Imagination Developer Connection
October 15, 2013, Imagination Developer Connection, San Francisco—Imagination Technologies held one in a series of conferences to address the information needs for their developers. Topics included hardware architectures, explain their ray tracing technologies, development tools, and a market analysis to justify the efforts from developers.
Imagination has a number of processors, graphics engines, and software available a IP. The existing volumes of devices with their IOP is growing and moving into new platforms including home, storage, automotive, and the Internet of Things. The various processors all have extensive development support including SDK, development kits, and a community of experienced developers.
The graphics hardware is evolving from render engines to full ray tracing. The graphics hardware architectures are helping to change the “brute force” image development to a nuanced tiling and on-chip buffer to render the rasterized tiles to allow all compute to take place in the buffers.
Another tool is deferred render. Fragments are done in two stages, where hidden surface removal is performed to eliminate non-visible surfaces, and then the remaining surfaces are rendered. This process reduces over draw and power consumption for objects that don’t even show on the display. All these changes are native to OpenGL ES 2.0
The tiling-based coprocessor in the PowerVR processors is the key to the better performance and lower power consumption. The architecture allows writes to memory, which are referred to a pixel data master, and then to the pixel co-processor in overlapping pipelines. The latest Series 5 XT versions have scalable multicore processors and extra glue logic to enable a 4-wide SIMD flow. The improvements promote greater geometry and pixel throughput plus MSAA speedup and texture compression. The combination of features gives nearly linear performance improvements on all functions with more cores.
The Series 6 architecture changed the compute engine into sharing clusters and have a tessellation engine. The scalar ALU does not require hand optimization of the instructions for high throughput. Some of the Direct X 10.0 functions are now native, and lossless geometry and image compression is included. The multi-threaded modes hide latency and delays in the processing streams.
Three types of compression are integrated: RTC for lossy compression, and lossless geometry, and image compression. Increasing the compression reduces the bandwidth requirements for the whole system. In addition, OpenGL ES 3.0 contributes to the efficiency by introducing multiple render targets which eliminates multiple passes and reduces API calls. OpenCL shares many of the same features with Open GL and helps both work well in mobile functions. The latest versions of the open graphics languages add anti-aliasing and texture anisotropic filtering. The future will see hybrid ray tracing for increased shadow detail and pixilated shadow elimination.
To incorporate all of the latest features, the SDK and other utilities for the Power VR family are available to assist development and creation and provide other support in IDEs for any platform. Graphics can be exported to the PowerVR Object Data (pod) format and other tools can interoperate with the full IDE or any other graphics tools. Some of the utilities include profiling compilers for the shaders, debug, trace, log, and hardware profilers. Performance analyzers help to identify processing bottlenecks and a performance analysis library provides additional tools.
