CoreTiming: Document initial startup behavior
Events don't update the downcount until after the first Advance(), thus Advance() must be called once before scheduling works normally.
This commit is contained in:
1 parent
59465911d7
commit
fb5537213a
2 files changed
+17
No files matched your search
@@ -129,6 +129,11 @@ void Init()
|
||||
g_slice_length = MAX_SLICE_LENGTH;
|
||||
g_global_timer = 0;
|
||||
s_idled_cycles = 0;
|
||||
|
||||
// The time between CoreTiming being intialized and the first call to Advance() is considered
|
||||
// the slice boundary between slice -1 and slice 0. Dispatcher loops must call Advance() before
|
||||
// executing the first PPC cycle of each slice to prepare the slice length and downcount for
|
||||
// that slice.
|
||||
s_is_global_timer_sane = true;
|
||||
|
||||
s_ev_lost = RegisterEvent("_lost_event", &EmptyTimedCallback);
|
||||
|
||||
Reference in new issue
Block a user