de Brebisson, Cyrille (Calculator Division) writes:
> Hello,
>
> I hope that this is the best location to ask this question, if not, please accept my apologize and redirect me where needed.
>
> I am trying to write a fast byte code interpreter, but the compiler optimizer just 'does not get it' and generates bad code (it does not realize that they are jumps everywhere and optimizes out the code out)...
>
> Here is a simplified version of the code:
>
> static int rom[]=
> { 0, 1, 2, 3, 4, 5, 6, 7, 8,
> 9, 10, 11, 12, 13, 14, }; // the 'program'
>
>
> void execute()
>
> {
>
> const void * const jumps[] =
> { &&ins000, &&ins001, &&ins002, &&ins003,
> &&ins004, &&ins005, &&ins006, &&ins007 }; // table of jumps
>
> register int carry asm ("r0");
> register int instr asm("r1"); // currently executed instruction
> register int *pc asm ("r4"); // program counter, points on next instr.
> register const void * const * jm asm ("r5") = jumps; //pointer jump table
>
> int a=0, b=0; // virtual machine registers
>
> // this macro does a fast carry=0; goto *jumps[*pc++];
> #define next asm ("ldrh %2, [%0], #2\n\t" \
> "mov %1, #0\n\t" \
> "ldr pc, [%4, %2, asl #2]" :
> "=r" (pc),
> "=r" (carry),
> "=r" (instr):
> "0" (pc),
> "r" (jm))
>
> // this macro does a fast goto *jumps[*pc++];
> #define nextnocarry asm ("ldrh %1, [%0], #2\n\t"\
> "ldr pc, [%3, %1, asl #2]" :
> "=r" (pc),
> "=r" (instr) :
> "0" (pc),
> "r" (jm))
This is the crucial mistake: you can't jump out of an inline asm.
Andrew.
