[Relax] Add FInferMixedPrecision and FRelaxInferLayout for conv transpose ops #18629

guan404ming · 2025-12-31T15:00:59Z

Why

The conv1d_transpose and conv2d_transpose operators were missing FInferMixedPrecision and FRelaxInferLayout attribute implementations, which are needed for:

Mixed precision training/inference support (e.g., float16 inputs with float32 outputs)
Layout transformation optimizations during compilation
Consistency with conv1d and conv2d operators which already have these attributes

How

Implemented InferLayoutConv1dTranspose and InferMixedPrecisionConv1dTranspose
Implemented InferLayoutConv2dTranspose and InferMixedPrecisionConv2dTranspose

gemini-code-assist · 2025-12-31T15:01:14Z

Summary of Changes

Hello @guan404ming, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request enhances the conv1d_transpose and conv2d_transpose operators by integrating essential functionalities for mixed-precision processing and optimized data layout handling. These additions are crucial for improving the efficiency and flexibility of these operations within the framework, aligning them with the capabilities of their non-transpose counterparts.

Highlights

Mixed Precision Support: Implemented FInferMixedPrecision for conv1d_transpose and conv2d_transpose operators, enabling mixed precision training and inference (e.g., float16 inputs with float32 outputs).
Layout Transformation Optimization: Added FRelaxInferLayout for conv1d_transpose and conv2d_transpose, which facilitates layout transformation optimizations during compilation.
Consistency and Testing: Ensured consistency with existing conv1d and conv2d operators by adding these attributes and included new unit tests to verify the mixed precision inference for both transpose operations.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request adds FInferMixedPrecision and FRelaxInferLayout support for conv1d_transpose and conv2d_transpose operators, which is a great addition for consistency and enabling optimizations. The implementations look correct and are accompanied by new tests. I've provided a few suggestions to improve code clarity and reduce duplication in the C++ implementation, and to enhance readability in the new Python tests.

gemini-code-assist · 2025-12-31T15:04:42Z

src/relax/op/nn/convolution.cc

+    Layout desired_data_layout = (*it).second[0];
+    Layout desired_weight_layout = (*it).second[1];
+    Layout desired_output_layout = (*it).second.size() == 3 ? (*it).second[2] : (*it).second[0];


There is repeated code for accessing desired layouts. To improve readability and maintainability, you can store the desired layout strings in local variables and reuse them. This avoids repeating (*it).second access and the ternary logic for the output layout. You can then use these variables on lines 730-732.

gemini-code-assist · 2025-12-31T15:04:42Z

src/relax/op/nn/convolution.cc

+    Layout desired_data_layout = (*it).second[0];
+    Layout desired_weight_layout = (*it).second[1];
+    Layout desired_output_layout = (*it).second.size() == 3 ? (*it).second[2] : (*it).second[0];


There is repeated code for accessing desired layouts. To improve readability and maintainability, you can store the desired layout strings in local variables at the beginning of the if block and reuse them. This avoids repeating (*it).second access and the ternary logic for the output layout. You can then use these variables to update new_attrs in lines 933-935 and 957-959.

gemini-code-assist · 2025-12-31T15:04:42Z

tests/python/relax/test_op_nn_convolution.py

+    bb = relax.BlockBuilder()
+    x0 = relax.Var("x", R.Tensor((2, 3, 28), "float16"))
+    w0 = relax.Var("w", R.Tensor((3, 4, 3), "float16"))
+    x1 = relax.Var("x", R.Tensor((2, 3, 28), "int8"))
+    w1 = relax.Var("w", R.Tensor((3, 4, 3), "int8"))
+
+    _check_inference(
+        bb,
+        relax.op.nn.conv1d_transpose(x0, w0, out_dtype="float32"),
+        relax.TensorStructInfo((2, 4, 30), "float32"),
+    )
+    _check_inference(
+        bb,
+        relax.op.nn.conv1d_transpose(x1, w1, out_dtype="int32"),
+        relax.TensorStructInfo((2, 4, 30), "int32"),
+    )


For better readability, consider using more descriptive variable names that indicate the data type, like x_f16, w_f16, x_i8, and w_i8.

Suggested change

bb = relax.BlockBuilder()

x0 = relax.Var("x", R.Tensor((2, 3, 28), "float16"))

w0 = relax.Var("w", R.Tensor((3, 4, 3), "float16"))

x1 = relax.Var("x", R.Tensor((2, 3, 28), "int8"))

w1 = relax.Var("w", R.Tensor((3, 4, 3), "int8"))

_check_inference(

bb,

relax.op.nn.conv1d_transpose(x0, w0, out_dtype="float32"),

relax.TensorStructInfo((2, 4, 30), "float32"),

)

_check_inference(

bb,

relax.op.nn.conv1d_transpose(x1, w1, out_dtype="int32"),

relax.TensorStructInfo((2, 4, 30), "int32"),

)

bb = relax.BlockBuilder()

x_f16 = relax.Var("x", R.Tensor((2, 3, 28), "float16"))

w_f16 = relax.Var("w", R.Tensor((3, 4, 3), "float16"))

x_i8 = relax.Var("x", R.Tensor((2, 3, 28), "int8"))

w_i8 = relax.Var("w", R.Tensor((3, 4, 3), "int8"))

_check_inference(

bb,

relax.op.nn.conv1d_transpose(x_f16, w_f16, out_dtype="float32"),

relax.TensorStructInfo((2, 4, 30), "float32"),

)

_check_inference(

bb,

relax.op.nn.conv1d_transpose(x_i8, w_i8, out_dtype="int32"),

relax.TensorStructInfo((2, 4, 30), "int32"),

)

gemini-code-assist · 2025-12-31T15:04:42Z

tests/python/relax/test_op_nn_convolution.py

+    bb = relax.BlockBuilder()
+    x0 = relax.Var("x", R.Tensor((2, 3, 28, 28), "float16"))
+    w0 = relax.Var("w", R.Tensor((3, 4, 3, 3), "float16"))
+    x1 = relax.Var("x", R.Tensor((2, 3, 28, 28), "int8"))
+    w1 = relax.Var("w", R.Tensor((3, 4, 3, 3), "int8"))
+
+    _check_inference(
+        bb,
+        relax.op.nn.conv2d_transpose(x0, w0, out_dtype="float32"),
+        relax.TensorStructInfo((2, 4, 30, 30), "float32"),
+    )
+    _check_inference(
+        bb,
+        relax.op.nn.conv2d_transpose(x1, w1, out_dtype="int32"),
+        relax.TensorStructInfo((2, 4, 30, 30), "int32"),
+    )


For better readability, consider using more descriptive variable names that indicate the data type, like x_f16, w_f16, x_i8, and w_i8.

Suggested change

bb = relax.BlockBuilder()

x0 = relax.Var("x", R.Tensor((2, 3, 28, 28), "float16"))

w0 = relax.Var("w", R.Tensor((3, 4, 3, 3), "float16"))

x1 = relax.Var("x", R.Tensor((2, 3, 28, 28), "int8"))

w1 = relax.Var("w", R.Tensor((3, 4, 3, 3), "int8"))

_check_inference(

bb,

relax.op.nn.conv2d_transpose(x0, w0, out_dtype="float32"),

relax.TensorStructInfo((2, 4, 30, 30), "float32"),

)

_check_inference(

bb,

relax.op.nn.conv2d_transpose(x1, w1, out_dtype="int32"),

relax.TensorStructInfo((2, 4, 30, 30), "int32"),

)

bb = relax.BlockBuilder()

x_f16 = relax.Var("x", R.Tensor((2, 3, 28, 28), "float16"))

w_f16 = relax.Var("w", R.Tensor((3, 4, 3, 3), "float16"))

x_i8 = relax.Var("x", R.Tensor((2, 3, 28, 28), "int8"))

w_i8 = relax.Var("w", R.Tensor((3, 4, 3, 3), "int8"))

_check_inference(

bb,

relax.op.nn.conv2d_transpose(x_f16, w_f16, out_dtype="float32"),

relax.TensorStructInfo((2, 4, 30, 30), "float32"),

)

_check_inference(

bb,

relax.op.nn.conv2d_transpose(x_i8, w_i8, out_dtype="int32"),

relax.TensorStructInfo((2, 4, 30, 30), "int32"),

)

guan404ming · 2026-01-01T10:58:13Z

cc @tlopex @mshr-h

Add FInferMixedPrecision and FRelaxInferLayout for conv transpose ops

5756731

guan404ming force-pushed the add-conv-transpose-infer-attrs branch from 2adc83b to 5756731 Compare December 31, 2025 15:01

gemini-code-assist bot reviewed Dec 31, 2025

View reviewed changes

Fix lint error

f0ce8fb

guan404ming force-pushed the add-conv-transpose-infer-attrs branch from 5040008 to f0ce8fb Compare January 1, 2026 08:40

guan404ming marked this pull request as ready for review January 1, 2026 10:32

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Relax] Add FInferMixedPrecision and FRelaxInferLayout for conv transpose ops #18629

[Relax] Add FInferMixedPrecision and FRelaxInferLayout for conv transpose ops #18629

guan404ming commented Dec 31, 2025

Uh oh!

gemini-code-assist bot commented Dec 31, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Dec 31, 2025

Uh oh!

gemini-code-assist bot Dec 31, 2025

Uh oh!

gemini-code-assist bot Dec 31, 2025

Uh oh!

gemini-code-assist bot Dec 31, 2025

Uh oh!

guan404ming commented Jan 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

[Relax] Add FInferMixedPrecision and FRelaxInferLayout for conv transpose ops #18629

Are you sure you want to change the base?

[Relax] Add FInferMixedPrecision and FRelaxInferLayout for conv transpose ops #18629

Conversation

guan404ming commented Dec 31, 2025

Why

How

Uh oh!

gemini-code-assist bot commented Dec 31, 2025

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Dec 31, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Dec 31, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Dec 31, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Dec 31, 2025

Choose a reason for hiding this comment

Uh oh!

guan404ming commented Jan 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant