/******************************************************************************* * * MIT License * * Copyright (c) 2024 Advanced Micro Devices, Inc. * * Permission is hereby granted, free of charge, to any person obtaining a copy * of this software and associated documentation files (the "Software"), to deal * in the Software without restriction, including without limitation the rights * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell * copies of the Software, and to permit persons to whom the Software is * furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice shall be included in all * copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. * *******************************************************************************/ #include #include #include #include "bn_test_data.hpp" #include "get_handle.hpp" // Test that batch norm training APIs reject invalid inputs matching PyTorch behavior // PyTorch errors: // - Spatial: "Expected more than 1 value per channel when training" // - PerActivation: "Expected more than 1 value per channel when training" (N > 1) struct GPU_BNInputValidation2D_FP32 : public testing::TestWithParam< std::tuple> { }; struct GPU_BNInputValidation3D_FP32 : public testing::TestWithParam< std::tuple> { }; TEST_P(GPU_BNInputValidation2D_FP32, RejectsInvalidInput) { auto [config, layout, bn_mode] = GetParam(); auto&& handle = get_handle(); float alpha = 1.0f, beta = 0.0f; if(config.Direction == miopen::batchnorm::Direction::ForwardTraining) { BNFwdTrainTestData data; data.SetUpImpl(config, bn_mode, layout); auto status = miopenBatchNormalizationForwardTraining(&handle, bn_mode, &alpha, &beta, &data.input.desc, data.in_dev.get(), &data.output.desc, data.out_dev.get(), &data.scale.desc, data.scale_dev.get(), data.shift_dev.get(), data.averageFactor, data.runMean_dev.get(), data.runVariance_dev.get(), data.epsilon, data.saveMean_dev.get(), data.saveVariance_dev.get()); const char* mode_str = (bn_mode == miopenBNSpatial) ? "Spatial" : "PerActivation"; const char* layout_str = (layout == miopenTensorNCHW) ? "NCHW" : "NHWC"; EXPECT_EQ(status, miopenStatusBadParm) << "ForwardTraining should reject " << mode_str << " mode with " << layout_str << " layout: N=" << config.N << " H=" << config.H << " W=" << config.W << " (N*H*W=" << (config.N * config.H * config.W) << ")"; } else if(config.Direction == miopen::batchnorm::Direction::Backward) { BNBwdTestData data; data.SetUpImpl(config, bn_mode, layout); auto status = miopenBatchNormalizationBackward(&handle, bn_mode, &alpha, &beta, &alpha, &beta, &data.input.desc, data.in_dev.get(), &data.dy.desc, data.dy_dev.get(), &data.output.desc, data.out_dev.get(), &data.bnScale.desc, data.bnScale_dev.get(), data.dScale_dev.get(), data.dBias_dev.get(), data.epsilon, data.savedMean_dev.get(), data.savedInvVar_dev.get()); const char* mode_str = (bn_mode == miopenBNSpatial) ? "Spatial" : "PerActivation"; const char* layout_str = (layout == miopenTensorNCHW) ? "NCHW" : "NHWC"; EXPECT_EQ(status, miopenStatusBadParm) << "Backward should reject " << mode_str << " mode with " << layout_str << " layout: N=" << config.N << " H=" << config.H << " W=" << config.W << " (N*H*W=" << (config.N * config.H * config.W) << ")"; } } TEST_P(GPU_BNInputValidation3D_FP32, RejectsInvalidInput) { auto [config, layout, bn_mode] = GetParam(); auto&& handle = get_handle(); float alpha = 1.0f, beta = 0.0f; if(config.Direction == miopen::batchnorm::Direction::ForwardTraining) { BNFwdTrainTestData data; data.SetUpImpl(config, bn_mode, layout); auto status = miopenBatchNormalizationForwardTraining(&handle, bn_mode, &alpha, &beta, &data.input.desc, data.in_dev.get(), &data.output.desc, data.out_dev.get(), &data.scale.desc, data.scale_dev.get(), data.shift_dev.get(), data.averageFactor, data.runMean_dev.get(), data.runVariance_dev.get(), data.epsilon, data.saveMean_dev.get(), data.saveVariance_dev.get()); const char* mode_str = (bn_mode == miopenBNSpatial) ? "Spatial" : "PerActivation"; const char* layout_str = (layout == miopenTensorNCDHW) ? "NCDHW" : "NDHWC"; EXPECT_EQ(status, miopenStatusBadParm) << "ForwardTraining 3D should reject " << mode_str << " mode with " << layout_str << " layout: N=" << config.N << " D=" << config.D << " H=" << config.H << " W=" << config.W << " (N*D*H*W=" << (config.N * config.D * config.H * config.W) << ")"; } else if(config.Direction == miopen::batchnorm::Direction::Backward) { BNBwdTestData data; data.SetUpImpl(config, bn_mode, layout); auto status = miopenBatchNormalizationBackward(&handle, bn_mode, &alpha, &beta, &alpha, &beta, &data.input.desc, data.in_dev.get(), &data.dy.desc, data.dy_dev.get(), &data.output.desc, data.out_dev.get(), &data.bnScale.desc, data.bnScale_dev.get(), data.dScale_dev.get(), data.dBias_dev.get(), data.epsilon, data.savedMean_dev.get(), data.savedInvVar_dev.get()); const char* mode_str = (bn_mode == miopenBNSpatial) ? "Spatial" : "PerActivation"; const char* layout_str = (layout == miopenTensorNCDHW) ? "NCDHW" : "NDHWC"; EXPECT_EQ(status, miopenStatusBadParm) << "Backward 3D should reject " << mode_str << " mode with " << layout_str << " layout: N=" << config.N << " D=" << config.D << " H=" << config.H << " W=" << config.W << " (N*D*H*W=" << (config.N * config.D * config.H * config.W) << ")"; } } // Instantiate tests for 2D with all combinations of layouts and modes INSTANTIATE_TEST_SUITE_P( Smoke, GPU_BNInputValidation2D_FP32, testing::Combine(testing::ValuesIn(Network2DInvalidTraining()), testing::ValuesIn({miopenTensorNCHW, miopenTensorNHWC}), testing::ValuesIn({miopenBNSpatial, miopenBNPerActivation}))); // Instantiate tests for 3D with all combinations of layouts and modes INSTANTIATE_TEST_SUITE_P( Smoke, GPU_BNInputValidation3D_FP32, testing::Combine(testing::ValuesIn(Network3DInvalidTraining()), testing::ValuesIn({miopenTensorNCDHW, miopenTensorNDHWC}), testing::ValuesIn({miopenBNSpatial, miopenBNPerActivation})));