From 554fa09a0f3d6c9c572634c9d2de9bfb6c3218b0 Mon Sep 17 00:00:00 2001 From: Keith Davis Date: Tue, 20 Jul 2021 11:25:22 +0100 Subject: IVGCVSW-5980 JSON profiling output * Add new ProfilingDetails class to construct operator details string * Add new macro which helps append layer details to ostream * Add ProfilingEnabled to NetworkProperties so that profiling can be realised when loading the network * Add further optional info to WorkloadInfo specific to convolutions * Generalise some JsonPrinter functions into JsonUtils for reusability * Remove explicit enabling of profiling within InferenceModel as it is done when loading network * Add ProfilingDetails macros to ConvolutionWorkloads for validation Signed-off-by: Keith Davis Change-Id: Ie84bc7dc667e72e6bcb635544f9ead7af1765690 --- .../neon/workloads/NeonConvolution2dWorkload.cpp | 36 ++++++++++++++++++++-- 1 file changed, 34 insertions(+), 2 deletions(-) (limited to 'src/backends/neon/workloads/NeonConvolution2dWorkload.cpp') diff --git a/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp b/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp index 32af3f853a..1e12e13357 100644 --- a/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp +++ b/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp @@ -74,8 +74,6 @@ NeonConvolution2dWorkload::NeonConvolution2dWorkload( m_Data.ValidateInputsOutputs("NeonConvolution2dWorkload", 1, 1); - // todo: check tensor shapes match. - arm_compute::ITensor& input = PolymorphicDowncast(m_Data.m_Inputs[0])->GetTensor(); arm_compute::ITensor& output = PolymorphicDowncast(m_Data.m_Outputs[0])->GetTensor(); @@ -120,6 +118,23 @@ NeonConvolution2dWorkload::NeonConvolution2dWorkload( activationInfo, isFastMathEnabled); + // Add details for profiling output + std::string workloadName = "NeonConvolution2dWorkload_Execute_Guid" + std::to_string(this->GetGuid()); + + WorkloadInfo detailsInfo; + + detailsInfo.m_InputTensorInfos = info.m_InputTensorInfos; + detailsInfo.m_OutputTensorInfos = info.m_OutputTensorInfos; + detailsInfo.m_WeightsTensorInfo = armnn::Optional(descriptor.m_Weight->GetTensorInfo()); + detailsInfo.m_ConvolutionMethod = armnn::Optional(GetConvolutionMethodString()); + if (descriptor.m_Parameters.m_BiasEnabled) + { + detailsInfo.m_BiasTensorInfo = armnn::Optional(descriptor.m_Bias->GetTensorInfo()); + } + + // Report Profiling Details + ARMNN_REPORT_PROFILING_WORKLOAD_DESC(workloadName, descriptor.m_Parameters, detailsInfo); + m_ConvolutionLayer.reset(convolutionLayer.release()); ARMNN_ASSERT(m_ConvolutionLayer); @@ -146,6 +161,23 @@ arm_compute::ConvolutionMethod NeonConvolution2dWorkload::GetConvolutionMethod() return m_ConvolutionMethod; } +std::string NeonConvolution2dWorkload::GetConvolutionMethodString() +{ + switch ( m_ConvolutionMethod ) + { + case arm_compute::ConvolutionMethod::FFT: + return "FFT"; + case arm_compute::ConvolutionMethod::DIRECT: + return "Direct"; + case arm_compute::ConvolutionMethod::GEMM: + return "GEMM"; + case arm_compute::ConvolutionMethod::WINOGRAD: + return "Winograd"; + default: + return "Unknown"; + } +} + void NeonConvolution2dWorkload::FreeUnusedTensors() { FreeTensorIfUnused(m_KernelTensor); -- cgit v1.2.1