would simply pass --debug_decode_stage 1 as a scarab arg. Make sense?

Sometimes you aren't looking for a specific bug in a file, but rather just trying to get a sense of what is going on in the pipeline. For that, the most important parameter is --debug_model 1. This will print a very helpful view of the pipeline and show ops as they move through the stages.

When I am debugging, I usually start with a mix of several models that provide enough information for most issues. Here is an example list that I have used in the past:

--debug_model                      1
--debug_map                         1
--debug_node_stage              1
--debug_thread                      1
--debug_icache_stage            1
--debug_exec_stage               1
--debug_pin_exec_driven       1

This certainly is not a complete list, or maybe not even a good list for whatever you are doing. I simply show it here as an example of how multiple parameters can be combined to form a very robust debug output.

from scarab.

Closing this issue as it has been over a week since the last activity.

from scarab.