#include <WorkloadData.hpp>

Inheritance diagram for QLstmQueueDescriptor:

Public Member Functions
	QLstmQueueDescriptor ()

void	Validate (const WorkloadInfo &workloadInfo) const

Public Member Functions inherited from QueueDescriptor
void	ValidateInputsOutputs (const std::string &descName, unsigned int numExpectedIn, unsigned int numExpectedOut) const

Public Attributes
const ConstCpuTensorHandle *	m_InputToInputWeights

const ConstCpuTensorHandle *	m_InputToForgetWeights

const ConstCpuTensorHandle *	m_InputToCellWeights

const ConstCpuTensorHandle *	m_InputToOutputWeights

const ConstCpuTensorHandle *	m_RecurrentToInputWeights

const ConstCpuTensorHandle *	m_RecurrentToForgetWeights

const ConstCpuTensorHandle *	m_RecurrentToCellWeights

const ConstCpuTensorHandle *	m_RecurrentToOutputWeights

const ConstCpuTensorHandle *	m_CellToInputWeights

const ConstCpuTensorHandle *	m_CellToForgetWeights

const ConstCpuTensorHandle *	m_CellToOutputWeights

const ConstCpuTensorHandle *	m_InputGateBias

const ConstCpuTensorHandle *	m_ForgetGateBias

const ConstCpuTensorHandle *	m_CellBias

const ConstCpuTensorHandle *	m_OutputGateBias

const ConstCpuTensorHandle *	m_ProjectionWeights

const ConstCpuTensorHandle *	m_ProjectionBias

const ConstCpuTensorHandle *	m_InputLayerNormWeights

const ConstCpuTensorHandle *	m_ForgetLayerNormWeights

const ConstCpuTensorHandle *	m_CellLayerNormWeights

const ConstCpuTensorHandle *	m_OutputLayerNormWeights

Public Attributes inherited from QueueDescriptorWithParameters< QLstmDescriptor >
QLstmDescriptor	m_Parameters

Public Attributes inherited from QueueDescriptor
std::vector< ITensorHandle * >	m_Inputs

std::vector< ITensorHandle * >	m_Outputs

Additional Inherited Members
Protected Member Functions inherited from QueueDescriptorWithParameters< QLstmDescriptor >
	~QueueDescriptorWithParameters ()=default

	QueueDescriptorWithParameters ()=default

	QueueDescriptorWithParameters (QueueDescriptorWithParameters const &)=default

QueueDescriptorWithParameters &	operator= (QueueDescriptorWithParameters const &)=default

Protected Member Functions inherited from QueueDescriptor
	~QueueDescriptor ()=default

	QueueDescriptor ()=default

	QueueDescriptor (QueueDescriptor const &)=default

QueueDescriptor &	operator= (QueueDescriptor const &)=default

Detailed Description

Definition at line 534 of file WorkloadData.hpp.

Constructor & Destructor Documentation

◆ QLstmQueueDescriptor()

QLstmQueueDescriptor ( )

inline

Definition at line 536 of file WorkloadData.hpp.

             : m_InputToInputWeights(nullptr)
             , m_InputToForgetWeights(nullptr)
             , m_InputToCellWeights(nullptr)
             , m_InputToOutputWeights(nullptr)
             , m_RecurrentToInputWeights(nullptr)
             , m_RecurrentToForgetWeights(nullptr)
             , m_RecurrentToCellWeights(nullptr)
             , m_RecurrentToOutputWeights(nullptr)
             , m_CellToInputWeights(nullptr)
             , m_CellToForgetWeights(nullptr)
             , m_CellToOutputWeights(nullptr)
             , m_InputGateBias(nullptr)
             , m_ForgetGateBias(nullptr)
             , m_CellBias(nullptr)
             , m_OutputGateBias(nullptr)
             , m_ProjectionWeights(nullptr)
             , m_ProjectionBias(nullptr)
             , m_InputLayerNormWeights(nullptr)
             , m_ForgetLayerNormWeights(nullptr)
             , m_CellLayerNormWeights(nullptr)
             , m_OutputLayerNormWeights(nullptr)
     {
     }

Member Function Documentation

◆ Validate()

void Validate ( const WorkloadInfo & workloadInfo ) const

Definition at line 2875 of file WorkloadData.cpp.

References WorkloadInfo::m_InputTensorInfos, WorkloadInfo::m_OutputTensorInfos, armnn::QAsymmS8, armnn::QSymmS16, armnn::QSymmS8, and armnn::Signed32.

 {
     const std::string descriptorName{"QLstmQueueDescriptor"};
 
     // Validate number of inputs/outputs
     ValidateNumInputs(workloadInfo,  descriptorName, 3);
     ValidateNumOutputs(workloadInfo, descriptorName, 3);
 
     // Input/output tensor info
     auto inputInfo = workloadInfo.m_InputTensorInfos[0];
     auto outputStateInInfo = workloadInfo.m_InputTensorInfos[1];
     auto cellStateInInfo = workloadInfo.m_InputTensorInfos[2];
 
     auto outputStateOutInfo = workloadInfo.m_OutputTensorInfos[0];
     auto cellStateOutInfo = workloadInfo.m_OutputTensorInfos[1];
     auto outputInfo = workloadInfo.m_OutputTensorInfos[2];
 
     // Supported types for various tensors in QLSTM
     std::vector<DataType> inputOutputSupportedTypes =
     {
         DataType::QAsymmS8
     };
 
     std::vector<DataType> cellStateSupportedTypes =
     {
         DataType::QSymmS16
     };
 
     std::vector<DataType> weightsSupportedTypes =
     {
         DataType::QSymmS8
     };
 
     std::vector<DataType> layerNormPeepholeWeightsSupportedTypes =
     {
         DataType::QSymmS16
     };
 
     std::vector<DataType> biasSupportedTypes =
     {
         DataType::Signed32
     };
 
     // Validate types of input/output tensors
     ValidateDataTypes(inputInfo, inputOutputSupportedTypes, descriptorName);
     ValidateDataTypes(outputStateInInfo, inputOutputSupportedTypes, descriptorName);
     ValidateDataTypes(cellStateInInfo, cellStateSupportedTypes, descriptorName);
 
     ValidateDataTypes(outputStateOutInfo, inputOutputSupportedTypes, descriptorName);
     ValidateDataTypes(cellStateOutInfo, cellStateSupportedTypes, descriptorName);
     ValidateDataTypes(outputInfo, inputOutputSupportedTypes, descriptorName);
 
     // Validate matching types of input/output tensors
     ValidateTensorDataTypesMatch(inputInfo, outputStateInInfo, descriptorName, "input", "outputStateIn");
     ValidateTensorDataTypesMatch(outputStateInInfo, outputStateOutInfo, descriptorName,
                                  "outputStateIn", "outputStateOut");
     ValidateTensorDataTypesMatch(cellStateInInfo, cellStateOutInfo, descriptorName, "cellStateIn", "cellStateOut");
 
     // Infer number of batches, number of units, input size and output size from tensor dimensions
     const uint32_t numBatches = inputInfo.GetShape()[0];
     const uint32_t inputSize  = inputInfo.GetShape()[1];
     const uint32_t outputSize = outputStateInInfo.GetShape()[1];
     const uint32_t numUnits = cellStateInInfo.GetShape()[1];
 
     // Validate number of dimensions and number of elements for input/output tensors
     ValidateTensorNumDimNumElem(inputInfo, 2, (numBatches * inputSize), descriptorName + " input");
     ValidateTensorNumDimNumElem(outputStateInInfo, 2, (numBatches * outputSize), descriptorName + " outputStateIn");
     ValidateTensorNumDimNumElem(cellStateInInfo, 2, (numBatches * numUnits), descriptorName + " cellStateIn");
 
     ValidateTensorNumDimNumElem(outputStateOutInfo, 2, (numBatches * outputSize), descriptorName + " outputStateOut");
     ValidateTensorNumDimNumElem(cellStateOutInfo, 2, (numBatches * numUnits), descriptorName + " cellStateOut");
     ValidateTensorNumDimNumElem(outputInfo, 2, (numBatches * outputSize), descriptorName + " output");
 
     // Validate number of dimensions and number of elements for MANDATORY weight tensors
     ValidatePointer(m_InputToForgetWeights, descriptorName, "InputToForgetWeights");
     auto inputToForgetWeightsInfo = m_InputToForgetWeights->GetTensorInfo();
     ValidateTensorNumDimNumElem(inputToForgetWeightsInfo, 2, (numUnits * inputSize), " InputToForgetWeights");
 
     ValidatePointer(m_InputToCellWeights, descriptorName, "InputToCellWeights");
     auto inputToCellWeightsInfo = m_InputToCellWeights->GetTensorInfo();
     ValidateTensorNumDimNumElem(inputToCellWeightsInfo, 2, (numUnits * inputSize), " InputToCellWeights");
 
     ValidatePointer(m_InputToOutputWeights, descriptorName, "InputToOutputWeights");
     auto inputToOutputWeightsInfo = m_InputToOutputWeights->GetTensorInfo();
     ValidateTensorNumDimNumElem(inputToOutputWeightsInfo, 2, (numUnits * inputSize), " InputToOutputWeights");
 
     ValidatePointer(m_RecurrentToForgetWeights, descriptorName, "RecurrentToForgetWeights");
     auto recurrentToForgetWeightsInfo = m_RecurrentToForgetWeights->GetTensorInfo();
     ValidateTensorNumDimNumElem(recurrentToForgetWeightsInfo, 2, (numUnits * outputSize),
                                 " RecurrentToForgetWeights");
 
     ValidatePointer(m_RecurrentToCellWeights, descriptorName, "RecurrentToCellWeights");
     auto recurrentToCellWeightsInfo = m_RecurrentToCellWeights->GetTensorInfo();
     ValidateTensorNumDimNumElem(recurrentToCellWeightsInfo, 2, (numUnits * outputSize), " RecurrentToCellWeights");
 
     ValidatePointer(m_RecurrentToOutputWeights, descriptorName, "RecurrentToOutputWeights");
     auto recurrentToOutputWeightsInfo = m_RecurrentToOutputWeights->GetTensorInfo();
     ValidateTensorNumDimNumElem(recurrentToOutputWeightsInfo, 2, (numUnits * outputSize), " RecurrentToCellWeights");
 
     // Validate data types for MANDATORY weights tensors (all should match each other)
     ValidateDataTypes(inputToForgetWeightsInfo, weightsSupportedTypes, descriptorName);
 
     ValidateTensorDataTypesMatch(inputToForgetWeightsInfo, inputToCellWeightsInfo, descriptorName,
                                  "inputToForgetWeights", "inputToCellWeights");
     ValidateTensorDataTypesMatch(inputToForgetWeightsInfo, inputToOutputWeightsInfo, descriptorName,
                                  "inputToForgetWeights", "inputToOutputWeights");
 
     ValidateTensorDataTypesMatch(inputToForgetWeightsInfo, recurrentToForgetWeightsInfo, descriptorName,
                                  "inputToForgetWeights", "recurrentToForgeteights");
     ValidateTensorDataTypesMatch(inputToForgetWeightsInfo, recurrentToCellWeightsInfo, descriptorName,
                                  "inputToForgetWeights", "recurrentToCellWeights");
     ValidateTensorDataTypesMatch(inputToForgetWeightsInfo, recurrentToOutputWeightsInfo, descriptorName,
                                  "inputToForgetWeights", "recurrentToOutputWeights");
 
     // Validate number of dimensions and number of elements for MANDATORY bias tensors
     ValidatePointer(m_ForgetGateBias, descriptorName, "ForgetGateBias");
     auto forgetGateBiasInfo = m_ForgetGateBias->GetTensorInfo();
     ValidateTensorNumDimNumElem(forgetGateBiasInfo, 1, numUnits, " ForgetGateBias");
 
     ValidatePointer(m_CellBias, descriptorName, "CellBias");
     auto cellBiasInfo = m_CellBias->GetTensorInfo();
     ValidateTensorNumDimNumElem(cellBiasInfo, 1, numUnits, " CellBias");
 
     ValidatePointer(m_OutputGateBias, descriptorName, "OutputGateBias");
     auto outputGateBiasInfo = m_OutputGateBias->GetTensorInfo();
     ValidateTensorNumDimNumElem(outputGateBiasInfo, 1, numUnits, " OutputGateBias");
 
     // Validate data types for MANDATORY bias tensors
     ValidateDataTypes(forgetGateBiasInfo, biasSupportedTypes, descriptorName);
 
     ValidateTensorDataTypesMatch(forgetGateBiasInfo, cellBiasInfo, descriptorName,
                                  "forgetGateBias", "cellBias");
     ValidateTensorDataTypesMatch(forgetGateBiasInfo, outputGateBiasInfo, descriptorName,
                                  "forgetGateBias", "outputGateBias");
 
     // Validate OPTIONAL params: CIFG (inputToInputWeights, recurrentToInputWeights, inputGateBias)
     const bool allCifgParamsPresentOrNot = ((m_InputToInputWeights && m_RecurrentToInputWeights && m_InputGateBias &&
                                              !m_Parameters.m_CifgEnabled) ||
                                             (!m_InputToInputWeights && !m_RecurrentToInputWeights &&
                                              !m_InputGateBias && m_Parameters.m_CifgEnabled));
 
     if (!allCifgParamsPresentOrNot)
     {
         throw InvalidArgumentException(descriptorName +
                 ": InputToInputWeights, RecurrentToInputWeights and InputGateBias must either all be present "
                 "(CIFG disabled) or not be present at all (CIFG enabled). m_Parameters.m_CifgEnabled should be "
                 "set appropriately.");
     }
 
     if (!m_Parameters.m_CifgEnabled)
     {
         // Validate number of dimensions and number of elements
         auto inputToInputWeightsInfo = m_InputToInputWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(inputToInputWeightsInfo, 2, (numUnits * inputSize), " InputToInputWeights");
 
         auto recurrentToInputWeightsInfo = m_RecurrentToInputWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(recurrentToInputWeightsInfo, 2, (numUnits * outputSize),
                                     " RecurrentToInputWeights");
 
         auto inputGateBiasInfo = m_InputGateBias->GetTensorInfo();
         ValidateTensorNumDimNumElem(inputGateBiasInfo, 1, numUnits, " InputGateBias");
 
         // Validate data types
         ValidateTensorDataTypesMatch(inputToForgetWeightsInfo, inputToInputWeightsInfo, descriptorName,
                                      "inputToForgetWeights", "inputToInputWeights");
         ValidateTensorDataTypesMatch(inputToForgetWeightsInfo, recurrentToInputWeightsInfo, descriptorName,
                                      "inputToForgetWeights", "recurrentToInputWeights");
         ValidateTensorDataTypesMatch(forgetGateBiasInfo, inputGateBiasInfo, descriptorName,
                                      "forgetGateBias", "inputGateBias");
     }
 
     // Validate OPTIONAL params: Peephole (cellToInputWeights, cellToForgetWeights, cellToOutputWeights)
     bool allPeepholeWeightsPresentOrNot =
             (((m_CellToInputWeights || m_Parameters.m_CifgEnabled) && m_CellToForgetWeights
               && m_CellToOutputWeights && m_Parameters.m_PeepholeEnabled)
              || (!m_CellToInputWeights && !m_CellToForgetWeights
                  && !m_CellToOutputWeights && !m_Parameters.m_PeepholeEnabled));
 
     if (!allPeepholeWeightsPresentOrNot)
     {
         throw InvalidArgumentException(descriptorName +
                 ": CellToInputWeights, CellToForgetWeights and CellToOutputWeights should all be present (Peephole "
                 "enabled) or not be present at all (Peephole disabled). CellToInputWeights should only be present "
                 "when Peephole is enabled and CIFG is disabled. m_Parameters.m_PeepholeEnabled should be set "
                 "appropriately.");
     }
 
     if (m_Parameters.m_PeepholeEnabled)
     {
         auto cellToForgetWeightsInfo = m_CellToForgetWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(cellToForgetWeightsInfo, 1, numUnits, " cellToForgetWeights");
         ValidateDataTypes(cellToForgetWeightsInfo, layerNormPeepholeWeightsSupportedTypes, descriptorName);
 
         auto cellToOutputWeightsInfo = m_CellToOutputWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(cellToOutputWeightsInfo, 1, numUnits, " cellToOutputWeights");
         ValidateTensorDataTypesMatch(cellToForgetWeightsInfo, cellToOutputWeightsInfo, descriptorName,
                                      "cellToForgetWeight", "cellToOutputWeights");
 
         if (!m_Parameters.m_CifgEnabled)
         {
             auto cellToInputWeightsInfo = m_CellToInputWeights->GetTensorInfo();
             ValidateTensorNumDimNumElem(cellToInputWeightsInfo, 1, numUnits, " cellToInputWeights");
             ValidateTensorDataTypesMatch(cellToForgetWeightsInfo, cellToInputWeightsInfo, descriptorName,
                                          "cellToForgetWeights", "cellToInputWeights");
         }
     }
 
     // Validate OPTIONAL params: Layer Norm Weights
     bool allLayerNormWeightsPresentOrNot =
             (((m_InputLayerNormWeights || m_Parameters.m_CifgEnabled) && m_ForgetLayerNormWeights
               && m_CellLayerNormWeights && m_OutputLayerNormWeights && m_Parameters.m_LayerNormEnabled)
              || (!m_InputLayerNormWeights && !m_ForgetLayerNormWeights && !m_CellLayerNormWeights
                  && !m_OutputLayerNormWeights && !m_Parameters.m_LayerNormEnabled));
 
     if (!allLayerNormWeightsPresentOrNot)
     {
         throw InvalidArgumentException(descriptorName +
                                        ": InputLayerNormWeights, ForgetLayerNormWeights, m_OutputLayerNormWeights "
                                        "and CellLayerNormWeights should all be present (Layer Norm enabled) or not "
                                        "be present at all (Layer Norm disabled). InputLayerNormWeights should "
                                        "only be present when Layer Norm is enabled and CIFG is disabled. "
                                        "m_Parameters.m_LayerNormEnabled should be set appropriately.");
     }
 
     if (m_Parameters.m_LayerNormEnabled)
     {
         auto forgetLayerNormWeightsInfo = m_ForgetLayerNormWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(forgetLayerNormWeightsInfo, 1, numUnits, " forgetLayerNormWeights");
         ValidateDataTypes(forgetLayerNormWeightsInfo, layerNormPeepholeWeightsSupportedTypes, descriptorName);
 
         auto cellLayerNormWeightsInfo = m_CellLayerNormWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(cellLayerNormWeightsInfo, 1, numUnits, " cellLayerNormWeights");
         ValidateTensorDataTypesMatch(forgetLayerNormWeightsInfo, cellLayerNormWeightsInfo, descriptorName,
                                      "forgetLayerNormWeights", "cellLayerNormWeights");
 
         auto outputLayerNormWeightsInfo = m_OutputLayerNormWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(outputLayerNormWeightsInfo, 1, numUnits, " outputLayerNormWeights");
         ValidateTensorDataTypesMatch(forgetLayerNormWeightsInfo, outputLayerNormWeightsInfo, descriptorName,
                                      "forgetLayerNormWeights", "outputLayerNormWeights");
 
         if (!m_Parameters.m_CifgEnabled)
         {
             auto inputLayerNormWeightsInfo = m_InputLayerNormWeights->GetTensorInfo();
             ValidateTensorNumDimNumElem(inputLayerNormWeightsInfo, 1, numUnits, " inputLayerNormWeights");
             ValidateTensorDataTypesMatch(forgetLayerNormWeightsInfo, inputLayerNormWeightsInfo, descriptorName,
                                          "forgetLayerNormWeights", "inputLayerNormWeights");
         }
     }
 
     // Validate OPTIONAL params: Projection (projectionWeights, projectionBias)
     bool correctProjectionTensorsPresent =
             ((!m_ProjectionWeights && !m_ProjectionBias && !m_Parameters.m_ProjectionEnabled) ||
             (m_ProjectionWeights && !m_ProjectionBias && m_Parameters.m_ProjectionEnabled) ||
             (m_ProjectionWeights && m_ProjectionBias && m_Parameters.m_ProjectionEnabled));
 
     if (!correctProjectionTensorsPresent)
     {
         throw InvalidArgumentException(descriptorName +
                                        ": If projection is enabled, ProjectionWeights should be present and "
                                        "ProjectionBias is optional. If projection is disabled, neither "
                                        "ProjectionWeights nor ProjectionBias should be present.");
     }
 
     if (m_Parameters.m_ProjectionEnabled)
     {
         auto projectionWeightsInfo = m_ProjectionWeights->GetTensorInfo();
         ValidateTensorNumDimNumElem(projectionWeightsInfo, 2, (numUnits * outputSize), "ProjectionWeights");
         ValidateDataTypes(projectionWeightsInfo, weightsSupportedTypes, descriptorName);
 
         if (m_ProjectionBias)
         {
             auto projectionBiasInfo = m_ProjectionBias->GetTensorInfo();
             ValidateTensorNumDimNumElem(projectionBiasInfo, 1, outputSize, "ProjectionBias");
             ValidateDataTypes(projectionBiasInfo, biasSupportedTypes, descriptorName);
         }
 
     }
     else if ((outputInfo.GetQuantizationScale() != m_Parameters.m_HiddenStateScale) &&
               outputInfo.GetQuantizationOffset() != m_Parameters.m_HiddenStateZeroPoint) {
         throw InvalidArgumentException(descriptorName +
                                        ": If projection is disabled, output quantization info (scale, offset) "
                                        "should match HiddenStateScale and HiddenStateZeroPoint.");
     }
 
 }