#16171: Preload kernels before receiving go message #16680
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
jbaumanTT
requested review from
aliuTT,
ubcheema,
abhullar-tt,
pgkeller,
tt-aho,
tt-dma and
tt-asaigal
as code owners
jbaumanTT
force-pushed
the
jbauman/preloadkernel3
branch
from
abe1525 to
5d0bc7b
Compare
arikTT
approved these changes
Jan 14, 2025
tt-aho
approved these changes
Jan 14, 2025
jbaumanTT
force-pushed
the
jbauman/preloadkernel3
branch
from
5d0bc7b to
cbb748a
Compare
pgkeller
reviewed
Jan 14, 2025
jbaumanTT
force-pushed
the
jbauman/preloadkernel3
branch
2 times, most recently
from
5863780 to
c98e90b
Compare
Add a flag that lets brisc.cc start loading kernels before receiving a go message. Fast dispatch ensures that that the flag will be set only after all necessary program data is sent to the core. This allows preparation for the following kernel (including loading NCRISC IRAM, setting up CBs, and initializing local memory) to happen in parallel with the round-trip to the dispatcher_s to sync up with the other kernels and ensure that they're all launched at the same time.
jbaumanTT
force-pushed
the
jbauman/preloadkernel3
branch
from
c98e90b to
7d95eed
Compare
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Outdated suggestions cannot be applied.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.
Ticket
#16171
Problem description
Currently it takes around 750 ns (ideally) between a core sending its acknowledgement that it's finished a kernel and receiving the GO message for the next kernel. That time could better be spent preparing to load the next kernel.
What's changed
Add a flag that lets brisc.cc and erisc.cc start loading kernels before receiving a go message. Fast dispatch ensures that that the flag will be set only after all necessary program data is sent to the core.
This allows preparation for the following kernel (including loading NCRISC IRAM, setting up CBs, and initializing local memory) to happen in parallel with the round-trip to the dispatcher_s to sync up with the other kernels and ensure that they're all launched at the same time.
Checklist