trunk/acea0ff1beb8d211cf591ffd5d6836dc5d59cd9c: [ROCm] Unskip AOTI copy, flex guard and FFT stride tests (#197469)
- PyTorch: 1124 events in the last 90 days
- PyTorch: 1110th Release in the last 90 days
- Previous: earlier the same day · ciflow/trunk/197289: [UPDATE] Update
What happened
Four ROCm exclusions no longer describe the code. test_small_constant_pinned_async_copy was marked "CUDA-only", but the pinned staging path in the AOTI runtime (model_base.h) sits under USE_CUDA in the source tree, which the ROCm build hipifies to USE_ROCM (the define the AOTI compile step sets on ROCm), and on gfx950 the test logs the PinnedStagingPool allocation and matches eager. test_cuda_to_cuda_device_copy che…
Summary assembled by rule from the sources below