Symbiotic radio, which exploits the benefits of passive communications and cognitive radio via backscattering or ambient reflecting, is a promising paradigm to support a large amount of Internet of Things devices with high spectrum efficiency and energy efficiency. For edge intelligence, data collection from heterogeneous devices including wireless sensing devices and wired sensing devices brings challenge for efficient model training. Intelligent reflecting surface (IRS) based symbiotic radio, which exploits IRS in backscatter systems to passively modulate information in the radio-frequency domain and strengthen both the primary links and the backscatter links, provides a promising solution to address this issue. In this paper, we investigate the joint sensing, communication and computation for edge intelligence oriented symbiotic radio with IRS. Specifically, IRS not only enhances the primary communication to deliver the sensing data for wireless sensing devices, but also delivers the data from wired sensing device to the edge server by modulating the data on the primary communication signal. We formulate a problem to jointly optimize the sensing strategies (i.e., the sensing durations and the sensing rates), the communication strategies (i.e., the IRS reflecting matrix and the data uploading durations) and the computing strategies (i.e., the computing capacities of the wireless sensing devices and the edge server), with the objective of maximizing the energy efficiency of the whole system. An efficient algorithm is proposed to solve the formulated optimization problem, with sufficient numerical results provided to demonstrate the performance advantages.