Re: PR rtl-optimization/15248 -- semi-latent reload bug

[Jeffrey A Law <law at redhat dot com>]

On Wed, 2005-04-20 at 19:45 -0600, Roger Sayle wrote:

> Perhaps you can clear up a confusion of mine that I've had since
> we first ran into this spilling to read-only memory issue, a while
> back.
> 
> Isn't it a good thing to equivalence a register to a read-only memory
> address? 
If it weren't for the problems we're having in our *implementation*,
yes, absolutely.




>  My (perhaps misguided) understanding is that if we equivalence
> a pseudo to such a memory address, then that pseudo always holds the
> equivalenced memory value.
Precisely.  




>   As the value in read-only memory can never
> change, this makes spilling that pseudo relatively cheap.  i.e. the
> register never has to be written to memory, only re-read.
Right.


>   For example,
> if a pseudo only ever holds the value of a constant pool entry, when
> register pressure gets to high, the spill can be rematerialized without
> previously writing it to memory.
Right.

> 
> Perhaps I'm getting the issues confused, and this bug/fix involves a
> different problem.  However, in reload spilling constants and read-only
> memory should be prioritized over other classes of spill.
What you're missing is that when the pseudo doesn't get a hard register,
we must eliminate all assignments to the pseudo as any assignment to the
pseudo will generate a write to a read-only memory location.

What makes that particular problem non-trivial to solve is PARALLELs and
the bookkeeping problems in the reg_equiv_XXX arrays if we have a pseudo
equivalenced to a read-only memory location which was assigned a hard
reg which we later spilled.

Jeff