Re: [graphite] Add -fgraphite-identity

[Jack Howarth <howarth at bromo dot msbb dot uc dot edu>]

On Thu, Oct 23, 2008 at 01:09:51PM -0200, Tobias Grosser wrote:
> Hi,
> 
> after testing "-fgraphite-identity" in the graphite branch for a week, I
> would like to add it to trunk.
> 
> This patch contains only the addition of "-fgraphite-identity".
> 
> What is ï"-fgraphite-identity"?
> ï"-fgraphite-identity" allows us to test graphite without any
> optimizations enabled. So we can just check the GIMPLE->GRAPHITE->GIMPLE
> identity transformation.
> 
> Benefits:
> -	Easier debugging/testing of front and backend, without any
> 	optimizations changing the test results.
> -	The backend can be exposed to more code, so we more test cases
> 	for the backend.
> 
> Because of this, we will use it in the planned nightly tests.
> 
> See you
> Tobi
> 
> 

> 2008-10-10  Tobias Grosser  <grosser@fim.uni-passau.de>
> 
> 	* doc/invoke.texi: Add -fgraphite-identity.
> 	* graphite.c (graphite_apply_transformations): Check for
> 	-fgraphite-identity.
> 	* toplev.c: (process_options): Add flag_graphite_identity.
> 	* tree-ssa-loop.c: Add flag_graphite_identity.
> 
> diff --git a/gcc/common.opt b/gcc/common.opt
> index 6f09dfd..4e68067 100644
> --- a/gcc/common.opt
> +++ b/gcc/common.opt
> @@ -567,6 +567,10 @@ floop-block
>  Common Report Var(flag_loop_block) Optimization
>  Enable Loop Blocking transformation
>  
> +fgraphite-identity
> +Common Report Var(flag_graphite_identity) Optimization
> +Enable Graphite Identity transformation
> +
>  fguess-branch-probability
>  Common Report Var(flag_guess_branch_prob) Optimization
>  Enable guessing of branch probabilities
> diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
> index 755c422..4fcf975 100644
> --- a/gcc/doc/invoke.texi
> +++ b/gcc/doc/invoke.texi
> @@ -340,7 +340,7 @@ Objective-C and Objective-C++ Dialects}.
>  -fira-coalesce -fno-ira-share-save-slots @gol
>  -fno-ira-share-spill-slots -fira-verbose=@var{n} @gol
>  -fivopts -fkeep-inline-functions -fkeep-static-consts @gol
> --floop-block -floop-interchange -floop-strip-mine @gol
> +-floop-block -floop-interchange -floop-strip-mine -fgraphite-identity @gol
>  -fmerge-all-constants -fmerge-constants -fmodulo-sched @gol
>  -fmodulo-sched-allow-regmoves -fmove-loop-invariants -fmudflap @gol
>  -fmudflapir -fmudflapth -fno-branch-count-reg -fno-default-inline @gol
> @@ -6116,6 +6116,13 @@ because the innermost loop will iterate over a smaller amount of data
>  that can be kept in the caches.  This optimization applies to all the
>  languages supported by GCC and is not limited to Fortran.
>  
> +@item -fgraphite-identity
> +@opindex fgraphite-identity
> +Enable the identity transformation for graphite. For every SCoP we generate
> +the polyhedral representation and transform it back to gimple. Using
> +-fgraphite-identity we can check the costs or benefits of the
> +GIMPLE -> GRAPHITE -> GIMPLE transformation.
> +
>  @item -fcheck-data-deps
>  @opindex fcheck-data-deps
>  Compare the results of several data dependence analyzers.  This option
> diff --git a/gcc/graphite.c b/gcc/graphite.c
> index a615e2c..ab386d7 100644
> --- a/gcc/graphite.c
> +++ b/gcc/graphite.c
> @@ -4912,17 +4912,13 @@ graphite_apply_transformations (scop_p scop)
>    if (flag_loop_block)
>      transform_done = graphite_trans_scop_block (scop);
>  
> -#if 0 && ENABLE_CHECKING
> -  /* When the compiler is configured with ENABLE_CHECKING, always
> -     generate code, even if we did not apply any transformation.  This
> -     provides better code coverage of the backend code generator.
> -
> -     This also allows to check the performance for an identity
> -     transform: GIMPLE -> GRAPHITE -> GIMPLE; and the output of CLooG
> -     is never an identity: if CLooG optimizations are not disabled,
> -     the CLooG output is always optimized in control flow.  */
> -  transform_done = true;
> -#endif
> +  /* Generate code, even if we did not apply any real transformation.
> +     This also allows to check the performance for the identity
> +     transformation: GIMPLE -> GRAPHITE -> GIMPLE
> +     Keep in mind, that CLooG optimizes in control, so the loop structure
> +     may change, even if we only use -fgraphite-identity.  */ 
> +  if (flag_graphite_identity)
> +    transform_done = true;
>  
>    return transform_done;
>  }
> diff --git a/gcc/toplev.c b/gcc/toplev.c
> index 24e4df7..42ad2a4 100644
> --- a/gcc/toplev.c
> +++ b/gcc/toplev.c
> @@ -1707,7 +1707,8 @@ process_options (void)
>    if (flag_graphite
>        || flag_loop_block
>        || flag_loop_interchange
> -      || flag_loop_strip_mine)
> +      || flag_loop_strip_mine
> +      || flag_graphite_identity)
>      sorry ("Graphite loop optimizations cannot be used");
>  #endif
>  
> diff --git a/gcc/tree-ssa-loop.c b/gcc/tree-ssa-loop.c
> index 51fc07c..2ea58f6 100644
> --- a/gcc/tree-ssa-loop.c
> +++ b/gcc/tree-ssa-loop.c
> @@ -305,7 +305,8 @@ gate_graphite_transforms (void)
>  {
>    /* Enable -fgraphite pass if any one of the graphite optimization flags 
>       is turned on.  */
> -  if (flag_loop_block || flag_loop_interchange || flag_loop_strip_mine)
> +  if (flag_loop_block || flag_loop_interchange || flag_loop_strip_mine
> +      || flag_graphite_identity)
>      flag_graphite = 1;
>  
>    return flag_graphite != 0;

Tobias,
    The -fgraphite-identity option does generate code though from the
GIMPLE->GRAPHITE->GIMPLE transformations, right? Also, I thought it was
said that these transformations might improve the ability of GIMPLE to
optimize loops. Is that a correct assumption?
                     Jack