Re: What is byte-code ?

Hi Anton,


Anton Ertl <anton@mips.complang.tuwien.ac.at> wrote:
> Nathan Moore <nathan.moore@sdc.cox.net> writes:
> > This will save you about 2 jumps per instructoin since the jump tables
> > that are usually spit out by compilers are:
> >
> > jump ip+(constant*case)
> > jump CODE_THAT_ACTUALLY_DOES_CASE_0
> > jump CODE_THAT_ACTUALLY_DOES_CASE_1
> > ...
>
> I have never seen that kind of code generated. Which compiler does it
> that way?


I've seen the Norcroft and gcc C compilers targetted at ARM generate
such code.


                ldrb r1, [r11, #0]
                cmp r1, #7
                addls pc, pc, r1, lsl #2 ;
                b case_8
                b case_0
                b case_1
                b case_2
                b case_3
                b case_4
                b case_5
                b case_6


        .case_7
                ldr r1, [r11, #&028]
                mov r0, r11


With a 32-bit word, pc is already two words ahead of the current
instruction due to pipe-lining. This is the idiomatic jump table in ARM
and assembly programmers use it too.


> Big, dense switch statements have been translated into a range check,
> an array access (for the target address), and an indirect jump to the
> target in the code I looked at (usually by gcc).


The assembler above only works when the relative jump destination can be
encoded along with the b instruction in the 32-bit word.


Cheers,


--
Ralph Corderoy. http://inputplus.co.uk/ralph/ http://troff.org/

Return to the comp.compilers page.
Search the comp.compilers archives again.