introduce _to_dim_order_copy op to runtime #1970
Closed
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/1970
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 7090428 with merge base 53078c4 ( 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: D53747744 |
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
force-pushed
the
export-D53747744
branch
from
175d288 to
892098b
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
force-pushed
the
export-D53747744
branch
from
892098b to
6d60a22
Compare
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Feb 24, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
pushed a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Feb 29, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: https://internalfb.com/D53747744
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Feb 29, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
6d60a22 to
1ba0fa6
Compare
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 1, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
1ba0fa6 to
0238c2c
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 1, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 1, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
0238c2c to
001c37f
Compare
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 4, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 4, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
001c37f to
f6d1369
Compare
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 4, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 4, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 4, 2024
Summary: Pull Request resolved: pytorch#1970 Differential Revision: D53747744
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
Mar 5, 2024
Summary: bypass-github-export-checks Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
f6d1369 to
beaa4db
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Reviewed By: larryliu0820 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
c8046cc to
edbc9ad
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
pushed a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: Pull Request resolved: pytorch#1970 This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Differential Revision: https://internalfb.com/D53747744
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Reviewed By: larryliu0820 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
edbc9ad to
a874034
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
pushed a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: Pull Request resolved: pytorch#1970 This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Differential Revision: https://internalfb.com/D53747744
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Reviewed By: larryliu0820 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
a874034 to
c7cc8f4
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Reviewed By: larryliu0820 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
c7cc8f4 to
d889fd7
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
pushed a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: Pull Request resolved: pytorch#1970 This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Differential Revision: https://internalfb.com/D53747744
Summary: This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Reviewed By: larryliu0820 Differential Revision: D53747744
Gasoonjia
force-pushed
the
export-D53747744
branch
from
d889fd7 to
7090428
Compare
|
This pull request was exported from Phabricator. Differential Revision: D53747744 |
Gasoonjia
added a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Reviewed By: larryliu0820 Differential Revision: D53747744
Gasoonjia
pushed a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 3, 2024
Summary: Pull Request resolved: pytorch#1970 This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Differential Revision: https://internalfb.com/D53747744
Gasoonjia
pushed a commit
to Gasoonjia/executorch-1
that referenced
this pull request
May 4, 2024
Summary: Pull Request resolved: pytorch#1970 This diff creates a new and special operator, `_to_dim_order_copy`. This new operator introduces two critical attributes to runtime system: 1. Extract memory_format information from tensor based on dim_order, instead of stride. 2. Support both channal_last and contiguous memory_format in runtime. Please note that memory format here is a parallel concept with memory layout, and supporting new format does not violate our contract on only supporting contiguous memory layout tensor. Details can be found in [here](https://discuss.pytorch.org/t/contigious-vs-non-contigious-tensor/30107) and [here](https://pytorch.org/blog/tensor-memory-format-matters/). Furthermore, dim order is a specifial operator, which does not have a native aten variant but is needed by most model in edge dialect, so it can not directly be put in `kernels/portable/custom_ops.yaml` (need manually registered everytime, not work for an operator needed by many models), or `kernels/portable/functions.yaml` (should have native aten variant). To overcome that, this diff puts `_to_dim_order_copy`'s aten mode under `kernels/aten`, while lean mode under `kernels/portable/functions.yaml`. Also update dependencies and utils. Differential Revision: https://internalfb.com/D53747744
|
This pull request has been merged in cf40792. |
nathanaelsee
added a commit
to nathanaelsee/executorch
that referenced
this pull request
Jun 12, 2024
Summary:
When converting an ExportedProgram to EdgeProgramManager via to_edge, _to_dim_order_copy or _to_copy will be used depending on the passed EdgeCompileConfig having _skip_dim_order set to true or false. For example:
```
compile_config=EdgeCompileConfig(
_check_ir_validity=False,
_skip_dim_order=True,
),
```
The newer _to_dim_order_copy op was added in D53747744/pytorch#1970.
Adding a param to the pass init to determine if we use the newer op or not.
Differential Revision: D58395616
nathanaelsee
added a commit
to nathanaelsee/executorch
that referenced
this pull request
Jun 12, 2024
Summary: Pull Request resolved: pytorch#3968 When converting an ExportedProgram to EdgeProgramManager via to_edge, _to_dim_order_copy or _to_copy will be used depending on the passed EdgeCompileConfig having _skip_dim_order set to true or false. For example: ``` compile_config=EdgeCompileConfig( _check_ir_validity=False, _skip_dim_order=True, ), ``` The newer _to_dim_order_copy op was added in D53747744/pytorch#1970. Adding a param to the pass init to determine if we use the newer op or not. Reviewed By: copyrightly Differential Revision: D58395616
nathanaelsee
added a commit
to nathanaelsee/executorch
that referenced
this pull request
Jun 13, 2024
Summary: Pull Request resolved: pytorch#3968 When converting an ExportedProgram to EdgeProgramManager via to_edge, _to_dim_order_copy or _to_copy will be used depending on the passed EdgeCompileConfig having _skip_dim_order set to true or false. For example: ``` compile_config=EdgeCompileConfig( _check_ir_validity=False, _skip_dim_order=True, ), ``` The newer _to_dim_order_copy op was added in D53747744/pytorch#1970. Adding a param to the pass init to determine if we use the newer op or not. Reviewed By: copyrightly Differential Revision: D58395616
facebook-github-bot
pushed a commit
that referenced
this pull request
Jun 13, 2024
Summary: Pull Request resolved: #3968 When converting an ExportedProgram to EdgeProgramManager via to_edge, _to_dim_order_copy or _to_copy will be used depending on the passed EdgeCompileConfig having _skip_dim_order set to true or false. For example: ``` compile_config=EdgeCompileConfig( _check_ir_validity=False, _skip_dim_order=True, ), ``` The newer _to_dim_order_copy op was added in D53747744/#1970. Adding a param to the pass init to determine if we use the newer op or not. Reviewed By: copyrightly Differential Revision: D58395616 fbshipit-source-id: 8f5289f8ed62a5c3814ec32b91fb055a10944fdb
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: D53747744