Ensure adjoint allocates memory for max concurrent observables only #221
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
|
Hello. You may have forgotten to update the changelog!
|
Codecov Report
@@ Coverage Diff @@
## master #221 +/- ##
==========================================
+ Coverage 99.70% 99.72% +0.01%
==========================================
Files 4 4
Lines 344 358 +14
==========================================
+ Hits 343 357 +14
Misses 1 1
Continue to review full report at Codecov.
|
This reverts commit 2f48486. Favour chunking at Python layer due to configurability.
AmintorDusko
reviewed
Feb 3, 2022
| @@ -65,6 +67,12 @@ | |||
| ) | |||
|
|
|||
|
|
|||
| def _chunk_iterable(it, num_chunks): | |||
| "Lazy-evaluted chunking of given iterable from https://stackoverflow.com/a/22045226" | |||
There was a problem hiding this comment.
Suggested change
| "Lazy-evaluted chunking of given iterable from https://stackoverflow.com/a/22045226" | |
| "Lazy-evaluated chunking of given iterable from https://stackoverflow.com/a/22045226" |
AmintorDusko
approved these changes
Feb 3, 2022
maliasadi
requested changes
Feb 3, 2022
Co-authored-by: Ali Asadi <[email protected]>
Co-authored-by: Ali Asadi <[email protected]>
trevor-vincent
approved these changes
Feb 4, 2022
maliasadi
approved these changes
Feb 4, 2022
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Outdated suggestions cannot be applied.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.
All new features must include a unit test.
If you've fixed a bug or added code that should be tested, add a test to the
testsdirectory!All new functions and code must be clearly commented and documented.
If you do make documentation changes, make sure that the docs build and
render correctly by running
make docs.Ensure that the test suite passes, by running
make test.Add a new entry to the
.github/CHANGELOG.mdfile, summarizing thechange, and including a link back to the PR.
Ensure that code is properly formatted by running
make format.When all the above are checked, delete everything above the dashed
line and fill in the pull request template.
Context: The current implementation of the adjoint Jacobian method in Lightning parallelizes over observables, with each observable handled by a separate OpenMP thread. This works fine in practice unless a large number of observables are required, wherein a new statevector memory block is allocated upfront for the total number of observables. This causes OOM errors for large numbers of observables, even for modest qubit counts. The solution is to restrict the requested number of observables into batches of OMP_NUM_THREADS concurrent executions, where the number of statevectors allocated is limited by the number of executing threads. This imposes a small overhead in requiring additional computation across batches, but ensures the user can reach much higher qubit counts for large numbers of concurrent executions.
Description of the Change: The adjoint Jacobian calculation is now batched at the maximum number of OpenMP threads, and allocates enough memory for a given batch only.
Benefits: Reduces overall memory footprint within a given series of concurrent calculations, allowing larger numbers of qubits in a given calculation, prevents OOM errors encountered for large workflows.
Possible Drawbacks: Imposes additional overheads due to repetition between batches.
Related GitHub Issues: