Hi. I used asyncifiy in default version of JS-DOS. However I found way
how to increase performance of it on 25%. While dosbox is doing
emulation it uses one loop that works forever:
while (1) {
static lastSleep = GetTicks();
if (GetTicks() - lastSleep > 16) {
emscripten_sleep(0);
lastSleep = GetTicks();
}
PIC_RunQueue();
}
16 is a period of continuous emulation (== 60 FPS).
BUT, this implementation is not so effective as it can be.
Implementation of emscripten_sleep is based on setTimeout function. As
known minimum timeout delay is near ~4ms. So real execution flow is
follow:
| 16ms emulation | 4 ms idle | 16ms emulation | 4ms idle | ....
4ms is 25%. I checked it in profilers of Chrome and Firefox, and as
expected I see idle self time ~ 20-25%.
This 4m ms idle can be removed if we switch implementation from
setTimeout to postMessage. I tried it, and results are very good I see
improvement of Dhrystone 2 test on 60% (sic!), and also profilers
reports about 1-2% idle time.
This is very good, but have some questions:
1. Is any chance to change implementation of emscripten_sleep to
post/recive message (in case when sleep time <4ms)?
2. Even though it works, I don't like fact that I am checking
emulation period in c++, I want to do in this way:
while (1) {
syncSleep();
PIC_RunQueue();
}
//...
EM_JS(void, syncSleep, (), {
if (Date.now() - Module.last_sync_time <= 16) {
return;
}
return Asyncify.handleSleep(function(wakeUp) {
// post/recive impl
});
});
BUT, looks like it's not possible to not call Asyncify.handleSleep in
ASYNCIFY_IMPORT function. If I return 0/undfeinied from it execution
always terminate with message "Unreachable executed...". Is there are
any workaround for this?
3. Can you explain how asyncify handle recursion. Looks it not valid
case for asyncify. DOSBOX can run emulation loop inside emulation
loop, like this:
void loop() {
while(1) {
syncSleep();
// <...>
printf("int21 handler\n");
loop();
}
}
For this case browser hangs. I added console.log into syncSleep
function, and typical log is:
syncSleep (1593)
int21 handler called
// no any syncSleep anymore
I think maybe in case of recursion call stack is overwritten or
something similar, what do you think?
Cause I don't understand why log stops.
--
You received this message because you are subscribed to the Google Groups
"emscripten-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/emscripten-discuss/CAKOm%3DVFaCX1Y--QjVP1Rxqs3TTf7ETZ-%2B%3DmHqZ_UyRDVbkqJ6g%40mail.gmail.com.
