On 24-Jul-12 14:04, Walter Bright wrote:
On 7/24/2012 12:58 AM, Dmitry Olshansky wrote:
Now if I use final switches there is still:
A) jump table per switch, or maybe less but there is no guarantees
(= depend on another optimization that's not here)
B) it's an ugly and a roundabout way to do this
However I think that always requiring tail call optimization or
providing syntax
to enforce it would work:
void op_1()
{
...//some code for instruction 1
opcode = cast(function void ())code[pc++];
goto opcode(); //opcode is pointer to function op_xx
}
//can do without goto fn() iff tail call is GUARANTEED
I believe you can do this with:
switch (pc++)
and there are the same number of indirections.
And how is pc is supposed to be an opcode? It's a counter after all...
The trick is that it must be switch(code[pc++])...
It's just if code contains function pointers (or goto jump pointers)
then separate jump table table is not needed:
code = [ &op_1, &op_2, &op_1, ... ]; //generated somewhere
pc = 0;
code[pc]();
// op_1 increments ( or changes somehow) pc, decodes next op and jumps to it
So I still of opinion that enforced tail call is the cleanest way to
support this idiom.
--
Dmitry Olshansky
