Reorganize CUDAScopedContext #355
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
Motivation is that the acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier.
Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics).
Validation summaryReference release CMSSW_10_6_0 at b45186e
|
|
No changes in physics performance, as expected, |
|
No changes in performance observed on the T4:
|
fwyzard
approved these changes
Jun 20, 2019
I agree that it's a bit confusing. I could add |
|
Yes, I think that would make things even clear, thank you. |
fwyzard
pushed a commit
that referenced
this pull request
Oct 8, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Oct 19, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Oct 20, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Oct 23, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Nov 6, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Nov 16, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
added a commit
that referenced
this pull request
Nov 27, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Dec 25, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Dec 29, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
fwyzard
pushed a commit
that referenced
this pull request
Dec 29, 2020
* Split CUDAScopedContext to *Acquire and *Produce The motivation is that acquire() and produce() need a different functionality, and are constructed differently (e.g. acquire version always needs the edm::WaitingTaskWithArenaHolder). This split should make it more difficult to make mistakes. It should also make future evolution, e.g. towards chains of TBB tasks alternating in CPU and GPU work, easier. * Rename CUDAContextToken to CUDAContextState, and change semantics Now CUDAScopedContextAcquire takes it as a parameter to constructor, and stores the state in its destructor (yielding RAII semantics). * Document the constructors.
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.
PR description:
This PR reorganizes
CUDAScopedContextin two waysCUDAScopedContexttoCUDAScopedContextAcquireandCUDAScopedContextProduceCUDAScopedContexthas a mixture of possible operations that is based on how it was constructed. This is rather error prone, e.g. I remember forgetting several times to passedm::WaitingTaskWithArenaHolderto it leading to weird crashes. NowCUDAScopedContextAcquirealways requires the holder.CUDAContextTokentoCUDAContextState, and instead of explicitly saving the state ofCUDAScopedContextAcquireinto it, passCUDAContextStateto the constructor ofCUDAScopedContextAcquireso that the latter can save the state on its destructor.std::mutex(as member variable) and locks (local variable)PR validation:
Profiling workflow runs.