scriptcomp: Optimize typical for loop construct instructions

[mtijanic]

The common loop pattern of for (i = 0; i < 1000; ++i) gets compiled into

     8  RUNSTACK_ADD, TYPE_STRING               
    10  CONSTANT, TYPE_INTEGER, 0               
    16  RUNSTACK_COPY, TYPE_VOID, -4, 4         
    24  CONSTANT, TYPE_INTEGER, 1000            
    30  LT, TYPETYPE_INTEGER_INTEGER            
    32  JZ, 32                                  
    38  INCREMENT, TYPE_INTEGER, -4             
    44  RUNSTACK_COPY, TYPE_VOID, -4, 4         
    52  MODIFY_STACK_POINTER, -4                
    58  JMP, -42                                

The inc/dec operations can reach all across the stack and don't need to be copied over to top of stack. However, since ++i returns a value we have to emit the RUNSTACK_COPY. But, when we don't use that value, we just immediately clear it with MODIFY_STACK_POINTER. Just don't do either.

Testing

Added an extra test case with ++i and runtime check on iteration count.
Verified disassembly of

void main() {
    int i;
    for (i = 0; i < 1000; ++i) { }
}

is sane.

Changelog

Performance Improvements

• Optimized the common for loop construct for (i = 0; i < N; ++i)

Licence

• I am licencing my change under the project's MIT licence, including all changes to GPL-3.0 licenced parts of the codebase.