Method run

src/cpu/operators/CpuAddMulAdd.cpp:107–146

}

void CpuAddMulAdd::run(ITensorPack &tensors)
{
    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuAddMulAdd::run");
    const DataType data_type = tensors.get_const_tensor(TensorType::ACL_SRC_0)->info()->data_type();

    if (is_data_type_quantized(data_type))
    {
        const ITensor *bn_mul = tensors.get_const_tensor(TensorType::ACL_SRC_2);
        const ITensor *bn_add = tensors.get_const_tensor(TensorType::ACL_SRC_3);

        CpuAuxTensorHandler dequantized_bn_mul_handler(offset_int_vec(DequantizedBnMul), _dequantized_bn_mul, tensors,
                                                       true);
        CpuAuxTensorHandler dequantized_bn_add_handler(offset_int_vec(DequantizedBnAdd), _dequantized_bn_add, tensors,
                                                       true);

        ITensorPack dequantize_mul_pack = {{TensorType::ACL_SRC_0, bn_mul},
                                           {TensorType::ACL_DST_0, dequantized_bn_mul_handler.get()}};

        ITensorPack dequantize_add_pack = {{TensorType::ACL_SRC_0, bn_add},
                                           {TensorType::ACL_DST_0, dequantized_bn_add_handler.get()}};

        _dequantize_bn_mul.run(dequantize_mul_pack);
        _dequantize_bn_add.run(dequantize_add_pack);

        ITensorPack add_mul_add_pack = {
            {TensorType::ACL_SRC_0, tensors.get_const_tensor(TensorType::ACL_SRC_0)},
            {TensorType::ACL_SRC_1, tensors.get_const_tensor(TensorType::ACL_SRC_1)},
            {TensorType::ACL_SRC_2, dequantized_bn_mul_handler.get()},
            {TensorType::ACL_SRC_3, dequantized_bn_add_handler.get()},
            {TensorType::ACL_DST_0, tensors.get_tensor(TensorType::ACL_DST_0)},
            {TensorType::ACL_DST_1, tensors.get_tensor(TensorType::ACL_DST_1)},
        };

        NEScheduler::get().schedule_op(_kernel.get(), Window::DimY, _kernel->window(), add_mul_add_pack);
    }
    else
    {
        NEScheduler::get().schedule_op(_kernel.get(), Window::DimY, _kernel->window(), tensors);
    }
}

experimental::MemoryRequirements CpuAddMulAdd::workspace() const
{