Cache sensitive code arrangement for virtual machine
Resource
Transactions on High-Performance Embedded Architectures and Compilers,3(1),22-41.
Journal
Transactions on High-Performance Embedded Architectures and Compilers III (Lecture Notes in Computer Science)
Pages
24-42
Date Issued
2008-12
Date
2008-12
Author(s)
Chen, C. C.
Lin, C. L.
Abstract
This paper proposes a systematic approach to optimize the code layout of a Java ME virtual machine for an embedded system with a cache-sensitive architecture. A practice example is to run JVM directly (execution-in-place) in NAND flash memory, for which cache miss penalty is too high to endure. The refined virtual machine generated cache misses 96% less than the original version. We developed a mathematical approach helping to predict the flow of the interpreter inside the virtual machine. This approach analyzed both the static control flow graph and the pattern of bytecode instruction streams, since we found the input sequence drives the program flow of the virtual machine interpreter. Then we proposed a rule to model the execution flows of Java instructions of real applications. Furthermore, we used a graph partition algorithm as a tool to deal with the mathematical model, and this finding helped the relocation process to move program blocks to proper memory pages. The refinement approach dramatically improved the locality of the virtual machine thus reduced cache miss rates. Our technique can help Java ME-enabled devices to run faster and extend longer battery life. The approach also brings potential for designers to integrate the XIP function into System-on-Chip thanks to lower demand for cache memory. © 2011 Springer-Verlag Berlin Heidelberg.
Type
journal article
File(s)![Thumbnail Image]()
Loading...
Name
903.pdf
Size
2.03 MB
Format
Adobe PDF
Checksum
(MD5):dbfe74162efc1c002f9f681e32eda018