Hi;
> Hilary Cheng wrote:
> 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, jjos+decaf *is* slow, but it isn't due to this.
The compiler is actually emitting a jump table for the switch statement. For
your convenience, I have attached excerpts from my not-too-out-of-date version
of interp.cc (compiled with the optimizer OFF) -- The jump table proper begins
at label L1562. The first entry (for value zero) jumps to label L1535, which,
as included below, you can see is the call to the NOP instruction. Having said
this, there is still a lot of things we could do to speed the system up in its
entirety. I have run the host version with profiling enabled, and have learned
lots of interesting stuff (well, to me, anyways). I would be happy to discuss
this with anybody who was willing to concentrate upon performance.
runOpcode__5frameR11java_object:
.LFB208:
pushl %ebp
.LCFI1871:
movl %esp,%ebp
.LCFI1872:
subl $4,%esp
.LCFI1873:
pushl %ebx
.LCFI1874:
movl 8(%ebp),%ebx
movb $1,-1(%ebp)
cmpl $0,debug
je .L1359
movl 4(%ebx),%eax
pushl %eax
pushl %ebx
movl 32(%ebx),%eax
pushl %eax
pushl $.LC263
.LCFI1875:
call kprintf__FPCce
addl $16,%esp
.L1359:
movl 24(%ebx),%eax
movl 16(%eax),%edx
movl 4(%ebx),%eax
addl %eax,%edx
movzbl (%edx),%eax
cmpl $201,%eax
ja .L1561
movl .L1562(,%eax,4),%eax
jmp *%eax
.p2align 4,,7
.section .rodata
.align 4
.align 4
.L1562:
.long .L1535
.long .L1363
.long .L1456
.long .L1450
.long .L1451
.long .L1452
.long .L1453
.long .L1454
.long .L1455
.long .L1507
.long .L1508
.long .L1418
.long .L1419
.long .L1420
.long .L1392
.long .L1393
.long .L1380
[ The rest of the jump table deleted for brevity's sake. -jm ]
.L1535:
movl 12(%ebp),%eax
pushl %eax
pushl %ebx
call nop__5frameR11java_object
addl $8,%esp
movb %al,%al
movb %al,-1(%ebp)
jmp .L1360
.p2align 4,,7
[ Lots more stuff deleted -jm ]
-jm
--
==== John Morrison ==== MaK Technologies, Inc.
==== Chief Technology Officer ==== 185 Alewife Brook Pkwy, Cambridge, MA 02138
==== [EMAIL PROTECTED] ==== http://www.mak.com/
==== vox:617-876-8085 x115 ==== fax:617-876-9208
_______________________________________________
Kernel maillist - [EMAIL PROTECTED]
http://jos.org/mailman/listinfo/kernel
