A Method-Based Ahead-of-Time Compiler For Android Applications

Size: px

Start display at page:

Download "A Method-Based Ahead-of-Time Compiler For Android Applications"

Helen Lillian White
5 years ago
Views:

1 A Method-Based Ahead-of-Time Compiler For Android Applications Fatma Deli Computer Science & Software Engineering University of Washington Bothell November, 2012

2 2 Introduction This paper proposes a method-based ahead-of-time compiler (AOTC) which is called Icing. Purpose of AOTC is to speed up the execution of Android applications without modifying any components of Android framework.

3 3 Introduction Icing converts hot methods of an application program from DEX code to C code and uses the GCC compiler to translate the C code to the corresponding native code. With the Java Native Interface (JNI) library, the translated native code can be called by DVM.

4 4 Introduction Icing converts hot methods of an application program from DEX code to C code and uses the GCC compiler to translate the C code to the corresponding native code. With the Java Native Interface (JNI) library, the translated native code can be called by DVM.

5 5 Icing Icing is a mixed-mode AOTC. Icing framework's components are static profiling model, Icing AOTC, and Bridge library. The execution flow of Icing starts with identifying hot methods of a DEX bytecode. This is done with the help of static profiling model.

6 6 Icing In order to find hot methods, first, the execution frequency for each method is calculated. Then, the high frequency ones are selected as hot methods. Then, methods of Android application are divided into two groups: AOTC and JITC lists. DEX bytecodes of methods in the AOTC list are fed into the Icing AOTC as input.

7 7 Icing Icing modifies the original DEX bytecode by replacing the original hot method s bytecode with a native header after compilation. Package with the modified DEX code and the native code generated by AOTC are built into a new application (.apk). Icing AOTC translates DEX bytecode to C code and it uses the COINS compiler infrastructure. This basically first does a DEX code preprocessing.

8 8 Icing The preprocessing disassembles the DEX binary into human readable format using backsmali and dexdump. Smali/backsmali is an open source tool which serves as the assembler and disassembler for DEX bytecode. Dexdump is a DEX disassembler provided by Google.

9 9 Icing After the preprocessing, the DEX code is parsed into the High-level Intermediate Representation (HIR) of the COINS compiler infrastructure, and the corresponding C code is generated with the support of the COINS s HIR-to-C translator. Eventually, the C code is compiled into a shared library with the arm-gcc by the Android NDK.

10 10 Optimization Java can call functions written in other languages with the help of JNI but this brings a time and space overhead, which are categorized as call-out and callback operations. The call-out operations are used by the VM to invoke a native function through JNI. The overheads of the call-out operations include argument passing, native initialization, returning from native code, etc.

11 11 Optimization The call-back operations involve more overhead because they are used by the native code to access VM's resources. Call-back operation jumps to JNI environment variable which brings an overhead. Icing introduces three optimizations in order to solve the overhead : Ahead of time resolution, Caching, and Method cloning

12 12 Experimental Results and Conclusion After optimizations, performance if Icing is measured by utilizing four benchmarks, namely CaffeineMark 3.0, Checkers game, Linpack, and the BenchmarkPi. Results show that, with Icing, the execution time of an application is two to three times faster than that without JITC, and 25% to 110% faster than that with JITC. As a conclusion, it can be said that Icing, as a newly introduced compiler, has a much better performance than the current JITC in DVM.

H.-S. Oh, B.-J. Kim, H.-K. Choi, S.-M. Moon. School of Electrical Engineering and Computer Science Seoul National University, Korea

H.-S. Oh, B.-J. Kim, H.-K. Choi, S.-M. Moon School of Electrical Engineering and Computer Science Seoul National University, Korea Android apps are programmed using Java Android uses DVM instead of JVM