Parrot - Running




This document describes Parrot's command line options.


 parrot [-options] <file> [arguments ...]


If this environment variable is set, parrot will use this path as its runtime prefix instead of the compiled in path.
Turn on the --gc-debug flag.


Assembler options

-a, --pasm
Assume PASM input on stdin.
-c, --pbc
Assume PBC file on stdin, run it.
-d, --imcc-debug [hexbits]
The -d switch takes an optional argument which is considered to hold a hex value of debug bits. Without a value, debug is set to 1.The individual bits can be listed on the command line by use of the --help-debug switch.To produce really huge output on stderr run "parrot -d 0ffff ...". Note: If the argument is separated by whitespace from the -d switch, it has to start with a number.
-h, --help
Print command line option summary.
Print debugging and tracing flag bits summary.
-o outputfile, --output=outputfile
Act like an assembler. Don't run code, unless -r is given too. If the outputfile ends with .pbc, a PBC file is written. If it ends with .pasm, a PASM output is generated, even from PASM input. This can be handy to check various optimizations, including -Op.
Act like an assembler, but always output bytecode, even if the output file does not end in .pbc
-r, --run-pbc
Only useful after -o or --output-pbc. Run the program from the compiled in-memory image. If two -r options are given, the .pbc file is read from disc and run. This is mainly needed for tests.
-v, --verbose
One -v shows which files are worked on and prints a summary over register usage and optimization stats per subroutine. With two -v switches, parrot prints a line per individual processing step too.
-y, --yydebug
Turn on yydebug in yacc/bison.
-V, --version
Print version information and exit.
 -O0 no optimization (default)
 -O1 optimizations without life info (e.g. branches)
 -O  same
 -O2 optimizations with life info
 -Op rewrite I and N PASM registers most used first
 -Ot select fastest runcore (default with -O1 and -O2)
 -Oc turns on the optional/experimental tail call optimizations
See docs/dev/optimizer.pod for more information on the optimizer. Note that optimization is currently experimental and these options are likely to change.
-E, --pre-process-only
Preprocess source file (expand macros) and print result to stdout:
  $ parrot -E t/op/macro_10.pasm
  $ parrot -E t/op/macro_10.pasm | parrot -- -

Runcore Options

These options select the runcore, which is useful for performance tuning and debugging. See "About runcores" for details.

-R, --runcore CORE
Select the runcore. The following cores are available in Parrot, but not all may be available on your system:
  slow, bounds  bounds checking core (default)
  cgoto         computed goto core
  cgp           computed goto-predereferenced core
  cgp-jit       computed goto-predereferenced core with JIT
  exec          exec core (uses JIT at compile time to generate native code)
  fast          fast core (no bounds checking, profiling, or tracing)
  gcdebug       performs a full GC run before every op dispatch (good for
                debugging GC problems)
  jit           JIT core
  switch        switch core
  switch-jit    switch core with JIT
  trace         bounds checking core w/ trace info (see 'parrot --help-debug')
--bounds-checks, --slow-core
Select the bounds-checking slow core (default).
Select the CGoto core (if available).
Select the CGP (CGoto Predereferenced) core (if available).
Select the fast (or function) core. Bounds checking, profiling, and tracing are not available. (That's why it's so fast.)
Run with the JIT subsystem (if available).
-p, --profile
Run with the slow core and print an execution profile.
TODO: This needs to be documented briefly here and also in glossary.pod.
-t, --trace
Run with the slow core and print trace information to stderr. See parrot --help-debug for available flag bits.

VM Options

-w, --warnings
Turn on warnings. See parrot --help-debug for available flag bits.
-D, --parrot-debug
Turn on interpreter debug flag. See parrot --help-debug for available flag bits.
Turn on GC (Garbage Collection) debugging. This imposes some stress on the GC subsystem and can slow down execution considerably.
-G, --no-gc
This turns off GC. This may be useful to find GC related bugs. Don't use this option for longer running programs: as memory is no longer recycled, it may quickly become exhausted.
--leak-test, --destroy-at-end
Free all memory of the last interpreter. This is useful when running leak checkers.
-., --wait
Read a keystroke before starting. This is useful when you want to attach a debugger on platforms such as Windows.
Print the runtime prefix path and exit.


If the file ends in .pbc it will be interpreted immediately.

If the file ends in .pasm, then it is parsed as PASM code. Otherwise, it is parsed as PIR code. In both cases, it will then be run, unless the -o flag was given.

If the file is a single dash, input from stdin is read.

[arguments ...]

Optional arguments passed to the running program as ARGV. The program is assumed to know what to do with these.

Generated files

If JIT debugging is enabled (e.g. via --parrot-debug 04), the following additional output files are generated:

  F<file.stabs.s>     stabsfile for the program
  F<file.o>           object file with debug information
  F<EVAL_n>           source of C<compile> op number I<n>
  F<EVAL_n.stabs.s>   stabsfile for this block
  F<EVAL_n.o>         object file with debug information

See docs/jit.pod for further information.

About runcores

The runcore (or runloop) tells Parrot how to find the C code that implements each instruction. Parrot provides more than one way to do this, partly because no single runcore will perform optimally on all architectures (or even for all problems on a given architecture), and partly because some of the runcores have specific debugging and tracing capabilities.

In the default "slow" runcore, each opcode is a separate C function. That's pretty easy in pseudocode:

    slow_runcore( op ):
        while ( op ):
            op = op_function( op )

The GC debugging runcore is similar:

    gcdebug_runcore( op ):
        while ( op ):
            op = op_function( op )

Of course, this is much slower, but is extremely helpful for pinning memory corruption problems that affect GC down to single-instruction resolution. See for more information.

The trace and profile cores are also based on the "slow" core, doing full bounds checking, and also printing runtime information to stderr.

The switched core eschews these tiny op functions in favor of cases in a large switch statement:

    switch_runcore( op ):
        while ( op ):
            switch *op:
                case NOP:
                case STORE:

Depending on the C compiler implementation, this may be faster than function calling. On older systems, it may fail to compile altogether.

The computed-goto ("cgoto") runcore avoids the overhead of function calls by jumping directly to the address where each opcode's function starts. The computed-goto-prederef ("CGP") core takes this one step further by replacing opcode numbers in the bytecode with those opfunc addresses. See "Predereferencing" in docs/glossary.pod for a fuller explanation.

Finally, the JIT runcore uses the "slow" core, but also creates and jumps to JIT-compiled native code for supported opcodes. "cgp-jit" and "switched-jit" are variations that use the CGP or switched core but run JIT code when possible.

Operation table

 Command Line          Action         Output
 parrot x.pir          run
 parrot x.pasm         run
 parrot x.pbc          run
 -o x.pasm x.pir       ass            x.pasm
 -o x.pasm y.pasm      ass            x.pasm
 -o x.pbc  x.pir       ass            x.pbc
 -o x.pbc  x.pasm      ass            x.pbc
 -o x.pbc -r x.pasm    ass/run pasm   x.pbc
 -o x.pbc -r -r x.pasm ass/run pbc    x.pbc
 -o x.o    x.pbc       obj

... where the possible actions are:

  run ... yes, run the program
  ass ... assemble sourcefile
  obj ..  produce native (ELF) object file for the EXEC subsystem