The table below gives individual bytecode instruction performance increases measured for PersonalJava. Results are for EBCI Interpreters on PowerPC, MIPS and StrongARM. All have final averages near or better than three times faster execution over the best optimized 'C' compiled interpreters we could find. |
aconst_null iconst_m1 thru iconst_5 lconst_0, lconst_1 bipush sipush ldc iload, aload iload_0 thru iload_3 aload_0 thru aload_3 iaload aaload istore astore istore_0 thru istore_3 astore_0 thru astore_3 pop, l2i dup iadd,isub,iand,ior,ixor ineg imul iinc ifeq thru ifle if_icmpeq thru if_acmpne goto
Bytecode Speedup Ratio Over Optimized
'C'
C Compilers used were, for PowerPC: Diab/GNU/Metrowerks CodeWarrior, for MIPS: Diab, for StrongARM: Microsoft eMbedded Visual C/C++. All compilers had full optimization options enabled. |