Thyra DefaultMultipliedLinearOp: Caching of intermediate vectors #13738
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
|
Status Flag 'Pre-Test Inspection' - Auto Inspected - Inspection is Not Necessary for this Pull Request. |
|
Status Flag 'Pull Request AutoTester' - Testing Jenkins Projects: Pull Request Auto Testing STARTING (click to expand)Build InformationTest Name: PR_gcc-openmpi-openmp
Jenkins Parameters
Build InformationTest Name: PR_gcc
Jenkins Parameters
Build InformationTest Name: PR_gcc-openmpi_debug
Jenkins Parameters
Build InformationTest Name: PR_clang
Jenkins Parameters
Build InformationTest Name: PR_cuda
Jenkins Parameters
Build InformationTest Name: PR_intel
Jenkins Parameters
Build InformationTest Name: PR_cuda-uvm
Jenkins Parameters
Using Repos:
Pull Request Author: cgcgcg |
Signed-off-by: Christian Glusa <[email protected]>
|
Status Flag 'Pull Request AutoTester' - Error: Jenkins Jobs - A user has pushed a change to the PR before testing completed. NEW EVENT 'committed', ID C_kwDOAsJyMdoAKGE3OWZmZTMxYTkxZjcyMjNmMGZiNjE1MWVhOTc5YmJjNWQ5YTkyYjI... The Jenkins Jobs will be shutdown; Testing of this PR must occur again. |
|
Status Flag 'Pull Request AutoTester' - Jenkins Testing: 1 or more Jobs FAILED Note: Testing will normally be attempted again in approx. 2 Hrs 30 Mins. If a change to the PR source branch occurs, the testing will be attempted again on next available autotester run. Pull Request Auto Testing has FAILED (click to expand)Build InformationTest Name: PR_gcc-openmpi-openmp
Jenkins Parameters
Build InformationTest Name: PR_gcc
Jenkins Parameters
Build InformationTest Name: PR_gcc-openmpi_debug
Jenkins Parameters
Build InformationTest Name: PR_clang
Jenkins Parameters
Build InformationTest Name: PR_cuda
Jenkins Parameters
Build InformationTest Name: PR_intel
Jenkins Parameters
Build InformationTest Name: PR_cuda-uvm
Jenkins Parameters
|
cgcgcg
added
the
AT: RETEST
|
Status Flag 'Pull Request AutoTester' - User Requested Retest - Label AT: RETEST will be reset after testing. |
|
Status Flag 'Pre-Test Inspection' - Auto Inspected - Inspection is Not Necessary for this Pull Request. |
|
Status Flag 'Pull Request AutoTester' - Testing Jenkins Projects: Pull Request Auto Testing STARTING (click to expand)Build InformationTest Name: PR_gcc-openmpi-openmp
Jenkins Parameters
Build InformationTest Name: PR_gcc
Jenkins Parameters
Build InformationTest Name: PR_gcc-openmpi_debug
Jenkins Parameters
Build InformationTest Name: PR_clang
Jenkins Parameters
Build InformationTest Name: PR_cuda
Jenkins Parameters
Build InformationTest Name: PR_intel
Jenkins Parameters
Build InformationTest Name: PR_cuda-uvm
Jenkins Parameters
Using Repos:
Pull Request Author: cgcgcg |
|
Status Flag 'Pull Request AutoTester' - Jenkins Testing: all Jobs PASSED Pull Request Auto Testing has PASSED (click to expand)Build InformationTest Name: PR_gcc-openmpi-openmp
Jenkins Parameters
Build InformationTest Name: PR_gcc
Jenkins Parameters
Build InformationTest Name: PR_gcc-openmpi_debug
Jenkins Parameters
Build InformationTest Name: PR_clang
Jenkins Parameters
Build InformationTest Name: PR_cuda
Jenkins Parameters
Build InformationTest Name: PR_intel
Jenkins Parameters
Build InformationTest Name: PR_cuda-uvm
Jenkins Parameters
|
trilinos-autotester
removed
the
AT: RETEST
|
Status Flag 'Pre-Merge Inspection' - - This Pull Request Requires Inspection... The code must be inspected by a member of the Team before Testing/Merging |
|
All Jobs Finished; status = PASSED, However Inspection must be performed before merge can occur... |
cgcgcg
added
the
AT: AUTOMERGE
bartlettroscoe
approved these changes
Jan 21, 2025
There was a problem hiding this comment.
Looks reasonable. What situations show this to be a performance problem and what is the impact on improved performance with this update? It would just be good to document that somewhere as clear motivation for the need to add caching and extra complexity like this.
My plan for this sort of thing was for the vector space itself to maintain a cache of vectors that got allocated and then released, so that resuse of allocated vectors could be shared across the entire application (that had access to that vector space). But that is way more work and has additional complexity (which is why it was never implemented).
|
Status Flag 'Pre-Merge Inspection' - SUCCESS: The last commit to this Pull Request has been INSPECTED AND APPROVED by [ bartlettroscoe ]! |
|
Status Flag 'Pull Request AutoTester' - Pull Request will be Automerged |
|
Merge on Pull Request# 13738: IS A SUCCESS - Pull Request successfully merged |
trilinos-autotester
removed
the
AT: AUTOMERGE
I made a similar change in Belos recently ( #13469 ). My experience is that generally it is just the GPU platforms where many of these allocations become an issue, and we saw 2-3x speedups of Belos GMRES for SPARC's use case on H100 with that PR. I think having a consistent Trilinos-wide caching/pooling strategy for temporary device memory could be worth looking into. |
It would take some design and refactoring work and creating various knobs and doing profiling to balance reducing allocations/deallocations against a larger memory footprint (which @rppawlo pointed out can be an issue with GPUs). There are pros and cons to local vs. global approaches caching/pooling vectors. (But the abstract interfaces in Thyra would make that much easier to achieve and customize than the concrete Tpetra classes.) |
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.
@trilinos/thyra
Motivation
Caching of intermediate vectors for
MultipliedLinearOpto avoid reallocation during timestepping.Update: Internal customer that proposed this change reported up to a 2x speedup on some time-stepping problems on GPUs.