// clang-format off // Generated file (from: dequantize_relaxed.mod.py). Do not edit void CreateModel(Model *model) { OperandType type0(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 1.0f, 0); OperandType type1(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); // Phase 1, operands auto op1 = model->addOperand(&type0); auto op2 = model->addOperand(&type1); // Phase 2, operations model->addOperation(ANEURALNETWORKS_DEQUANTIZE, {op1}, {op2}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op2}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape(Model *model) { OperandType type0(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 1.0f, 0); OperandType type2(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); // Phase 1, operands auto op1 = model->addOperand(&type0); auto op2 = model->addOperand(&type2); // Phase 2, operations model->addOperation(ANEURALNETWORKS_DEQUANTIZE, {op1}, {op2}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op2}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); }