Description

Summary: Test for roi_align with GPU wrapper and dynamic shapes failing on CPU and sGPU platforms.

Test Class: inductor/test_gpu_cpp_wrapper.py::DynamicShapesGpuWrapperGpuTests
Number of Failing Tests: 1
Platform: CPU, sGPU (CUDA)
Test Type: Unit Test

Version Information:

• PyTorch Commit: 6bdd8c9
• Branch: main
• Test Date: 2026-01-14
• Sprint: Sprint 24

Failure Pattern:
Single root cause - torchvision roi_align operator not available for CUDA backend

Common Error:
code
NotImplementedError: Could not run 'torchvision::roi_align' with arguments from the 'CUDA' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process.
'torchvision::roi_align' is only available for these backends: [CPU, Meta, QuantizedCPU, ...]
code

Failing Tests:
1. test_roi_align_cuda_dynamic_shapes_gpu_wrapper

Steps to Reproduce:
code
TEST_CONFIG=cpu python3 test/run_test.py -i inductor/test_gpu_cpp_wrapper -k test_roi_align_cuda_dynamic_shapes_gpu_wrapper
code

Expected Result:
Test should pass with torchvision CUDA backend available

Actual Result:
Test fails because torchvision::roi_align is not registered for CUDA backend

Root Cause Analysis:
The torchvision operator roi_align is missing CUDA kernel registration. This is likely due to incomplete build of torchvision with CUDA support or selective build excluding torchvision CUDA operators.

Priority: P2