Re: Release of RosAsm V.2.025a

Randy wrote:
> Elsewhere in this thread, Rene has mentioned that he uses a "start
> long, optimize short" algorithm for branch displacement optimization.
> This scheme has the advantage that if you ever prematurely stop during
> the optimization process, the program will still run correctly.
> However, this algorithm often produces results that are not as optimal
> as they could be.

<snipped>

> Back to RosAsm's algorithms, I see a couple of problems that exist:
>
> 1) RosAsm only optimizes conditional jumps. It does not optimize JMP
> instructions or any other variable-length instructions whose size can
> change based on various factors.
>
> 2) There is a bug in the RosAsm algorithm -- it doesn't always
> optimizes even when it has the chance (e.g., the example given
> earlier).
>
> 3) RosAsm only applies the branch displacement optimization algorithm
> to condition jumps whose target label is a local symbol. If you use a
> regular target symbol, RosAsm always encodes this as a large
> displacement instruction. This makes no sense at all. If you can
> optimize one label type, why not both?  Just a bad design decision on
> Rene's part.

Who was this written for/to?

greetz
Jakob Wrigley