RFC: Move build tools to Perl5, developer tools to Perl6
rurban at x-ray.at
Sat Feb 23 22:17:17 UTC 2013
On Fri, Feb 22, 2013 at 6:57 AM, Jimmy Zhuo <jimmy.zhuo at gmail.com> wrote:
>> Moving opsc back to perl5 is the biggest mistake you can make. Upgrade
>> to modern nqp is the way.
>> Keywords: JIT, compile-time optimisations, PBC emitting, kill PIR with
> I'm not optimistic about JIT, since parrot's ops is so easy to extend.
> Java/Lua VM can't JIT well too if we can easily add some new ops, me thinks.
> I'd like to see jit m0.
I heard this argument from Whiteknight before. I don't buy it.
whiteknight said something like "that with so many ops you have to jit
all of them. This does not scale.
Only m0 will help, ..."
That's not true.
You only have to jit the ops you want to jit. The others are compiled
as object code,
and just need to be called.
Such a generic ops call wrapper is ~5 lines assembly code.
You only jit ops where a function call is too much overhead (
int/double arith and cmp, method calls ),
which you want to inline,
or where you want to apply compile-time stack logic.
(i.e. saving stack-space as in callcc/yield, coroutines, thread switching.)
More information about the parrot-dev