Research on ARM Architecture

As screens and touch screens become larger and larger, pixels, frames, and resolutions become higher and higher in the consumer electronics market, and various wireless communications are expanding one after another, the processing power of processors is increasingly being tested by these multimedia applications. In this context, ARM chips have launched their own solution - NEON.

NEON technology is a media and signal processing technology released by ARM in 2004. It can accelerate various types of multimedia applications and is suitable for mobile phones and consumer entertainment electronics. It can flexibly implement various video encoding/decoding, 3D imaging, voice processing, audio decoding, image processing and baseband functions. This technology has been applied to the Cortex-A series of ARM processors and has received extensive support from ARM and third-party tool providers. It is suitable for high-end multimedia devices.

NEON technology uses 64/128-bit single instruction multiple data (SIMD) instruction set, which can execute up to 16 operations at the same time, and is used to accelerate the next generation of media and signal processing applications. Its programming interface is simple, the performance is excellent, and the power consumption is very small. When executing MP3 audio decoder under NEON technology, the CPU frequency can be lower than 10MHZ; when running GSM AMR voice digital codec, the CPU frequency is only 13MHZ, and other professional


Figure 2.5 Schematic diagram of 16-bit vector addition in NEON

The figure above is a logical diagram of adding the D1 vector to the D2 vector and then storing the value into the D0 vector. In the instruction, VADD.I16 indicates that each operand of the operation is 16 bits.

       In programming, we often encounter data bit conversion. Sometimes you need to downgrade a long data type to an int, and sometimes you need to upgrade an int type to a long type. The NEON system provides instruction set support for data conversion. Here is an example. Instruction: VSHR.I16.I32 D0, Q1, #5

Figure 2.6 Schematic diagram of demotion in NEON

This example shifts each element of a 128-bit register of four 32-bit elements right by 5 bits and then truncates 16 bits to store in 64-bit D0. The I16.I32 in the instruction specifies this data demotion.

       NEON also supports a cache-friendly structure load/store function, which can load a structure array at a time, so as to better utilize the cache and make the structure logic of the code clear. Let's take an example to illustrate. Instruction VLD3.16 {D0, D1, D2}, {R0}

Figure 2.6 NEON instruction set support for structures

In the figure, a structure array in memory is loaded into three register vectors respectively. The same index bit between vectors constitutes a complete structure (the figure shows three pixels of the screen).

2.2      Comparison between ARM architecture and Atom

The Atom processor is a product launched by Intel in 2008 for netbooks, entry-level PCs,

Figure 2.8 Atom and Cortex Benchmark performance test comparison

         Since Atom was initially invested in the Netbook market, the benefits brought by previous compatibility were relatively advantageous, but since Steve Jobs stood on the stage of the PC market with an iPad in his hand, the Netbook era came to an end. At this time, ARM and Apple took the lead.

Notes:

http://www.hyqb.sh.cn/publish/portal2/tab270/info1268.htm  Technology to improve the performance of Java applications in embedded systems - Jazelle DBX.

2.          http://www.arm.com/products/processors/technologies/jazelle.php  —— ARM homepage.

3.          Detailed explanation of ARM instruction set - from WWW.

4.          http://www.hudong.com/wiki/arm#5 ——Hudong  Encyclopedia.

5.          http://www.cnblogs.com/leaven/archive/2010/10/28/1863744.htm l  —— Introduction to Cortex series ARM core and architecture

6.          http://en.wikipedia.org/wiki/ARM_architecture#Advanced_SIMD_.28NEON.29  —— Wikipedia.

7.          NEON Support in the ARM - White Paper - Sep. 2008.

8.          http://www.intel.com/technology/atom/ ——Intel Home Page