Korea Android - www.kandroid.org

댓글 소통 부탁합니다.

2. 세부

최근에 안드로이드 버젼에서 JIT(Just-In-Time) Compiler가 성능관련 핫이슈로 거론되었었습니다.
아래는 Google I/O 행사에서 발표되었던 Dalvik & JIT에 대한 구글 직원의 발표 동영상입니다.

* YouTube: Google I/O 2010 - A JIT Compiler for Android's Dalvik VM
http://www.youtube.com/watch?v=Ls0tM-c4Vfo&feature=player_embedded

* YouTube: Google I/O 2008 - Virtual Machines for Android Apps
http://www.youtube.com/watch?v=ptjedOZEXPM

* Dalvik JIT 컴파일러 방식: Trace-granuality JIT방식
(바이트 코드중 반복이 많은 코드블럭 영역만을 골라서 컴파일처리)
관련 논문 - Incremental Dynamic Code Generation with Trace rees
http://www.ics.uci.edu/~franz/Site/pubs-pdf/ICS-TR-06-16.pdf

프로요 버젼에서 JIT Compiler가 성능 개선의 방법으로 제안이 되었다가,
성능관련하여 엇는 잇점대비 잃는 부분들이 더 많은것이 주 이유라 할수 있는데
이에 대한 구글의 공식적인 글입니다.

* Performance Myths
( http://developer.android.com/guide/practices/design/performance.html )

Previous versions of this document made various misleading claims.
We address some of them here.

On devices without a JIT, it is true that invoking methods
via a variable with an exact type rather than an interface is
slightly more efficient.

(So, for example, it was cheaper to invoke methods on a HashMap map
than a Map map, even though in both cases the map was a HashMap.)
It was not the case that this was 2x slower; the actual difference was
more like 6% slower. Furthermore, the JIT makes the two effectively
indistinguishable.

On devices without a JIT, caching field accesses is about 20% faster
than repeatedly accesssing the field. With a JIT, field access costs
about the same as local access, so this isn't a worthwhile
optimization unless you feel it makes your code easier to read.
(This is true of final, static, and static final fields too.)

Dalvik JIT : What Makes This Possible?
( http://android-developers.blogspot.com/2010/05/dalvik-jit.html )

To be clear, the differences aren’t always dramatic, nor do they apply uniformly to all applications. Code that is written to run the CPU all-out can now do more in the same amount of time (running faster), and code that is written to be rate-limited can get its work done using less time and less of the CPU (using less battery). On the performance front in particular, we have seen realistic improvements of 2x to 5x for CPU-bound code, compared to the previous version of the Dalvik VM. This is equivalent to about 4x to 10x faster than a more traditional interpreter implementation

참고로, GIT컴파일러를 분기시키는 구현은 ./android-ics-4.0.3/dalvik/vm/Init.cpp 파일내의
ProcessOptions()함수에서 처리하고 있습니다.

                   . . . . . 윗부분 생략 . . . . .
915                 if (strcmp(argv[i] + 6, "portable") == 0) <- Desktop 정복을 위한 야망의 문장.ㅡ.ㅡ;
916                     gDvm.executionMode = kExecutionModeInterpPortable;
917                 else if (strcmp(argv[i] + 6, "fast") == 0)
918                     gDvm.executionMode = kExecutionModeInterpFast;
919 #ifdef WITH_JIT
920                 else if (strcmp(argv[i] + 6, "jit") == 0)
921                     gDvm.executionMode = kExecutionModeJit;
922 #endif
923                 else {
924                     dvmFprintf(stderr,
925                         "Warning: Unrecognized interpreter mode %sn",argv[i]);
926                     /* keep going */
927                 }
                    . . . . . 중 간 생 략 . . . . .
1073
1074     /*
1075      * Default execution mode.
1076      *
1077      * This should probably interact with the mterp code somehow, e.g. if
1078      * we know we're using the "desktop" build we should probably be
1079      * using "portable" rather than "fast".
1080      */
1081 #if defined(WITH_JIT)
1082     gDvm.executionMode = kExecutionModeJit;
1083 #else
1084     gDvm.executionMode = kExecutionModeInterpFast;
1085 #endif
1086

JIT 동작을 위해 필요한 바이너리들은
1) /system/bin/dalvikvm
2) /system/lib/libdvm.so
3) /system/lib/libnativehelper.so
로 구성이 되고 있습니다. 따라서, JIT를 위해 재작업이 되었다면 위의 바이너리를 직접 교체해도 된다는 의미입니다.

이상.