MLCE-530 Add support of int8 weight for UnidirectionalSequenceLstm
to Ref backend and armnn delegate
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Change-Id: I203d0029c12221228ffe229acda3c90594394e9b
diff --git a/delegate/src/test/UnidirectionalSequenceLstmTest.cpp b/delegate/src/test/UnidirectionalSequenceLstmTest.cpp
index f0a96da..4bee715 100644
--- a/delegate/src/test/UnidirectionalSequenceLstmTest.cpp
+++ b/delegate/src/test/UnidirectionalSequenceLstmTest.cpp
@@ -786,7 +786,615 @@
isTimeMajor);
}
+void UnidirectionalSequenceLstmInt8Test(std::vector<armnn::BackendId>& backends)
+{
+ int32_t batchSize = 3;
+ int32_t timeSize = 2;
+ int32_t inputSize = 3;
+ int32_t outputSize = 4;
+ // cellSize and outputSize have the same size when there is no projection.
+ int32_t numUnits = outputSize;
+ //tensorInfo12
+ bool hasInputToInputWeights = true;
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3 };
+
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ //tensorInfo16
+ bool hasRecurrentToInputWeights = true;
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -1, -1 };
+
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+ // tensorInfo4
+ bool hasCellToInputWeights = false;
+ std::vector<int8_t> cellToInputWeights;
+ bool hasCellToForgetWeights = false;
+ std::vector<int8_t> cellToForgetWeights;
+ bool hasCellToOutputWeights = false;
+ std::vector<int8_t> cellToOutputWeights;
+
+ bool hasInputGateBias = true;
+ std::vector<float> inputGateBias = { 0., 0., 0., 0. };
+ std::vector<float> forgetGateBias = { 1., 1., 1., 1. };
+ std::vector<float> cellBias = { 0., 0., 0., 0. };
+ std::vector<float> outputGateBias = { 0., 0., 0., 0. };
+
+ bool hasProjectionWeights = false;
+ std::vector<int8_t> projectionWeights;
+ bool hasProjectionBias = false;
+ std::vector<float> projectionBias;
+
+ bool hasInputLayerNormWeights = false;
+ std::vector<float> inputLayerNormWeights;
+ bool hasForgetLayerNormWeights = false;
+ std::vector<float> forgetLayerNormWeights;
+ bool hasCellLayerNormWeights = false;
+ std::vector<float> cellLayerNormWeights;
+ bool hasOutputLayerNormWeights = false;
+ std::vector<float> outputLayerNormWeights;
+
+ std::vector<float> inputValues = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> expectedOutputValues = { -0.0142517f, -0.0198845f, -0.0120569f, -0.0116868f,
+ -0.0350714f, -0.0343202f, -0.047504f, -0.0569789f,
+ -0.0146346f, 0.0106663f, -0.0247238f, -0.0319502f,
+ -0.0294759f, -0.0129935f, -0.0444175f, -0.0444354f,
+ -0.0280855f, 0.00545101f, -0.051422f, -0.0463838f,
+ -0.0310702f, 0.00915739f, -0.0625207f, -0.0482648f };
+
+ tflite::ActivationFunctionType activationFunction = tflite::ActivationFunctionType_TANH;
+ float clippingThresCell = 10.f;
+ float clippingThresProj = 0.f;
+ bool isTimeMajor = false;
+
+ UnidirectionalSequenceLstmTestImpl<int8_t>(backends,
+ ::tflite::TensorType_INT8,
+ batchSize,
+ timeSize,
+ inputSize,
+ outputSize,
+ numUnits,
+ hasInputToInputWeights,
+ inputToInputWeights,
+ inputToForgetWeights,
+ inputToCellWeights,
+ inputToOutputWeights,
+ hasRecurrentToInputWeights,
+ recurrentToInputWeights,
+ recurrentToForgetWeights,
+ recurrentToCellWeights,
+ recurrentToOutputWeights,
+ hasCellToInputWeights,
+ cellToInputWeights,
+ hasCellToForgetWeights,
+ cellToForgetWeights,
+ hasCellToOutputWeights,
+ cellToOutputWeights,
+ hasInputGateBias,
+ inputGateBias,
+ forgetGateBias,
+ cellBias,
+ outputGateBias,
+ hasProjectionWeights,
+ projectionWeights,
+ hasProjectionBias,
+ projectionBias,
+ hasInputLayerNormWeights,
+ inputLayerNormWeights,
+ hasForgetLayerNormWeights,
+ forgetLayerNormWeights,
+ hasCellLayerNormWeights,
+ cellLayerNormWeights,
+ hasOutputLayerNormWeights,
+ outputLayerNormWeights,
+ inputValues,
+ expectedOutputValues,
+ activationFunction,
+ clippingThresCell,
+ clippingThresProj,
+ isTimeMajor,
+ 0.1f);
+}
+
+void UnidirectionalSequenceLstmInt8TimeMajorTest(std::vector<armnn::BackendId>& backends)
+{
+ int32_t batchSize = 3;
+ int32_t timeSize = 2;
+ int32_t inputSize = 3;
+ int32_t outputSize = 4;
+ // cellSize and outputSize have the same size when there is no projection.
+ int32_t numUnits = outputSize;
+
+ //tensorInfo12
+ bool hasInputToInputWeights = true;
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3 };
+
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ //tensorInfo16
+ bool hasRecurrentToInputWeights = true;
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -1, -1 };
+
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+ // tensorInfo4
+ bool hasCellToInputWeights = false;
+ std::vector<int8_t> cellToInputWeights;
+ bool hasCellToForgetWeights = false;
+ std::vector<int8_t> cellToForgetWeights;
+ bool hasCellToOutputWeights = false;
+ std::vector<int8_t> cellToOutputWeights;
+
+ bool hasInputGateBias = true;
+ std::vector<float> inputGateBias = { 0., 0., 0., 0. };
+ std::vector<float> forgetGateBias = { 1., 1., 1., 1. };
+ std::vector<float> cellBias = { 0., 0., 0., 0. };
+ std::vector<float> outputGateBias = { 0., 0., 0., 0. };
+
+ bool hasProjectionWeights = false;
+ std::vector<int8_t> projectionWeights;
+ bool hasProjectionBias = false;
+ std::vector<float> projectionBias;
+
+ bool hasInputLayerNormWeights = false;
+ std::vector<float> inputLayerNormWeights;
+ bool hasForgetLayerNormWeights = false;
+ std::vector<float> forgetLayerNormWeights;
+ bool hasCellLayerNormWeights = false;
+ std::vector<float> cellLayerNormWeights;
+ bool hasOutputLayerNormWeights = false;
+ std::vector<float> outputLayerNormWeights;
+
+ std::vector<float> inputValues = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> expectedOutputValues = { -0.0142517f, -0.0198845f, -0.0120122f, -0.0116868f,
+ -0.0261295f, -0.0188487f, -0.0345463f, -0.049733f,
+ -0.0146346f, 0.0106663f, -0.0247238f, -0.0319502f,
+ -0.0291863f, -0.0369402f, -0.0354071f, -0.0296529f,
+ -0.0419539f, -0.00617731f, -0.0814796f, -0.0804005f,
+ -0.0244737f, 0.0119905f, -0.0457527f, -0.0331862f };
+
+ tflite::ActivationFunctionType activationFunction = tflite::ActivationFunctionType_TANH;
+ float clippingThresCell = 10.f;
+ float clippingThresProj = 0.f;
+ bool isTimeMajor = true;
+
+ UnidirectionalSequenceLstmTestImpl<int8_t>(backends,
+ ::tflite::TensorType_INT8,
+ batchSize,
+ timeSize,
+ inputSize,
+ outputSize,
+ numUnits,
+ hasInputToInputWeights,
+ inputToInputWeights,
+ inputToForgetWeights,
+ inputToCellWeights,
+ inputToOutputWeights,
+ hasRecurrentToInputWeights,
+ recurrentToInputWeights,
+ recurrentToForgetWeights,
+ recurrentToCellWeights,
+ recurrentToOutputWeights,
+ hasCellToInputWeights,
+ cellToInputWeights,
+ hasCellToForgetWeights,
+ cellToForgetWeights,
+ hasCellToOutputWeights,
+ cellToOutputWeights,
+ hasInputGateBias,
+ inputGateBias,
+ forgetGateBias,
+ cellBias,
+ outputGateBias,
+ hasProjectionWeights,
+ projectionWeights,
+ hasProjectionBias,
+ projectionBias,
+ hasInputLayerNormWeights,
+ inputLayerNormWeights,
+ hasForgetLayerNormWeights,
+ forgetLayerNormWeights,
+ hasCellLayerNormWeights,
+ cellLayerNormWeights,
+ hasOutputLayerNormWeights,
+ outputLayerNormWeights,
+ inputValues,
+ expectedOutputValues,
+ activationFunction,
+ clippingThresCell,
+ clippingThresProj,
+ isTimeMajor,
+ 0.1);
+}
+
+void UnidirectionalSequenceLstmInt8NoCifgWithPeepholeWithProjectionTest(std::vector<armnn::BackendId>& backends)
+{
+ int32_t batchSize = 3;
+ int32_t timeSize = 2;
+ int32_t inputSize = 3;
+ int32_t outputSize = 4;
+ int32_t numUnits = 4;
+
+ bool hasInputToInputWeights = true;
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3 };
+
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ //tensorInfo16
+ bool hasRecurrentToInputWeights = true;
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -1, -1 };
+
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+ // tensorInfo4
+ bool hasCellToInputWeights = true;
+ std::vector<int8_t> cellToInputWeights = { 5, 10, 25, 15 };
+ bool hasCellToForgetWeights = true;
+ std::vector<int8_t> cellToForgetWeights = { -5, 15, 25, 3 };
+ bool hasCellToOutputWeights = true;
+ std::vector<int8_t> cellToOutputWeights = { 10, -10, -5, 50 };
+
+ bool hasInputGateBias = true;
+ std::vector<float> inputGateBias = { 0.02234832f, 0.14757581f, 0.18176508f, 0.10380666f};
+ std::vector<float> forgetGateBias = { 0.035185695f, -0.042891346f, -0.3032477f, 0.23027696f};
+ std::vector<float> cellBias = { -0.124379363f, 0.55531194f, 0.23377132f, 0.033463873f };
+ std::vector<float> outputGateBias = { 0.046159424f, -0.12809046f, 0.03563469f, 0.12648113f };
+
+ bool hasProjectionWeights = true;
+ std::vector<int8_t> projectionWeights = { -25, 51, 3, -5, 25, 127, 77, 20, 18, 51, -10, 51, -25, 88, 77, -13 };
+ bool hasProjectionBias = true;
+ std::vector<float> projectionBias(outputSize, 0.f);
+
+ bool hasInputLayerNormWeights = false;
+ std::vector<float> inputLayerNormWeights;
+ bool hasForgetLayerNormWeights = false;
+ std::vector<float> forgetLayerNormWeights;
+ bool hasCellLayerNormWeights = false;
+ std::vector<float> cellLayerNormWeights;
+ bool hasOutputLayerNormWeights = false;
+ std::vector<float> outputLayerNormWeights;
+
+ std::vector<float> inputValues = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> expectedOutputValues = { 0.612103f, 1.56788f, 0.31966f, 1.42956f,
+ 0.909718f, 3.07916f, -0.560586f, 3.8907f,
+ 0.753671f, 1.77485f, 0.365122f, 1.60077f,
+ 0.812644f, 2.79092f, -0.605396f, 3.61742f,
+ 0.791857f, 1.64353f, 0.316588f, 1.55192f,
+ 0.807265f, 2.47012f, -0.539598f, 3.25654f };
+
+ tflite::ActivationFunctionType activationFunction = tflite::ActivationFunctionType_TANH;
+ float clippingThresCell = 10.f;
+ float clippingThresProj = 0.f;
+ bool isTimeMajor = false;
+
+ UnidirectionalSequenceLstmTestImpl<int8_t>(backends,
+ ::tflite::TensorType_INT8,
+ batchSize,
+ timeSize,
+ inputSize,
+ outputSize,
+ numUnits,
+ hasInputToInputWeights,
+ inputToInputWeights,
+ inputToForgetWeights,
+ inputToCellWeights,
+ inputToOutputWeights,
+ hasRecurrentToInputWeights,
+ recurrentToInputWeights,
+ recurrentToForgetWeights,
+ recurrentToCellWeights,
+ recurrentToOutputWeights,
+ hasCellToInputWeights,
+ cellToInputWeights,
+ hasCellToForgetWeights,
+ cellToForgetWeights,
+ hasCellToOutputWeights,
+ cellToOutputWeights,
+ hasInputGateBias,
+ inputGateBias,
+ forgetGateBias,
+ cellBias,
+ outputGateBias,
+ hasProjectionWeights,
+ projectionWeights,
+ hasProjectionBias,
+ projectionBias,
+ hasInputLayerNormWeights,
+ inputLayerNormWeights,
+ hasForgetLayerNormWeights,
+ forgetLayerNormWeights,
+ hasCellLayerNormWeights,
+ cellLayerNormWeights,
+ hasOutputLayerNormWeights,
+ outputLayerNormWeights,
+ inputValues,
+ expectedOutputValues,
+ activationFunction,
+ clippingThresCell,
+ clippingThresProj,
+ isTimeMajor,
+ 0.1f);
+}
+
+void UnidirectionalSequenceLstmInt8WithCifgWithPeepholeNoProjectionTest(std::vector<armnn::BackendId>& backends)
+{
+ int32_t batchSize = 3;
+ int32_t timeSize = 2;
+ int32_t inputSize = 3;
+ int32_t outputSize = 4;
+ // cellSize and outputSize have the same size when there is no projection.
+ int32_t numUnits = outputSize;
+
+ //tensorInfo12,
+ bool hasInputToInputWeights = false;
+ std::vector<int8_t> inputToInputWeights;
+
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ //tensorInfo16,
+ bool hasRecurrentToInputWeights = false;
+ std::vector<int8_t> recurrentToInputWeights;
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+ // tensorInfo4
+ bool hasCellToInputWeights = false;
+ std::vector<int8_t> cellToInputWeights;
+ bool hasCellToForgetWeights = true;
+ std::vector<int8_t> cellToForgetWeights = { 47, -52, -24, 31 };
+ bool hasCellToOutputWeights = true;
+ std::vector<int8_t> cellToOutputWeights = { -17, 82, 85, -77 };
+
+ bool hasInputGateBias = false;
+ std::vector<float> inputGateBias;
+ std::vector<float> forgetGateBias = { 1., 1., 1., 1. };
+ std::vector<float> cellBias = { 0., 0., 0., 0. };
+ std::vector<float> outputGateBias = { 0., 0., 0., 0. };
+
+ bool hasProjectionWeights = false;
+ std::vector<int8_t> projectionWeights;
+ bool hasProjectionBias = false;
+ std::vector<float> projectionBias;
+
+ bool hasInputLayerNormWeights = false;
+ std::vector<float> inputLayerNormWeights;
+ bool hasForgetLayerNormWeights = false;
+ std::vector<float> forgetLayerNormWeights;
+ bool hasCellLayerNormWeights = false;
+ std::vector<float> cellLayerNormWeights;
+ bool hasOutputLayerNormWeights = false;
+ std::vector<float> outputLayerNormWeights;
+
+ std::vector<float> inputValues = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> expectedOutputValues = { -0.0072104f, -0.00991171f, -0.00650478f, -0.00713055f,
+ -0.0191782f, -0.0161269f, -0.0233683f, -0.054299f,
+ -0.00783725f, 0.00635271f, -0.0126718f, -0.022613f,
+ -0.0161351f, -0.00775868f, -0.021054f, -0.0339778f,
+ -0.0146392f, 0.00330261f, -0.0258733f, -0.0407797f,
+ -0.0174297f, 0.0050105f, -0.0266275f, -0.0362564f };
+
+ tflite::ActivationFunctionType activationFunction = tflite::ActivationFunctionType_TANH;
+ float clippingThresCell = 10.f;
+ float clippingThresProj = 0.f;
+ bool isTimeMajor = false;
+
+ UnidirectionalSequenceLstmTestImpl<int8_t>(backends,
+ ::tflite::TensorType_INT8,
+ batchSize,
+ timeSize,
+ inputSize,
+ outputSize,
+ numUnits,
+ hasInputToInputWeights,
+ inputToInputWeights,
+ inputToForgetWeights,
+ inputToCellWeights,
+ inputToOutputWeights,
+ hasRecurrentToInputWeights,
+ recurrentToInputWeights,
+ recurrentToForgetWeights,
+ recurrentToCellWeights,
+ recurrentToOutputWeights,
+ hasCellToInputWeights,
+ cellToInputWeights,
+ hasCellToForgetWeights,
+ cellToForgetWeights,
+ hasCellToOutputWeights,
+ cellToOutputWeights,
+ hasInputGateBias,
+ inputGateBias,
+ forgetGateBias,
+ cellBias,
+ outputGateBias,
+ hasProjectionWeights,
+ projectionWeights,
+ hasProjectionBias,
+ projectionBias,
+ hasInputLayerNormWeights,
+ inputLayerNormWeights,
+ hasForgetLayerNormWeights,
+ forgetLayerNormWeights,
+ hasCellLayerNormWeights,
+ cellLayerNormWeights,
+ hasOutputLayerNormWeights,
+ outputLayerNormWeights,
+ inputValues,
+ expectedOutputValues,
+ activationFunction,
+ clippingThresCell,
+ clippingThresProj,
+ isTimeMajor,
+ 0.1);
+}
+
+void UnidirectionalSequenceLstmInt8NoCifgWithPeepholeWithProjectionWithLayerNormTest(
+ std::vector<armnn::BackendId>& backends)
+{
+ int32_t batchSize = 3;
+ int32_t timeSize = 2;
+ int32_t inputSize = 3;
+ int32_t outputSize = 4;
+ int32_t numUnits = 5;
+
+ bool hasInputToInputWeights = true;
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3, 2, 2, -4 };
+
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1, -3, -2, -4 };
+
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3, 2, 5, -4 };
+
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4, -4, -1, -1 };
+
+ bool hasRecurrentToInputWeights = true;
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3,
+ 5, -1, 1, 3, -1, -1, -1, 4, 2, 3 };
+
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3,
+ 5, -1, 1, 3, -2, -1, -1, 2, 2, 1 };
+
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2,
+ 1, 2, 3, -2, 3, -3, -1, -5, 1, 3 };
+
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3,
+ -4, -1, -1, -1, 2, -1, 5, 1, -3, -4 };
+
+ // tensorInfo5
+ bool hasCellToInputWeights = true;
+ std::vector<int8_t> cellToInputWeights = { 5, 3, 8, -5, 2 };
+ bool hasCellToForgetWeights = true;
+ std::vector<int8_t> cellToForgetWeights = { -2, -7, 5, -3, 4 };
+ bool hasCellToOutputWeights = true;
+ std::vector<int8_t> cellToOutputWeights = { 9, -10 , -5, 5, 1 };
+
+ bool hasInputGateBias = true;
+ std::vector<float> inputGateBias = { 0.03f, 0.15f, 0.22f, 0.38f, 0.05f };
+ std::vector<float> forgetGateBias = { 0.1f, -0.3f, -0.2f, 0.1f, 0.4f };
+ std::vector<float> cellBias = { -0.05f, 0.72f, 0.25f, 0.08f, 0.1f };
+ std::vector<float> outputGateBias = { 0.05f, -0.01f, 0.2f, 0.1f, -0.2f };
+
+ bool hasProjectionWeights = true;
+ std::vector<int8_t> projectionWeights = { -1, 2, 1, -2, 1, 5, 3, 8, 7, 2,
+ -4, 2, 5, -4, 3, -2, 3, 8, -7, 2 };
+ bool hasProjectionBias = true;
+ std::vector<float> projectionBias(outputSize, 0.f);
+
+ bool hasInputLayerNormWeights = true;
+ std::vector<float> inputLayerNormWeights = { 0.1f, 0.2f, -0.3f, -0.1f, 0.5f };
+ bool hasForgetLayerNormWeights = true;
+ std::vector<float> forgetLayerNormWeights = { -0.1f, 0.2f, 0.3f, 0.5f, 0.2f };
+ bool hasCellLayerNormWeights = true;
+ std::vector<float> cellLayerNormWeights = { 0.5f, 0.2f, 0.3f, 0.4f, -0.5f };
+ bool hasOutputLayerNormWeights = true;
+ std::vector<float> outputLayerNormWeights = { 0.6f, -0.2f, -0.2f, 0.5f, 0.1f };
+
+ std::vector<float> inputValues = { 1., 8., 3., 4., 5., 4.,
+ 3., 2., 1., 2., 3., 4.,
+ 5., 4., 3., 2., 1., 2. };
+
+ std::vector<float> expectedOutputValues = { 0.0471276f, 0.0168155f, 0.0789885f, 0.16550f,
+ 0.0643133f, -0.0400722f, 0.100593f, 0.197722f,
+ 0.0465562f, -0.0600682f, 0.0622087f, 0.115053f,
+ 0.056287f, -0.0566218f, 0.0856832f, 0.148484f,
+ 0.0457859f, -0.0588112f, 0.0623636f, 0.114333f,
+ 0.0509271f, -0.0754262f, 0.058600f, 0.0801288f };
+
+ tflite::ActivationFunctionType activationFunction = tflite::ActivationFunctionType_TANH;
+ float clippingThresCell = 10.f;
+ float clippingThresProj = 0.f;
+ bool isTimeMajor = false;
+
+ UnidirectionalSequenceLstmTestImpl<int8_t>(backends,
+ ::tflite::TensorType_INT8,
+ batchSize,
+ timeSize,
+ inputSize,
+ outputSize,
+ numUnits,
+ hasInputToInputWeights,
+ inputToInputWeights,
+ inputToForgetWeights,
+ inputToCellWeights,
+ inputToOutputWeights,
+ hasRecurrentToInputWeights,
+ recurrentToInputWeights,
+ recurrentToForgetWeights,
+ recurrentToCellWeights,
+ recurrentToOutputWeights,
+ hasCellToInputWeights,
+ cellToInputWeights,
+ hasCellToForgetWeights,
+ cellToForgetWeights,
+ hasCellToOutputWeights,
+ cellToOutputWeights,
+ hasInputGateBias,
+ inputGateBias,
+ forgetGateBias,
+ cellBias,
+ outputGateBias,
+ hasProjectionWeights,
+ projectionWeights,
+ hasProjectionBias,
+ projectionBias,
+ hasInputLayerNormWeights,
+ inputLayerNormWeights,
+ hasForgetLayerNormWeights,
+ forgetLayerNormWeights,
+ hasCellLayerNormWeights,
+ cellLayerNormWeights,
+ hasOutputLayerNormWeights,
+ outputLayerNormWeights,
+ inputValues,
+ expectedOutputValues,
+ activationFunction,
+ clippingThresCell,
+ clippingThresProj,
+ isTimeMajor,
+ 0.1);
+}
TEST_SUITE("UnidirectionalSequenceLstmTest_CpuRefTests")
{
@@ -821,6 +1429,36 @@
UnidirectionalSequenceLstmNoCifgWithPeepholeWithProjectionWithLayerNormTest(backends);
}
+TEST_CASE ("UnidirectionalSequenceLstmInt8Test_CpuRef_Test")
+{
+ std::vector <armnn::BackendId> backends = {armnn::Compute::CpuRef};
+ UnidirectionalSequenceLstmInt8Test(backends);
+}
+
+TEST_CASE ("UnidirectionalSequenceLstmTimeInt8TimeMajorTest_CpuRef_Test")
+{
+ std::vector <armnn::BackendId> backends = {armnn::Compute::CpuRef};
+ UnidirectionalSequenceLstmInt8TimeMajorTest(backends);
+}
+
+TEST_CASE ("UnidirectionalSequenceLstmInt8NoCifgWithPeepholeWithProjectionTest_CpuRef_Test")
+{
+ std::vector <armnn::BackendId> backends = {armnn::Compute::CpuRef};
+ UnidirectionalSequenceLstmInt8NoCifgWithPeepholeWithProjectionTest(backends);
+}
+
+TEST_CASE ("UnidirectionalSequenceLstmInt8WithCifgWithPeepholeNoProjectionTest_CpuRef_Test")
+{
+ std::vector <armnn::BackendId> backends = {armnn::Compute::CpuRef};
+ UnidirectionalSequenceLstmInt8WithCifgWithPeepholeNoProjectionTest(backends);
+}
+
+TEST_CASE ("UnidirectionalSequenceLstmInt8NoCifgWithPeepholeWithProjectionWithLayerNormTest_CpuRef_Test")
+{
+ std::vector <armnn::BackendId> backends = {armnn::Compute::CpuRef};
+ UnidirectionalSequenceLstmInt8NoCifgWithPeepholeWithProjectionWithLayerNormTest(backends);
+}
+
} //End of TEST_SUITE("UnidirectionalSequenceLstmTest_CpuRef")
} // namespace armnnDelegate
\ No newline at end of file
diff --git a/docs/01_03_delegate.dox b/docs/01_03_delegate.dox
index 92cdf6d..04f216a 100644
--- a/docs/01_03_delegate.dox
+++ b/docs/01_03_delegate.dox
@@ -165,6 +165,8 @@
- TRANSPOSE_CONV
+- UNIDIRECTIONAL_SEQUENCE_LSTM
+
- UNPACK
More machine learning operators will be supported in future releases.
diff --git a/docs/05_operator_list.dox b/docs/05_operator_list.dox
index 4c4f6d1..1f4b43f 100644
--- a/docs/05_operator_list.dox
+++ b/docs/05_operator_list.dox
@@ -3173,7 +3173,7 @@
</table>
<tr>
<td rowspan="3">UnidirectionalSquenceLstmLayer
- <td rowspan="3" style="width:200px;"> Layer to perform unidirectional LSTM operation.
+ <td rowspan="3" style="width:200px;"> Layer to perform unidirectional sequence LSTM operation.
<td rowspan="3">
<ul>
<li>ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_LSTM
@@ -3185,43 +3185,14 @@
</ul>
<td>
<table>
- <tr><th>
- <tr><td>All
+ <tr><th>Input Types
+ <tr><td>FLOAT32
</table>
-<tr>
- <td>CpuAcc
- <td>
- <ul>
- <li>NHWC
- <li>NCHW
- </ul>
- <td>
- <table>
- <tr><th>
- <tr><td>SIGNED32
- <tr><td>FLOAT16
- <tr><td>FLOAT32
- <tr><td>QASYMMU8
- <tr><td>QASYMMS8
- <tr><td>QUANTIZEDSYMM8PERAXIS
- </table>
-<tr>
- <td>GpuAcc
- <td>
- <ul>
- <li>NHWC
- <li>NCHW
- </ul>
- <td>
- <table>
- <tr><th>
- <tr><td>SIGNED32
- <tr><td>FLOAT16
- <tr><td>FLOAT32
- <tr><td>QASYMMU8
- <tr><td>QASYMMS8
- <tr><td>QUANTIZEDSYMM8PERAXIS
- </table>
+ <table>
+ <tr><th>Weight Types
+ <tr><td>FLOAT32
+ <tr><td>QASYMMS8
+ </table>
<tr>
<td rowspan="3">UnmapLayer
<td rowspan="3" style="width:200px;"> Layer to perform unmap operation on tensor.
diff --git a/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.cpp b/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.cpp
index ac22d5d..d17dceb 100644
--- a/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.cpp
+++ b/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.cpp
@@ -1028,3 +1028,771 @@
outputHandle->GetShape(),
outputTensorInfo.GetShape());
}
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8Test(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory)
+{
+ IgnoreUnused(memoryManager);
+ unsigned int batchSize = 3;
+ unsigned int timeSize = 2;
+ unsigned int inputSize = 3;
+ unsigned int outputSize = 4;
+ unsigned numUnits = outputSize;
+
+ armnn::TensorInfo inputTensorInfo({batchSize, timeSize, inputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo cellStateInTensorInfo({batchSize, numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo outputStateInTensorInfo({batchSize, outputSize}, armnn::DataType::Float32);
+
+ armnn::TensorInfo outputTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
+
+ const std::vector<float> inputVector = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> cellStateInVector(batchSize * numUnits, 0.f);
+ std::vector<float> outputStateInVector(batchSize * outputSize, 0.f);
+
+ std::vector<float> actualOutput(outputTensorInfo.GetNumElements());
+
+ const std::vector<float> outputVector = { -0.0142517f, -0.0198845f, -0.0120569f, -0.0116868f,
+ -0.0350714f, -0.0343202f, -0.047504f, -0.0569789f,
+ -0.0146346f, 0.0106663f, -0.0247238f, -0.0319502f,
+ -0.0294759f, -0.0129935f, -0.0444175f, -0.0444354f,
+ -0.0280855f, 0.00545101f, -0.051422f, -0.0463838f,
+ -0.0310702f, 0.00915739f, -0.0625207f, -0.0482648f };
+
+ std::unique_ptr<armnn::ITensorHandle> inputHandle = tensorHandleFactory.CreateTensorHandle(inputTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> cellStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(cellStateInTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> outputStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(outputStateInTensorInfo);
+
+ std::unique_ptr<armnn::ITensorHandle> outputHandle = tensorHandleFactory.CreateTensorHandle(outputTensorInfo);
+
+ armnn::UnidirectionalSequenceLstmQueueDescriptor data;
+ armnn::WorkloadInfo info;
+
+ AddInputToWorkload(data, info, inputTensorInfo, inputHandle.get());
+ AddInputToWorkload(data, info, outputStateInTensorInfo, outputStateInHandle.get());
+ AddInputToWorkload(data, info, cellStateInTensorInfo, cellStateInHandle.get());
+
+ AddOutputToWorkload(data, info, outputTensorInfo, outputHandle.get());
+
+ armnn::TensorInfo tensorInfoNumFp({numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo tensorInfoNumInput({numUnits, inputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumOutput({numUnits, outputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3 };
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -1, -1 };
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+ std::vector<float> inputGateBias = { 0., 0., 0., 0. };
+ std::vector<float> forgetGateBias = { 1., 1., 1., 1. };
+ std::vector<float> cellBias = { 0., 0., 0., 0. };
+ std::vector<float> outputGateBias = { 0., 0., 0., 0. };
+
+ armnn::ScopedTensorHandle inputToInputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToForgetWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToCellWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToOutputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle recurrentToInputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToForgetWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToCellWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToOutputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle inputGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle forgetGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle outputGateBiasTensor(tensorInfoNumFp);
+
+ AllocateAndCopyDataToITensorHandle(&inputToInputWeightsTensor, inputToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToForgetWeightsTensor, inputToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToCellWeightsTensor, inputToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToOutputWeightsTensor, inputToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToInputWeightsTensor, recurrentToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToForgetWeightsTensor, recurrentToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToCellWeightsTensor, recurrentToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToOutputWeightsTensor, recurrentToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputGateBiasTensor, inputGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&forgetGateBiasTensor, forgetGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&cellBiasTensor, cellBias.data());
+ AllocateAndCopyDataToITensorHandle(&outputGateBiasTensor, outputGateBias.data());
+
+ data.m_InputToInputWeights = &inputToInputWeightsTensor;
+ data.m_InputToForgetWeights = &inputToForgetWeightsTensor;
+ data.m_InputToCellWeights = &inputToCellWeightsTensor;
+ data.m_InputToOutputWeights = &inputToOutputWeightsTensor;
+ data.m_RecurrentToInputWeights = &recurrentToInputWeightsTensor;
+ data.m_RecurrentToForgetWeights = &recurrentToForgetWeightsTensor;
+ data.m_RecurrentToCellWeights = &recurrentToCellWeightsTensor;
+ data.m_RecurrentToOutputWeights = &recurrentToOutputWeightsTensor;
+ data.m_InputGateBias = &inputGateBiasTensor;
+ data.m_ForgetGateBias = &forgetGateBiasTensor;
+ data.m_CellBias = &cellBiasTensor;
+ data.m_OutputGateBias = &outputGateBiasTensor;
+
+ // Flags to set test configuration
+ data.m_Parameters.m_ClippingThresCell = 10;
+ data.m_Parameters.m_ClippingThresProj = 0;
+ data.m_Parameters.m_ActivationFunc = 4;
+ data.m_Parameters.m_CifgEnabled = false;
+ data.m_Parameters.m_PeepholeEnabled = false;
+ data.m_Parameters.m_ProjectionEnabled = false;
+ data.m_Parameters.m_TimeMajor = false;
+
+ std::unique_ptr<armnn::IWorkload> workload = workloadFactory.CreateUnidirectionalSequenceLstm(data, info);
+ inputHandle->Allocate();
+ outputStateInHandle->Allocate();
+ cellStateInHandle->Allocate();
+
+ outputHandle->Allocate();
+
+ CopyDataToITensorHandle(inputHandle.get(), inputVector.data());
+ CopyDataToITensorHandle(outputStateInHandle.get(), outputStateInVector.data());
+ CopyDataToITensorHandle(cellStateInHandle.get(), cellStateInVector.data());
+
+ workload->Execute();
+
+ CopyDataFromITensorHandle(actualOutput.data(), outputHandle.get());
+
+ return LayerTestResult<float, 3>(actualOutput,
+ outputVector,
+ outputHandle->GetShape(),
+ outputTensorInfo.GetShape());
+}
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8TimeMajorTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory)
+{
+ IgnoreUnused(memoryManager);
+ unsigned int batchSize = 3;
+ unsigned int timeSize = 2;
+ unsigned int inputSize = 3;
+ unsigned int outputSize = 4;
+ unsigned numUnits = outputSize;
+
+ armnn::TensorInfo inputTensorInfo({timeSize, batchSize, inputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo cellStateInTensorInfo({batchSize, numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo outputStateInTensorInfo({batchSize, outputSize}, armnn::DataType::Float32);
+
+ armnn::TensorInfo outputTensorInfo({timeSize, batchSize, outputSize}, armnn::DataType::Float32);
+
+ const std::vector<float> inputVector = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> cellStateInVector(batchSize * numUnits, 0.f);
+ std::vector<float> outputStateInVector(batchSize * outputSize, 0.f);
+
+ std::vector<float> actualOutput(outputTensorInfo.GetNumElements());
+
+ const std::vector<float> outputVector = { -0.0142517f, -0.0198845f, -0.0120122f, -0.0116868f,
+ -0.0261295f, -0.0188487f, -0.0345463f, -0.049733f,
+ -0.0146346f, 0.0106663f, -0.0247238f, -0.0319502f,
+ -0.0291863f, -0.0369402f, -0.0354071f, -0.0296529f,
+ -0.0419539f, -0.00617731f, -0.0814796f, -0.0804005f,
+ -0.0244737f, 0.0119905f, -0.0457527f, -0.0331862f };
+ std::unique_ptr<armnn::ITensorHandle> inputHandle = tensorHandleFactory.CreateTensorHandle(inputTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> cellStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(cellStateInTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> outputStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(outputStateInTensorInfo);
+
+ std::unique_ptr<armnn::ITensorHandle> outputHandle = tensorHandleFactory.CreateTensorHandle(outputTensorInfo);
+
+ armnn::UnidirectionalSequenceLstmQueueDescriptor data;
+ armnn::WorkloadInfo info;
+
+ AddInputToWorkload(data, info, inputTensorInfo, inputHandle.get());
+ AddInputToWorkload(data, info, outputStateInTensorInfo, outputStateInHandle.get());
+ AddInputToWorkload(data, info, cellStateInTensorInfo, cellStateInHandle.get());
+
+ AddOutputToWorkload(data, info, outputTensorInfo, outputHandle.get());
+
+ armnn::TensorInfo tensorInfoNumFp({numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo tensorInfoNumInput({numUnits, inputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumOutput({numUnits, outputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3 };
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -1, -1 };
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+
+ std::vector<float> inputGateBias = { 0., 0., 0., 0. };
+ std::vector<float> forgetGateBias = { 1., 1., 1., 1. };
+ std::vector<float> cellBias = { 0., 0., 0., 0. };
+ std::vector<float> outputGateBias = { 0., 0., 0., 0. };
+
+ armnn::ScopedTensorHandle inputToInputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToForgetWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToCellWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToOutputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle recurrentToInputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToForgetWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToCellWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToOutputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle inputGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle forgetGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle outputGateBiasTensor(tensorInfoNumFp);
+
+ AllocateAndCopyDataToITensorHandle(&inputToInputWeightsTensor, inputToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToForgetWeightsTensor, inputToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToCellWeightsTensor, inputToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToOutputWeightsTensor, inputToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToInputWeightsTensor, recurrentToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToForgetWeightsTensor, recurrentToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToCellWeightsTensor, recurrentToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToOutputWeightsTensor, recurrentToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputGateBiasTensor, inputGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&forgetGateBiasTensor, forgetGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&cellBiasTensor, cellBias.data());
+ AllocateAndCopyDataToITensorHandle(&outputGateBiasTensor, outputGateBias.data());
+
+ data.m_InputToInputWeights = &inputToInputWeightsTensor;
+ data.m_InputToForgetWeights = &inputToForgetWeightsTensor;
+ data.m_InputToCellWeights = &inputToCellWeightsTensor;
+ data.m_InputToOutputWeights = &inputToOutputWeightsTensor;
+ data.m_RecurrentToInputWeights = &recurrentToInputWeightsTensor;
+ data.m_RecurrentToForgetWeights = &recurrentToForgetWeightsTensor;
+ data.m_RecurrentToCellWeights = &recurrentToCellWeightsTensor;
+ data.m_RecurrentToOutputWeights = &recurrentToOutputWeightsTensor;
+ data.m_InputGateBias = &inputGateBiasTensor;
+ data.m_ForgetGateBias = &forgetGateBiasTensor;
+ data.m_CellBias = &cellBiasTensor;
+ data.m_OutputGateBias = &outputGateBiasTensor;
+
+ // Flags to set test configuration
+ data.m_Parameters.m_ClippingThresCell = 10;
+ data.m_Parameters.m_ClippingThresProj = 0;
+ data.m_Parameters.m_ActivationFunc = 4;
+ data.m_Parameters.m_CifgEnabled = false;
+ data.m_Parameters.m_PeepholeEnabled = false;
+ data.m_Parameters.m_ProjectionEnabled = false;
+ data.m_Parameters.m_TimeMajor = true;
+
+ std::unique_ptr<armnn::IWorkload> workload = workloadFactory.CreateUnidirectionalSequenceLstm(data, info);
+ inputHandle->Allocate();
+ outputStateInHandle->Allocate();
+ cellStateInHandle->Allocate();
+
+ outputHandle->Allocate();
+
+ CopyDataToITensorHandle(inputHandle.get(), inputVector.data());
+ CopyDataToITensorHandle(outputStateInHandle.get(), outputStateInVector.data());
+ CopyDataToITensorHandle(cellStateInHandle.get(), cellStateInVector.data());
+
+ workload->Execute();
+
+ CopyDataFromITensorHandle(actualOutput.data(), outputHandle.get());
+
+ return LayerTestResult<float, 3>(actualOutput,
+ outputVector,
+ outputHandle->GetShape(),
+ outputTensorInfo.GetShape());
+}
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjectionTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory)
+{
+ IgnoreUnused(memoryManager);
+ unsigned int batchSize = 3;
+ unsigned int timeSize = 2;
+ unsigned int outputSize = 4;
+ unsigned int inputSize = 3;
+ unsigned numUnits = 4;
+
+ armnn::TensorInfo inputTensorInfo({batchSize, timeSize, inputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo cellStateInTensorInfo({batchSize , numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo outputStateInTensorInfo({batchSize , outputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo outputTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
+
+ const std::vector<float> inputVector = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> cellStateInVector(batchSize * numUnits, 0.f);
+ std::vector<float> outputStateInVector(batchSize * outputSize, 0.f);
+
+ std::vector<float> actualOutput(outputTensorInfo.GetNumElements());
+
+ const std::vector<float> expectedOutput = { 0.612103f, 1.56788f, 0.31966f, 1.42956f,
+ 0.909718f, 3.07916f, -0.560586f, 3.8907f,
+ 0.753671f, 1.77485f, 0.365122f, 1.60077f,
+ 0.812644f, 2.79092f, -0.605396f, 3.61742f,
+ 0.791857f, 1.64353f, 0.316588f, 1.55192f,
+ 0.807265f, 2.47012f, -0.539598f, 3.25654f };
+
+ std::unique_ptr<armnn::ITensorHandle> inputHandle = tensorHandleFactory.CreateTensorHandle(inputTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> cellStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(cellStateInTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> outputStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(outputStateInTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> outputHandle = tensorHandleFactory.CreateTensorHandle(outputTensorInfo);
+
+ armnn::UnidirectionalSequenceLstmQueueDescriptor data;
+ armnn::WorkloadInfo info;
+
+ AddInputToWorkload(data, info, inputTensorInfo, inputHandle.get());
+ AddInputToWorkload(data, info, outputStateInTensorInfo, outputStateInHandle.get());
+ AddInputToWorkload(data, info, cellStateInTensorInfo, cellStateInHandle.get());
+ AddOutputToWorkload(data, info, outputTensorInfo, outputHandle.get());
+
+ armnn::TensorInfo tensorInfoOut({outputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo tensorInfoNumFp({numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo tensorInfoNum({numUnits}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumInput({numUnits, inputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumOutput({numUnits, outputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoOutNum({outputSize, numUnits}, armnn::DataType::QAsymmS8, 0.1f, 0);
+
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3 };
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -1, -1 };
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+ std::vector<float> inputGateBias = { 0.02234832f, 0.14757581f, 0.18176508f, 0.10380666f};
+ std::vector<float> forgetGateBias = { 0.035185695f, -0.042891346f, -0.3032477f, 0.23027696f};
+ std::vector<float> cellBias = { -0.124379363f, 0.55531194f, 0.23377132f, 0.033463873f };
+ std::vector<float> outputGateBias = { 0.046159424f, -0.12809046f, 0.03563469f, 0.12648113f };
+
+ std::vector<int8_t> cellToInputWeights = { 5, 10, 25, 15 };
+ std::vector<int8_t> cellToForgetWeights = { -5, 15, 25, 3 };
+ std::vector<int8_t> cellToOutputWeights = { 10, -10, -5, 50 };
+
+ std::vector<int8_t> projectionWeights = { -25, 51, 3, -5, 25, 127, 77, 20, 18, 51, -10, 51, -25, 88, 77, -13 };
+
+ std::vector<float> projectionBiasVector(outputSize, 0.f); //{outputSize}
+
+ armnn::ScopedTensorHandle inputToInputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToForgetWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToCellWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToOutputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle recurrentToForgetWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToInputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToCellWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToOutputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle cellToInputWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle inputGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle forgetGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle outputGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellToForgetWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle cellToOutputWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle projectionWeightsTensor(tensorInfoOutNum);
+ armnn::ScopedTensorHandle projectionBiasTensor(tensorInfoOut);
+
+ AllocateAndCopyDataToITensorHandle(&inputToInputWeightsTensor, inputToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToForgetWeightsTensor, inputToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToCellWeightsTensor, inputToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToOutputWeightsTensor, inputToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToInputWeightsTensor, recurrentToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToForgetWeightsTensor, recurrentToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToCellWeightsTensor, recurrentToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToOutputWeightsTensor, recurrentToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&cellToInputWeightsTensor, cellToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputGateBiasTensor, inputGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&forgetGateBiasTensor, forgetGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&cellBiasTensor, cellBias.data());
+ AllocateAndCopyDataToITensorHandle(&outputGateBiasTensor, outputGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&cellToForgetWeightsTensor, cellToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&cellToOutputWeightsTensor, cellToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&projectionWeightsTensor, projectionWeights.data());
+ AllocateAndCopyDataToITensorHandle(&projectionBiasTensor, projectionBiasVector.data());
+
+ data.m_InputToInputWeights = &inputToInputWeightsTensor;
+ data.m_InputToForgetWeights = &inputToForgetWeightsTensor;
+ data.m_InputToCellWeights = &inputToCellWeightsTensor;
+ data.m_InputToOutputWeights = &inputToOutputWeightsTensor;
+ data.m_RecurrentToInputWeights = &recurrentToInputWeightsTensor;
+ data.m_RecurrentToForgetWeights = &recurrentToForgetWeightsTensor;
+ data.m_RecurrentToCellWeights = &recurrentToCellWeightsTensor;
+ data.m_RecurrentToOutputWeights = &recurrentToOutputWeightsTensor;
+ data.m_CellToInputWeights = &cellToInputWeightsTensor;
+ data.m_InputGateBias = &inputGateBiasTensor;
+ data.m_ForgetGateBias = &forgetGateBiasTensor;
+ data.m_CellBias = &cellBiasTensor;
+ data.m_OutputGateBias = &outputGateBiasTensor;
+ data.m_CellToForgetWeights = &cellToForgetWeightsTensor;
+ data.m_CellToOutputWeights = &cellToOutputWeightsTensor;
+ data.m_ProjectionWeights = &projectionWeightsTensor;
+ data.m_ProjectionBias = &projectionBiasTensor;
+
+ // Flags to set test configuration
+ data.m_Parameters.m_ActivationFunc = 4;
+ data.m_Parameters.m_CifgEnabled = false;
+ data.m_Parameters.m_PeepholeEnabled = true;
+ data.m_Parameters.m_ProjectionEnabled = true;
+ data.m_Parameters.m_LayerNormEnabled = false;
+ data.m_Parameters.m_TimeMajor = false;
+ data.m_Parameters.m_ClippingThresCell = 10.0f;
+
+
+ std::unique_ptr<armnn::IWorkload> workload = workloadFactory.CreateUnidirectionalSequenceLstm(data, info);
+ inputHandle->Allocate();
+ outputStateInHandle->Allocate();
+ cellStateInHandle->Allocate();
+ outputHandle->Allocate();
+
+ CopyDataToITensorHandle(inputHandle.get(), inputVector.data());
+ CopyDataToITensorHandle(outputStateInHandle.get(), outputStateInVector.data());
+ CopyDataToITensorHandle(cellStateInHandle.get(), cellStateInVector.data());
+
+ workload->Execute();
+
+ CopyDataFromITensorHandle(actualOutput.data(), outputHandle.get());
+
+ return LayerTestResult<float, 3>(actualOutput,
+ expectedOutput,
+ outputHandle->GetShape(),
+ outputTensorInfo.GetShape());
+}
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjectionWithLayerNormTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory)
+{
+ IgnoreUnused(memoryManager);
+ unsigned int batchSize = 3;
+ unsigned int timeSize = 2;
+ unsigned int outputSize = 4;
+ unsigned int inputSize = 3;
+ unsigned numUnits = 5;
+
+ armnn::TensorInfo inputTensorInfo({batchSize, timeSize, inputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo cellStateInTensorInfo({batchSize , numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo outputStateInTensorInfo({batchSize , outputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo outputTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
+
+ const std::vector<float> inputVector = { 1., 8., 3., 4., 5., 4.,
+ 3., 2., 1., 2., 3., 4.,
+ 5., 4., 3., 2., 1., 2. };
+
+ std::vector<float> cellStateInVector(batchSize * numUnits, 0.f);
+ std::vector<float> outputStateInVector(batchSize * outputSize, 0.f);
+
+ std::vector<float> actualOutput(outputTensorInfo.GetNumElements());
+
+ const std::vector<float> expectedOutput = { 0.0471276f, 0.0168155f, 0.0789885f, 0.16550f,
+ 0.0643133f, -0.0400722f, 0.100593f, 0.197722f,
+ 0.0465562f, -0.0600682f, 0.0622087f, 0.115053f,
+ 0.056287f, -0.0566218f, 0.0856832f, 0.148484f,
+ 0.0457859f, -0.0588112f, 0.0623636f, 0.114333f,
+ 0.0509271f, -0.0754262f, 0.058600f, 0.0801288f };
+
+ std::unique_ptr<armnn::ITensorHandle> inputHandle = tensorHandleFactory.CreateTensorHandle(inputTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> cellStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(cellStateInTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> outputStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(outputStateInTensorInfo);
+
+ std::unique_ptr<armnn::ITensorHandle> outputHandle = tensorHandleFactory.CreateTensorHandle(outputTensorInfo);
+
+ armnn::UnidirectionalSequenceLstmQueueDescriptor data;
+ armnn::WorkloadInfo info;
+
+ AddInputToWorkload(data, info, inputTensorInfo, inputHandle.get());
+ AddInputToWorkload(data, info, outputStateInTensorInfo, outputStateInHandle.get());
+ AddInputToWorkload(data, info, cellStateInTensorInfo, cellStateInHandle.get());
+
+ AddOutputToWorkload(data, info, outputTensorInfo, outputHandle.get());
+
+ armnn::TensorInfo tensorInfoOut({outputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo tensorInfoNumFp({numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo tensorInfoNum({numUnits}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumInput({numUnits, inputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumOutput({numUnits, outputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoOutNum({outputSize, numUnits}, armnn::DataType::QAsymmS8, 0.1f, 0);
+
+ std::vector<int8_t> inputToInputWeights = { -4, -1, -1, -2, 3, -2, 2, 4, 1, -4, -2, 3, 2, 2, -4 };
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1, -3, -2, -4 };
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3, 2, 5, -4 };
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4, -4, -1, -1 };
+
+ std::vector<float> inputGateBias = { 0.03f, 0.15f, 0.22f, 0.38f, 0.05f };
+ std::vector<float> forgetGateBias = { 0.1f, -0.3f, -0.2f, 0.1f, 0.4f };
+ std::vector<float> cellBias = { -0.05f, 0.72f, 0.25f, 0.08f, 0.1f };
+ std::vector<float> outputGateBias = { 0.05f, -0.01f, 0.2f, 0.1f, -0.2f };
+
+ std::vector<int8_t> recurrentToInputWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3,
+ 5, -1, 1, 3, -1, -1, -1, 4, 2, 3 };
+
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3,
+ 5, -1, 1, 3, -2, -1, -1, 2, 2, 1 };
+
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2,
+ 1, 2, 3, -2, 3, -3, -1, -5, 1, 3 };
+
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3,
+ -4, -1, -1, -1, 2, -1, 5, 1, -3, -4 };
+
+ std::vector<int8_t> cellToInputWeights = { 5, 3, 8, -5, 2 };
+ std::vector<int8_t> cellToForgetWeights = { -2, -7, 5, -3, 4 };
+ std::vector<int8_t> cellToOutputWeights = { 9, -10 , -5, 5, 1 };
+
+ std::vector<int8_t> projectionWeights = { -1, 2, 1, -2, 1, 5, 3, 8, 7, 2,
+ -4, 2, 5, -4, 3, -2, 3, 8, -7, 2 };
+
+ std::vector<float> projectionBiasVector(outputSize, 0.f); //{outputSize}
+
+ std::vector<float> inputLayerNormWeights = { 0.1f, 0.2f, -0.3f, -0.1f, 0.5f };
+ std::vector<float> forgetLayerNormWeights = { -0.1f, 0.2f, 0.3f, 0.5f, 0.2f };
+ std::vector<float> cellLayerNormWeights = { 0.5f, 0.2f, 0.3f, 0.4f, -0.5f };
+ std::vector<float> outputLayerNormWeights = { 0.6f, -0.2f, -0.2f, 0.5f, 0.1f };
+
+ armnn::ScopedTensorHandle inputToInputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToForgetWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToCellWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToOutputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle recurrentToForgetWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToInputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToCellWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToOutputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle cellToInputWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle inputGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle forgetGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle outputGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellToForgetWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle cellToOutputWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle projectionWeightsTensor(tensorInfoOutNum);
+ armnn::ScopedTensorHandle projectionBiasTensor(tensorInfoOut);
+
+ armnn::ScopedTensorHandle inputLayerNormWeightsTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle forgetLayerNormWeightsTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellLayerNormWeightsTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle outputLayerNormWeightsTensor(tensorInfoNumFp);
+
+ AllocateAndCopyDataToITensorHandle(&inputToInputWeightsTensor, inputToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToForgetWeightsTensor, inputToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToCellWeightsTensor, inputToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToOutputWeightsTensor, inputToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToInputWeightsTensor, recurrentToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToForgetWeightsTensor, recurrentToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToCellWeightsTensor, recurrentToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToOutputWeightsTensor, recurrentToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&cellToInputWeightsTensor, cellToInputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputGateBiasTensor, inputGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&forgetGateBiasTensor, forgetGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&cellBiasTensor, cellBias.data());
+ AllocateAndCopyDataToITensorHandle(&outputGateBiasTensor, outputGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&cellToForgetWeightsTensor, cellToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&cellToOutputWeightsTensor, cellToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&projectionWeightsTensor, projectionWeights.data());
+ AllocateAndCopyDataToITensorHandle(&projectionBiasTensor, projectionBiasVector.data());
+
+ AllocateAndCopyDataToITensorHandle(&inputLayerNormWeightsTensor, inputLayerNormWeights.data());
+ AllocateAndCopyDataToITensorHandle(&forgetLayerNormWeightsTensor, forgetLayerNormWeights.data());
+ AllocateAndCopyDataToITensorHandle(&cellLayerNormWeightsTensor, cellLayerNormWeights.data());
+ AllocateAndCopyDataToITensorHandle(&outputLayerNormWeightsTensor, outputLayerNormWeights.data());
+
+ data.m_InputToInputWeights = &inputToInputWeightsTensor;
+ data.m_InputToForgetWeights = &inputToForgetWeightsTensor;
+ data.m_InputToCellWeights = &inputToCellWeightsTensor;
+ data.m_InputToOutputWeights = &inputToOutputWeightsTensor;
+ data.m_RecurrentToInputWeights = &recurrentToInputWeightsTensor;
+ data.m_RecurrentToForgetWeights = &recurrentToForgetWeightsTensor;
+ data.m_RecurrentToCellWeights = &recurrentToCellWeightsTensor;
+ data.m_RecurrentToOutputWeights = &recurrentToOutputWeightsTensor;
+ data.m_CellToInputWeights = &cellToInputWeightsTensor;
+ data.m_InputGateBias = &inputGateBiasTensor;
+ data.m_ForgetGateBias = &forgetGateBiasTensor;
+ data.m_CellBias = &cellBiasTensor;
+ data.m_OutputGateBias = &outputGateBiasTensor;
+ data.m_CellToForgetWeights = &cellToForgetWeightsTensor;
+ data.m_CellToOutputWeights = &cellToOutputWeightsTensor;
+ data.m_ProjectionWeights = &projectionWeightsTensor;
+ data.m_ProjectionBias = &projectionBiasTensor;
+
+ data.m_InputLayerNormWeights = &inputLayerNormWeightsTensor;
+ data.m_ForgetLayerNormWeights = &forgetLayerNormWeightsTensor;
+ data.m_CellLayerNormWeights = &cellLayerNormWeightsTensor;
+ data.m_OutputLayerNormWeights = &outputLayerNormWeightsTensor;
+
+ // Flags to set test configuration
+ data.m_Parameters.m_ActivationFunc = 4;
+ data.m_Parameters.m_CifgEnabled = false;
+ data.m_Parameters.m_PeepholeEnabled = true;
+ data.m_Parameters.m_ProjectionEnabled = true;
+ data.m_Parameters.m_LayerNormEnabled = true;
+ data.m_Parameters.m_TimeMajor = false;
+ data.m_Parameters.m_ClippingThresCell = 10.0f;
+
+ std::unique_ptr<armnn::IWorkload> workload = workloadFactory.CreateUnidirectionalSequenceLstm(data, info);
+ inputHandle->Allocate();
+ outputStateInHandle->Allocate();
+ cellStateInHandle->Allocate();
+ outputHandle->Allocate();
+
+ CopyDataToITensorHandle(inputHandle.get(), inputVector.data());
+ CopyDataToITensorHandle(outputStateInHandle.get(), outputStateInVector.data());
+ CopyDataToITensorHandle(cellStateInHandle.get(), cellStateInVector.data());
+
+ workload->Execute();
+
+ CopyDataFromITensorHandle(actualOutput.data(), outputHandle.get());
+
+ return LayerTestResult<float, 3>(actualOutput,
+ expectedOutput,
+ outputHandle->GetShape(),
+ outputTensorInfo.GetShape());
+}
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmInt8WithCifgWithPeepholeNoProjectionTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory)
+{
+ IgnoreUnused(memoryManager);
+ unsigned int batchSize = 3;
+ unsigned int timeSize = 2;
+ unsigned int inputSize = 3;
+ unsigned int outputSize = 4;
+ unsigned numUnits = outputSize;
+
+ armnn::TensorInfo inputTensorInfo({batchSize, timeSize, inputSize}, armnn::DataType::Float32);
+ armnn::TensorInfo cellStateInTensorInfo({batchSize, numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo outputStateInTensorInfo({batchSize, outputSize}, armnn::DataType::Float32);
+
+ armnn::TensorInfo outputTensorInfo({batchSize, timeSize, outputSize}, armnn::DataType::Float32);
+
+ const std::vector<float> inputVector = { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.4f,
+ 0.3f, 0.2f, 0.1f, 0.2f, 0.3f, 0.4f,
+ 0.5f, 0.4f, 0.3f, 0.2f, 0.1f, 0.2f };
+
+ std::vector<float> cellStateInVector(batchSize * numUnits, 0.f);
+ std::vector<float> outputStateInVector(batchSize * outputSize, 0.f);
+
+ std::vector<float> actualOutput(outputTensorInfo.GetNumElements());
+
+ const std::vector<float> outputVector = { -0.0072104f, -0.00991171f, -0.00650478f, -0.00713055f,
+ -0.0191782f, -0.0161269f, -0.0233683f, -0.054299f,
+ -0.00783725f, 0.00635271f, -0.0126718f, -0.022613f,
+ -0.0161351f, -0.00775868f, -0.021054f, -0.0339778f,
+ -0.0146392f, 0.00330261f, -0.0258733f, -0.0407797f,
+ -0.0174297f, 0.0050105f, -0.0266275f, -0.0362564f };
+
+ std::unique_ptr<armnn::ITensorHandle> inputHandle = tensorHandleFactory.CreateTensorHandle(inputTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> cellStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(cellStateInTensorInfo);
+ std::unique_ptr<armnn::ITensorHandle> outputStateInHandle =
+ tensorHandleFactory.CreateTensorHandle(outputStateInTensorInfo);
+
+ std::unique_ptr<armnn::ITensorHandle> outputHandle = tensorHandleFactory.CreateTensorHandle(outputTensorInfo);
+
+ armnn::UnidirectionalSequenceLstmQueueDescriptor data;
+ armnn::WorkloadInfo info;
+
+ AddInputToWorkload(data, info, inputTensorInfo, inputHandle.get());
+ AddInputToWorkload(data, info, outputStateInTensorInfo, outputStateInHandle.get());
+ AddInputToWorkload(data, info, cellStateInTensorInfo, cellStateInHandle.get());
+
+ AddOutputToWorkload(data, info, outputTensorInfo, outputHandle.get());
+
+ armnn::TensorInfo tensorInfoNumFp({numUnits}, armnn::DataType::Float32);
+ armnn::TensorInfo tensorInfoNum({numUnits}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumInput({numUnits, inputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+ armnn::TensorInfo tensorInfoNumOutput({numUnits, outputSize}, armnn::DataType::QAsymmS8, 0.1f, 0);
+
+ std::vector<int8_t> inputToForgetWeights = { 2, 1, 4, -4, 3, -1, -3, -2, -3, 1, -4, -1 };
+ std::vector<int8_t> inputToCellWeights = { -2, 1, -2, 4, -3, -2, -4, 3, -2, -2, -6, 3 };
+ std::vector<int8_t> inputToOutputWeights = { 2, 5, -4, 5, 2, -3, 5, 7, 3, -5, 1, -4 };
+
+ std::vector<int8_t> recurrentToForgetWeights = { -1, 1, -1, 1, -3, -4, -1, 4, 2, 3, 5, -1, 1, 3, -2, -1 };
+ std::vector<int8_t> recurrentToCellWeights = { -2, -3, -1, -3, -4, 2, 1, -1, 2, 2, 1, 2, 3, -2, 3, -3 };
+ std::vector<int8_t> recurrentToOutputWeights = { -3, 3, -1, -2, -2, -2, -1, -5, 1, 3, -4, -1, -1, -1, 2, -1 };
+
+ std::vector<int8_t> cellToForgetWeights = { 47, -52, -24, 31 };
+ std::vector<int8_t> cellToOutputWeights = { -17, 82, 85, -77 };
+
+ std::vector<float> forgetGateBias = { 1., 1., 1., 1. };
+ std::vector<float> cellBias = { 0., 0., 0., 0. };
+ std::vector<float> outputGateBias = { 0., 0., 0., 0. };
+
+ armnn::ScopedTensorHandle inputToForgetWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToCellWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle inputToOutputWeightsTensor(tensorInfoNumInput);
+ armnn::ScopedTensorHandle recurrentToForgetWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToCellWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle recurrentToOutputWeightsTensor(tensorInfoNumOutput);
+ armnn::ScopedTensorHandle cellToForgetWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle cellToOutputWeightsTensor(tensorInfoNum);
+ armnn::ScopedTensorHandle forgetGateBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle cellBiasTensor(tensorInfoNumFp);
+ armnn::ScopedTensorHandle outputGateBiasTensor(tensorInfoNumFp);
+
+ AllocateAndCopyDataToITensorHandle(&inputToForgetWeightsTensor, inputToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToCellWeightsTensor, inputToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&inputToOutputWeightsTensor, inputToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToForgetWeightsTensor, recurrentToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToCellWeightsTensor, recurrentToCellWeights.data());
+ AllocateAndCopyDataToITensorHandle(&recurrentToOutputWeightsTensor, recurrentToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&cellToForgetWeightsTensor, cellToForgetWeights.data());
+ AllocateAndCopyDataToITensorHandle(&cellToOutputWeightsTensor, cellToOutputWeights.data());
+ AllocateAndCopyDataToITensorHandle(&forgetGateBiasTensor, forgetGateBias.data());
+ AllocateAndCopyDataToITensorHandle(&cellBiasTensor, cellBias.data());
+ AllocateAndCopyDataToITensorHandle(&outputGateBiasTensor, outputGateBias.data());
+
+ data.m_InputToForgetWeights = &inputToForgetWeightsTensor;
+ data.m_InputToCellWeights = &inputToCellWeightsTensor;
+ data.m_InputToOutputWeights = &inputToOutputWeightsTensor;
+ data.m_RecurrentToForgetWeights = &recurrentToForgetWeightsTensor;
+ data.m_RecurrentToCellWeights = &recurrentToCellWeightsTensor;
+ data.m_RecurrentToOutputWeights = &recurrentToOutputWeightsTensor;
+ data.m_CellToForgetWeights = &cellToForgetWeightsTensor;
+ data.m_CellToOutputWeights = &cellToOutputWeightsTensor;
+ data.m_ForgetGateBias = &forgetGateBiasTensor;
+ data.m_CellBias = &cellBiasTensor;
+ data.m_OutputGateBias = &outputGateBiasTensor;
+
+ // Flags to set test configuration
+ data.m_Parameters.m_ClippingThresCell = 10;
+ data.m_Parameters.m_ClippingThresProj = 0;
+ data.m_Parameters.m_ActivationFunc = 4;
+ data.m_Parameters.m_CifgEnabled = true;
+ data.m_Parameters.m_PeepholeEnabled = true;
+ data.m_Parameters.m_ProjectionEnabled = false;
+ data.m_Parameters.m_TimeMajor = false;
+
+ std::unique_ptr<armnn::IWorkload> workload = workloadFactory.CreateUnidirectionalSequenceLstm(data, info);
+ inputHandle->Allocate();
+ outputStateInHandle->Allocate();
+ cellStateInHandle->Allocate();
+
+ outputHandle->Allocate();
+
+ CopyDataToITensorHandle(inputHandle.get(), inputVector.data());
+ CopyDataToITensorHandle(outputStateInHandle.get(), outputStateInVector.data());
+ CopyDataToITensorHandle(cellStateInHandle.get(), cellStateInVector.data());
+
+ workload->Execute();
+
+ CopyDataFromITensorHandle(actualOutput.data(), outputHandle.get());
+
+ return LayerTestResult<float, 3>(actualOutput,
+ outputVector,
+ outputHandle->GetShape(),
+ outputTensorInfo.GetShape());
+}
\ No newline at end of file
diff --git a/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.hpp b/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.hpp
index 7b14065..20ac313 100644
--- a/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.hpp
+++ b/src/backends/backendsCommon/test/layerTests/UnidirectionalSequenceLstmTestImpl.hpp
@@ -33,4 +33,29 @@
LayerTestResult<float, 3> UnidirectionalSequenceLstmWithCifgWithPeepholeNoProjectionTest(
armnn::IWorkloadFactory& workloadFactory,
const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory);
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8Test(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory);
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8TimeMajorTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory);
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjectionTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory);
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjectionWithLayerNormTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
+ const armnn::ITensorHandleFactory& tensorHandleFactory);
+
+LayerTestResult<float, 3> UnidirectionalSequenceLstmInt8WithCifgWithPeepholeNoProjectionTest(
+ armnn::IWorkloadFactory& workloadFactory,
+ const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
const armnn::ITensorHandleFactory& tensorHandleFactory);
\ No newline at end of file
diff --git a/src/backends/reference/RefLayerSupport.cpp b/src/backends/reference/RefLayerSupport.cpp
index 2603371..5eba3e5 100644
--- a/src/backends/reference/RefLayerSupport.cpp
+++ b/src/backends/reference/RefLayerSupport.cpp
@@ -2317,9 +2317,10 @@
DataType::Float32
};
- std::array<DataType, 1> supportedWeightTypes =
+ std::array<DataType, 2> supportedWeightTypes =
{
- DataType::Float32
+ DataType::Float32,
+ DataType::QAsymmS8
};
// check inputs and outputs
diff --git a/src/backends/reference/test/RefLayerTests.cpp b/src/backends/reference/test/RefLayerTests.cpp
index 0cf36f2..e906b29 100644
--- a/src/backends/reference/test/RefLayerTests.cpp
+++ b/src/backends/reference/test/RefLayerTests.cpp
@@ -2341,5 +2341,15 @@
UnidirectionalSequenceLstmLayerNoCifgWithPeepholeWithProjectionWithLayerNormTest)
ARMNN_AUTO_TEST_CASE_WITH_THF(UnidirectionalSequenceLstmWithCifgWithPeepholeNoProjection,
UnidirectionalSequenceLstmWithCifgWithPeepholeNoProjectionTest)
+ARMNN_AUTO_TEST_CASE_WITH_THF(UnidirectionalSequenceLstmLayerInt8,
+ UnidirectionalSequenceLstmLayerInt8Test)
+ARMNN_AUTO_TEST_CASE_WITH_THF(UnidirectionalSequenceLstmLayerInt8TimeMajor,
+ UnidirectionalSequenceLstmLayerInt8TimeMajorTest)
+ARMNN_AUTO_TEST_CASE_WITH_THF(UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjection,
+ UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjectionTest)
+ARMNN_AUTO_TEST_CASE_WITH_THF(UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjectionWithLayerNorm,
+ UnidirectionalSequenceLstmLayerInt8NoCifgWithPeepholeWithProjectionWithLayerNormTest)
+ARMNN_AUTO_TEST_CASE_WITH_THF(UnidirectionalSequenceLstmInt8WithCifgWithPeepholeNoProjection,
+ UnidirectionalSequenceLstmInt8WithCifgWithPeepholeNoProjectionTest)
}
\ No newline at end of file