Make the scheduler tickless #201
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
davidchisnall
force-pushed
the
tickless
branch
5 times, most recently
from
7066e1d to
26bd4a5
Compare
rmn30
reviewed
Jun 5, 2024
| @@ -335,7 +336,13 @@ debug_log_message_write(const char *context, | |||
| DebugPrinter printer; | |||
| printer.write("\x1b[35m"); | |||
| printer.write(context); | |||
| #if 0 | |||
There was a problem hiding this comment.
Make make this a compile time option?
Suggested change
| #if 0 | |
| #if defined(DEBUG_SHOW_THREAD_ID) |
There was a problem hiding this comment.
This probably wants wiring up in an option, it's mostly disabled for now so that I can do timing checks without it there (it slows things down quite a bit with a real UART).
nwf-msr
reviewed
Jun 5, 2024
davidchisnall
force-pushed
the
tickless
branch
3 times, most recently
from
4d12dee to
cd95d17
Compare
nwf-msr
reviewed
Jun 5, 2024
nwf-msr
reviewed
Jun 5, 2024
nwf-msr
reviewed
Jun 5, 2024
sdk/core/scheduler/thread.h
Outdated
| ThreadState state : 2; | ||
| /** | ||
| * If the thread is yielding, it may be scheduled before its timeout | ||
| * expires, as long as no other threads are runnable. |
There was a problem hiding this comment.
I might leave off the "as long as..." part here and just point at the docs for ThreadSleepNoEarlyWake or the comment down in the actual wake code.
There was a problem hiding this comment.
I think it's fine to have duplicated explanations, especially in doc comments. If you mouse over a use of this field in VS Code (for example), you'll get this doc comment pop up. Having it tell you to go and read other docs can be annoying.
There was a problem hiding this comment.
That's fair. In which case...
Suggested change
| * expires, as long as no other threads are runnable. | |
| * expires, as long as no other threads are runnable or | |
| * sleeping with shorter timeouts. |
nwf-msr
approved these changes
Jun 6, 2024
| /** | ||
| * Wake any threads that were sleeping until a timeout before the | ||
| * current time. This also wakes yielded threads if there are no | ||
| * runnable threads. |
There was a problem hiding this comment.
Suggested change
| * runnable threads. | |
| * threads runnable or sleeping with earlier timeouts. |
sdk/core/scheduler/thread.h
Outdated
| ThreadState state : 2; | ||
| /** | ||
| * If the thread is yielding, it may be scheduled before its timeout | ||
| * expires, as long as no other threads are runnable. |
There was a problem hiding this comment.
That's fair. In which case...
Suggested change
| * expires, as long as no other threads are runnable. | |
| * expires, as long as no other threads are runnable or | |
| * sleeping with shorter timeouts. |
This commit incorporates several changes: Timers are now not set for a regular tick, they are set when a thread may be preempted. Specifically, the timer is set to the next timeout that will trigger a scheduling operation. This avoids timers triggering a switch to the scheduler to do nothing (resume the currently running thread). This means that if a thread sleeps for ten ticks while another runs, we will get one timer interrupt ten ticks in the future, rather than ten interrupts one tick apart. This means that ticks are now calculated retroactively based on elapsed time, rather than counted on each context switch. This, in turn, necessitates some small API changes. We previously conflated two things: - Sleep for N * (tick duration) - Yield and allow lower-priority threads to run for, at most, N * (tick duration) These are now deconflated by adding a second parameter to thread_sleep. Most sleeps are of the second form and so this is the default. This reduces the time taken to run the test suite on Sonata by around 30% and in the Ibex SAFE simulator by 13%.
rmn30
reviewed
Jun 7, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Timers are now not set for a regular tick, they are set when a thread
may be preempted. Specifically, the timer is set to the next timeout
that will trigger a scheduling operation. This avoids timers
triggering a switch to the scheduler to do nothing (resume the currently
running thread).
This means that if a thread sleeps for ten ticks while another runs, we
will get one timer interrupt ten ticks in the future, rather than ten
interrupts one tick apart.
This means that ticks are now calculated retroactively based on elapsed
time, rather than counted on each context switch.
This, in turn, necessitates some small API changes. We previously
conflated two things:
duration)
These are now deconflated by adding a second parameter to thread_sleep.
Most sleeps are of the second form and so this is the default.
This reduces the time taken to run the test suite on Sonata by around
30% and in the Ibex SAFE simulator by 13%.