[ET-VK][Ops] aten.convolution (SlidingWindow)#2812
Closed
junpi3 wants to merge 5 commits intogh/jorgep31415/37/basefrom
Closed
[ET-VK][Ops] aten.convolution (SlidingWindow)#2812junpi3 wants to merge 5 commits intogh/jorgep31415/37/basefrom
junpi3 wants to merge 5 commits intogh/jorgep31415/37/basefrom
Conversation
## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon For our CUNET model, the first two are required and the third is useful. 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/2812
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 8fee216 with merge base d3326a2 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D55346778 |
junpi3
pushed a commit
that referenced
this pull request
Apr 2, 2024
## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon For our CUNET model, the first two are required and the third is useful. 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/) ghstack-source-id: 220997713 Pull Request resolved: #2812
## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon For our CUNET model, the first two are required and the third is useful. 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D55346778 |
junpi3
pushed a commit
that referenced
this pull request
Apr 3, 2024
Pull Request resolved: #2812 ## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon For our CUNET model, the first two are required and the third is useful. 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias ghstack-source-id: 221059609 @exported-using-ghexport Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/)
## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/) [ghstack-poisoned]
This was referenced Apr 3, 2024
junpi3
pushed a commit
that referenced
this pull request
Apr 4, 2024
Pull Request resolved: #2812 ## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias Internal: For our CUNET model, the first two are required and the third is useful. ghstack-source-id: 221233078 @exported-using-ghexport Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/)
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D55346778 |
SS-JIA
approved these changes
Apr 4, 2024
## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D55346778 |
This was referenced Apr 5, 2024
Closed
## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. ## Coming soon 1. Transpose convolution 2. Depthwise convolution (for completeness) 3. Pointwise convolution (for optimization) 4. Null bias Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/) [ghstack-poisoned]
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D55346778 |
junpi3
pushed a commit
that referenced
this pull request
Apr 8, 2024
Pull Request resolved: #2812 ## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. We will handle Transpose, Depthwise, and Pointwise in the rest of this stack. Internal: For our CUNET model, SlidingWindow, Transpose, Depthwise are required and Pointwise is a useful optimization. ghstack-source-id: 221721750 @exported-using-ghexport Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/)
Contributor
|
This pull request has been merged in 5d299fe. |
Closed
kedarnath03
pushed a commit
to kedarnath03/executorch
that referenced
this pull request
Jun 25, 2025
Pull Request resolved: pytorch/executorch#2812 ## The Operator `nn.Module` invocations of [`nn.Conv2d`](https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html#torch.nn.Conv2d) and [`nn.ConvTranspose2d`](https://pytorch.org/docs/stable/generated/torch.nn.ConvTranspose2d.html#torch.nn.ConvTranspose2d) get compiled to `aten.convolution.default` in the Edge Dialect, which carries the signature ``` - func: convolution(Tensor input, Tensor weight, Tensor? bias, int[] stride, SymInt[] padding, int[] dilation, bool transposed, SymInt[] output_padding, int groups) -> Tensor ``` ## Summary (cases handled) We introduce support for the convolution cases covered by [ATen-VK's default SlidingWindow implementation](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L73). This is achieved by - reusing the [existing `conv2d.glsl`](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/glsl/conv2d.glsl), and - [moving special weights prepacking from CPU](https://github.com/pytorch/pytorch/blob/09c72eaa3f69f90402c86a30abf4fc621298578c/aten/src/ATen/native/vulkan/ops/Convolution.cpp#L134-L235) to the GPU in `conv2d_prepack_weights.glsl`. We also include resizing support for dynamic shapes. Note that only height and width of the input can vary. ## Cases not handled The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing: 1. **Groups G > 1.** Largely not covered by ATen-VK. `G = in_channels` is covered by ATen-VK's Depthwise impl and will be added soon. 2. **Batch (input) N > 1.** Not covered by ATen-VK. 3. **Padding > 0 while Dilation, Kernel > 1.** Not covered by ATen-VK. We will handle Transpose, Depthwise, and Pointwise in the rest of this stack. Internal: For our CUNET model, SlidingWindow, Transpose, Depthwise are required and Pointwise is a useful optimization. ghstack-source-id: 221526243 @exported-using-ghexport Differential Revision: [D55346778](https://our.internmc.facebook.com/intern/diff/D55346778/)
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Stack from ghstack (oldest at bottom):
The Operator
nn.Moduleinvocations ofnn.Conv2dandnn.ConvTranspose2dget compiled toaten.convolution.defaultin the Edge Dialect, which carries the signatureSummary (cases handled)
We introduce support for the convolution cases covered by ATen-VK's default SlidingWindow implementation. This is achieved by
conv2d.glsl, andconv2d_prepack_weights.glsl.We also include resizing support for dynamic shapes. Note that only height and width of the input can vary.
Cases not handled
The implementation is on-par with ATen-VK's SlidingWindow. This means the following cases are missing:
G = in_channelsis covered by ATen-VK's Depthwise impl and will be added soon.Coming soon
Differential Revision: D55346778