Re: [GXemul-devel] [GXemul-users] Gxemul

[Anders G. <ga...@gm...>]

Hi Vincent,

Tis 2010-05-11 klockan 01:20 -0400 skrev Vincent Mirian:
> Hi all,
> 
> I have a few questions... :-)
> 
> I would like to create execution traces of a mips processor from
> gxemul. I would like to run the simulation for a while (until a
> breakpoint let's say) and then save the _execution_ traces in a file.

I didn't quite understand which of the following two you want to do:

a) run for a while (tracing instructions), hit a breakpoint, exit
(saving the traced instruction)

or

b) run for a while without instruction trace, hit a breakpoint, turn on
instruction trace, and then continue (saving instruction trace from here
on).

Both should be possible by hacking the source code. A brutal one-off
approach could be to just hack in something in
DYNTRANS_RUN_INSTR_DEF(struct cpu *cpu) in cpu_dyntrans.cc, like

	if (single_step || cpu->machine->instruction_trace
	    || cpu->machine->register_dump) {
		/*
		 *  Single-step:
		 */
FILE *old = stdout;	// HACK!
stdout = fopen("trace-output.txt", "a");	// HACK!

		struct DYNTRANS_IC *ic = cpu->cd.DYNTRANS_ARCH.next_ic;

...

		/*  Execute just one instruction:  */
		I;

		n_instrs = 1;

fclose(stdout);	// HACK!
stdout = old;	// HACK!

	} else if (cpu->machine->statistics.enabled) {

to redirect stdout to a file during single step and instruction trace.
But it would probably be horribly slow.

>  Furthermore, I do not want to trace the cycle time that the processor
> is halted for memory requests.

That's easy. (Because GXemul currently does not simulate caches or
memory latencies. :-))


Anders