[PATCH] Fix PR52298

[Richard Guenther]

On Thu, 23 Feb 2012, Ulrich Weigand wrote:

> Richard Guenther wrote:
> 
> > 	PR tree-optimization/52298
> > 	* tree-vect-stmts.c (vectorizable_store): Properly use
> > 	STMT_VINFO_DR_STEP instead of DR_STEP when vectorizing
> > 	outer loops.
> > 	(vectorizable_load): Likewise.
> > 	* tree-vect-data-refs.c (vect_analyze_data_ref_access):
> > 	Access DR_STEP after ensuring it is not NULL.
> 
> This causes a bunch of regressions on SPU:
> 
> FAIL: gcc.dg/vect/vect-outer-fir-big-array.c (internal compiler error)
> FAIL: gcc.dg/vect/vect-outer-fir-big-array.c (test for excess errors)
> WARNING: gcc.dg/vect/vect-outer-fir-big-array.c compilation failed to produce executable
> FAIL: gcc.dg/vect/vect-outer-fir-big-array.c scan-tree-dump-times vect "OUTER LOOP VECTORIZED" 2
> FAIL: gcc.dg/vect/vect-outer-fir-lb-big-array.c (internal compiler error)
> FAIL: gcc.dg/vect/vect-outer-fir-lb-big-array.c (test for excess errors)
> WARNING: gcc.dg/vect/vect-outer-fir-lb-big-array.c compilation failed to produce executable
> FAIL: gcc.dg/vect/vect-outer-fir-lb-big-array.c scan-tree-dump-times vect "OUTER LOOP VECTORIZED" 2
> FAIL: gcc.dg/vect/vect-outer-fir-lb.c (internal compiler error)
> FAIL: gcc.dg/vect/vect-outer-fir-lb.c (test for excess errors)
> WARNING: gcc.dg/vect/vect-outer-fir-lb.c compilation failed to produce executable
> FAIL: gcc.dg/vect/vect-outer-fir-lb.c scan-tree-dump-times vect "OUTER LOOP VECTORIZED" 2
> FAIL: gcc.dg/vect/vect-outer-fir.c (internal compiler error)
> FAIL: gcc.dg/vect/vect-outer-fir.c (test for excess errors)
> WARNING: gcc.dg/vect/vect-outer-fir.c compilation failed to produce executable
> FAIL: gcc.dg/vect/vect-outer-fir.c scan-tree-dump-times vect "OUTER LOOP VECTORIZED" 2
> 
> all due to ICEs of the same type:
> 
>  internal compiler error: in vectorizable_load, at tree-vect-stmts.c:4665
> 
> The assert in question looks like:
> 
>   if (nested_in_vect_loop
>       && (TREE_INT_CST_LOW (STMT_VINFO_DR_STEP (stmt_info))
>           % GET_MODE_SIZE (TYPE_MODE (vectype)) != 0))
>     { 
>       gcc_assert (alignment_support_scheme != dr_explicit_realign_optimized);
>       compute_in_loop = true;
>     }
> 
> where your patch changed DR_STEP to STMT_VINFO_DR_STEP (reverting just this
> one change makes the ICEs go away).
> 
> However, at the place where the decision to use the dr_explicit_realign_optimized 
> strategy is made (tree-vect-data-refs.c:vect_supportable_dr_alignment), we still
> have:
> 
>           if ((nested_in_vect_loop
>                && (TREE_INT_CST_LOW (DR_STEP (dr))
>                    != GET_MODE_SIZE (TYPE_MODE (vectype))))
>               || !loop_vinfo)
>             return dr_explicit_realign;
>           else
>             return dr_explicit_realign_optimized;
> 
> Should this now also use STMT_VINFO_DR_STEP?

Yes, I think so.

Richard.

> Bye,
> Ulrich
> 
> 

-- 
Richard Guenther <rguenther@suse.de>
SUSE / SUSE Labs
SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer