The Conv alpaka GPU tests cover the SAME_LOWER autopad branch (ConvWithAutopadSameLower) but not SAME_UPPER. The codegen handles both modes in Initialize, so the SAME_UPPER path is generated but never exercised. Upstream ROOT tests both lower and upper
ConvWithAutopadSameUpper.onnx + reference already exist there); the fork has only the lower model and test.
This is a coverage gap, not a known bug: add a ConvWithAutopadSameUpper model (portable from ROOT's input_models) with its reference, and a TEST_F in TestCustomModelsFromONNXForAlpakaCuda.cxx following the existing ConvWithAutopadSameLower pattern. The fork's CPU test (TestCustomModelsFromONNX.cxx) is also missing it and can be added in the same PR.
The Conv alpaka GPU tests cover the SAME_LOWER autopad branch (ConvWithAutopadSameLower) but not SAME_UPPER. The codegen handles both modes in Initialize, so the SAME_UPPER path is generated but never exercised. Upstream ROOT tests both lower and upper
ConvWithAutopadSameUpper.onnx + reference already exist there); the fork has only the lower model and test.
This is a coverage gap, not a known bug: add a ConvWithAutopadSameUpper model (portable from ROOT's input_models) with its reference, and a
TEST_FinTestCustomModelsFromONNXForAlpakaCuda.cxxfollowing the existing ConvWithAutopadSameLower pattern. The fork's CPU test (TestCustomModelsFromONNX.cxx) is also missing it and can be added in the same PR.