Re: Autoincrement examples

[Joern Rennecke <amylaar at cygnus dot co dot uk>]

>  > - By processing each register separately, you miss register-register
>  >   copies and three-address adds.  Optimizing code with such constructs
>  >   was actually the main objective of my patch.  It actually reduces
>  >   register pressure.
> 
> I have rarely found these cases to be important for autoinc address
> generation.  However, reduction of register pressure is a good thing.

It gets pretty important with my loop patches applied.

>  > - Your code can generate PRE_MODIFY / POST_MODIFY, something which mine
>  >   can't.
> 
> Yes, this is important for DSP architectures; especially things like
> matrix multiplication where you want a large stride.

My patch can also modified to accomodate this, but for an efficient
implementation, this wants a balanced tree instead of a hash table.
Just like rewriting (mem (plus (reg ..) (const_int ..))) .
I have some AVL code, but I haven't assigned the deletion to the FSF and
I really need the code somewhere else, too.  So maybe it'll be best to go
with splay trees.  I understand we already have code for that in the source
code tree.

> With the information that I collect, these transformations would be
> straightforward.  More importantly, I would like to reorder some of
> the memory references to improve autoinc generation.

Which gets us into aliasing problems... well, I suppose it's easier these
days, now that the alias code is centralized.

> Does your code maintain the LOG_LINKS?  If so I would like to try

No, it doesn't.

> running the instruction combiner after the regmove pass.

> Alternatively, could your code be easily modified to remove redundant
> load instructions once an autoinc address was generated?

Could you sketch an rtl example of the transformation you have in mind?