嵌入式ARM开发基础bak.pptVIP

Hardware/software tradeoff balances speed and power consumption. Full Java in silicon is very expensive in terms of power and area - so ARM has implemented only the most important bytecodes in hardware. 140 bytecodes (out of 210) executed in hardware In Jazelle state, the processor doesn’t perform 8-bit fetches from memory. Instead it does aligned 32-bit fetches (4-byte prefetching) which is more efficient. Note we don’t mention the PC in Jazelle state because the ‘Jazelle PC’ is actually stored in r14 - this is technical detail that is not relevant as it is completely hidden by the Jazelle support code. Key parameter for battery-powered embedded applications is energy per caffeinemark. A Jazelle-enabled core typically uses 87% less energy than a non-optimized JVM. Jazelle is better than a JIT (Just-In-Time compiler) because it does not have a compute-intensive compilation phase, and a JIT is not appropriate for embedded applications because it can require at least 100Kbyte of code and much RAM to hold the recompiled classes (can expand up to 8x). Caffeinemark = benchmark which measures the speed of Java execution and simple 2-D graphics. Pre-3.0 versions of this benchmark lost some credibility due to huge scores for JITs which could optimize certain tests. Results given are for Caffeinemark 3.0 which doesn’t have these problems. Some customers might have seen 10x MIDP-mark improvement in our marketing figures - this is another benchmark MIDP = Mobile Information Device Profile ARM JTEK support code is an essential part of the target s/w to enable use of the Jazelle extensions. This is separately licensed by ARM, and can only be provided to Sun licensees. It will typically be licensed to O.S. vendors that provide Java support (e.g. PDA O.S vendors, cellphone O.S. manufacturers) 端格式决定寄存器在存储器中存放的关系： ARM 寄存器是16位宽度； ARM 地址总线也是16位 ARM 处理器是小端格式： 但可以配置成大端格式存储的系统。 小端格式存储的系统 低字节在低地址 大端格式存储的系统 高字节在低地址 ARM 支持三种模式的端格式： LE Little-Endian BE-32 Word Invariant B