[JOS-Kernel] Some Idea About Speed Up JJOS

[Hilary Cheng]

Hi All,       I got a idea to speed up JJOS a little bit.       For the right now, we are using switch statement to interrupt the Java Bytecodes. i.e.   swtich (bytecodes) { case 0: do_opcode_0(); break; case 1: do_opcode_1(); break; ............ ............ ............ ............ }   It is slow. every opcodes needs one to hundred comparsion so that they can perform the operation.   well, yup, we can store all the bytecode operation function into a array. It can run a little faster. i.e.   function_list = { do_opcode_0, do_opcode_1, ...........};   (*function_list[bytecodes * sizeof(function_list[]])();   But It needs several mulitplation(*) and addition(+). and then we make a call. It is not fast enough.   maybe we can do it in another way.   For the right now, Java bytecode in JJOS are a serial of i386 "Call" instruction (in general). i.e.   push variable_a Call Do_OpCode_0 mov ax, [bp+12] mov variable_a , ax push variable_b Call Do_OpCode_1 mov ax, [bp+10] mov variable_b , ax push variable_c Call Do_OpCode_1 .............   sths like that. When we load the class from Disk/Memory into JVM, we convert each bytecodes into a i386 "Call" Instruction and store into Memory. When a Java method call, we simple jump to the start of code. Will It be faster ? I think it will. Since We Remove All the Comparsion / Multiplation(*) / Addition (+) for each bytecodes. For Each Java Method Call , we just simply jump to instruction set that we are already compiled. it will be faster. But we need to precompile each Class when Class is uploaded into memory. It is the time we need to pay.   Any Opinion ?   Regards,   Hilary