[GPU] Fix invalid input format of fused ops for resample opt #32654
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
+ Reorder caused unexpected format change Signed-off-by: Min,Byungil <[email protected]>
Signed-off-by: Min,Byungil <[email protected]>
Signed-off-by: Min,Byungil <[email protected]>
Signed-off-by: Min,Byungil <[email protected]>
Lyamin-Roman
reviewed
Nov 4, 2025
src/plugins/intel_gpu/src/kernel_selector/kernels/resample/resample_kernel_opt.cpp
Show resolved
Hide resolved
byungilm
changed the title
hyunback
reviewed
Nov 5, 2025
src/plugins/intel_gpu/src/graph/graph_optimizer/remove_redundant_reorders.cpp
Outdated
Show resolved
Hide resolved
+ Not to add Reorder for resample in reorder_inputs + Not to add logic into remove_redundant_reorder Signed-off-by: Min,Byungil <[email protected]>
Signed-off-by: Min,Byungil <[email protected]>
hyunback
reviewed
Nov 5, 2025
Signed-off-by: Min,Byungil <[email protected]>
isanghao
changed the title
Signed-off-by: Min,Byungil <[email protected]>
Signed-off-by: Min,Byungil <[email protected]>
hyunback
reviewed
Nov 6, 2025
isanghao
reviewed
Nov 6, 2025
isanghao
reviewed
Nov 6, 2025
Signed-off-by: Min,Byungil <[email protected]>
isanghao
reviewed
Nov 6, 2025
| std::vector<size_t> get_shape_infer_dependencies() const override { return {1, 2}; } | ||
| std::vector<size_t> get_shape_infer_dependencies() const override { | ||
| // if vector of sizes or scales exists, resample in CreateInterpolateOp generates no dependency for inputs of sizes and scales | ||
| if (typed_desc()->sizes.size() != 0 || typed_desc()->scales.size() != 0) |
Contributor
There was a problem hiding this comment.
nit: could you add an assertion so that we can prevent mistake in the future?
// Both sizes and scales should be fed as attribute
OPENVINO_ASSERT(typed_desc()->sizes.size() != 0 && typed_desc()->scales.size() != 0)
Contributor
Author
There was a problem hiding this comment.
Applied assertion to prevent mistake
Contributor
Author
There was a problem hiding this comment.
Applied assertion to the if condition.
+ add assertion to prevent possible mistake Signed-off-by: Min,Byungil <[email protected]>
isanghao
approved these changes
Nov 6, 2025
Signed-off-by: Min,Byungil <[email protected]>
byungilm
added a commit
to byungilm/openvino
that referenced
this pull request
Nov 7, 2025
…otoolkit#32654) + Reorder caused unexpected input format for resample_opt kernel's fused ops ### Description of the issue - A Reorder is added between the Resample and its dep node, 'conv.81', at reorder_inputs. Input preferred format of the Resample was simple format while 'conv.81' has blocked format output. <img width="1145" height="39" alt="image" src="https://github.com/user-attachments/assets/5af9325d-0d76-439b-9a24-5650cf088951" /> - The reason of simple format selection is that get_shape_infer_dependencies() of Resample returned dependency of input1,2 However, this resample primitive did not have input1,2 as the shape_infer_dependencies because it is already converted to attribute. Input1 was actually an input from fused operation. - Dependency result from get_shape_infer_dependencies() of Resample was invalid. If vector values of sizes and scales exists, it means Resample has no dependency for input1 and input2. #### The code and line that caused this issue - When Resample primitive is created, input1 and 2 may be converted to attribute from this code src/plugins/intel_gpu/src/plugin/ops/interpolate.cpp - missing logic in resample_opt.cpp to prevent simple format post-ops #### Reproduction step and snapshot - Reproduced by benchmark `./benchmark_app -m inference.xml -d GPU.1 -nireq 1 -niter 1 -data_shape x[1,3,736,608]` #### Checklist - [x] Is it a proper fix? - [X] Did you include test case for this fix, if necessary? - [x] Did you review existing test that can be extended to cover this scenario? Passed llm_bench ### Tickets: - CVS-175824 --------- Signed-off-by: Min,Byungil <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
Description of the issue
Input preferred format of the Resample was simple format while 'conv.81' has blocked format output.
However, this resample primitive did not have input1,2 as the shape_infer_dependencies because it is already converted to attribute. Input1 was actually an input from fused operation.
If vector values of sizes and scales exists, it means Resample has no dependency for input1 and input2.
The code and line that caused this issue
src/plugins/intel_gpu/src/plugin/ops/interpolate.cpp
Reproduction step and snapshot
./benchmark_app -m inference.xml -d GPU.1 -nireq 1 -niter 1 -data_shape x[1,3,736,608]Checklist
Tickets: