Fix placeholder lifetime bug in memory planning #6971
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/6971
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 7c8271e with merge base a347665 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
facebook-github-bot
force-pushed
the
export-D66184849
branch
from
5abeae6
to
8bb10bb
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
tarun292
added a commit
that referenced
this pull request
Nov 21, 2024
Summary:
If we have a node that is inserted between placeholders via a pass after to_edge then there is a bug in memory planning lifetime calculations which results in two placeholders being allocated the same memory segment.
Two placeholders should never be using the same memory segment and to prevent this we set the beginning lifetime of the placeholder to be always 0 in the memory planning pass.
In the test case added this is the graph:
```
graph():
%a : [num_users=1] = placeholder[target=a]
%aten_permute_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%a, [1, 0, 2]), kwargs = {})
%b : [num_users=2] = placeholder[target=b]
%aten_add_tensor : [num_users=2] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_permute_copy_default, %b), kwargs = {})
%aten_add_tensor_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_add_tensor, %b), kwargs = {})
return (aten_add_tensor, aten_add_tensor_1)
```
Without this fix the lifetimes of placeholders a and b are:
`a => [0, 2]`
`b => [3, 7]`
Thus the same memory segment is allocated for both of them.
After this fix the lifetimes of the placeholders a and b are:
`a => [0, 2]`
`b => [0, 7]`
Differential Revision: D66184849
tarun292
force-pushed
the
export-D66184849
branch
from
8bb10bb
to
bb3f155
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
JacobSzwejbka
approved these changes
Nov 22, 2024
tarun292
added a commit
that referenced
this pull request
Nov 27, 2024
Summary:
If we have a node that is inserted between placeholders via a pass after to_edge then there is a bug in memory planning lifetime calculations which results in two placeholders being allocated the same memory segment.
Two placeholders should never be using the same memory segment and to prevent this we set the beginning lifetime of the placeholder to be always 0 in the memory planning pass.
In the test case added this is the graph:
```
graph():
%a : [num_users=1] = placeholder[target=a]
%aten_permute_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%a, [1, 0, 2]), kwargs = {})
%b : [num_users=2] = placeholder[target=b]
%aten_add_tensor : [num_users=2] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_permute_copy_default, %b), kwargs = {})
%aten_add_tensor_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_add_tensor, %b), kwargs = {})
return (aten_add_tensor, aten_add_tensor_1)
```
Without this fix the lifetimes of placeholders a and b are:
`a => [0, 2]`
`b => [3, 7]`
Thus the same memory segment is allocated for both of them.
After this fix the lifetimes of the placeholders a and b are:
`a => [0, 2]`
`b => [0, 7]`
Reviewed By: JacobSzwejbka
Differential Revision: D66184849
tarun292
force-pushed
the
export-D66184849
branch
from
bb3f155
to
b9ad5a6
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
facebook-github-bot
pushed a commit
that referenced
this pull request
Nov 27, 2024
Summary:
If we have a node that is inserted between placeholders via a pass after to_edge then there is a bug in memory planning lifetime calculations which results in two placeholders being allocated the same memory segment.
Two placeholders should never be using the same memory segment and to prevent this we set the beginning lifetime of the placeholder to be always 0 in the memory planning pass.
In the test case added this is the graph:
```
graph():
%a : [num_users=1] = placeholder[target=a]
%aten_permute_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%a, [1, 0, 2]), kwargs = {})
%b : [num_users=2] = placeholder[target=b]
%aten_add_tensor : [num_users=2] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_permute_copy_default, %b), kwargs = {})
%aten_add_tensor_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_add_tensor, %b), kwargs = {})
return (aten_add_tensor, aten_add_tensor_1)
```
Without this fix the lifetimes of placeholders a and b are:
`a => [0, 2]`
`b => [3, 7]`
Thus the same memory segment is allocated for both of them.
After this fix the lifetimes of the placeholders a and b are:
`a => [0, 2]`
`b => [0, 7]`
Reviewed By: JacobSzwejbka
Differential Revision: D66184849
facebook-github-bot
force-pushed
the
export-D66184849
branch
from
b9ad5a6
to
a7c7528
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
facebook-github-bot
pushed a commit
that referenced
this pull request
Nov 27, 2024
Summary:
If we have a node that is inserted between placeholders via a pass after to_edge then there is a bug in memory planning lifetime calculations which results in two placeholders being allocated the same memory segment.
Two placeholders should never be using the same memory segment and to prevent this we set the beginning lifetime of the placeholder to be always 0 in the memory planning pass.
In the test case added this is the graph:
```
graph():
%a : [num_users=1] = placeholder[target=a]
%aten_permute_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%a, [1, 0, 2]), kwargs = {})
%b : [num_users=2] = placeholder[target=b]
%aten_add_tensor : [num_users=2] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_permute_copy_default, %b), kwargs = {})
%aten_add_tensor_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_add_tensor, %b), kwargs = {})
return (aten_add_tensor, aten_add_tensor_1)
```
Without this fix the lifetimes of placeholders a and b are:
`a => [0, 2]`
`b => [3, 7]`
Thus the same memory segment is allocated for both of them.
After this fix the lifetimes of the placeholders a and b are:
`a => [0, 2]`
`b => [0, 7]`
Reviewed By: JacobSzwejbka
Differential Revision: D66184849
facebook-github-bot
force-pushed
the
export-D66184849
branch
from
a7c7528
to
9aa41fe
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
Summary:
If we have a node that is inserted between placeholders via a pass after to_edge then there is a bug in memory planning lifetime calculations which results in two placeholders being allocated the same memory segment.
Two placeholders should never be using the same memory segment and to prevent this we set the beginning lifetime of the placeholder to be always 0 in the memory planning pass.
In the test case added this is the graph:
```
graph():
%a : [num_users=1] = placeholder[target=a]
%aten_permute_copy_default : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.permute_copy.default](args = (%a, [1, 0, 2]), kwargs = {})
%b : [num_users=2] = placeholder[target=b]
%aten_add_tensor : [num_users=2] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_permute_copy_default, %b), kwargs = {})
%aten_add_tensor_1 : [num_users=1] = call_function[target=executorch.exir.dialects.edge._ops.aten.add.Tensor](args = (%aten_add_tensor, %b), kwargs = {})
return (aten_add_tensor, aten_add_tensor_1)
```
Without this fix the lifetimes of placeholders a and b are:
`a => [0, 2]`
`b => [3, 7]`
Thus the same memory segment is allocated for both of them.
After this fix the lifetimes of the placeholders a and b are:
`a => [0, 2]`
`b => [0, 7]`
Reviewed By: JacobSzwejbka
Differential Revision: D66184849
facebook-github-bot
force-pushed
the
export-D66184849
branch
from
9aa41fe
to
7c8271e
Compare
|
This pull request was exported from Phabricator. Differential Revision: D66184849 |
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.
Differential Revision: D66184849