diff options
3 files changed, 27 insertions, 21 deletions
diff --git a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp index 62dabc8d32..09e4acda5f 100644 --- a/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp +++ b/src/core/NEON/kernels/NEDepthwiseConvolutionLayer3x3Kernel.cpp @@ -150,13 +150,14 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights, ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F32); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights); - if(is_optimized) - { - ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(1) != 3 || weights->dimension(2) != 3); - } - else + const DataLayout data_layout = input->data_layout(); + const unsigned int width_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH); + const unsigned int height_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT); + + ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(width_idx) != 3 || weights->dimension(height_idx) != 3); + + if(!is_optimized) { - ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(0) != 3 || weights->dimension(1) != 3); ARM_COMPUTE_RETURN_ERROR_ON(conv_info.stride().first < 1 || conv_info.stride().first > 3); } diff --git a/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp b/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp index 8d22f03d7f..24b12f4969 100644 --- a/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp +++ b/src/runtime/NEON/functions/NEDepthwiseConvolutionLayer.cpp @@ -163,9 +163,14 @@ Status NEDepthwiseConvolutionLayer3x3::validate(const ITensorInfo *input, const unsigned int depth_multiplier) { ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output); - ARM_COMPUTE_UNUSED(biases); ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() != DataLayout::NCHW && input->data_layout() != DataLayout::NHWC); + if(biases != nullptr) + { + ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1); + ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != weights->dimension(3)); + } + return NEDepthwiseConvolutionLayer3x3Kernel::validate(input, weights, output, conv_info, depth_multiplier); } diff --git a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp index 6b3411b965..fe7bba365a 100644 --- a/tests/validation/NEON/DepthwiseConvolutionLayer.cpp +++ b/tests/validation/NEON/DepthwiseConvolutionLayer.cpp @@ -61,27 +61,27 @@ DATA_TEST_CASE(Validate3x3, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip TensorInfo(TensorShape(32U, 18U, 3U), 1, DataType::F32), // Mismatching input feature maps TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::F32), // Unsupported weights dimensions TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::F32), // Mismatching depth multiplier - TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::F32), // Invalid stride + TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::QASYMM8), // Invalid stride TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::F32), // Invalid biases size TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::F32), // Invalid biases dimensions TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::F32), // Invalid output size - TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32), // Window shrink + TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32), }), - framework::dataset::make("WeightsInfo", { TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F16), - TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(5U, 5U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(3U, 3U, 2U), 1, DataType::F32), + framework::dataset::make("WeightsInfo", { TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F16), + TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(5U, 5U, 2U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::QASYMM8), + TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(3U, 3U, 2U, 2U), 1, DataType::F32), })), framework::dataset::make("BiasesInfo", { TensorInfo(TensorShape(2U), 1, DataType::F32), TensorInfo(TensorShape(2U), 1, DataType::F32), TensorInfo(TensorShape(2U), 1, DataType::F32), TensorInfo(TensorShape(2U), 1, DataType::F32), - TensorInfo(TensorShape(2U), 1, DataType::F32), + TensorInfo(TensorShape(2U), 1, DataType::S32), TensorInfo(TensorShape(4U), 1, DataType::F32), TensorInfo(TensorShape(2U, 2U), 1, DataType::F32), TensorInfo(TensorShape(2U), 1, DataType::F32), @@ -91,7 +91,7 @@ DATA_TEST_CASE(Validate3x3, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip TensorInfo(TensorShape(30U, 16U, 2U), 1, DataType::F32), TensorInfo(TensorShape(30U, 16U, 2U), 1, DataType::F32), TensorInfo(TensorShape(30U, 16U, 2U), 1, DataType::F32), - TensorInfo(TensorShape(30U, 16U, 2U), 1, DataType::F32), + TensorInfo(TensorShape(30U, 16U, 2U), 1, DataType::QASYMM8), TensorInfo(TensorShape(30U, 16U, 2U), 1, DataType::F32), TensorInfo(TensorShape(30U, 16U, 2U), 1, DataType::F32), TensorInfo(TensorShape(32U, 18U, 2U), 1, DataType::F32), @@ -117,7 +117,7 @@ DATA_TEST_CASE(Validate3x3, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip 1, 1, })), - framework::dataset::make("Expected", { false, false, false, false, false, false, false, false, false })), + framework::dataset::make("Expected", { false, false, false, false, false, false, false, false, true })), input_info, weights_info, biases_info, output_info, conv_info, depth_multiplier, expected) { bool is_valid = bool(NEDepthwiseConvolutionLayer3x3::validate(&input_info.clone()->set_is_resizable(false), &weights_info.clone()->set_is_resizable(false), &biases_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false), conv_info, depth_multiplier)); |