Will Coleda wrote:
On Feb 13, 2008 1:41 PM, chromatic <[EMAIL PROTECTED]> wrote:
On Wednesday 13 February 2008 06:21:32 [EMAIL PROTECTED] wrote:
Should we perhaps add a performance benchmark to the tests?
Normalising it to account for hardware variations might be a problem.
That's somewhat difficult, as it's the performance of languages hosted on
Parrot that's most important.
I profile Rakudo once in a while. If I ever get the time and energy and
brains to figure out where the weird Lua segfaults come from, I'll profile it
too.
If we were really fortunate, someone would write a Parrot runloop that emits
callgrind-compatible output, so that we could profile Parrot at the sub and
op level with callgrind_annotate.
-- c
One of the thing tcl needs to be fully supported is the ability to add
sub hooks that execute on enter/exit of a particular sub[1]; adding
this would give us the ability to profile which PIR subs we were
spending most of our time (by using the pre-post hooks to capture
timing information), which as more people are using PCT, would help us
concentrate our efforts on speedups that could potentially improve
things for many HLLs at once.
Lua has too this kind of hooks in his Debug library (see
http://www.lua.org/manual/5.1/manual.html#5.9).
These features could be very VM specific.
François.
[1] http://www.tcl.tk/man/tcl8.5/TclCmd/trace.htm
