Re: Dynamic Typing Efficiency

clearm@comcast.net wrote:
> I am writing a small, python-like scripting language that uses dynamic
> typing. I am also writing a stack-based virtual machine to go with
> it.
>
> I have read many papers and posts about improving the efficiency of
> VMs - for example by using threaded dispatch instead of switch
> statements.
>
> The problem I have is that dynamic typing seems to be extremely
> inefficient when you have a large number of types. For example, if I
> have integers, doubles, strings, booleans, and various compound types,
> then an ADD instruction would have to look like this:
>
> [snip]
>
> switch INSTRUCTION:
> {
>
> case ADD:
>
> if (operand1->type == INTEGER && operand2->type == INTEGER)
> {
> value.i = operand1->val.i + operand2->val.i;
> type = INTEGER;
> }
> else if (operand1->type == INTEGER && operand2->type == DOUBLE)
> {
> value.d = (double)operand1->val.i + operand2->val.d;
> type = DOUBLE;
> }
> else if (operand1->type == DOUBLE && operand2->type == INTEGER)
> {
> value.d = operand1->val.d + (double)operand2->val.i;
> type = DOUBLE;
> }
> else if (operand1->type == DOUBLE && operand2->type == DOUBLE)
> {
> value.d = operand1->val.d + operand2->val.i;
> type = DOUBLE;
> }
> else if... blah blah blah
>
> break;
>
> case SUB: ...
>
> ...
> }
>
> [snip]


I'm thinking that you could do something sneaky with C++ template
meta-programming. You could construct a template


template<int type1, int type2>
struct plus
{
          static variant eval(variant, variant) ...
};


and statically generate the appropriate function. Then you use
meta-programming again to generate an array of these things, and put
them into a large jump-table.


Disclaimer: I haven't tried this....


Calum




--
Fast object persistence in C++ http://lightwave2.com/persist
A fast malloc/new replacement http://lightwave2.com
Home http://visula.org/calum

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