Re: Valid gimple for MEM_REF

[Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>]

Hello,

> I noticed that gcc.dg/tree-ssa/loop-19.c was failing on both
> powerpc-linux-gnu and powerpc64-linux-gnu:
> FAIL: gcc.dg/tree-ssa/loop-19.c scan-tree-dump-times MEM.(base: &|symbol: 
> )a, 2
> FAIL: gcc.dg/tree-ssa/loop-19.c scan-tree-dump-times MEM.(base: &|symbol: 
> )c, 2
> 
> The reason why they are failing is because we produce:
> MEM[base: (double *) &c, index: ivtmp.34] = MEM[base: (double *) &a,
> index: ivtmp.34];
> Which does not match the regex as there is a cast there.
> Now the real question comes down to, is the following valid gimple
> that IV-OPTS produces:
> MEM[base: (double *) &a, index: ivtmp.34_12];
> 
> base is now a non gimple invariant but instead is a full expression.
> If we decide to do any other optimizations with MEM_REF, we might run
> into more of these issues?
> 
> So what are the constraints on MEM_REF's base argument, is it a simple
> expression (SSA_name or invariant) or can it be a complex expression?

only gimple_vals (name or invariant).  However, the expressions are
matched in final_cleanup dump (after out-of-ssa and ter), so this no
longer is the case.  I think just the regular expressions need to be
updated.

Zdenek