-
Notifications
You must be signed in to change notification settings - Fork 3.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[METAL] Fix memory leaks in Metal runtime #7714
Merged
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
1. In case when we build runtime without ARC, we can have problems with memory releasing. Due to some of Objective-C methods returns autoreleased pointers, we should specify `autoreleasepool` blocks to determine life cycle of these pointers. 2. Added workaround for problem with work group size. Sometimes auto scheduler generates parameters when work group size is more than possible. And in this case we got assert from Metal library. Added check for this situation and it helps to avoid assert. 3. Fixed memory leak problem when fill tensor by random data. DLManagedTensor increases reference counter in NDArray but nobody delete this DLManagedTensor in proper way. This is why memory which was allocated by NDArray was never released. 4. Removed unnecessary retains. It is not necessary use retain in some places where they were used, due to we build metal runtime without ARC.
tqchen
requested changes
Mar 21, 2021
tqchen
requested changes
Mar 21, 2021
tqchen
approved these changes
Mar 23, 2021
Thanks @echuraev |
mehrdadh
pushed a commit
to mehrdadh/tvm
that referenced
this pull request
Mar 23, 2021
* [METAL] Fix memory leaks in Metal runtime 1. In case when we build runtime without ARC, we can have problems with memory releasing. Due to some of Objective-C methods returns autoreleased pointers, we should specify `autoreleasepool` blocks to determine life cycle of these pointers. 2. Added workaround for problem with work group size. Sometimes auto scheduler generates parameters when work group size is more than possible. And in this case we got assert from Metal library. Added check for this situation and it helps to avoid assert. 3. Fixed memory leak problem when fill tensor by random data. DLManagedTensor increases reference counter in NDArray but nobody delete this DLManagedTensor in proper way. This is why memory which was allocated by NDArray was never released. 4. Removed unnecessary retains. It is not necessary use retain in some places where they were used, due to we build metal runtime without ARC. * Use const_cast instead of creation DLManagedTensor
mehrdadh
added a commit
to mehrdadh/tvm
that referenced
this pull request
Mar 23, 2021
This reverts commit 917b8b3.
mehrdadh
added a commit
to mehrdadh/tvm
that referenced
this pull request
Mar 23, 2021
This reverts commit 917b8b3.
echuraev
added a commit
to Deelvin/tvm
that referenced
this pull request
Apr 23, 2021
* [METAL] Fix memory leaks in Metal runtime 1. In case when we build runtime without ARC, we can have problems with memory releasing. Due to some of Objective-C methods returns autoreleased pointers, we should specify `autoreleasepool` blocks to determine life cycle of these pointers. 2. Added workaround for problem with work group size. Sometimes auto scheduler generates parameters when work group size is more than possible. And in this case we got assert from Metal library. Added check for this situation and it helps to avoid assert. 3. Fixed memory leak problem when fill tensor by random data. DLManagedTensor increases reference counter in NDArray but nobody delete this DLManagedTensor in proper way. This is why memory which was allocated by NDArray was never released. 4. Removed unnecessary retains. It is not necessary use retain in some places where they were used, due to we build metal runtime without ARC. * Use const_cast instead of creation DLManagedTensor
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
May 6, 2021
* [METAL] Fix memory leaks in Metal runtime 1. In case when we build runtime without ARC, we can have problems with memory releasing. Due to some of Objective-C methods returns autoreleased pointers, we should specify `autoreleasepool` blocks to determine life cycle of these pointers. 2. Added workaround for problem with work group size. Sometimes auto scheduler generates parameters when work group size is more than possible. And in this case we got assert from Metal library. Added check for this situation and it helps to avoid assert. 3. Fixed memory leak problem when fill tensor by random data. DLManagedTensor increases reference counter in NDArray but nobody delete this DLManagedTensor in proper way. This is why memory which was allocated by NDArray was never released. 4. Removed unnecessary retains. It is not necessary use retain in some places where they were used, due to we build metal runtime without ARC. * Use const_cast instead of creation DLManagedTensor
trevor-m
pushed a commit
to neo-ai/tvm
that referenced
this pull request
May 11, 2021
* [METAL] Fix memory leaks in Metal runtime 1. In case when we build runtime without ARC, we can have problems with memory releasing. Due to some of Objective-C methods returns autoreleased pointers, we should specify `autoreleasepool` blocks to determine life cycle of these pointers. 2. Added workaround for problem with work group size. Sometimes auto scheduler generates parameters when work group size is more than possible. And in this case we got assert from Metal library. Added check for this situation and it helps to avoid assert. 3. Fixed memory leak problem when fill tensor by random data. DLManagedTensor increases reference counter in NDArray but nobody delete this DLManagedTensor in proper way. This is why memory which was allocated by NDArray was never released. 4. Removed unnecessary retains. It is not necessary use retain in some places where they were used, due to we build metal runtime without ARC. * Use const_cast instead of creation DLManagedTensor
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.
memory releasing. Due to some of Objective-C methods returns
autoreleased pointers, we should specify
autoreleasepool
blocks todetermine life cycle of these pointers.
Sometimes auto scheduler generates parameters when work group size
is more than possible. And in this case we got assert from Metal
library. Added check for this situation and it helps to avoid
assert.
DLManagedTensor increases reference counter in NDArray but nobody
delete this DLManagedTensor in proper way. This is why memory which
was allocated by NDArray was never released.
places where they were used, due to we build metal runtime without
ARC.
Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from Reviewers by @ them in the pull request thread.