[patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Tue, Mar 16, 2010 at 11:26:26PM +0100, Dominique Dhumieres wrote:
> With the following patch on top of r157496
> 
> diff -up /opt/gcc/_clean/gcc/passes.c /opt/gcc/p_work/gcc/passes.c
> --- /opt/gcc/_clean/gcc/passes.c	2010-02-22 15:28:45.000000000 +0100
> +++ /opt/gcc/p_work/gcc/passes.c	2010-03-16 21:28:43.000000000 +0100
> @@ -885,6 +885,7 @@ init_optimization_passes (void)
>  	{
>  	  struct opt_pass **p = &pass_tree_loop.pass.sub;
>  	  NEXT_PASS (pass_tree_loop_init);
> +	  /* NEXT_PASS (pass_lim); */
>  	  NEXT_PASS (pass_copy_prop);
>  	  NEXT_PASS (pass_dce_loop);
>  	  NEXT_PASS (pass_lim);
> @@ -897,6 +898,7 @@ init_optimization_passes (void)
>  	  NEXT_PASS (pass_graphite_transforms);
>  	    {
>  	      struct opt_pass **p = &pass_graphite_transforms.pass.sub;
> +	      NEXT_PASS (pass_copy_prop);
>  	      NEXT_PASS (pass_dce_loop);
>  	      NEXT_PASS (pass_lim);
>  	    }
> diff -up /opt/gcc/_clean/gcc/tree-ssa-loop-manip.c /opt/gcc/p_work/gcc/tree-ssa-loop-manip.c
> --- /opt/gcc/_clean/gcc/tree-ssa-loop-manip.c	2009-11-25 18:20:33.000000000 +0100
> +++ /opt/gcc/p_work/gcc/tree-ssa-loop-manip.c	2010-03-16 20:10:04.000000000 +0100
> @@ -1215,9 +1215,9 @@ canonicalize_loop_ivs (struct loop *loop
>  	gsi_insert_seq_on_edge_immediate (loop_preheader_edge (loop), stmts);
>      }
>  
> -  gsi = gsi_last_bb (loop->latch);
> +  gsi = gsi_last_bb (loop->header);
>    create_iv (build_int_cst_type (type, 0), build_int_cst (type, 1), NULL_TREE,
> -	     loop, &gsi, true, &var_before, NULL);
> +	     loop, &gsi, false, &var_before, NULL);
>  
>    rewrite_all_phi_nodes_with_iv (loop, var_before);
>  
> I get
> 
> [macbook] lin/test% gfcp -O3 -ffast-math gas_dyn.f90 -ftree-vectorizer-verbose=2 -fgraphite-identity > & tmp1
> [macbook] lin/test% grep VECTORIZED tmp1 | wc
>       33     132    1345
> 
> Sebastian are you sure that you don't have other patches 
> in your build?
> 
> Dominique

Dominique,
    Despite the discrepancy, these patches are a real improvement. I did a complete build of xplor-nih 2.25
tonight against gcc trunk with -O3 -ffast-math -funroll-loops on x86_64-apple-darwin10. The benchmarks for
that build compared to one with -fgraphite-identity added are identical. It will be really interesting to see what
the graphite loop optimizations can actually do now that vectorization is being retained.
          Jack