// clang-format off // Generated file (from: depthwise_conv2d_dilation.mod.py). Do not edit void CreateModel_nhwc(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type2); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_nhwc(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type2); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nhwc_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type10); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type11); // Phase 2, operations static _Float16 op2_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(_Float16) * 16); static _Float16 op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(_Float16) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_nhwc_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type16(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type13); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type16); // Phase 2, operations static uint8_t op2_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op2, op2_init, sizeof(uint8_t) * 16); static int32_t op3_init[] = {200, 400, 600, 800}; model->setOperandValue(op3, op3_init, sizeof(int32_t) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_nhwc_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type2); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type2); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type10); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type11); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type16(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type13); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type16); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type18); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_nchw(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type18); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nchw_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type20(Type::TENSOR_FLOAT16, {1, 4, 2, 2}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type19); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type20); // Phase 2, operations static _Float16 op2_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(_Float16) * 16); static _Float16 op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(_Float16) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_nchw_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type22(Type::TENSOR_QUANT8_ASYMM, {1, 4, 2, 2}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type21); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type22); // Phase 2, operations static uint8_t op2_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op2, op2_init, sizeof(uint8_t) * 16); static int32_t op3_init[] = {200, 400, 600, 800}; model->setOperandValue(op3, op3_init, sizeof(int32_t) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_nchw_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type18); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type18); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type20(Type::TENSOR_FLOAT16, {1, 4, 2, 2}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type19); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type20); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type22(Type::TENSOR_QUANT8_ASYMM, {1, 4, 2, 2}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type21); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type22); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type10); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type24); // Phase 2, operations static _Float16 op2_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(_Float16) * 16); static _Float16 op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(_Float16) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type13); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type25); // Phase 2, operations static uint8_t op2_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op2, op2_init, sizeof(uint8_t) * 16); static int32_t op3_init[] = {200, 400, 600, 800}; model->setOperandValue(op3, op3_init, sizeof(int32_t) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type1); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type10); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type24); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type13); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type25); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static float op2_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(float) * 16); static float op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(float) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type19); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type24); // Phase 2, operations static _Float16 op2_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op2, op2_init, sizeof(_Float16) * 16); static _Float16 op3_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op3, op3_init, sizeof(_Float16) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type21); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type25); // Phase 2, operations static uint8_t op2_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op2, op2_init, sizeof(uint8_t) * 16); static int32_t op3_init[] = {200, 400, 600, 800}; model->setOperandValue(op3, op3_init, sizeof(int32_t) * 4); static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_relaxed(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type17); auto op2 = model->addOperand(&type2); auto op3 = model->addOperand(&type3); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type23); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_relaxed(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_float16(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type19); auto op2 = model->addOperand(&type11); auto op3 = model->addOperand(&type12); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type24); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_float16(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_quant8(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op1 = model->addOperand(&type21); auto op2 = model->addOperand(&type14); auto op3 = model->addOperand(&type15); auto param = model->addOperand(&type4); auto param1 = model->addOperand(&type4); auto param2 = model->addOperand(&type4); auto param3 = model->addOperand(&type4); auto param4 = model->addOperand(&type4); auto param5 = model->addOperand(&type4); auto param6 = model->addOperand(&type4); auto param7 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param8 = model->addOperand(&type4); auto param9 = model->addOperand(&type4); auto op4 = model->addOperand(&type25); // Phase 2, operations static int32_t param_init[] = {0}; model->setOperandValue(param, param_init, sizeof(int32_t) * 1); static int32_t param1_init[] = {0}; model->setOperandValue(param1, param1_init, sizeof(int32_t) * 1); static int32_t param2_init[] = {0}; model->setOperandValue(param2, param2_init, sizeof(int32_t) * 1); static int32_t param3_init[] = {0}; model->setOperandValue(param3, param3_init, sizeof(int32_t) * 1); static int32_t param4_init[] = {1}; model->setOperandValue(param4, param4_init, sizeof(int32_t) * 1); static int32_t param5_init[] = {1}; model->setOperandValue(param5, param5_init, sizeof(int32_t) * 1); static int32_t param6_init[] = {2}; model->setOperandValue(param6, param6_init, sizeof(int32_t) * 1); static int32_t param7_init[] = {0}; model->setOperandValue(param7, param7_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param8_init[] = {1}; model->setOperandValue(param8, param8_init, sizeof(int32_t) * 1); static int32_t param9_init[] = {1}; model->setOperandValue(param9, param9_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op1, op2, op3, param, param1, param2, param3, param4, param5, param6, param7, layout, param8, param9}, {op4}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op1, op2, op3}, {op4}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_quant8(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op11 = model->addOperand(&type5); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type2); // Phase 2, operations static float op21_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op21, op21_init, sizeof(float) * 16); static float op31_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op31, op31_init, sizeof(float) * 4); static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11}, {op41}); assert(model->isValid()); } inline bool is_ignored_nhwc_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op11 = model->addOperand(&type5); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type2); // Phase 2, operations static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11, op21, op31}, {op41}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op11 = model->addOperand(&type26); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type18); // Phase 2, operations static float op21_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op21, op21_init, sizeof(float) * 16); static float op31_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op31, op31_init, sizeof(float) * 4); static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11}, {op41}); assert(model->isValid()); } inline bool is_ignored_nchw_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op11 = model->addOperand(&type26); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type18); // Phase 2, operations static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11, op21, op31}, {op41}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op11 = model->addOperand(&type5); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type23); // Phase 2, operations static float op21_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op21, op21_init, sizeof(float) * 16); static float op31_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op31, op31_init, sizeof(float) * 4); static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11}, {op41}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op11 = model->addOperand(&type5); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type23); // Phase 2, operations static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11, op21, op31}, {op41}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op11 = model->addOperand(&type26); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type23); // Phase 2, operations static float op21_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op21, op21_init, sizeof(float) * 16); static float op31_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op31, op31_init, sizeof(float) * 4); static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11}, {op41}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op11 = model->addOperand(&type26); auto op21 = model->addOperand(&type2); auto op31 = model->addOperand(&type3); auto param10 = model->addOperand(&type4); auto param11 = model->addOperand(&type4); auto param12 = model->addOperand(&type4); auto param13 = model->addOperand(&type4); auto param14 = model->addOperand(&type4); auto param15 = model->addOperand(&type4); auto param16 = model->addOperand(&type4); auto param17 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param18 = model->addOperand(&type4); auto param19 = model->addOperand(&type4); auto op41 = model->addOperand(&type23); // Phase 2, operations static int32_t param10_init[] = {0}; model->setOperandValue(param10, param10_init, sizeof(int32_t) * 1); static int32_t param11_init[] = {0}; model->setOperandValue(param11, param11_init, sizeof(int32_t) * 1); static int32_t param12_init[] = {0}; model->setOperandValue(param12, param12_init, sizeof(int32_t) * 1); static int32_t param13_init[] = {0}; model->setOperandValue(param13, param13_init, sizeof(int32_t) * 1); static int32_t param14_init[] = {1}; model->setOperandValue(param14, param14_init, sizeof(int32_t) * 1); static int32_t param15_init[] = {1}; model->setOperandValue(param15, param15_init, sizeof(int32_t) * 1); static int32_t param16_init[] = {2}; model->setOperandValue(param16, param16_init, sizeof(int32_t) * 1); static int32_t param17_init[] = {0}; model->setOperandValue(param17, param17_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param18_init[] = {2}; model->setOperandValue(param18, param18_init, sizeof(int32_t) * 1); static int32_t param19_init[] = {2}; model->setOperandValue(param19, param19_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op11, op21, op31, param10, param11, param12, param13, param14, param15, param16, param17, layout, param18, param19}, {op41}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op11, op21, op31}, {op41}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type2); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_nhwc_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type2); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nhwc_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type10); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type11); // Phase 2, operations static _Float16 op22_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(_Float16) * 16); static _Float16 op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(_Float16) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_nhwc_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type16(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type13); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type16); // Phase 2, operations static uint8_t op22_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op22, op22_init, sizeof(uint8_t) * 16); static int32_t op32_init[] = {200, 400, 600, 800}; model->setOperandValue(op32, op32_init, sizeof(int32_t) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_nhwc_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type2); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type2); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type10); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type11); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type16(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type13); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type16); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type18); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_nchw_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type18); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nchw_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type20(Type::TENSOR_FLOAT16, {1, 4, 2, 2}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type19); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type20); // Phase 2, operations static _Float16 op22_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(_Float16) * 16); static _Float16 op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(_Float16) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_nchw_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type22(Type::TENSOR_QUANT8_ASYMM, {1, 4, 2, 2}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type21); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type22); // Phase 2, operations static uint8_t op22_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op22, op22_init, sizeof(uint8_t) * 16); static int32_t op32_init[] = {200, 400, 600, 800}; model->setOperandValue(op32, op32_init, sizeof(int32_t) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_nchw_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type18); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type18); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type20(Type::TENSOR_FLOAT16, {1, 4, 2, 2}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type19); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type20); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type22(Type::TENSOR_QUANT8_ASYMM, {1, 4, 2, 2}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type21); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type22); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type10); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type24); // Phase 2, operations static _Float16 op22_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(_Float16) * 16); static _Float16 op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(_Float16) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type13); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type25); // Phase 2, operations static uint8_t op22_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op22, op22_init, sizeof(uint8_t) * 16); static int32_t op32_init[] = {200, 400, 600, 800}; model->setOperandValue(op32, op32_init, sizeof(int32_t) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type1(Type::TENSOR_FLOAT32, {1, 3, 3, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type1); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type10(Type::TENSOR_FLOAT16, {1, 3, 3, 2}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type10); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type24); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type13(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 2}, 0.5f, 0); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type13); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type25); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static float op22_init[] = {0.25f, 0.0f, 0.2f, 0.0f, 0.25f, 0.0f, 0.0f, 0.3f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.1f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(float) * 16); static float op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(float) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type19); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type24); // Phase 2, operations static _Float16 op22_init[] = {0.25f, 0.0f, 0.20000000298023224f, 0.0f, 0.25f, 0.0f, 0.0f, 0.30000001192092896f, 0.25f, 0.0f, 0.0f, 0.0f, 0.25f, 0.10000000149011612f, 0.0f, 0.0f}; model->setOperandValue(op22, op22_init, sizeof(_Float16) * 16); static _Float16 op32_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op32, op32_init, sizeof(_Float16) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type21); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type25); // Phase 2, operations static uint8_t op22_init[] = {25, 0, 20, 0, 25, 0, 0, 30, 25, 0, 0, 0, 25, 10, 0, 0}; model->setOperandValue(op22, op22_init, sizeof(uint8_t) * 16); static int32_t op32_init[] = {200, 400, 600, 800}; model->setOperandValue(op32, op32_init, sizeof(int32_t) * 4); static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_relaxed_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type17(Type::TENSOR_FLOAT32, {1, 2, 3, 3}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type17); auto op22 = model->addOperand(&type2); auto op32 = model->addOperand(&type3); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type23); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_relaxed_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_float16_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type11(Type::TENSOR_FLOAT16, {1, 2, 2, 4}); OperandType type12(Type::TENSOR_FLOAT16, {4}); OperandType type19(Type::TENSOR_FLOAT16, {1, 2, 3, 3}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type19); auto op22 = model->addOperand(&type11); auto op32 = model->addOperand(&type12); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type24); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_float16_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_quant8_2(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type14(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 4}, 0.01f, 0); OperandType type15(Type::TENSOR_INT32, {4}, 0.005f, 0); OperandType type21(Type::TENSOR_QUANT8_ASYMM, {1, 2, 3, 3}, 0.5f, 0); OperandType type25(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.1f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op12 = model->addOperand(&type21); auto op22 = model->addOperand(&type14); auto op32 = model->addOperand(&type15); auto param20 = model->addOperand(&type4); auto param21 = model->addOperand(&type4); auto param22 = model->addOperand(&type4); auto param23 = model->addOperand(&type4); auto param24 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param25 = model->addOperand(&type4); auto param26 = model->addOperand(&type4); auto op42 = model->addOperand(&type25); // Phase 2, operations static int32_t param20_init[] = {2}; model->setOperandValue(param20, param20_init, sizeof(int32_t) * 1); static int32_t param21_init[] = {1}; model->setOperandValue(param21, param21_init, sizeof(int32_t) * 1); static int32_t param22_init[] = {1}; model->setOperandValue(param22, param22_init, sizeof(int32_t) * 1); static int32_t param23_init[] = {2}; model->setOperandValue(param23, param23_init, sizeof(int32_t) * 1); static int32_t param24_init[] = {0}; model->setOperandValue(param24, param24_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param25_init[] = {1}; model->setOperandValue(param25, param25_init, sizeof(int32_t) * 1); static int32_t param26_init[] = {1}; model->setOperandValue(param26, param26_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op12, op22, op32, param20, param21, param22, param23, param24, layout, param25, param26}, {op42}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op12, op22, op32}, {op42}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_quant8_2(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op13 = model->addOperand(&type5); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type2); // Phase 2, operations static float op23_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op23, op23_init, sizeof(float) * 16); static float op33_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op33, op33_init, sizeof(float) * 4); static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13}, {op43}); assert(model->isValid()); } inline bool is_ignored_nhwc_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op13 = model->addOperand(&type5); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type2); // Phase 2, operations static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13, op23, op33}, {op43}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op13 = model->addOperand(&type26); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type18); // Phase 2, operations static float op23_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op23, op23_init, sizeof(float) * 16); static float op33_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op33, op33_init, sizeof(float) * 4); static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13}, {op43}); assert(model->isValid()); } inline bool is_ignored_nchw_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type18(Type::TENSOR_FLOAT32, {1, 4, 2, 2}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op13 = model->addOperand(&type26); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type18); // Phase 2, operations static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13, op23, op33}, {op43}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op13 = model->addOperand(&type5); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type23); // Phase 2, operations static float op23_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op23, op23_init, sizeof(float) * 16); static float op33_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op33, op33_init, sizeof(float) * 4); static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13}, {op43}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); OperandType type5(Type::TENSOR_FLOAT32, {1, 4, 4, 2}); // Phase 1, operands auto op13 = model->addOperand(&type5); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type23); // Phase 2, operations static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13, op23, op33}, {op43}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op13 = model->addOperand(&type26); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type23); // Phase 2, operations static float op23_init[] = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f, 16.0f}; model->setOperandValue(op23, op23_init, sizeof(float) * 16); static float op33_init[] = {0.0f, 0.0f, 0.0f, 0.0f}; model->setOperandValue(op33, op33_init, sizeof(float) * 4); static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13}, {op43}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_4(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type2(Type::TENSOR_FLOAT32, {1, 2, 2, 4}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type26(Type::TENSOR_FLOAT32, {1, 2, 4, 4}); OperandType type3(Type::TENSOR_FLOAT32, {4}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op13 = model->addOperand(&type26); auto op23 = model->addOperand(&type2); auto op33 = model->addOperand(&type3); auto param27 = model->addOperand(&type4); auto param28 = model->addOperand(&type4); auto param29 = model->addOperand(&type4); auto param30 = model->addOperand(&type4); auto param31 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param32 = model->addOperand(&type4); auto param33 = model->addOperand(&type4); auto op43 = model->addOperand(&type23); // Phase 2, operations static int32_t param27_init[] = {2}; model->setOperandValue(param27, param27_init, sizeof(int32_t) * 1); static int32_t param28_init[] = {1}; model->setOperandValue(param28, param28_init, sizeof(int32_t) * 1); static int32_t param29_init[] = {1}; model->setOperandValue(param29, param29_init, sizeof(int32_t) * 1); static int32_t param30_init[] = {2}; model->setOperandValue(param30, param30_init, sizeof(int32_t) * 1); static int32_t param31_init[] = {0}; model->setOperandValue(param31, param31_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param32_init[] = {2}; model->setOperandValue(param32, param32_init, sizeof(int32_t) * 1); static int32_t param33_init[] = {2}; model->setOperandValue(param33, param33_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op13, op23, op33, param27, param28, param29, param30, param31, layout, param32, param33}, {op43}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op13, op23, op33}, {op43}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_4(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); OperandType type9(Type::TENSOR_FLOAT32, {1, 3, 3, 1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type9); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_nhwc_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); OperandType type9(Type::TENSOR_FLOAT32, {1, 3, 3, 1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type9); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nhwc_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type27(Type::TENSOR_QUANT8_ASYMM, {1, 6, 6, 1}, 0.5f, 0); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type30(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 1}, 0.125f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type27); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type30); // Phase 2, operations static uint8_t op24_init[] = {8, 16, 24, 32}; model->setOperandValue(op24, op24_init, sizeof(uint8_t) * 4); static int32_t op34_init[] = {0}; model->setOperandValue(op34, op34_init, sizeof(int32_t) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_nhwc_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type31(Type::TENSOR_FLOAT16, {1, 6, 6, 1}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type34(Type::TENSOR_FLOAT16, {1, 3, 3, 1}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type31); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type34); // Phase 2, operations static _Float16 op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(_Float16) * 4); static _Float16 op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(_Float16) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_nhwc_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); OperandType type9(Type::TENSOR_FLOAT32, {1, 3, 3, 1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type9); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); OperandType type9(Type::TENSOR_FLOAT32, {1, 3, 3, 1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type9); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type27(Type::TENSOR_QUANT8_ASYMM, {1, 6, 6, 1}, 0.5f, 0); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type30(Type::TENSOR_QUANT8_ASYMM, {1, 3, 3, 1}, 0.125f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type27); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type30); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nhwc_weight_as_input_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type31(Type::TENSOR_FLOAT16, {1, 6, 6, 1}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type34(Type::TENSOR_FLOAT16, {1, 3, 3, 1}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type31); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type34); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_nhwc_weight_as_input_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type36(Type::TENSOR_FLOAT32, {1, 1, 3, 3}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type36); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_nchw_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type36(Type::TENSOR_FLOAT32, {1, 1, 3, 3}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type36); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nchw_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type37(Type::TENSOR_QUANT8_ASYMM, {1, 1, 6, 6}, 0.5f, 0); OperandType type38(Type::TENSOR_QUANT8_ASYMM, {1, 1, 3, 3}, 0.125f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type37); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type38); // Phase 2, operations static uint8_t op24_init[] = {8, 16, 24, 32}; model->setOperandValue(op24, op24_init, sizeof(uint8_t) * 4); static int32_t op34_init[] = {0}; model->setOperandValue(op34, op34_init, sizeof(int32_t) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_nchw_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type39(Type::TENSOR_FLOAT16, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); OperandType type40(Type::TENSOR_FLOAT16, {1, 1, 3, 3}); // Phase 1, operands auto op14 = model->addOperand(&type39); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type40); // Phase 2, operations static _Float16 op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(_Float16) * 4); static _Float16 op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(_Float16) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_nchw_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type36(Type::TENSOR_FLOAT32, {1, 1, 3, 3}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type36); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type36(Type::TENSOR_FLOAT32, {1, 1, 3, 3}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type36); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type37(Type::TENSOR_QUANT8_ASYMM, {1, 1, 6, 6}, 0.5f, 0); OperandType type38(Type::TENSOR_QUANT8_ASYMM, {1, 1, 3, 3}, 0.125f, 0); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type37); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type38); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_nchw_weight_as_input_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type39(Type::TENSOR_FLOAT16, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); OperandType type40(Type::TENSOR_FLOAT16, {1, 1, 3, 3}); // Phase 1, operands auto op14 = model->addOperand(&type39); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type40); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_nchw_weight_as_input_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type27(Type::TENSOR_QUANT8_ASYMM, {1, 6, 6, 1}, 0.5f, 0); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type4(Type::INT32, {}); OperandType type41(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.125f, 0); // Phase 1, operands auto op14 = model->addOperand(&type27); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type41); // Phase 2, operations static uint8_t op24_init[] = {8, 16, 24, 32}; model->setOperandValue(op24, op24_init, sizeof(uint8_t) * 4); static int32_t op34_init[] = {0}; model->setOperandValue(op34, op34_init, sizeof(int32_t) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type31(Type::TENSOR_FLOAT16, {1, 6, 6, 1}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type31); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type24); // Phase 2, operations static _Float16 op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(_Float16) * 4); static _Float16 op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(_Float16) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type4(Type::INT32, {}); OperandType type6(Type::TENSOR_FLOAT32, {1, 6, 6, 1}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type6); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type27(Type::TENSOR_QUANT8_ASYMM, {1, 6, 6, 1}, 0.5f, 0); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type4(Type::INT32, {}); OperandType type41(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.125f, 0); // Phase 1, operands auto op14 = model->addOperand(&type27); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type41); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nhwc_weight_as_input_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type31(Type::TENSOR_FLOAT16, {1, 6, 6, 1}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type31); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type24); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {false}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nhwc_weight_as_input_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static float op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(float) * 4); static float op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(float) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type37(Type::TENSOR_QUANT8_ASYMM, {1, 1, 6, 6}, 0.5f, 0); OperandType type4(Type::INT32, {}); OperandType type41(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.125f, 0); // Phase 1, operands auto op14 = model->addOperand(&type37); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type41); // Phase 2, operations static uint8_t op24_init[] = {8, 16, 24, 32}; model->setOperandValue(op24, op24_init, sizeof(uint8_t) * 4); static int32_t op34_init[] = {0}; model->setOperandValue(op34, op34_init, sizeof(int32_t) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type39(Type::TENSOR_FLOAT16, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type39); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type24); // Phase 2, operations static _Float16 op24_init[] = {1.0f, 2.0f, 3.0f, 4.0f}; model->setOperandValue(op24, op24_init, sizeof(_Float16) * 4); static _Float16 op34_init[] = {0.0f}; model->setOperandValue(op34, op34_init, sizeof(_Float16) * 1); static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_5(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_5(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_relaxed_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type23(Type::TENSOR_FLOAT32, {0, 0, 0, 0}); OperandType type35(Type::TENSOR_FLOAT32, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); OperandType type7(Type::TENSOR_FLOAT32, {1, 2, 2, 1}); OperandType type8(Type::TENSOR_FLOAT32, {1}); // Phase 1, operands auto op14 = model->addOperand(&type35); auto op24 = model->addOperand(&type7); auto op34 = model->addOperand(&type8); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type23); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); // Phase 4: set relaxed execution model->relaxComputationFloat32toFloat16(true); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_relaxed_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_quant8_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type28(Type::TENSOR_QUANT8_ASYMM, {1, 2, 2, 1}, 0.125f, 0); OperandType type29(Type::TENSOR_INT32, {1}, 0.0625f, 0); OperandType type37(Type::TENSOR_QUANT8_ASYMM, {1, 1, 6, 6}, 0.5f, 0); OperandType type4(Type::INT32, {}); OperandType type41(Type::TENSOR_QUANT8_ASYMM, {0, 0, 0, 0}, 0.125f, 0); // Phase 1, operands auto op14 = model->addOperand(&type37); auto op24 = model->addOperand(&type28); auto op34 = model->addOperand(&type29); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type41); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_quant8_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); } void CreateModel_dynamic_output_shape_nchw_weight_as_input_float16_3(Model *model) { OperandType type0(Type::BOOL, {}); OperandType type24(Type::TENSOR_FLOAT16, {0, 0, 0, 0}); OperandType type32(Type::TENSOR_FLOAT16, {1, 2, 2, 1}); OperandType type33(Type::TENSOR_FLOAT16, {1}); OperandType type39(Type::TENSOR_FLOAT16, {1, 1, 6, 6}); OperandType type4(Type::INT32, {}); // Phase 1, operands auto op14 = model->addOperand(&type39); auto op24 = model->addOperand(&type32); auto op34 = model->addOperand(&type33); auto param34 = model->addOperand(&type4); auto param35 = model->addOperand(&type4); auto param36 = model->addOperand(&type4); auto param37 = model->addOperand(&type4); auto param38 = model->addOperand(&type4); auto layout = model->addOperand(&type0); auto param39 = model->addOperand(&type4); auto param40 = model->addOperand(&type4); auto op44 = model->addOperand(&type24); // Phase 2, operations static int32_t param34_init[] = {1}; model->setOperandValue(param34, param34_init, sizeof(int32_t) * 1); static int32_t param35_init[] = {2}; model->setOperandValue(param35, param35_init, sizeof(int32_t) * 1); static int32_t param36_init[] = {2}; model->setOperandValue(param36, param36_init, sizeof(int32_t) * 1); static int32_t param37_init[] = {1}; model->setOperandValue(param37, param37_init, sizeof(int32_t) * 1); static int32_t param38_init[] = {0}; model->setOperandValue(param38, param38_init, sizeof(int32_t) * 1); static bool8 layout_init[] = {true}; model->setOperandValue(layout, layout_init, sizeof(bool8) * 1); static int32_t param39_init[] = {3}; model->setOperandValue(param39, param39_init, sizeof(int32_t) * 1); static int32_t param40_init[] = {3}; model->setOperandValue(param40, param40_init, sizeof(int32_t) * 1); model->addOperation(ANEURALNETWORKS_DEPTHWISE_CONV_2D, {op14, op24, op34, param34, param35, param36, param37, param38, layout, param39, param40}, {op44}); // Phase 3, inputs and outputs model->identifyInputsAndOutputs( {op14, op24, op34}, {op44}); assert(model->isValid()); } inline bool is_ignored_dynamic_output_shape_nchw_weight_as_input_float16_3(int i) { static std::set ignore = {}; return ignore.find(i) != ignore.end(); }