[JIT] Separate GPU implementation of frozen_conv_add_relu_fusion.cpp (#68149) #69253
Conversation
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
…68149) JIT optimization passes are part of the CPU-only build (i.e. necessary GPU flags are not passed in). This separates the implementation of frozen_conv_add_relu_fusion so that the GPU-enabled implementation is registered at runtime (if it is available) ghstack-source-id: 143676384 Test Plan: In the following script, conv_add_relu fusion is not observed without this change, but is observed when this change is added. ``` from typing import List, Optional import torch class Model(torch.nn.Module): def __init__(self): super().__init__() self.weight = torch.nn.Parameter(torch.rand((3, 3, 7, 7), device="cuda")) self.add_tensor = torch.nn.Parameter(torch.rand((3, 3, 7, 7), device="cuda")) def forward( self, inp: torch.Tensor, bias: Optional[torch.Tensor], stride: List[int], padding: List[int], dilation: List[int], groups: int, ): # weight = torch.zeros((3, 3, 7, 7), device="cuda") inp = inp.to("cuda") conv_result = torch.conv2d( inp, self.weight, bias, stride, padding, dilation, groups ) add_result = conv_result.add_(self.add_tensor) return add_result.relu_() torch.jit.export def make_prediction(self, inp: torch.Tensor): bias = None groups = 1 stride = (1, 1) padding = (0, 0) dilation = (1, 1) return self.forward(inp, bias, stride, padding, dilation, groups) if __name__ == "__main__": # generate some sample input groups = 1 channels_in = 3 channels_out = 3 kernel_size = (7, 7) stride = (1, 1) padding = (0, 0) dilation = (1, 1) inp = torch.rand((64, 3, 432, 432)) weight = torch.rand( (channels_out, channels_in, kernel_size[0], kernel_size[1]), device="cuda" ) bias = None model = Model() model.eval() script = torch.jit.script(model) script = torch.jit.freeze(script) script = torch.jit.optimize_for_inference(script) print("~~~~ FORWARD ~~~~") print(script.graph) print("with preserved_attrs") print(torch.sum(script.forward(inp, bias, stride, padding, dilation, groups))) ``` fbshipit-source-id: c0f10da4b9540c588819efe3ec540baa0fae4b35 [ghstack-poisoned]
CI Flow Status⚛️ CI FlowRuleset - Version:
You can add a comment to the PR and tag @pytorchbot with the following commands: # ciflow rerun, "ciflow/default" will always be added automatically
@pytorchbot ciflow rerun
# ciflow rerun with additional labels "-l <ciflow/label_name>", which is equivalent to adding these labels manually and trigger the rerun
@pytorchbot ciflow rerun -l ciflow/scheduled -l ciflow/slowFor more information, please take a look at the CI Flow Wiki. |
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit 16e51bf (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Please report bugs/suggestions to the (internal) Dr. CI Users group. |
davidberard98
added a commit
that referenced
this pull request
Dec 1, 2021
…68149) JIT optimization passes are part of the CPU-only build (i.e. necessary GPU flags are not passed in). This separates the implementation of frozen_conv_add_relu_fusion so that the GPU-enabled implementation is registered at runtime (if it is available) ghstack-source-id: 143676384 Test Plan: In the following script, conv_add_relu fusion is not observed without this change, but is observed when this change is added. ``` from typing import List, Optional import torch class Model(torch.nn.Module): def __init__(self): super().__init__() self.weight = torch.nn.Parameter(torch.rand((3, 3, 7, 7), device="cuda")) self.add_tensor = torch.nn.Parameter(torch.rand((3, 3, 7, 7), device="cuda")) def forward( self, inp: torch.Tensor, bias: Optional[torch.Tensor], stride: List[int], padding: List[int], dilation: List[int], groups: int, ): # weight = torch.zeros((3, 3, 7, 7), device="cuda") inp = inp.to("cuda") conv_result = torch.conv2d( inp, self.weight, bias, stride, padding, dilation, groups ) add_result = conv_result.add_(self.add_tensor) return add_result.relu_() torch.jit.export def make_prediction(self, inp: torch.Tensor): bias = None groups = 1 stride = (1, 1) padding = (0, 0) dilation = (1, 1) return self.forward(inp, bias, stride, padding, dilation, groups) if __name__ == "__main__": # generate some sample input groups = 1 channels_in = 3 channels_out = 3 kernel_size = (7, 7) stride = (1, 1) padding = (0, 0) dilation = (1, 1) inp = torch.rand((64, 3, 432, 432)) weight = torch.rand( (channels_out, channels_in, kernel_size[0], kernel_size[1]), device="cuda" ) bias = None model = Model() model.eval() script = torch.jit.script(model) script = torch.jit.freeze(script) script = torch.jit.optimize_for_inference(script) print("~~~~ FORWARD ~~~~") print(script.graph) print("with preserved_attrs") print(torch.sum(script.forward(inp, bias, stride, padding, dilation, groups))) ``` fbshipit-source-id: c0f10da4b9540c588819efe3ec540baa0fae4b35 ghstack-source-id: e921804 Pull Request resolved: #69253
facebook-github-bot
added
the
oncall: jit
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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:
JIT optimization passes are part of the CPU-only build (i.e. necessary GPU flags are not passed in). This separates the implementation of frozen_conv_add_relu_fusion so that the GPU-enabled implementation is registered at runtime (if it is available)
Test Plan:
In the following script, conv_add_relu fusion is not observed without this change, but is observed when this change is added.
fbshipit-source-id: c0f10da4b9540c588819efe3ec540baa0fae4b35