|
Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes. Set by lizmat on 24 May 2021. |
|||
|
12:36
JimmyZhuo joined
|
|||
| lizmat | Rakudo compiling setting time performance analysis pictures by Visual Studio, posted to github.com/rakudo/rakudo/issues/5851 JimmyZhuo++ | 13:28 | |
| hope someone will be able something with that here | 13:29 | ||
|
13:43
JimmyZhuo left
|
|||
| Woodi | frame dispatch used a lot - heavy functions calls ? maybe functions are too small and too specialized ? :) | 14:24 | |
| lizmat | isn't that what gets called on *any* dispatch ? | 14:28 | |
| timo | MVM_frame_dispatch is for dispatching execution to a frame with some arguments | 14:31 | |
| lizmat | as opposed to? | ||
| I mean, that would cover any method call, no ? | |||
| timo | as opposed to other uses of nqp::dispatch i think? | 14:34 | |
| lizmat | I'm not sure ? | 14:40 | |
| timo | so it turns out there actually is a way to ask libuv to callback when a numbered fd is ready for reading and/or writing and/or has an error of some kind | 14:51 | |
| nine | As I said before I'm very skeptical that such low level profiles will be very helpful at this point. I recommend looking for places in the code where we calculate the same information multiple times where a simple cache would safe us a lot of work. | 14:55 | |
| Or even before that: I have noticed that in some cases we can end up referencing the same child nodes on multiple paths (in visit-children). That means everything we do including begin and check processing is done multiple times as well. | 14:56 | ||
| japhb | I guess an interesting report would be all caller/callee pairs where the callee count is 2x or more the caller count. Even a simple filtering that doesn't account for the callee possibly being called by many different callers would still give a useful view I'd think. | 15:02 | |
|
21:59
MasterDuke joined
23:48
MasterDuke left
|
|||