Re: [Lightning] Are tail calls possible in GNU lightning?

[Paulo César Pereira de Andrade]

2015-01-10 5:23 GMT-02:00 Marc Nieper-Wißkirchen <address@hidden>:

>> > jit_node_t *compile() {
>> >   jit_state_t *_jit;
>> >   _jit = jit_new_state();
>> >   jit_prolog();
>> >   jit_node_t *label = jit_indirect();
>> >   jit_reti(42);
>> >   jit_emit();
>> >   jit_clear_state();
>> >   return jit_address(label);
>> > }

> For my purposes (using a virtual stack), I will never have to do a real
> C-style function call (except when doing FFI calls). Instead there will be
> (indirect) jumps everywhere and the real stack will not be touched. Thus the
> assumption that all generated functions take the same number of parameters
> (namely 0) would be fine for me.
>
>>
>> >   jit_retval(JIT_R0);
>> >   jit_jmpr(JIT_R0);
>>
>>   This will not work on some backends, like ia64 or big endian
>> powerpc (function descriptors), or, mips or little endian powerpc
>> where the function pointer must be in a specific register, to
>> compute GOT, PIC, etc. hppa also needs some special
>> handling. Also, for the sake of simplicity/safety, lightning would
>> still need to spill/reload modified callee save registers.
>>
>
> How to do indirect jumps then? In the manual to GNU lightning, the code
> jit_jmpr(R0) is used in several examples. How can I use an indirect label
> then?

  It is useful for computed gotos, e.g. something like C
switch/case or choosing what virtual method to call in
a vm that has a stack in the heap.

  Your specific example would be better done using
jit_frame and jit_tramp. These calls are not in a
released lightning version, but were added for
GNU Smalltalk. Examples:

http://git.savannah.gnu.org/cgit/lightning.git/tree/check/tramp.tst
http://git.savannah.gnu.org/cgit/lightning.git/tree/check/ctramp.c

Basically, "jit_frame(int size)" creates a stack frame
saving all callee save registers, and allocates at
least "size" bytes, that have as a good value, 8*
maximum number of arguments used in a C call.

After that, trampolines, but work as tail call in the
way you need it,  created with "jit_tramp(int size)"
are like a normal function body, just that while it
needs calling jit_prolog and jit_epilog, it does not
generate any code, so, it is very "light" and allow
jumping from code generated by one jit_context_t
to another.

I suggest you test it. Lightning 2.0.6 should be
released shortly, and will have the feature described
above.

I will also make some experiments with "true" tail
call. For a vm that could use the real stack, and
would need to pass arguments in the proper call
sequence, the feature would be very useful, and
if available, would be used...

> Marc

Thanks,
Paulo