Small optimizations on iggen buffer handling #317
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
fknorr
force-pushed
the
iggen-buffer-opt
branch
from
e59eacf to
9785173
Compare
fknorr
force-pushed
the
iggen-buffer-opt
branch
from
9785173 to
0bd705c
Compare
This also avoids an unordered_map by transposing the perform_task_buffer_accesses loop.
fknorr
force-pushed
the
iggen-buffer-opt
branch
from
0bd705c to
a46a214
Compare
|
Check-perf-impact results: (c8fb992b35322012b54e351345fdf71a) ✔️ No significant performance change in the microbenchmark set. You are good to go! Relative execution time per category: (mean of relative medians)
|
Pull Request Test Coverage Report for Build 12158729604Details
💛 - Coveralls |
psalz
approved these changes
Dec 17, 2024
| } | ||
| } | ||
|
|
||
| // 3. Clear tracking structures for all regions that are being written to. We gracefully handle overlapping writes by treating the set of all conflicting | ||
| // writers as last writers of an allocation. | ||
| // 3. To gracefully handle overlapping writes, clear tracking structures for all regions that are being written to, so we can treat the set of all |
There was a problem hiding this comment.
Suggested change
| // 3. To gracefully handle overlapping writes, clear tracking structures for all regions that are being written to, so we can treat the set of all | |
| // 3. To gracefully handle overlapping writes if detection is disabled via the error policy, clear tracking structures for all regions that are being written to, so we can treat the set of all |
| } | ||
| buffer.track_original_write(concurrent_writes[i], command_instructions[i], concurrent_chunks[i].memory_id); |
There was a problem hiding this comment.
Suggested change
| buffer.track_original_write(concurrent_writes[i], command_instructions[i], concurrent_chunks[i].memory_id); | |
| // On the buffer level, there is no special handling of overlapping writes: The last chunk that writes to a given | |
| // region becomes its last writer, and subsequent reads on other devices will require a copy. | |
| buffer.track_original_write(concurrent_writes[i], command_instructions[i], concurrent_chunks[i].memory_id); |
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.
perform_task_buffer_accessesupdates last-writers twice to gracefully handle overlapping writes, which is an edge case. This PR quickly checks if overlapping writes are present, and sticks to a single update if there are not. By transposing the loop nest fromchunk -> bidtobid -> chunk, we can also save avoid constructing anotherunordered_map.Results are not looking too impressive in the benchmark report, but I do get a consistent 4% speedup for RSim
room_small, which is scheduler bound on gpuc3.