| Summary: | 碩士 === 國立臺灣科技大學 === 電子工程系 === 100 === Currently, multimedia applications have become the main part of every computing platform. Coupled with the growing diversity and complexity of multimedia informational content, the requirements for the hardware specifications are increasing as well. The presence of the difference between an embedded system and the common PC environment with its abundant hardware resources is evident to the extent that the embedded systems are regarded as high-performance, yet low cost and low power consumption solutions, etc. As the result of that, the main question that arises is how to enhance the performance of multimedia data processing while designing an embedded system with the limited availability of resources.
Since the requirements for making a practical use of embedded systems are growing, the ARM Corporation provides the inclusion of enhanced SIMD architecture in the ARMv7 instruction set. This architecture is also known as NEON multimedia processing engine. The NEON technology is created to be extremely competitive in contrast to the previous generations of ARM products. In general, audio and video data are appeared mostly to be based on specific fixed data types, for example: data of an image pixel are mostly based on 8bit color components, while the audio is placed on a 16bit level. In respect to a 32bit ARM CPU, the processing of audio and video data is relatively inefficient, and, therefore, for a support of the data processing, the NEON multimedia engine is applied. It is worth mentioning that the NEON technology supports the 128bit data vector in order to provide a good sustainability during a parallel computing and at the same time implements the features of the SIMD technology.
This thesis provides an analysis and discussion about the overall points of view regarding the process of how the NEON instruction set can be used in order to improve a system performance and optimization, also how to implement a NEON technology application platform in designing H.264/AVC software decoder that uses an application processor TI OMAP3530 based on Cortex-A8 core. In our experiment, we use ARM-Linux as an embedded operating system; we compare RGB-to-Gray color space conversion performance and H.264/AVC video playback with the NEON technology.
According to the experimental results, the processor TI OMAP3530 with the NEON instruction providing the resolution of 640x480 and the color depth of 24bits of RGB-to-Gray conversion, is able to reduce approximately 23.5% of processing time cost, and for the other 5 different resolutions, H.264 video playback can also force a considerable reduction up to 11%~13% of decoding time cost in average. In the system’s overall power consumption measurement, enabling NEON technology merely results in the 3.87% increase of maximum power consumption.
|
|---|
