The CogniVue APEX™ DSP Sub-System is the next-generation IP core technology incorporated into Maple-3 and Maple-3X product families. Virtually any existing computing application which can benefit from the massively parallel architecture of APEX can be run using this technology. The APEX technology was developed to deliver high-performance and maximum efficiency for a variety of compute-intensive applications including:
- Video and image processing supporting a wide variety of formats and sizes
- Video codecs including H.263, H.264, MPEG4, MPEG2, Real Video, DivX, XviD, WMV, VP6
- JPEG encode and decode
- Streaming video
- Pattern recognition/detection and video analytics
- 2D and 3D graphics
- GPS correlation
- DVB solution for digital TV
- Dual channel processing for picture in picture (PIP)
- Video telephony
The CogniVue Maple-3 and Maple-3X processors integrate a number of programmable resources including the APEX DSP Sub-System, a main RISC processor, and peripheral devices. The APEX foundation technology consists of a hardware bit acceleration block used to speed up bit manipulation required during variable length encoding and decoding operations, an ARM9 RISC processor, a massively parallel SIMD Array Processor Unit (APU) made up of a configurable number of computational units (CUs) each with dedicated memory, and a multi-channel DMA engine devised for efficient data movement into and out of the APU.
The APEX ADVANTAGE
The advantage of APEX over traditional DSP is power and speed. Traditional DSP increases the operating clock and adds parallelism by minimally increasing the number of parallel multiply-accumulate (MAC) units. This burns significantly more power as the clock speeds ramp higher and higher, but as importantly places an extreme load on the external memory to be able to keep up with the read and writes necessary to feed the MAC units. APEX solves these problems by locating memory in conjunction with the local CU. Data is fetched from external memory and streamed into the local CU memory which is located on-chip where all of the processing is performed locally before being stored back to external memory. Since there are many more CUs operating in parallel than the largest number of MAC units on DSPs, the operating frequency is significantly lower for more performance and the external memory load is greatly reduced as well. These combined features result in significant overall power savings.
The fully-programmable nature of the APEX DSP Sub-System is unique. Customers gain competitive advantages through:
- Fast turnaround when incorporating new features such as supporting new video codec algorithms
- Scalable solution allowing designers to adjusting to new requirements
- Support of new instructions for new functionality and for optimizing instruction sequences that are executed repeatedly and consume significant computational resources
- Development of a single platform that can be configured for multiple implementations to address different market segments
- Allowing bug fixes at all stages of the development cycle, from prototype to production.
These advantages are made possible through software changes as no hardware changes are required. In addition, applications are portable across product platforms, and scalable for use with future CogniVue video and image processors.
