From 303be90ee1f03f75309b421297ba16428ea98ea5 Mon Sep 17 00:00:00 2001 From: Ioan-Cristian Szabo Date: Mon, 27 Nov 2017 16:31:10 +0000 Subject: COMPMID-617: Add validate support for NEON BatchNormalizationLayer. Change-Id: I037ec6df7eee06bdd1381e908677803426fa614c Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/110788 Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com Reviewed-by: Anthony Barbier --- .../kernels/NEBatchNormalizationLayerKernel.cpp | 104 ++++++++++++++------- 1 file changed, 68 insertions(+), 36 deletions(-) (limited to 'src/core/NEON') diff --git a/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp b/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp index 1123f2c9ca..4bbf67d13b 100644 --- a/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp +++ b/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp @@ -33,9 +33,39 @@ using namespace arm_compute; -NEBatchNormalizationLayerKernel::NEBatchNormalizationLayerKernel() - : _func(nullptr), _input(nullptr), _output(nullptr), _mean(nullptr), _var(nullptr), _gamma(nullptr), _beta(nullptr), _epsilon() +namespace +{ +Error validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *mean, const ITensorInfo *var, const ITensorInfo *beta, const ITensorInfo *gamma, float epsilon) { + ARM_COMPUTE_UNUSED(epsilon); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32); + + if(nullptr != output) + { + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output); + } + + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, mean, var, beta, gamma); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, mean, var, beta, gamma); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(mean, var, beta, gamma); + ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(2) != mean->dimension(0)); + + return Error{}; +} + +std::pair validate_and_configure_window(ITensorInfo *input, ITensorInfo *output) +{ + unsigned int num_elems_processed_per_iteration = 16 / input->element_size(); + + Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration)); + AccessWindowHorizontal input_access(input, 0, num_elems_processed_per_iteration); + AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration); + bool window_changed = update_window_and_padding(win, input_access, output_access); + output_access.set_valid_region(win, input->valid_region()); + Error err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Error{}; + return std::make_pair(err, win); } void batch_normalization_q8(ITensor *in, ITensor *out, const ITensor *mean, const ITensor *var, const ITensor *beta, const ITensor *gamma, float epsilon, const Window &window) @@ -213,10 +243,28 @@ void batch_normalization_fp16(ITensor *in, ITensor *out, const ITensor *mean, co input, output); } #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ +} // namespace + +NEBatchNormalizationLayerKernel::NEBatchNormalizationLayerKernel() + : _func(nullptr), _input(nullptr), _output(nullptr), _mean(nullptr), _var(nullptr), _gamma(nullptr), _beta(nullptr), _epsilon() +{ +} void NEBatchNormalizationLayerKernel::configure(ITensor *input, ITensor *output, const ITensor *mean, const ITensor *var, const ITensor *beta, const ITensor *gamma, float epsilon) { - ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32); + ARM_COMPUTE_ERROR_ON_NULLPTR(input, mean, var, beta, gamma); + + ITensorInfo *output_info = nullptr; + + if(nullptr != output) + { + // Output tensor auto initialization if not yet initialized + auto_init_if_empty(*output->info(), *input->info()); + + output_info = output->info(); + } + + ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output_info, mean->info(), var->info(), beta->info(), gamma->info(), epsilon)); _input = input; _output = input; @@ -228,39 +276,23 @@ void NEBatchNormalizationLayerKernel::configure(ITensor *input, ITensor *output, if(output != nullptr) { - // Output tensor auto initialization if not yet initialized - auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, input->info()->data_type(), input->info()->fixed_point_position()); - - ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input, output); - _output = output; } - ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output, mean, var, beta, gamma); - ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output, mean, var, beta, gamma); - ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(mean, var, beta, gamma); - ARM_COMPUTE_ERROR_ON(input->info()->dimension(2) != mean->info()->dimension(0)); - - unsigned int num_elems_processed_per_iteration = 0; - switch(input->info()->data_type()) { case DataType::QS8: - _func = &batch_normalization_q8; - num_elems_processed_per_iteration = 16; + _func = &batch_normalization_q8; break; case DataType::QS16: - _func = &batch_normalization_q16; - num_elems_processed_per_iteration = 8; + _func = &batch_normalization_q16; break; case DataType::F32: - _func = &batch_normalization_fp32; - num_elems_processed_per_iteration = 4; + _func = &batch_normalization_fp32; break; case DataType::F16: #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC - _func = &batch_normalization_fp16; - num_elems_processed_per_iteration = 8; + _func = &batch_normalization_fp16; break; #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ default: @@ -268,19 +300,19 @@ void NEBatchNormalizationLayerKernel::configure(ITensor *input, ITensor *output, break; } - Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration)); - AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration); - if(output != nullptr) - { - AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration); - update_window_and_padding(win, input_access, output_access); - output_access.set_valid_region(win, input->info()->valid_region()); - } - else - { - update_window_and_padding(win, input_access); - } - INEKernel::configure(win); + // Configure kernel window + auto win_config = validate_and_configure_window(input->info(), output_info); + ARM_COMPUTE_ERROR_THROW_ON(win_config.first); + INEKernel::configure(win_config.second); +} + +Error NEBatchNormalizationLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *mean, const ITensorInfo *var, const ITensorInfo *beta, const ITensorInfo *gamma, + float epsilon) +{ + ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, mean, var, beta, gamma, epsilon)); + ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output ? output->clone().get() : nullptr).first); + + return Error{}; } void NEBatchNormalizationLayerKernel::run(const Window &window, const ThreadInfo &info) -- cgit v1.2.1