Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

scheduler: Add explanations for YieldWith and WithoutLoadBalancing

Browse source
This commit is contained in:
Zach Hilman 2018-11-22 00:33:53 -05:00
parent 409dcf0e0a
commit 820d81b9a5
6 changed files with 141 additions and 79 deletions
Download patch file Download diff file Expand all files Collapse all files

70
src/core/hle/kernel/scheduler.h
View file

@ -49,13 +49,79 @@ public:
void UnscheduleThread(Thread* thread, u32 priority);
/// Moves a thread to the back of the current priority queue
void RescheduleThread(Thread* thread, u32 priority);
void MoveThreadToBackOfPriorityQueue(Thread* thread, u32 priority);
/// Sets the priority of a thread in the scheduler
void SetThreadPriority(Thread* thread, u32 priority);
/// Gets the next suggested thread for load balancing
Thread* GetNextSuggestedThread(u32 core);
Thread* GetNextSuggestedThread(u32 core) const;
/**
* YieldWithoutLoadBalancing -- analogous to normal yield on a system
* Moves the thread to the end of the ready queue for its priority, and then reschedules the
* system to the new head of the queue.
*
* Example (Single Core -- but can be extrapolated to multi):
* ready_queue[prio=0]: ThreadA, ThreadB, ThreadC (->exec order->)
* Currently Running: ThreadR
*
* ThreadR calls YieldWithoutLoadBalancing
*
* ThreadR is moved to the end of ready_queue[prio=0]:
* ready_queue[prio=0]: ThreadA, ThreadB, ThreadC, ThreadR (->exec order->)
* Currently Running: Nothing
*
* System is rescheduled (ThreadA is popped off of queue):
* ready_queue[prio=0]: ThreadB, ThreadC, ThreadR (->exec order->)
* Currently Running: ThreadA
*
* If the queue is empty at time of call, no yielding occurs. This does not cross between cores
* or priorities at all.
*/
void YieldWithoutLoadBalancing(Thread* thread);
/**
* YieldWithLoadBalancing -- yield but with better selection of the new running thread
* Moves the current thread to the end of the ready queue for its priority, then selects a
* 'suggested thread' (a thread on a different core that could run on this core) from the
* scheduler, changes its core, and reschedules the current core to that thread.
*
* Example (Dual Core -- can be extrapolated to Quad Core, this is just normal yield if it were
* single core):
* ready_queue[core=0][prio=0]: ThreadA, ThreadB (affinities not pictured as irrelevant
* ready_queue[core=1][prio=0]: ThreadC[affinity=both], ThreadD[affinity=core1only]
* Currently Running: ThreadQ on Core 0 || ThreadP on Core 1
*
* ThreadQ calls YieldWithLoadBalancing
*
* ThreadQ is moved to the end of ready_queue[core=0][prio=0]:
* ready_queue[core=0][prio=0]: ThreadA, ThreadB
* ready_queue[core=1][prio=0]: ThreadC[affinity=both], ThreadD[affinity=core1only]
* Currently Running: ThreadQ on Core 0 || ThreadP on Core 1
*
* A list of suggested threads for each core is compiled
* Suggested Threads: {ThreadC on Core 1}
* If this were quad core (as the switch is), there could be between 0 and 3 threads in this
* list. If there are more than one, the thread is selected by highest prio.
*
* ThreadC is core changed to Core 0:
* ready_queue[core=0][prio=0]: ThreadC, ThreadA, ThreadB, ThreadQ
* ready_queue[core=1][prio=0]: ThreadD
* Currently Running: None on Core 0 || ThreadP on Core 1
*
* System is rescheduled (ThreadC is popped off of queue):
* ready_queue[core=0][prio=0]: ThreadA, ThreadB, ThreadQ
* ready_queue[core=1][prio=0]: ThreadD
* Currently Running: ThreadC on Core 0 || ThreadP on Core 1
*
* If no suggested threads can be found this will behave just as normal yield. If there are
* multiple candidates for the suggested thread on a core, the highest prio is taken.
*/
void YieldWithLoadBalancing(Thread* thread);
/// Currently unknown -- asserts as unimplemented on call
void YieldAndWaitForLoadBalancing(Thread* thread);
/// Returns a list of all threads managed by the scheduler
const std::vector<SharedPtr<Thread>>& GetThreadList() const {