[patch] mainline: Remove -fopenmp-ssa

[patch] mainline: Remove -fopenmp-ssa

[Aldy Hernandez]

Hi folks.

-fopenmp-ssa isn't needed any more.  Diego pre-approved its removal.

I'm committing the following patch to mainline (yes, mainline, not
tuples) :).

Tested on x86-linux.

Aldy

	* omp-low.c (gate_expand_omp_ssa): Remove.
	(pass_expand_omp_ssa): Remove.
	(gate_expand_omp): Do not check for flag_openmp_ssa.
	* common.opt (-fopenmp-ssa): Remove.
	* passes.c (init_optimization_passes): Remove pass_expand_omp_ssa.

Index: omp-low.c
===================================================================
--- omp-low.c	(revision 135785)
+++ omp-low.c	(working copy)
@@ -4217,40 +4217,12 @@ execute_expand_omp (void)
   return 0;
 }
 
-/* OMP expansion in SSA form.  For testing purposes only.  */
-
-static bool
-gate_expand_omp_ssa (void)
-{
-  return flag_openmp_ssa && flag_openmp != 0 && errorcount == 0;
-}
-
-struct gimple_opt_pass pass_expand_omp_ssa = 
-{
- {
-  GIMPLE_PASS,
-  "ompexpssa",				/* name */
-  gate_expand_omp_ssa,			/* gate */
-  execute_expand_omp,			/* execute */
-  NULL,					/* sub */
-  NULL,					/* next */
-  0,					/* static_pass_number */
-  0,					/* tv_id */
-  PROP_gimple_any,			/* properties_required */
-  PROP_gimple_lomp,			/* properties_provided */
-  0,					/* properties_destroyed */
-  0,					/* todo_flags_start */
-  TODO_dump_func			/* todo_flags_finish */
- }
-};
-
 /* OMP expansion -- the default pass, run before creation of SSA form.  */
 
 static bool
 gate_expand_omp (void)
 {
-  return ((!flag_openmp_ssa || !optimize)
-	  && flag_openmp != 0 && errorcount == 0);
+  return (flag_openmp != 0 && errorcount == 0);
 }
 
 struct gimple_opt_pass pass_expand_omp = 
Index: common.opt
===================================================================
--- common.opt	(revision 135785)
+++ common.opt	(working copy)
@@ -732,10 +732,6 @@ fomit-frame-pointer
 Common Report Var(flag_omit_frame_pointer) Optimization
 When possible do not generate stack frames
 
-fopenmp-ssa
-Common Report Var(flag_openmp_ssa)
-Expand OpenMP operations on SSA form
-
 foptimize-register-move
 Common Report Var(flag_regmove) Optimization
 Do the full register move optimization pass
Index: passes.c
===================================================================
--- passes.c	(revision 135785)
+++ passes.c	(working copy)
@@ -548,7 +548,6 @@ init_optimization_passes (void)
 	  NEXT_PASS (pass_referenced_vars);
 	  NEXT_PASS (pass_reset_cc_flags);
 	  NEXT_PASS (pass_build_ssa);
-	  NEXT_PASS (pass_expand_omp_ssa);
 	  NEXT_PASS (pass_early_warn_uninitialized);
 	  NEXT_PASS (pass_rebuild_cgraph_edges);
 	  NEXT_PASS (pass_early_inline);

Re: [patch] mainline: Remove -fopenmp-ssa

[Gerald Pfeifer]

On Thu, 22 May 2008, Aldy Hernandez wrote:
> -fopenmp-ssa isn't needed any more.  Diego pre-approved its removal.

Should we document this in gcc-4.4/changes.html?

Gerald

Re: [patch] mainline: Remove -fopenmp-ssa

[Diego Novillo]

2008/5/23 Gerald Pfeifer <gerald@pfeifer.com>:
> On Thu, 22 May 2008, Aldy Hernandez wrote:
>> -fopenmp-ssa isn't needed any more.  Diego pre-approved its removal.
>
> Should we document this in gcc-4.4/changes.html?

It's not really something that received much exposure.  It was an
undocumented flag that Zdenek introduced for testing purposes while
implementing the loop parallelizer.


Diego.

Re: [patch] mainline: Remove -fopenmp-ssa

[Aldy Hernandez]

On Fri, May 23, 2008 at 10:59:17AM +0200, Gerald Pfeifer wrote:
> On Thu, 22 May 2008, Aldy Hernandez wrote:
> > -fopenmp-ssa isn't needed any more.  Diego pre-approved its removal.
> 
> Should we document this in gcc-4.4/changes.html?

It was an undocumented feature and I don't see any bugs reported against
it.  Perhaps we can get away with silently dropping it?

Re: [patch] mainline: Remove -fopenmp-ssa

[Tom Tromey]

>>>>> "Aldy" == Aldy Hernandez <aldyh@redhat.com> writes:

>> Should we document this in gcc-4.4/changes.html?

Aldy> It was an undocumented feature and I don't see any bugs reported against
Aldy> it.  Perhaps we can get away with silently dropping it?

The only hit for "fopenmp-ssa" in google code search is
gcc/common.opt.  So, I think it is safe to say that it is not used.

Tom

Re: [patch] mainline: Remove -fopenmp-ssa

[Gerald Pfeifer]

On Wed, 28 May 2008, Aldy Hernandez wrote:
> It was an undocumented feature and I don't see any bugs reported against 
> it.  Perhaps we can get away with silently dropping it?

Diego and you have a very strong case. :-)  Sorry for the noise in this
case, just trying to keep an eye on the documentation in general.

Gerald