# New Ticket Created by Jonathan Sillito # Please include the string: [perl #22633] # in the subject line of all future correspondence about this issue. # <URL: http://rt.perl.org/rt2/Ticket/Display.html?id=22633 >
(This conflicts with "[perl #22592] [PATCH] Introduce macros for register
access." because I use the REG_PMC(x) approach. It would be easy to fix I
just don't know when or if #22592 is going to be committed.)
This patch converts parrot to a continuation passing style. This means that
when invoking a sub or a method the "continuation" to return to should be
placed in P1. If the 'cc' ops (invokecc and invokecc_method) are used the
current continuation is captured and placed in P1 automatically.
So for a callee to return:
invoke P1
for a tail call, just leave P1 (i.e. the continuation) as is and:
invoke
for returning from a sub that is to be used as a coroutine:
invokecc P1
nice, eh?
Specifically this patch applies the following changes:
- adds file completecontext.pmc: should go in classes/
- changes continuation.pmc - to not restore registers
- small changes to sub.pmc & coroutine.pmc
- supporting changes to sub.c, sub.h, register.h
- add invoke_method & invokecc & invokecc_method ops
- update tests and added tail call test
I am not satisfied with the time taken to make a call. I did some rough
benchmarking and the parrot implementation makes us slower than python
2.2.1. The most expensive part of our call is the saveall/restoreall that
wraps the invokecc. In some cases the caller can be more selective about
what is saved, if no saveall/restoreall are used parrot calls are faster
than python's. The PASM and python code I used to benchmark along with the
times are below.
#################################
### test.pasm
new P0, .Sub
set_addr I3, foo
set P0, I3
new P11, .PerlInt
set P11, 1000000
loop:
saveall
invokecc
restoreall
sub P11, 1
if P11, loop
end
foo:
invoke P1 # return
#################################
### test.py
def foo():
pass
p11 = 1000000
while p11:
foo()
p11 -= 1
$ time python test.py
real 0m4.401s
user 0m3.990s
sys 0m0.030s
$ perl assemble.pl test.pasm >test.pbc ; time ./parrot test.pbc
real 0m6.831s
user 0m6.820s
sys 0m0.010s
However if we remove the saveall/restore all we are much faster
$ perl assemble.pl test.pasm >test.pbc ; time ./parrot test.pbc
real 0m2.382s
user 0m2.380s
sys 0m0.000s
--
Jonathan Sillito
-- attachment 1 ------------------------------------------------------
url: http://rt.perl.org/rt2/attach/59206/43869/1cbffc/cps.patch
-- attachment 2 ------------------------------------------------------
url: http://rt.perl.org/rt2/attach/59206/43870/55771c/completecontext.pmc
cps.patch
Description: cps.patch
completecontext.pmc
Description: completecontext.pmc
