[patch] Merge to trunk from Graphite branch

[patch] Merge to trunk from Graphite branch

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 182 bytes --]

Hi,

I just merged the Graphite branch back to trunk.  This solved
PR43349 for which I just added the testcase, as I already have
taken care of it in the Graphite branch.

Sebastian

[-- Attachment #2: 0001-Limit-the-number-of-parameters-per-SCoP.patch --]
[-- Type: text/x-diff, Size: 3282 bytes --]

From c2e502a56a95d14a87d60f288c1aa8612fea911c Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:34:38 +0000
Subject: [PATCH 01/11] Limit the number of parameters per SCoP.

2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-sese-to-poly.c (build_poly_scop): Limit scops following
	the number of parameters in the scop.  Use as an upper bound
	PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS.
	* params.def (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS): Declared.
	* doc/invoke.texi: Document it.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157431 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite      |    8 ++++++++
 gcc/doc/invoke.texi         |    4 ++++
 gcc/graphite-sese-to-poly.c |    5 +++++
 gcc/params.def              |    7 +++++++
 4 files changed, 24 insertions(+), 0 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 9dcdcda..61b463b 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,11 @@
+2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (build_poly_scop): Limit scops following
+	the number of parameters in the scop.  Use as an upper bound
+	PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS.
+	* params.def (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS): Declared.
+	* doc/invoke.texi: Document it.
+
 2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
 
 	* graphite-sese-to-poly.c (add_param_constraints): Use
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index 060be88..2159701 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -8492,6 +8492,10 @@ parameters only when their cumulative size is less or equal to
 @option{ipa-sra-ptr-growth-factor} times the size of the original
 pointer parameter.
 
+@item graphite-max-nb-scop-params
+To avoid exponential effects in the Graphite loop transforms, the
+number of parameters in a SCoP is bounded by 10.
+
 @end table
 @end table
 
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 11bddf8..ae4a083 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -2937,6 +2937,7 @@ build_poly_scop (scop_p scop)
 {
   sese region = SCOP_REGION (scop);
   sbitmap reductions = sbitmap_alloc (last_basic_block * 2);
+  graphite_dim_t max_dim;
 
   sbitmap_zero (reductions);
   rewrite_commutative_reductions_out_of_ssa (region, reductions);
@@ -2960,6 +2961,10 @@ build_poly_scop (scop_p scop)
   build_sese_conditions (region);
   find_scop_parameters (scop);
 
+  max_dim = PARAM_VALUE (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS);
+  if (scop_nb_params (scop) > max_dim)
+    return false;
+
   build_scop_iteration_domain (scop);
   build_scop_context (scop);
 
diff --git a/gcc/params.def b/gcc/params.def
index 07bfb90..f6f549c 100644
--- a/gcc/params.def
+++ b/gcc/params.def
@@ -745,6 +745,13 @@ DEFPARAM (PARAM_LOOP_BLOCK_TILE_SIZE,
 	  "size of tiles for loop blocking",
 	  51, 0, 0)
 
+/* Maximal number of parameters that we allow in a SCoP.  */
+
+DEFPARAM (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS,
+	  "graphite-max-nb-scop-params",
+	  "maximal number of parameters in a SCoP",
+	  10, 0, 0)
+
 /* Avoid doing loop invariant motion on very large loops.  */
 
 DEFPARAM (PARAM_LOOP_INVARIANT_MAX_BBS_IN_LOOP,
-- 
1.6.3.3


[-- Attachment #3: 0002-Make-build_poly_scop-not-return-a-bool.patch --]
[-- Type: text/x-diff, Size: 3021 bytes --]

From f49215cea754cbd1f73143110bbc832c05fd597f Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:34:44 +0000
Subject: [PATCH 02/11] Make build_poly_scop not return a bool.

2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-sese-to-poly.c (build_poly_scop): Do not return bool.
	* graphite-sese-to-poly.h (build_poly_scop): Same.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157432 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite      |    5 +++++
 gcc/graphite-sese-to-poly.c |   14 +++++++-------
 gcc/graphite-sese-to-poly.h |    2 +-
 3 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 61b463b..2aa6fd5 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,10 @@
 2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* graphite-sese-to-poly.c (build_poly_scop): Do not return bool.
+	* graphite-sese-to-poly.h (build_poly_scop): Same.
+
+2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-sese-to-poly.c (build_poly_scop): Limit scops following
 	the number of parameters in the scop.  Use as an upper bound
 	PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS.
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index ae4a083..75be56d 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -2932,7 +2932,7 @@ scop_ivs_can_be_represented (scop_p scop)
 
 /* Builds the polyhedral representation for a SESE region.  */
 
-bool
+void
 build_poly_scop (scop_p scop)
 {
   sese region = SCOP_REGION (scop);
@@ -2950,12 +2950,11 @@ build_poly_scop (scop_p scop)
      sense to optimize a scop containing only PBBs that do not belong
      to any loops.  */
   if (nb_pbbs_in_loops (scop) == 0)
-    return false;
+    return;
 
   scop_canonicalize_loops (scop);
-
   if (!scop_ivs_can_be_represented (scop))
-    return false;
+    return;
 
   build_sese_loop_nests (region);
   build_sese_conditions (region);
@@ -2963,7 +2962,7 @@ build_poly_scop (scop_p scop)
 
   max_dim = PARAM_VALUE (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS);
   if (scop_nb_params (scop) > max_dim)
-    return false;
+    return;
 
   build_scop_iteration_domain (scop);
   build_scop_context (scop);
@@ -2972,9 +2971,10 @@ build_poly_scop (scop_p scop)
   scop_to_lst (scop);
   build_scop_scattering (scop);
   build_scop_drs (scop);
-  POLY_SCOP_P (scop) = true;
 
-  return true;
+  /* This SCoP has been translated to the polyhedral
+     representation.  */
+  POLY_SCOP_P (scop) = true;
 }
 
 /* Always return false.  Exercise the scop_to_clast function.  */
diff --git a/gcc/graphite-sese-to-poly.h b/gcc/graphite-sese-to-poly.h
index 0737c49..3213471 100644
--- a/gcc/graphite-sese-to-poly.h
+++ b/gcc/graphite-sese-to-poly.h
@@ -28,7 +28,7 @@ struct base_alias_pair
   int *alias_set;
 };
 
-bool build_poly_scop (scop_p);
+void build_poly_scop (scop_p);
 void check_poly_representation (scop_p);
 
 #endif
-- 
1.6.3.3


[-- Attachment #4: 0003-Add-PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION.patch --]
[-- Type: text/x-diff, Size: 3097 bytes --]

From 94ba21de76678316108cb184d517133a66622525 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:34:51 +0000
Subject: [PATCH 03/11] Add PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION.

2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite.c (graphite_initialize): To bound the number of bbs per
	function, use PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION.
	* params.def (PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION): Declared.
	* doc/invoke.texi: Document it.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157433 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite |    7 +++++++
 gcc/doc/invoke.texi    |    5 +++++
 gcc/graphite.c         |    2 +-
 gcc/params.def         |    7 +++++++
 4 files changed, 20 insertions(+), 1 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 2aa6fd5..3a94ee1 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,12 @@
 2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* graphite.c (graphite_initialize): To bound the number of bbs per
+	function, use PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION.
+	* params.def (PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION): Declared.
+	* doc/invoke.texi: Document it.
+
+2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-sese-to-poly.c (build_poly_scop): Do not return bool.
 	* graphite-sese-to-poly.h (build_poly_scop): Same.
 
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index 2159701..b856ca2 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -8496,6 +8496,11 @@ pointer parameter.
 To avoid exponential effects in the Graphite loop transforms, the
 number of parameters in a SCoP is bounded by 10.
 
+@item graphite-max-bbs-per-function
+To avoid exponential effects in the detection of SCoPs, the functions
+with more than 100 basic blocks are not handled by the Graphite loop
+transforms.
+
 @end table
 @end table
 
diff --git a/gcc/graphite.c b/gcc/graphite.c
index ba05cc7..a244b87 100644
--- a/gcc/graphite.c
+++ b/gcc/graphite.c
@@ -202,7 +202,7 @@ graphite_initialize (void)
   if (number_of_loops () <= 1
       /* FIXME: This limit on the number of basic blocks of a function
 	 should be removed when the SCOP detection is faster.  */
-      || n_basic_blocks > 100)
+      || n_basic_blocks > PARAM_VALUE (PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION))
     {
       if (dump_file && (dump_flags & TDF_DETAILS))
 	print_global_statistics (dump_file);
diff --git a/gcc/params.def b/gcc/params.def
index f6f549c..dca575d 100644
--- a/gcc/params.def
+++ b/gcc/params.def
@@ -752,6 +752,13 @@ DEFPARAM (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS,
 	  "maximal number of parameters in a SCoP",
 	  10, 0, 0)
 
+/* Maximal number of basic blocks in the functions analyzed by Graphite.  */
+
+DEFPARAM (PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION,
+	  "graphite-max-bbs-per-function",
+	  "maximal number of basic blocks per function to be analyzed by Graphite",
+	  100, 0, 0)
+
 /* Avoid doing loop invariant motion on very large loops.  */
 
 DEFPARAM (PARAM_LOOP_INVARIANT_MAX_BBS_IN_LOOP,
-- 
1.6.3.3


[-- Attachment #5: 0004-Fix-PR43306-correct-evolution_function_right_is_inte.patch --]
[-- Type: text/x-diff, Size: 2772 bytes --]

From c4ccbe87dae599b81683031c8b48e9f2459f668c Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:34:59 +0000
Subject: [PATCH 04/11] Fix PR43306: correct evolution_function_right_is_integer_cst.

2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/43306
	* tree-chrec.c (evolution_function_right_is_integer_cst): CHREC_RIGHT
	should be an INTEGER_CST.  Also handle CASE_CONVERT.
	* gcc.dg/graphite/pr43306.c: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157434 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                  |    7 +++++++
 gcc/testsuite/gcc.dg/graphite/pr43306.c |    9 +++++++++
 gcc/tree-chrec.c                        |   12 +++++-------
 3 files changed, 21 insertions(+), 7 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/graphite/pr43306.c

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 3a94ee1..837e4bc 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,12 @@
 2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
 
+	PR middle-end/43306
+	* tree-chrec.c (evolution_function_right_is_integer_cst): CHREC_RIGHT
+	should be an INTEGER_CST.  Also handle CASE_CONVERT.
+	* gcc.dg/graphite/pr43306.c: New.
+
+2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite.c (graphite_initialize): To bound the number of bbs per
 	function, use PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION.
 	* params.def (PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION): Declared.
diff --git a/gcc/testsuite/gcc.dg/graphite/pr43306.c b/gcc/testsuite/gcc.dg/graphite/pr43306.c
new file mode 100644
index 0000000..43195e4
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/pr43306.c
@@ -0,0 +1,9 @@
+/* { dg-options "-O1 -fstrict-overflow -fgraphite-identity" } */
+
+void foo(int x[])
+{
+  int i, j;
+  for (i = 0; i < 2; i++)
+    for (j = 0; j < 2; j++)
+      x[i] = x[i*j];
+}
diff --git a/gcc/tree-chrec.c b/gcc/tree-chrec.c
index c945f93..929bbc0 100644
--- a/gcc/tree-chrec.c
+++ b/gcc/tree-chrec.c
@@ -1529,14 +1529,12 @@ evolution_function_right_is_integer_cst (const_tree chrec)
       return true;
 
     case POLYNOMIAL_CHREC:
-      if (!evolution_function_right_is_integer_cst (CHREC_RIGHT (chrec)))
-	return false;
-
-      if (TREE_CODE (CHREC_LEFT (chrec)) == POLYNOMIAL_CHREC
-	&& !evolution_function_right_is_integer_cst (CHREC_LEFT (chrec)))
-	return false;
+      return TREE_CODE (CHREC_RIGHT (chrec)) == INTEGER_CST
+	&& (TREE_CODE (CHREC_LEFT (chrec)) != POLYNOMIAL_CHREC
+	    || evolution_function_right_is_integer_cst (CHREC_LEFT (chrec)));
 
-      return true;
+    CASE_CONVERT:
+      return evolution_function_right_is_integer_cst (TREE_OPERAND (chrec, 0));
 
     default:
       return false;
-- 
1.6.3.3


[-- Attachment #6: 0005-Document-PARAM_LOOP_BLOCK_TILE_SIZE.patch --]
[-- Type: text/x-diff, Size: 3276 bytes --]

From 89466e4339eb2a921006c0284a4da2dd7bf12c39 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:35:08 +0000
Subject: [PATCH 05/11] Document PARAM_LOOP_BLOCK_TILE_SIZE.

2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>

	* doc/invoke.texi (PARAM_LOOP_BLOCK_TILE_SIZE): Document.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157435 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite |    4 ++++
 gcc/doc/invoke.texi    |   25 +++++++++++++++++--------
 2 files changed, 21 insertions(+), 8 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 837e4bc..3a49ccc 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,9 @@
 2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* doc/invoke.texi (PARAM_LOOP_BLOCK_TILE_SIZE): Document.
+
+2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
+
 	PR middle-end/43306
 	* tree-chrec.c (evolution_function_right_is_integer_cst): CHREC_RIGHT
 	should be an INTEGER_CST.  Also handle CASE_CONVERT.
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index b856ca2..8648405 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -6675,7 +6675,9 @@ Graphite loop transformation infrastructure.
 Perform loop strip mining transformations on loops.  Strip mining
 splits a loop into two nested loops.  The outer loop has strides
 equal to the strip size and the inner loop has strides of the
-original loop within a strip.  For example, given a loop like:
+original loop within a strip.  The strip length can be changed
+using the @option{loop-block-tile-size} parameter.  For example,
+given a loop like:
 @smallexample
 DO I = 1, N
   A(I) = A(I) + C
@@ -6683,8 +6685,8 @@ ENDDO
 @end smallexample
 loop strip mining will transform the loop as if the user had written:
 @smallexample
-DO II = 1, N, 4
-  DO I = II, min (II + 3, N)
+DO II = 1, N, 51
+  DO I = II, min (II + 50, N)
     A(I) = A(I) + C
   ENDDO
 ENDDO
@@ -6697,7 +6699,9 @@ enable the Graphite loop transformation infrastructure.
 @item -floop-block
 Perform loop blocking transformations on loops.  Blocking strip mines
 each loop in the loop nest such that the memory accesses of the
-element loops fit inside caches.  For example, given a loop like:
+element loops fit inside caches.  The strip length can be changed
+using the @option{loop-block-tile-size} parameter.  For example, given
+a loop like:
 @smallexample
 DO I = 1, N
   DO J = 1, M
@@ -6707,10 +6711,10 @@ ENDDO
 @end smallexample
 loop blocking will transform the loop as if the user had written:
 @smallexample
-DO II = 1, N, 64
-  DO JJ = 1, M, 64
-    DO I = II, min (II + 63, N)
-      DO J = JJ, min (JJ + 63, M)
+DO II = 1, N, 51
+  DO JJ = 1, M, 51
+    DO I = II, min (II + 50, N)
+      DO J = JJ, min (JJ + 50, M)
         A(J, I) = B(I) + C(J)
       ENDDO
     ENDDO
@@ -8501,6 +8505,11 @@ To avoid exponential effects in the detection of SCoPs, the functions
 with more than 100 basic blocks are not handled by the Graphite loop
 transforms.
 
+@item loop-block-tile-size
+The default factor for the loop blocking or strip mining transforms,
+enabled with @option{-floop-block} or @option{-floop-strip-mine}, is
+51.
+
 @end table
 @end table
 
-- 
1.6.3.3


[-- Attachment #7: 0006-Cleanup-remove-FIXMEs-add-new-function.patch --]
[-- Type: text/x-diff, Size: 7456 bytes --]

From 5d92ac748a1816c48ce2cc86623ab0e70a3512b5 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:35:15 +0000
Subject: [PATCH 06/11] Cleanup: remove FIXMEs, add new function.

2010-03-10  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-clast-to-gimple.c (gcc_type_for_iv_of_clast_loop): Remove
	forward declaration.
	* graphite-sese-to-poly.c (reduction_phi_p): Remove FIXME comment.
	(add_upper_bounds_from_estimated_nit): New.
	(build_loop_iteration_domains): Use it.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157436 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite         |    8 +++
 gcc/graphite-clast-to-gimple.c |    3 -
 gcc/graphite-sese-to-poly.c    |  128 ++++++++++++++++++++++------------------
 3 files changed, 78 insertions(+), 61 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 3a49ccc..0ca526c 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,11 @@
+2010-03-10  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-clast-to-gimple.c (gcc_type_for_iv_of_clast_loop): Remove
+	forward declaration.
+	* graphite-sese-to-poly.c (reduction_phi_p): Remove FIXME comment.
+	(add_upper_bounds_from_estimated_nit): New.
+	(build_loop_iteration_domains): Use it.
+
 2010-03-09  Sebastian Pop  <sebastian.pop@amd.com>
 
 	* doc/invoke.texi (PARAM_LOOP_BLOCK_TILE_SIZE): Document.
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 64ddbb8..377ca8b 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -823,9 +823,6 @@ translate_clast_user (sese region, struct clast_user_stmt *stmt, edge next_e,
   return next_e;
 }
 
-static tree gcc_type_for_iv_of_clast_loop (struct clast_for *);
-
-
 /* Creates a new if region protecting the loop to be executed, if the execution
    count is zero (lb > ub).  */
 static edge
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 75be56d..28ed07c 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -180,7 +180,7 @@ reduction_phi_p (sese region, gimple_stmt_iterator *psi)
 
   if (simple_copy_phi_p (phi))
     {
-      /* FIXME: PRE introduces phi nodes like these, for an example,
+      /* PRE introduces phi nodes like these, for an example,
 	 see id-5.f in the fortran graphite testsuite:
 
 	 # prephitmp.85_265 = PHI <prephitmp.85_258(33), prephitmp.85_265(18)>
@@ -1038,6 +1038,74 @@ gbb_from_bb (basic_block bb)
   return (gimple_bb_p) bb->aux;
 }
 
+/* Insert in the SCOP context constraints from the estimation of the
+   number of iterations.  UB_EXPR is a linear expression describing
+   the number of iterations in a loop.  This expression is bounded by
+   the estimation NIT.  */
+
+static void
+add_upper_bounds_from_estimated_nit (scop_p scop, double_int nit,
+				     ppl_dimension_type dim,
+				     ppl_Linear_Expression_t ub_expr)
+{
+  Value val;
+  ppl_Linear_Expression_t nb_iters_le;
+  ppl_Polyhedron_t pol;
+  ppl_Coefficient_t coef;
+  ppl_Constraint_t ub;
+
+  ppl_new_Linear_Expression_with_dimension (&ub_expr, dim);
+  ppl_new_C_Polyhedron_from_space_dimension (&pol, dim, 0);
+  ppl_new_Linear_Expression_from_Linear_Expression (&nb_iters_le,
+						    ub_expr);
+
+  /* Construct the negated number of last iteration in VAL.  */
+  value_init (val);
+  mpz_set_double_int (val, nit, false);
+  value_sub_int (val, val, 1);
+  value_oppose (val, val);
+
+  /* NB_ITERS_LE holds the number of last iteration in
+     parametrical form.  Subtract estimated number of last
+     iteration and assert that result is not positive.  */
+  ppl_new_Coefficient_from_mpz_t (&coef, val);
+  ppl_Linear_Expression_add_to_inhomogeneous (nb_iters_le, coef);
+  ppl_delete_Coefficient (coef);
+  ppl_new_Constraint (&ub, nb_iters_le,
+		      PPL_CONSTRAINT_TYPE_LESS_OR_EQUAL);
+  ppl_Polyhedron_add_constraint (pol, ub);
+
+  /* Remove all but last GDIM dimensions from POL to obtain
+     only the constraints on the parameters.  */
+  {
+    graphite_dim_t gdim = scop_nb_params (scop);
+    ppl_dimension_type *dims = XNEWVEC (ppl_dimension_type, dim - gdim);
+    graphite_dim_t i;
+
+    for (i = 0; i < dim - gdim; i++)
+      dims[i] = i;
+
+    ppl_Polyhedron_remove_space_dimensions (pol, dims, dim - gdim);
+    XDELETEVEC (dims);
+  }
+
+  /* Add the constraints on the parameters to the SCoP context.  */
+  {
+    ppl_Pointset_Powerset_C_Polyhedron_t constraints_ps;
+
+    ppl_new_Pointset_Powerset_C_Polyhedron_from_C_Polyhedron
+      (&constraints_ps, pol);
+    ppl_Pointset_Powerset_C_Polyhedron_intersection_assign
+      (SCOP_CONTEXT (scop), constraints_ps);
+    ppl_delete_Pointset_Powerset_C_Polyhedron (constraints_ps);
+  }
+
+  ppl_delete_Polyhedron (pol);
+  ppl_delete_Linear_Expression (nb_iters_le);
+  ppl_delete_Constraint (ub);
+  value_clear (val);
+}
+
 /* Builds the constraint polyhedra for LOOP in SCOP.  OUTER_PH gives
    the constraints for the surrounding loops.  */
 
@@ -1113,64 +1181,8 @@ build_loop_iteration_domains (scop_p scop, struct loop *loop,
       scan_tree_for_params (SCOP_REGION (scop), nb_iters, ub_expr, one);
       value_clear (one);
 
-      /* N <= estimated_nb_iters
-
-	 FIXME: This is a workaround that should go away once we will
-	 have the PIP algorithm.  */
       if (estimated_loop_iterations (loop, true, &nit))
-	{
-	  Value val;
-	  ppl_Linear_Expression_t nb_iters_le;
-	  ppl_Polyhedron_t pol;
-	  graphite_dim_t n = scop_nb_params (scop);
-	  ppl_Coefficient_t coef;
-
-	  ppl_new_C_Polyhedron_from_space_dimension (&pol, dim, 0);
-	  ppl_new_Linear_Expression_from_Linear_Expression (&nb_iters_le,
-							    ub_expr);
-
-	  /* Construct the negated number of last iteration in VAL.  */
-	  value_init (val);
-	  mpz_set_double_int (val, nit, false);
-	  value_sub_int (val, val, 1);
-	  value_oppose (val, val);
-
-	  /* NB_ITERS_LE holds number of last iteration in parametrical form.
-	  Subtract estimated number of last iteration and assert that result
-	  is not positive.  */
-	  ppl_new_Coefficient_from_mpz_t (&coef, val);
-	  ppl_Linear_Expression_add_to_inhomogeneous (nb_iters_le, coef);
-	  ppl_delete_Coefficient (coef);
-	  ppl_new_Constraint (&ub, nb_iters_le,
-			      PPL_CONSTRAINT_TYPE_LESS_OR_EQUAL);
-	  ppl_Polyhedron_add_constraint (pol, ub);
-
-	  /* Remove all but last N dimensions from POL to obtain constraints
-	     on parameters.  */
-	    {
-	      ppl_dimension_type *dims = XNEWVEC (ppl_dimension_type, dim - n);
-	      graphite_dim_t i;
-	      for (i = 0; i < dim - n; i++)
-		dims[i] = i;
-	      ppl_Polyhedron_remove_space_dimensions (pol, dims, dim - n);
-	      XDELETEVEC (dims);
-	    }
-
-	  /* Add constraints on parameters to SCoP context.  */
-	    {
-	      ppl_Pointset_Powerset_C_Polyhedron_t constraints_ps;
-	      ppl_new_Pointset_Powerset_C_Polyhedron_from_C_Polyhedron
-	       (&constraints_ps, pol);
-	      ppl_Pointset_Powerset_C_Polyhedron_intersection_assign
-	       (SCOP_CONTEXT (scop), constraints_ps);
-	      ppl_delete_Pointset_Powerset_C_Polyhedron (constraints_ps);
-	    }
-
-	  ppl_delete_Polyhedron (pol);
-	  ppl_delete_Linear_Expression (nb_iters_le);
-	  ppl_delete_Constraint (ub);
-	  value_clear (val);
-	}
+	add_upper_bounds_from_estimated_nit (scop, nit, dim, ub_expr);
 
       /* loop_i <= expr_nb_iters */
       ppl_set_coef (ub_expr, nb, -1);
-- 
1.6.3.3


[-- Attachment #8: 0007-Fix-documentation-of-the-new-parameters.patch --]
[-- Type: text/x-diff, Size: 3704 bytes --]

From 578e34563f159ce091cda0b8f3fd8a2da30c297c Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:35:22 +0000
Subject: [PATCH 07/11] Fix documentation of the new parameters.

2010-03-10  Sebastian Pop  <sebastian.pop@amd.com>

	* doc/invoke.texi: Fix documentation of graphite-max-nb-scop-params,
	graphite-max-bbs-per-function, and loop-block-tile-size.
	* params.def (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS): Replace "maximal"
	with "maximum".
	(PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION): Same.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157437 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite |    8 ++++++++
 gcc/doc/invoke.texi    |   18 +++++++++++-------
 gcc/params.def         |    4 ++--
 3 files changed, 21 insertions(+), 9 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 0ca526c..75fe38f 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,13 @@
 2010-03-10  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* doc/invoke.texi: Fix documentation of graphite-max-nb-scop-params,
+	graphite-max-bbs-per-function, and loop-block-tile-size.
+	* params.def (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS): Replace "maximal"
+	with "maximum".
+	(PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION): Same.
+
+2010-03-10  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-clast-to-gimple.c (gcc_type_for_iv_of_clast_loop): Remove
 	forward declaration.
 	* graphite-sese-to-poly.c (reduction_phi_p): Remove FIXME comment.
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index 8648405..23a0f4a 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -8498,17 +8498,21 @@ pointer parameter.
 
 @item graphite-max-nb-scop-params
 To avoid exponential effects in the Graphite loop transforms, the
-number of parameters in a SCoP is bounded by 10.
+number of parameters in a Static Control Part (SCoP) is bounded.  The
+default value is 10 parameters.  A variable whose value is unknown at
+compile time and defined outside a SCoP is a parameter of the SCoP.
 
 @item graphite-max-bbs-per-function
-To avoid exponential effects in the detection of SCoPs, the functions
-with more than 100 basic blocks are not handled by the Graphite loop
-transforms.
+To avoid exponential effects in the detection of SCoPs, the size of
+the functions analyzed by Graphite is bounded.  The default value is
+100 basic blocks.
 
 @item loop-block-tile-size
-The default factor for the loop blocking or strip mining transforms,
-enabled with @option{-floop-block} or @option{-floop-strip-mine}, is
-51.
+Loop blocking or strip mining transforms, enabled with
+@option{-floop-block} or @option{-floop-strip-mine}, strip mine each
+loop in the loop nest by a given number of iterations.  The strip
+length can be changed using the @option{loop-block-tile-size}
+parameter.  The default value is 51 iterations.
 
 @end table
 @end table
diff --git a/gcc/params.def b/gcc/params.def
index dca575d..435a788 100644
--- a/gcc/params.def
+++ b/gcc/params.def
@@ -749,14 +749,14 @@ DEFPARAM (PARAM_LOOP_BLOCK_TILE_SIZE,
 
 DEFPARAM (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS,
 	  "graphite-max-nb-scop-params",
-	  "maximal number of parameters in a SCoP",
+	  "maximum number of parameters in a SCoP",
 	  10, 0, 0)
 
 /* Maximal number of basic blocks in the functions analyzed by Graphite.  */
 
 DEFPARAM (PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION,
 	  "graphite-max-bbs-per-function",
-	  "maximal number of basic blocks per function to be analyzed by Graphite",
+	  "maximum number of basic blocks per function to be analyzed by Graphite",
 	  100, 0, 0)
 
 /* Avoid doing loop invariant motion on very large loops.  */
-- 
1.6.3.3


[-- Attachment #9: 0008-Use-ssizetype-when-long_long_integer_type_node-is-NU.patch --]
[-- Type: text/x-diff, Size: 4158 bytes --]

From 119adb56544f6297218e0527ee3ebb18071fbb35 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:35:29 +0000
Subject: [PATCH 08/11] Use ssizetype when long_long_integer_type_node is NULL.

2010-03-11  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-clast-to-gimple.c (my_long_long): Defined.
	(gcc_type_for_cloog_iv): Use it instead of long_long_integer_type_node.
	* graphite-sese-to-poly.c (my_long_long): Defined.
	(scop_ivs_can_be_represented): Use it.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157438 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite         |    7 +++++++
 gcc/graphite-clast-to-gimple.c |   15 ++++++++++-----
 gcc/graphite-sese-to-poly.c    |    6 +++++-
 3 files changed, 22 insertions(+), 6 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 75fe38f..abd9888 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,10 @@
+2010-03-11  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-clast-to-gimple.c (my_long_long): Defined.
+	(gcc_type_for_cloog_iv): Use it instead of long_long_integer_type_node.
+	* graphite-sese-to-poly.c (my_long_long): Defined.
+	(scop_ivs_can_be_represented): Use it.
+
 2010-03-10  Sebastian Pop  <sebastian.pop@amd.com>
 
 	* doc/invoke.texi: Fix documentation of graphite-max-nb-scop-params,
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 377ca8b..f99a158 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -547,6 +547,9 @@ clast_get_body_of_loop (struct clast_stmt *stmt)
   gcc_unreachable ();
 }
 
+/* Java does not initialize long_long_integer_type_node.  */
+#define my_long_long (long_long_integer_type_node ? long_long_integer_type_node : ssizetype)
+
 /* Given a CLOOG_IV, return the type that CLOOG_IV should have in GCC
    land.  The selected type is big enough to include the original loop
    iteration variable, but signed to work with the subtractions CLooG
@@ -581,8 +584,8 @@ gcc_type_for_cloog_iv (const char *cloog_iv, gimple_bb_p gbb)
 	  if (type_precision <= TYPE_PRECISION (long_integer_type_node))
 	    return long_integer_type_node;
 
-	  if (type_precision <= TYPE_PRECISION (long_long_integer_type_node))
-	    return long_long_integer_type_node;
+	  if (type_precision <= TYPE_PRECISION (my_long_long))
+	    return my_long_long;
 
 	  gcc_unreachable ();
 	}
@@ -593,17 +596,19 @@ gcc_type_for_cloog_iv (const char *cloog_iv, gimple_bb_p gbb)
       if (type_precision < TYPE_PRECISION (long_integer_type_node))
 	return long_integer_type_node;
 
-      if (type_precision < TYPE_PRECISION (long_long_integer_type_node))
-	return long_long_integer_type_node;
+      if (type_precision < TYPE_PRECISION (my_long_long))
+	return my_long_long;
 
       /* There is no signed type available, that is large enough to hold the
 	 original value.  */
       gcc_unreachable ();
     }
 
-  return long_long_integer_type_node;
+  return my_long_long;
 }
 
+#undef my_long_long
+
 /* Returns the induction variable for the loop that gets translated to
    STMT.  */
 
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 28ed07c..0f370a8 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -2910,6 +2910,9 @@ scop_canonicalize_loops (scop_p scop)
       graphite_loop_normal_form (loop);
 }
 
+/* Java does not initialize long_long_integer_type_node.  */
+#define my_long_long (long_long_integer_type_node ? long_long_integer_type_node : ssizetype)
+
 /* Can all ivs be represented by a signed integer?
    As CLooG might generate negative values in its expressions, signed loop ivs
    are required in the backend. */
@@ -2934,13 +2937,14 @@ scop_ivs_can_be_represented (scop_p scop)
       precision = TYPE_PRECISION (type);
 
       if (TYPE_UNSIGNED (type)
-	  && precision >= TYPE_PRECISION (long_long_integer_type_node))
+	  && precision >= TYPE_PRECISION (my_long_long))
 	return false;
     }
 
   return true;
 }
 
+#undef my_long_long
 
 /* Builds the polyhedral representation for a SESE region.  */
 
-- 
1.6.3.3


[-- Attachment #10: 0009-Add-testcase-from-PR43349.patch --]
[-- Type: text/x-diff, Size: 2386 bytes --]

From 197de505f546b281cae1530deaccb7f7cf83071f Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:35:37 +0000
Subject: [PATCH 09/11] Add testcase from PR43349.

2010-03-11  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/43349
	* gfortran.dg/graphite/pr43349.f: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157439 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                       |    5 +++
 gcc/testsuite/gfortran.dg/graphite/pr43349.f |   35 ++++++++++++++++++++++++++
 2 files changed, 40 insertions(+), 0 deletions(-)
 create mode 100644 gcc/testsuite/gfortran.dg/graphite/pr43349.f

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index abd9888..331c080 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,10 @@
 2010-03-11  Sebastian Pop  <sebastian.pop@amd.com>
 
+	PR middle-end/43349
+	* gfortran.dg/graphite/pr43349.f: New.
+
+2010-03-11  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-clast-to-gimple.c (my_long_long): Defined.
 	(gcc_type_for_cloog_iv): Use it instead of long_long_integer_type_node.
 	* graphite-sese-to-poly.c (my_long_long): Defined.
diff --git a/gcc/testsuite/gfortran.dg/graphite/pr43349.f b/gcc/testsuite/gfortran.dg/graphite/pr43349.f
new file mode 100644
index 0000000..86e408f
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/graphite/pr43349.f
@@ -0,0 +1,35 @@
+! { dg-options "-O2 -floop-interchange" }
+
+      SUBROUTINE BUG(A,B,X,Y,Z,N)
+      IMPLICIT NONE
+      DOUBLE PRECISION A(*),B(*),X(*),Y(*),Z(*)
+      INTEGER N,J,K
+      K = 0
+      DO J = 1,N
+         K = K+1
+         X(K) = B(J+N*7)
+         Y(K) = B(J+N*8)
+         Z(K) = B(J+N*2)  + A(J+N*2)
+         K = K+1
+         X(K) = B(J+N*3)  + A(J+N*3)
+         Y(K) = B(J+N*9)  + A(J)
+         Z(K) = B(J+N*15)
+         K = K+1
+         X(K) = B(J+N*4)  + A(J+N*4)
+         Y(K) = B(J+N*15)
+         Z(K) = B(J+N*10) + A(J)
+         K = K+1
+         X(K) = B(J+N*11) + A(J+N)
+         Y(K) = B(J+N*5)  + A(J+N*5)
+         Z(K) = B(J+N*16)
+         K = K+1
+         X(K) = B(J+N*16)
+         Y(K) = B(J+N*6)  + A(J+N*6)
+         Z(K) = B(J+N*12) + A(J+N)
+         K = K+1
+         X(K) = B(J+N*13) + A(J+N*2)
+         Y(K) = B(J+N*17)
+         Z(K) = B(J+N*7)  + A(J+N*7)
+      ENDDO
+      RETURN
+      END
-- 
1.6.3.3


[-- Attachment #11: 0010-Fix-PR43354-Correctly-handle-default-definitions.patch --]
[-- Type: text/x-diff, Size: 2815 bytes --]

From 5a644317084d97fc349418e446682721319b5c06 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:35:45 +0000
Subject: [PATCH 10/11] Fix PR43354: Correctly handle default definitions.

2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/43354
	* graphite-sese-to-poly.c (rewrite_close_phi_out_of_ssa): Do not
	call insert_out_of_ssa_copy for default definitions.
	* gfortran.dg/graphite/id-pr43354.f: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157440 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                          |    9 ++++++++-
 gcc/graphite-sese-to-poly.c                     |    3 ++-
 gcc/testsuite/gfortran.dg/graphite/id-pr43354.f |   18 ++++++++++++++++++
 3 files changed, 28 insertions(+), 2 deletions(-)
 create mode 100644 gcc/testsuite/gfortran.dg/graphite/id-pr43354.f

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 331c080..1659de6 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,4 +1,11 @@
-2010-03-11  Sebastian Pop  <sebastian.pop@amd.com>
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43354
+	* graphite-sese-to-poly.c (rewrite_close_phi_out_of_ssa): Do not
+	call insert_out_of_ssa_copy for default definitions.
+	* gfortran.dg/graphite/id-pr43354.f: New.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
 
 	PR middle-end/43349
 	* gfortran.dg/graphite/pr43349.f: New.
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 0f370a8..b0bc348 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -2200,7 +2200,8 @@ rewrite_close_phi_out_of_ssa (gimple_stmt_iterator *psi)
   gimple stmt = gimple_build_assign (res, zero_dim_array);
   tree arg = gimple_phi_arg_def (phi, 0);
 
-  if (TREE_CODE (arg) == SSA_NAME)
+  if (TREE_CODE (arg) == SSA_NAME
+      && !SSA_NAME_IS_DEFAULT_DEF (arg))
     insert_out_of_ssa_copy (zero_dim_array, arg);
   else
     insert_out_of_ssa_copy_on_edge (single_pred_edge (gimple_bb (phi)),
diff --git a/gcc/testsuite/gfortran.dg/graphite/id-pr43354.f b/gcc/testsuite/gfortran.dg/graphite/id-pr43354.f
new file mode 100644
index 0000000..e614f91
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/graphite/id-pr43354.f
@@ -0,0 +1,18 @@
+      SUBROUTINE POFUN2(DIM,GRDENT,FPART,FPARTL)
+      DOUBLE PRECISION GRDENT(*)
+      DOUBLE COMPLEX FPART(*)
+      DOUBLE COMPLEX FPARTL(*)
+      INTEGER REFLCT,XRIREF
+      IF (DIM.GT.1) THEN
+         ABCS3=XRCELL(1)
+         IF (ABCS2.EQ.ABCS3) THEN
+         END IF
+      ELSE
+         DO REFLCT=1,XRIREF,1
+            FPARTL(REFLCT)=FPART(REFLCT)
+         END DO
+      END IF
+         IF (ABCS2.EQ.ABCS3) THEN
+            GRDENT(1)=GRDENT(3)
+         END IF
+      END
-- 
1.6.3.3


[-- Attachment #12: 0011-Add-ChangeLog-entries.patch --]
[-- Type: text/x-diff, Size: 3580 bytes --]

From 6c814203b659574418aab7ea0e4a6c42ff0d5b50 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 13 Mar 2010 17:35:52 +0000
Subject: [PATCH 11/11] Add ChangeLog entries.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@157441 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog           |   59 +++++++++++++++++++++++++++++++++++++++++++++++
 gcc/testsuite/ChangeLog |   15 ++++++++++++
 2 files changed, 74 insertions(+), 0 deletions(-)

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index f56c1e7..4d9c1c1 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,62 @@
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43354
+	* graphite-sese-to-poly.c (rewrite_close_phi_out_of_ssa): Do not
+	call insert_out_of_ssa_copy for default definitions.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-clast-to-gimple.c (my_long_long): Defined.
+	(gcc_type_for_cloog_iv): Use it instead of long_long_integer_type_node.
+	* graphite-sese-to-poly.c (my_long_long): Defined.
+	(scop_ivs_can_be_represented): Use it.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* doc/invoke.texi: Fix documentation of graphite-max-nb-scop-params,
+	graphite-max-bbs-per-function, and loop-block-tile-size.
+	* params.def (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS): Replace "maximal"
+	with "maximum".
+	(PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION): Same.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-clast-to-gimple.c (gcc_type_for_iv_of_clast_loop): Remove
+	forward declaration.
+	* graphite-sese-to-poly.c (reduction_phi_p): Remove FIXME comment.
+	(add_upper_bounds_from_estimated_nit): New.
+	(build_loop_iteration_domains): Use it.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* doc/invoke.texi (PARAM_LOOP_BLOCK_TILE_SIZE): Document.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43306
+	* tree-chrec.c (evolution_function_right_is_integer_cst): CHREC_RIGHT
+	should be an INTEGER_CST.  Also handle CASE_CONVERT.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite.c (graphite_initialize): To bound the number of bbs per
+	function, use PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION.
+	* params.def (PARAM_GRAPHITE_MAX_BBS_PER_FUNCTION): Declared.
+	* doc/invoke.texi: Document it.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (build_poly_scop): Do not return bool.
+	* graphite-sese-to-poly.h (build_poly_scop): Same.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (build_poly_scop): Limit scops following
+	the number of parameters in the scop.  Use as an upper bound
+	PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS.
+	* params.def (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS): Declared.
+	* doc/invoke.texi: Document it.
+
 2010-03-13  Jerry Quinn  <jlquinn@optonline.net>
 
 	* Makefile.in (TEXI_GCCINT_FILES): Remove c-tree.texi.
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 3f5e88b..336e2e3 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,18 @@
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43354
+	* gfortran.dg/graphite/id-pr43354.f: New.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43349
+	* gfortran.dg/graphite/pr43349.f: New.
+
+2010-03-13  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43306
+	* gcc.dg/graphite/pr43306.c: New.
+
 2010-03-12  David S. Miller  <davem@davemloft.net>
 
 	* gcc.dg/lto/20090313_0.c: Add -mcpu=v9 to dg-lto-options on
-- 
1.6.3.3

Re: [patch] Merge to trunk from Graphite branch

[Toon Moene]

Sebastian Pop wrote:

> I just merged the Graphite branch back to trunk.  This solved
> PR43349 for which I just added the testcase, as I already have
> taken care of it in the Graphite branch.

Hmmm, is it just me or ...

If I commpile:

       subroutine sum(a, b, c, n)
       integer i, n
       real a(n), b(n), c(n)
       do i = 1, n
          c(i) = a(i) + b(i)
       enddo
       end

with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get the 
impression that -fgraphite-identity turns off vectorization (which is on 
by default with -O3).

Or am I missing something ...

-- 
Toon Moene - e-mail: toon@moene.org - phone: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
At home: http://moene.org/~toon/
Progress of GNU Fortran: http://gcc.gnu.org/gcc-4.5/changes.html

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Sat, Mar 13, 2010 at 07:15:07PM +0100, Toon Moene wrote:
> Sebastian Pop wrote:
>
>> I just merged the Graphite branch back to trunk.  This solved
>> PR43349 for which I just added the testcase, as I already have
>> taken care of it in the Graphite branch.
>
> Hmmm, is it just me or ...
>
> If I commpile:
>
>       subroutine sum(a, b, c, n)
>       integer i, n
>       real a(n), b(n), c(n)
>       do i = 1, n
>          c(i) = a(i) + b(i)
>       enddo
>       end
>
> with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get the  
> impression that -fgraphite-identity turns off vectorization (which is on  
> by default with -O3).
>
> Or am I missing something ...
>
> -- 
> Toon Moene - e-mail: toon@moene.org - phone: +31 346 214290
> Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
> At home: http://moene.org/~toon/
> Progress of GNU Fortran: http://gcc.gnu.org/gcc-4.5/changes.html

  Speaking of vectorizations, what ever became of the efforts to 
address PR33113?

http://gcc.gnu.org/ml/gcc/2009-08/msg00057.html

The proposed patch that Dorit was impressed by...

http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html

doesn't seem to have gone anywhere.
                    Jack

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Sat, Mar 13, 2010 at 01:29:19PM -0500, Jack Howarth wrote:
> On Sat, Mar 13, 2010 at 07:15:07PM +0100, Toon Moene wrote:
> > Sebastian Pop wrote:
> >
> >> I just merged the Graphite branch back to trunk.  This solved
> >> PR43349 for which I just added the testcase, as I already have
> >> taken care of it in the Graphite branch.
> >
> > Hmmm, is it just me or ...
> >
> > If I commpile:
> >
> >       subroutine sum(a, b, c, n)
> >       integer i, n
> >       real a(n), b(n), c(n)
> >       do i = 1, n
> >          c(i) = a(i) + b(i)
> >       enddo
> >       end
> >
> > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get the  
> > impression that -fgraphite-identity turns off vectorization (which is on  
> > by default with -O3).
> >
> > Or am I missing something ...
> >
> > -- 
> > Toon Moene - e-mail: toon@moene.org - phone: +31 346 214290
> > Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
> > At home: http://moene.org/~toon/
> > Progress of GNU Fortran: http://gcc.gnu.org/gcc-4.5/changes.html
> 
>   Speaking of vectorizations, what ever became of the efforts to 
> address PR33113?
> 
> http://gcc.gnu.org/ml/gcc/2009-08/msg00057.html
> 
> The proposed patch that Dorit was impressed by...
> 
> http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html
> 
> doesn't seem to have gone anywhere.
>                     Jack

   FYI, the reason that I ask is that I just finished doing some
preliminary testing of a i686-apple-darwin10 build of xplor-nih 2.25
using various optimizations. On the positive side, I can build with...

-O3 -ffast-math -funroll-loops -fgraphite-identity -floop-interchange -floop-strip-mine -floop-block -floop-parallelize-all

without compile time or execution regressions compared to...

-O3 -ffast-math -funroll-loops

however on the downside, I don't get any speed improvements for the effort.
                Jack

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
> Sebastian Pop wrote:
>
>> I just merged the Graphite branch back to trunk.  This solved
>> PR43349 for which I just added the testcase, as I already have
>> taken care of it in the Graphite branch.
>
> Hmmm, is it just me or ...
>
> If I commpile:
>
>      subroutine sum(a, b, c, n)
>      integer i, n
>      real a(n), b(n), c(n)
>      do i = 1, n
>         c(i) = a(i) + b(i)
>      enddo
>      end
>
> with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get the
> impression that -fgraphite-identity turns off vectorization (which is on by
> default with -O3).
>
> Or am I missing something ...

The vectorizer should be fixed to handle the code generated by Graphite.
The vectorizer expects to see a very particular form of code, and if
anything differs from that pattern, it won't generate vector code.

Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Sat, Mar 13, 2010 at 12:54, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
>   FYI, the reason that I ask is that I just finished doing some
> preliminary testing of a i686-apple-darwin10 build of xplor-nih 2.25
> using various optimizations. On the positive side, I can build with...
>
> -O3 -ffast-math -funroll-loops -fgraphite-identity -floop-interchange -floop-strip-mine -floop-block -floop-parallelize-all
>
> without compile time or execution regressions compared to...
>
> -O3 -ffast-math -funroll-loops
>
> however on the downside, I don't get any speed improvements for the effort.

Looks like your code was well optimized by hand ;-)

Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Ira Rosen]

gcc-patches-owner@gcc.gnu.org wrote on 13/03/2010 09:19:00 PM:
>
> On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
> > Sebastian Pop wrote:
> >
> >> I just merged the Graphite branch back to trunk.  This solved
> >> PR43349 for which I just added the testcase, as I already have
> >> taken care of it in the Graphite branch.
> >
> > Hmmm, is it just me or ...
> >
> > If I commpile:
> >
> >      subroutine sum(a, b, c, n)
> >      integer i, n
> >      real a(n), b(n), c(n)
> >      do i = 1, n
> >         c(i) = a(i) + b(i)
> >      enddo
> >      end
> >
> > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get the
> > impression that -fgraphite-identity turns off vectorization (which is
on by
> > default with -O3).
> >
> > Or am I missing something ...
>
> The vectorizer should be fixed to handle the code generated by Graphite.
> The vectorizer expects to see a very particular form of code, and if
> anything differs from that pattern, it won't generate vector code.

AFAIR, adding the following two passes cleans the code generated by
GRAPHITE:

*** passes.c
--- passes.c
*************** init_optimization_passes (void)
*** 659,664 ****
--- 659,666 ----
  NEXT_PASS (pass_graphite_transforms);
  NEXT_PASS (pass_iv_canon);
+  NEXT_PASS (pass_copy_prop);
+  NEXT_PASS (pass_dce_loop);
  NEXT_PASS (pass_if_conversion);
  NEXT_PASS (pass_vectorize);

(I think, it was suggested by Richard to solve similar problem, but I can't
find the link).

Ira


>
> Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Sun, Mar 14, 2010 at 08:22:59AM +0200, Ira Rosen wrote:
> 
> 
> gcc-patches-owner@gcc.gnu.org wrote on 13/03/2010 09:19:00 PM:
> >
> > On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
> > > Sebastian Pop wrote:
> > >
> > >> I just merged the Graphite branch back to trunk.  This solved
> > >> PR43349 for which I just added the testcase, as I already have
> > >> taken care of it in the Graphite branch.
> > >
> > > Hmmm, is it just me or ...
> > >
> > > If I commpile:
> > >
> > >      subroutine sum(a, b, c, n)
> > >      integer i, n
> > >      real a(n), b(n), c(n)
> > >      do i = 1, n
> > >         c(i) = a(i) + b(i)
> > >      enddo
> > >      end
> > >
> > > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get the
> > > impression that -fgraphite-identity turns off vectorization (which is
> on by
> > > default with -O3).
> > >
> > > Or am I missing something ...
> >
> > The vectorizer should be fixed to handle the code generated by Graphite.
> > The vectorizer expects to see a very particular form of code, and if
> > anything differs from that pattern, it won't generate vector code.
> 
> AFAIR, adding the following two passes cleans the code generated by
> GRAPHITE:
> 
> *** passes.c
> --- passes.c
> *************** init_optimization_passes (void)
> *** 659,664 ****
> --- 659,666 ----
>   NEXT_PASS (pass_graphite_transforms);
>   NEXT_PASS (pass_iv_canon);
> +  NEXT_PASS (pass_copy_prop);
> +  NEXT_PASS (pass_dce_loop);
>   NEXT_PASS (pass_if_conversion);
>   NEXT_PASS (pass_vectorize);
> 
> (I think, it was suggested by Richard to solve similar problem, but I can't
> find the link).
> 
> Ira
> 
> 
> >
> > Sebastian

Ira,
  I believe you are thinking of Richard's proposed patch to add versioning 
for constant strides for vectorization...

http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html

This patch has severe bit rot against current gcc trunk but would
be nice to get into gcc trunk before gcc 4.5 if possible. I pinged
Richard on this yesterday...

http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00518.html

                  Jack

Re: [patch] Merge to trunk from Graphite branch

[Ira Rosen]

Jack Howarth <howarth@bromo.med.uc.edu> wrote on 14/03/2010 05:30:36 PM:

> On Sun, Mar 14, 2010 at 08:22:59AM +0200, Ira Rosen wrote:
> >
> >
> > gcc-patches-owner@gcc.gnu.org wrote on 13/03/2010 09:19:00 PM:
> > >
> > > On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
> > > > Sebastian Pop wrote:
> > > >
> > > >> I just merged the Graphite branch back to trunk.  This solved
> > > >> PR43349 for which I just added the testcase, as I already have
> > > >> taken care of it in the Graphite branch.
> > > >
> > > > Hmmm, is it just me or ...
> > > >
> > > > If I commpile:
> > > >
> > > >      subroutine sum(a, b, c, n)
> > > >      integer i, n
> > > >      real a(n), b(n), c(n)
> > > >      do i = 1, n
> > > >         c(i) = a(i) + b(i)
> > > >      enddo
> > > >      end
> > > >
> > > > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get
the
> > > > impression that -fgraphite-identity turns off vectorization (which
is
> > on by
> > > > default with -O3).
> > > >
> > > > Or am I missing something ...
> > >
> > > The vectorizer should be fixed to handle the code generated by
Graphite.
> > > The vectorizer expects to see a very particular form of code, and if
> > > anything differs from that pattern, it won't generate vector code.
> >
> > AFAIR, adding the following two passes cleans the code generated by
> > GRAPHITE:
> >
> > *** passes.c
> > --- passes.c
> > *************** init_optimization_passes (void)
> > *** 659,664 ****
> > --- 659,666 ----
> >   NEXT_PASS (pass_graphite_transforms);
> >   NEXT_PASS (pass_iv_canon);
> > +  NEXT_PASS (pass_copy_prop);
> > +  NEXT_PASS (pass_dce_loop);
> >   NEXT_PASS (pass_if_conversion);
> >   NEXT_PASS (pass_vectorize);
> >
> > (I think, it was suggested by Richard to solve similar problem, but I
can't
> > find the link).
> >
> > Ira
> >
> >
> > >
> > > Sebastian
>
> Ira,
>   I believe you are thinking of Richard's proposed patch to add
versioning
> for constant strides for vectorization...
>
> http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html
>
> This patch has severe bit rot against current gcc trunk but would
> be nice to get into gcc trunk before gcc 4.5 if possible. I pinged
> Richard on this yesterday...
>
> http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00518.html

Actually, I don't think that these patches are related.

Ira

>
>                   Jack

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Sun, Mar 14, 2010 at 05:56:22PM +0200, Ira Rosen wrote:
> 
> 
> Jack Howarth <howarth@bromo.med.uc.edu> wrote on 14/03/2010 05:30:36 PM:
> 
> > On Sun, Mar 14, 2010 at 08:22:59AM +0200, Ira Rosen wrote:
> > >
> > >
> > > gcc-patches-owner@gcc.gnu.org wrote on 13/03/2010 09:19:00 PM:
> > > >
> > > > On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
> > > > > Sebastian Pop wrote:
> > > > >
> > > > >> I just merged the Graphite branch back to trunk.  This solved
> > > > >> PR43349 for which I just added the testcase, as I already have
> > > > >> taken care of it in the Graphite branch.
> > > > >
> > > > > Hmmm, is it just me or ...
> > > > >
> > > > > If I commpile:
> > > > >
> > > > >      subroutine sum(a, b, c, n)
> > > > >      integer i, n
> > > > >      real a(n), b(n), c(n)
> > > > >      do i = 1, n
> > > > >         c(i) = a(i) + b(i)
> > > > >      enddo
> > > > >      end
> > > > >
> > > > > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get
> the
> > > > > impression that -fgraphite-identity turns off vectorization (which
> is
> > > on by
> > > > > default with -O3).
> > > > >
> > > > > Or am I missing something ...
> > > >
> > > > The vectorizer should be fixed to handle the code generated by
> Graphite.
> > > > The vectorizer expects to see a very particular form of code, and if
> > > > anything differs from that pattern, it won't generate vector code.
> > >
> > > AFAIR, adding the following two passes cleans the code generated by
> > > GRAPHITE:
> > >
> > > *** passes.c
> > > --- passes.c
> > > *************** init_optimization_passes (void)
> > > *** 659,664 ****
> > > --- 659,666 ----
> > >   NEXT_PASS (pass_graphite_transforms);
> > >   NEXT_PASS (pass_iv_canon);
> > > +  NEXT_PASS (pass_copy_prop);
> > > +  NEXT_PASS (pass_dce_loop);
> > >   NEXT_PASS (pass_if_conversion);
> > >   NEXT_PASS (pass_vectorize);
> > >
> > > (I think, it was suggested by Richard to solve similar problem, but I
> can't
> > > find the link).
> > >
> > > Ira
> > >
> > >
> > > >
> > > > Sebastian
> >
> > Ira,
> >   I believe you are thinking of Richard's proposed patch to add
> versioning
> > for constant strides for vectorization...
> >
> > http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html
> >
> > This patch has severe bit rot against current gcc trunk but would
> > be nice to get into gcc trunk before gcc 4.5 if possible. I pinged
> > Richard on this yesterday...
> >
> > http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00518.html
> 
> Actually, I don't think that these patches are related.
> 
> Ira

Ira,
    However they share the common change to gcc/passes.c, right? The
stride patch would have the advantage of eliminating the omission of
the current vectorizations while finding new vectorizations as well.
                  Jack
> 
> >
> >                   Jack

Re: [patch] Merge to trunk from Graphite branch

[Richard Guenther]

On Sun, Mar 14, 2010 at 5:28 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> On Sun, Mar 14, 2010 at 05:56:22PM +0200, Ira Rosen wrote:
>>
>>
>> Jack Howarth <howarth@bromo.med.uc.edu> wrote on 14/03/2010 05:30:36 PM:
>>
>> > On Sun, Mar 14, 2010 at 08:22:59AM +0200, Ira Rosen wrote:
>> > >
>> > >
>> > > gcc-patches-owner@gcc.gnu.org wrote on 13/03/2010 09:19:00 PM:
>> > > >
>> > > > On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
>> > > > > Sebastian Pop wrote:
>> > > > >
>> > > > >> I just merged the Graphite branch back to trunk.  This solved
>> > > > >> PR43349 for which I just added the testcase, as I already have
>> > > > >> taken care of it in the Graphite branch.
>> > > > >
>> > > > > Hmmm, is it just me or ...
>> > > > >
>> > > > > If I commpile:
>> > > > >
>> > > > >      subroutine sum(a, b, c, n)
>> > > > >      integer i, n
>> > > > >      real a(n), b(n), c(n)
>> > > > >      do i = 1, n
>> > > > >         c(i) = a(i) + b(i)
>> > > > >      enddo
>> > > > >      end
>> > > > >
>> > > > > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get
>> the
>> > > > > impression that -fgraphite-identity turns off vectorization (which
>> is
>> > > on by
>> > > > > default with -O3).
>> > > > >
>> > > > > Or am I missing something ...
>> > > >
>> > > > The vectorizer should be fixed to handle the code generated by
>> Graphite.
>> > > > The vectorizer expects to see a very particular form of code, and if
>> > > > anything differs from that pattern, it won't generate vector code.
>> > >
>> > > AFAIR, adding the following two passes cleans the code generated by
>> > > GRAPHITE:
>> > >
>> > > *** passes.c
>> > > --- passes.c
>> > > *************** init_optimization_passes (void)
>> > > *** 659,664 ****
>> > > --- 659,666 ----
>> > >   NEXT_PASS (pass_graphite_transforms);
>> > >   NEXT_PASS (pass_iv_canon);
>> > > +  NEXT_PASS (pass_copy_prop);
>> > > +  NEXT_PASS (pass_dce_loop);
>> > >   NEXT_PASS (pass_if_conversion);
>> > >   NEXT_PASS (pass_vectorize);
>> > >
>> > > (I think, it was suggested by Richard to solve similar problem, but I
>> can't
>> > > find the link).
>> > >
>> > > Ira
>> > >
>> > >
>> > > >
>> > > > Sebastian
>> >
>> > Ira,
>> >   I believe you are thinking of Richard's proposed patch to add
>> versioning
>> > for constant strides for vectorization...
>> >
>> > http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html
>> >
>> > This patch has severe bit rot against current gcc trunk but would
>> > be nice to get into gcc trunk before gcc 4.5 if possible. I pinged
>> > Richard on this yesterday...
>> >
>> > http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00518.html
>>
>> Actually, I don't think that these patches are related.
>>
>> Ira
>
> Ira,
>    However they share the common change to gcc/passes.c, right? The
> stride patch would have the advantage of eliminating the omission of
> the current vectorizations while finding new vectorizations as well.

Unconditionally adding two passes will just slow down compile.

Richard.

>                  Jack
>>
>> >
>> >                   Jack
>

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Mon, Mar 15, 2010 at 11:38:28AM +0100, Richard Guenther wrote:
> On Sun, Mar 14, 2010 at 5:28 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> > On Sun, Mar 14, 2010 at 05:56:22PM +0200, Ira Rosen wrote:
> >>
> >>
> >> Jack Howarth <howarth@bromo.med.uc.edu> wrote on 14/03/2010 05:30:36 PM:
> >>
> >> > On Sun, Mar 14, 2010 at 08:22:59AM +0200, Ira Rosen wrote:
> >> > >
> >> > >
> >> > > gcc-patches-owner@gcc.gnu.org wrote on 13/03/2010 09:19:00 PM:
> >> > > >
> >> > > > On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
> >> > > > > Sebastian Pop wrote:
> >> > > > >
> >> > > > >> I just merged the Graphite branch back to trunk.  This solved
> >> > > > >> PR43349 for which I just added the testcase, as I already have
> >> > > > >> taken care of it in the Graphite branch.
> >> > > > >
> >> > > > > Hmmm, is it just me or ...
> >> > > > >
> >> > > > > If I commpile:
> >> > > > >
> >> > > > >      subroutine sum(a, b, c, n)
> >> > > > >      integer i, n
> >> > > > >      real a(n), b(n), c(n)
> >> > > > >      do i = 1, n
> >> > > > >         c(i) = a(i) + b(i)
> >> > > > >      enddo
> >> > > > >      end
> >> > > > >
> >> > > > > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get
> >> the
> >> > > > > impression that -fgraphite-identity turns off vectorization (which
> >> is
> >> > > on by
> >> > > > > default with -O3).
> >> > > > >
> >> > > > > Or am I missing something ...
> >> > > >
> >> > > > The vectorizer should be fixed to handle the code generated by
> >> Graphite.
> >> > > > The vectorizer expects to see a very particular form of code, and if
> >> > > > anything differs from that pattern, it won't generate vector code.
> >> > >
> >> > > AFAIR, adding the following two passes cleans the code generated by
> >> > > GRAPHITE:
> >> > >
> >> > > *** passes.c
> >> > > --- passes.c
> >> > > *************** init_optimization_passes (void)
> >> > > *** 659,664 ****
> >> > > --- 659,666 ----
> >> > >   NEXT_PASS (pass_graphite_transforms);
> >> > >   NEXT_PASS (pass_iv_canon);
> >> > > +  NEXT_PASS (pass_copy_prop);
> >> > > +  NEXT_PASS (pass_dce_loop);
> >> > >   NEXT_PASS (pass_if_conversion);
> >> > >   NEXT_PASS (pass_vectorize);
> >> > >
> >> > > (I think, it was suggested by Richard to solve similar problem, but I
> >> can't
> >> > > find the link).
> >> > >
> >> > > Ira
> >> > >
> >> > >
> >> > > >
> >> > > > Sebastian
> >> >
> >> > Ira,
> >> >   I believe you are thinking of Richard's proposed patch to add
> >> versioning
> >> > for constant strides for vectorization...
> >> >
> >> > http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html
> >> >
> >> > This patch has severe bit rot against current gcc trunk but would
> >> > be nice to get into gcc trunk before gcc 4.5 if possible. I pinged
> >> > Richard on this yesterday...
> >> >
> >> > http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00518.html
> >>
> >> Actually, I don't think that these patches are related.
> >>
> >> Ira
> >
> > Ira,
> >    However they share the common change to gcc/passes.c, right? The
> > stride patch would have the advantage of eliminating the omission of
> > the current vectorizations while finding new vectorizations as well.
> 
> Unconditionally adding two passes will just slow down compile.
> 
> Richard.

Richard,
    Dominique Dhumieres tried adding those two additional passes and
they didn't help -fgraphite-identity stop dropping vectorizations.
Shouldn't PR43359 be a P1 since this represents a regression from
gcc 4.4?
                    Jack
> 
> >                  Jack
> >>
> >> >
> >> >                   Jack
> >

Re: [patch] Merge to trunk from Graphite branch

[Richard Guenther]

On Mon, Mar 15, 2010 at 2:12 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> On Mon, Mar 15, 2010 at 11:38:28AM +0100, Richard Guenther wrote:
>> On Sun, Mar 14, 2010 at 5:28 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
>> > On Sun, Mar 14, 2010 at 05:56:22PM +0200, Ira Rosen wrote:
>> >>
>> >>
>> >> Jack Howarth <howarth@bromo.med.uc.edu> wrote on 14/03/2010 05:30:36 PM:
>> >>
>> >> > On Sun, Mar 14, 2010 at 08:22:59AM +0200, Ira Rosen wrote:
>> >> > >
>> >> > >
>> >> > > gcc-patches-owner@gcc.gnu.org wrote on 13/03/2010 09:19:00 PM:
>> >> > > >
>> >> > > > On Sat, Mar 13, 2010 at 12:15, Toon Moene <toon@moene.org> wrote:
>> >> > > > > Sebastian Pop wrote:
>> >> > > > >
>> >> > > > >> I just merged the Graphite branch back to trunk.  This solved
>> >> > > > >> PR43349 for which I just added the testcase, as I already have
>> >> > > > >> taken care of it in the Graphite branch.
>> >> > > > >
>> >> > > > > Hmmm, is it just me or ...
>> >> > > > >
>> >> > > > > If I commpile:
>> >> > > > >
>> >> > > > >      subroutine sum(a, b, c, n)
>> >> > > > >      integer i, n
>> >> > > > >      real a(n), b(n), c(n)
>> >> > > > >      do i = 1, n
>> >> > > > >         c(i) = a(i) + b(i)
>> >> > > > >      enddo
>> >> > > > >      end
>> >> > > > >
>> >> > > > > with -g -O3 -S and with -g -O3 -fgraphite-identity I strongly get
>> >> the
>> >> > > > > impression that -fgraphite-identity turns off vectorization (which
>> >> is
>> >> > > on by
>> >> > > > > default with -O3).
>> >> > > > >
>> >> > > > > Or am I missing something ...
>> >> > > >
>> >> > > > The vectorizer should be fixed to handle the code generated by
>> >> Graphite.
>> >> > > > The vectorizer expects to see a very particular form of code, and if
>> >> > > > anything differs from that pattern, it won't generate vector code.
>> >> > >
>> >> > > AFAIR, adding the following two passes cleans the code generated by
>> >> > > GRAPHITE:
>> >> > >
>> >> > > *** passes.c
>> >> > > --- passes.c
>> >> > > *************** init_optimization_passes (void)
>> >> > > *** 659,664 ****
>> >> > > --- 659,666 ----
>> >> > >   NEXT_PASS (pass_graphite_transforms);
>> >> > >   NEXT_PASS (pass_iv_canon);
>> >> > > +  NEXT_PASS (pass_copy_prop);
>> >> > > +  NEXT_PASS (pass_dce_loop);
>> >> > >   NEXT_PASS (pass_if_conversion);
>> >> > >   NEXT_PASS (pass_vectorize);
>> >> > >
>> >> > > (I think, it was suggested by Richard to solve similar problem, but I
>> >> can't
>> >> > > find the link).
>> >> > >
>> >> > > Ira
>> >> > >
>> >> > >
>> >> > > >
>> >> > > > Sebastian
>> >> >
>> >> > Ira,
>> >> >   I believe you are thinking of Richard's proposed patch to add
>> >> versioning
>> >> > for constant strides for vectorization...
>> >> >
>> >> > http://gcc.gnu.org/ml/gcc-patches/2009-01/msg01174.html
>> >> >
>> >> > This patch has severe bit rot against current gcc trunk but would
>> >> > be nice to get into gcc trunk before gcc 4.5 if possible. I pinged
>> >> > Richard on this yesterday...
>> >> >
>> >> > http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00518.html
>> >>
>> >> Actually, I don't think that these patches are related.
>> >>
>> >> Ira
>> >
>> > Ira,
>> >    However they share the common change to gcc/passes.c, right? The
>> > stride patch would have the advantage of eliminating the omission of
>> > the current vectorizations while finding new vectorizations as well.
>>
>> Unconditionally adding two passes will just slow down compile.
>>
>> Richard.
>
> Richard,
>    Dominique Dhumieres tried adding those two additional passes and
> they didn't help -fgraphite-identity stop dropping vectorizations.
> Shouldn't PR43359 be a P1 since this represents a regression from
> gcc 4.4?

No.

Richard.

>                    Jack
>>
>> >                  Jack
>> >>
>> >> >
>> >> >                   Jack
>> >
>

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Mon, Mar 15, 2010 at 02:32:58PM +0100, Richard Guenther wrote:
> On Mon, Mar 15, 2010 at 2:12 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> > Richard,
> >    Dominique Dhumieres tried adding those two additional passes and
> > they didn't help -fgraphite-identity stop dropping vectorizations.
> > Shouldn't PR43359 be a P1 since this represents a regression from
> > gcc 4.4?
> 
> No.
> 
> Richard.

Richard,
    What exactly is the road-map for graphite to be enabled by
default in FSF gcc? It seems that vectorization improvements
have been treading water lately and that the biggest win so far
for 4.5 has been HJ Lu's update of the default x86 arch. Hopefully
we won't end up having to rely on milepost for future FSF gcc
code performance improvements.
> >                    Jack

Re: [patch] Merge to trunk from Graphite branch

[Richard Guenther]

On Mon, Mar 15, 2010 at 2:46 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> On Mon, Mar 15, 2010 at 02:32:58PM +0100, Richard Guenther wrote:
>> On Mon, Mar 15, 2010 at 2:12 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
>> > Richard,
>> >    Dominique Dhumieres tried adding those two additional passes and
>> > they didn't help -fgraphite-identity stop dropping vectorizations.
>> > Shouldn't PR43359 be a P1 since this represents a regression from
>> > gcc 4.4?
>>
>> No.
>>
>> Richard.
>
> Richard,
>    What exactly is the road-map for graphite to be enabled by
> default in FSF gcc? It seems that vectorization improvements
> have been treading water lately and that the biggest win so far
> for 4.5 has been HJ Lu's update of the default x86 arch. Hopefully
> we won't end up having to rely on milepost for future FSF gcc
> code performance improvements.

Sorry, but the biggest win for 4.5 is the various improvements
SUSE / AMD has contributed to improve SPEC CPU 2006 scores
(7% overall SPEC FP improvement).

Updating the default x86 arch does bring you nothing (well, if
you're not clueless in case you shouldn't build GCC yourself).

As far as I know Graphite does not bring any performance advantages
yet, so I see no reason to enable it by default.

If you want a specific benchmark to run faster you have to start
analyzing why it is slow (it helps to have a competing compiler
generate faster code), produce a testcase and at least file
a missed-optimization bugreport.  Of course even better if you
can think of a solution and implement it.

I am not aware of any major vectorization missed-optimizations
vs for example the Intel compiler for SPEC CPU 2006.

Richard.

>> >                    Jack
>

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Mon, Mar 15, 2010 at 09:11, Richard Guenther
<richard.guenther@gmail.com> wrote:
> As far as I know Graphite does not bring any performance advantages
> yet, so I see no reason to enable it by default.

Agreed.  There is no point to enable Graphite by default in GCC4.5.
We may reconsider this for 4.6 if by then Graphite will show some
performance improvements.

Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Mon, Mar 15, 2010 at 03:11:21PM +0100, Richard Guenther wrote:
> 
> Sorry, but the biggest win for 4.5 is the various improvements
> SUSE / AMD has contributed to improve SPEC CPU 2006 scores
> (7% overall SPEC FP improvement).
> 
> Updating the default x86 arch does bring you nothing (well, if
> you're not clueless in case you shouldn't build GCC yourself).

   I mentioned that because of the 2% improvement in the Polyhedron
2005 benchmarks which coincided with the x86 arch changes which
enabled -msse2 across the board...

http://gcc.gnu.org/ml/gcc/2010-03/msg00075.html

> 
> As far as I know Graphite does not bring any performance advantages
> yet, so I see no reason to enable it by default.
> 
> If you want a specific benchmark to run faster you have to start
> analyzing why it is slow (it helps to have a competing compiler
> generate faster code), produce a testcase and at least file
> a missed-optimization bugreport.  Of course even better if you
> can think of a solution and implement it.
> 
> I am not aware of any major vectorization missed-optimizations
> vs for example the Intel compiler for SPEC CPU 2006.

    Don't we still lack the ability to vectorize the cases described
here?

http://gcc.gnu.org/ml/gcc/2009-08/msg00057.html

Perhaps current graphite already would handle some of those if the
vectorization wasn't getting accidentally lost.
                     Jack


> 
> Richard.
> 
> >> >                    Jack
> >

Re: [patch] Merge to trunk from Graphite branch

[Richard Guenther]

On Mon, Mar 15, 2010 at 4:12 PM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> On Mon, Mar 15, 2010 at 03:11:21PM +0100, Richard Guenther wrote:
>>
>> Sorry, but the biggest win for 4.5 is the various improvements
>> SUSE / AMD has contributed to improve SPEC CPU 2006 scores
>> (7% overall SPEC FP improvement).
>>
>> Updating the default x86 arch does bring you nothing (well, if
>> you're not clueless in case you shouldn't build GCC yourself).
>
>   I mentioned that because of the 2% improvement in the Polyhedron
> 2005 benchmarks which coincided with the x86 arch changes which
> enabled -msse2 across the board...
>
> http://gcc.gnu.org/ml/gcc/2010-03/msg00075.html
>
>>
>> As far as I know Graphite does not bring any performance advantages
>> yet, so I see no reason to enable it by default.
>>
>> If you want a specific benchmark to run faster you have to start
>> analyzing why it is slow (it helps to have a competing compiler
>> generate faster code), produce a testcase and at least file
>> a missed-optimization bugreport.  Of course even better if you
>> can think of a solution and implement it.
>>
>> I am not aware of any major vectorization missed-optimizations
>> vs for example the Intel compiler for SPEC CPU 2006.
>
>    Don't we still lack the ability to vectorize the cases described
> here?
>
> http://gcc.gnu.org/ml/gcc/2009-08/msg00057.html
>
> Perhaps current graphite already would handle some of those if the
> vectorization wasn't getting accidentally lost.

This case doesn't happen in hot areas of SPEC CPU 2006.

Richard.

Re: [patch] Merge to trunk from Graphite branch

[Tobias Burnus]

On 03/15/2010 04:12 PM, Jack Howarth wrote:
> On Mon, Mar 15, 2010 at 03:11:21PM +0100, Richard Guenther wrote:
>> Updating the default x86 arch does bring you nothing (well, if
>> you're not clueless in case you shouldn't build GCC yourself).
> 
>    I mentioned that because of the 2% improvement in the Polyhedron
> 2005 benchmarks which coincided with the x86 arch changes which
> enabled -msse2 across the board...
> 
> http://gcc.gnu.org/ml/gcc/2010-03/msg00075.html

I somehow doubt that the changing of the default has anything to do with
it. I could well imagine that Toon uses -march=native or similar.

And he wrote "last week" thus I do not see which one can really call it
"coincide"; I think during that time there was at least one front-end
patch which removed some unnecessarily generated array temporaries which
can be very profitable in a hot loop (but had no visible influence on
the Polyhedron runs). Maybe also some middle-end fix helped with a hot
loop in Hirlam.

>> As far as I know Graphite does not bring any performance advantages
>> yet, so I see no reason to enable it by default.

I agree; I think there are cases, where it is very profitable, but
others, where it makes the program slower (cf. PR 38846). I hope that
graphite in 4.6 will be suitable for enabling it by default. :-)

>> I am not aware of any major vectorization missed-optimizations
>> vs for example the Intel compiler for SPEC CPU 2006.

Is there a place where one can find results for recent GCC and recent
versions of the Intel compiler for the same system? I have frankly no
idea how GCC and Intel compare on SPEC CPU 2000/2006 (or, for that
matter, how GCC and IBM's XL compiler compare).

(For the Polyhedron benchmark, Polyhedron itself shows that on geometric
average GCC 4.4 and 4.3 are between 6% to 17% slower than the latest
11.1 Intel compiler [1]; on my dated AMD it's about 9% w/ GCC 4.5 vs
ifort 11.1 and 6% w/ GCC 4.5 vs ifort 9.1; GCC gains 2% with -flto
-fwhole-program -fno-protect-parens. And pathscale is 2 per-cent points
faster than ifort 11.1 (i.e. 11% w.r.t. GCC).)

Tobias

[1] http://www.polyhedron.co.uk/compare0html

Re: [patch] Merge to trunk from Graphite branch

[Richard Guenther]

On Mon, Mar 15, 2010 at 5:22 PM, Tobias Burnus <burnus@net-b.de>
wrote:
> On 03/15/2010 04:12 PM, Jack Howarth wrote:
>> On Mon, Mar 15, 2010 at 03:11:21PM +0100, Richard Guenther wrote:
>>> Updating the default x86 arch does bring you nothing (well, if
>>> you're not clueless in case you shouldn't build GCC yourself).
>>
>>    I mentioned that because of the 2% improvement in the Polyhedron
>> 2005 benchmarks which coincided with the x86 arch changes which
>> enabled -msse2 across the board...
>>
>> http://gcc.gnu.org/ml/gcc/2010-03/msg00075.html
>
> I somehow doubt that the changing of the default has anything to do with
> it. I could well imagine that Toon uses -march=native or similar.
>
> And he wrote "last week" thus I do not see which one can really call it
> "coincide"; I think during that time there was at least one front-end
> patch which removed some unnecessarily generated array temporaries which
> can be very profitable in a hot loop (but had no visible influence on
> the Polyhedron runs). Maybe also some middle-end fix helped with a hot
> loop in Hirlam.
>
>>> As far as I know Graphite does not bring any performance advantages
>>> yet, so I see no reason to enable it by default.
>
> I agree; I think there are cases, where it is very profitable, but
> others, where it makes the program slower (cf. PR 38846). I hope that
> graphite in 4.6 will be suitable for enabling it by default. :-)
>
>>> I am not aware of any major vectorization missed-optimizations
>>> vs for example the Intel compiler for SPEC CPU 2006.
>
> Is there a place where one can find results for recent GCC and recent
> versions of the Intel compiler for the same system? I have frankly no
> idea how GCC and Intel compare on SPEC CPU 2000/2006 (or, for that
> matter, how GCC and IBM's XL compiler compare).
>
> (For the Polyhedron benchmark, Polyhedron itself shows that on geometric
> average GCC 4.4 and 4.3 are between 6% to 17% slower than the latest
> 11.1 Intel compiler [1]; on my dated AMD it's about 9% w/ GCC 4.5 vs
> ifort 11.1 and 6% w/ GCC 4.5 vs ifort 9.1; GCC gains 2% with -flto
> -fwhole-program -fno-protect-parens. And pathscale is 2 per-cent points
> faster than ifort 11.1 (i.e. 11% w.r.t. GCC).)

Note that a lot of the difference can be due to optimized math
routines that the Intel compiler uses by default.  You can eliminate
(most) of that difference by LD_PRELOADing its libimf.so also
for the GCC built executables.

And of course it's always worth to analyze the reason for the slowness
on the testcase with the largest difference.

Richard.

> Tobias
>
> [1] http://www.polyhedron.co.uk/compare0html
>

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 300 bytes --]

On Mon, Mar 15, 2010 at 05:38, Richard Guenther
<richard.guenther@gmail.com> wrote:
> Unconditionally adding two passes will just slow down compile.

The patch from Ira would not apply: here is an updated patch,
that anyways would execute conditionally based on running
the graphite pass.

Sebastian

[-- Attachment #2: 0001-Add-pass_copy_prop-after-Graphite.patch --]
[-- Type: text/x-diff, Size: 682 bytes --]

From edcd395abce6e7f59b4496794cc6da5f9b3fb3db Mon Sep 17 00:00:00 2001
From: Sebastian Pop <sebpop@gmail.com>
Date: Mon, 15 Mar 2010 03:25:42 -0500
Subject: [PATCH 1/2] Add pass_copy_prop after Graphite.

---
 gcc/passes.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/gcc/passes.c b/gcc/passes.c
index 9d0c55a..1ac8694 100644
--- a/gcc/passes.c
+++ b/gcc/passes.c
@@ -897,6 +897,7 @@ 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);
 	    }
-- 
1.6.3.3

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Mon, Mar 15, 2010 at 09:47:14AM -0500, Sebastian Pop wrote:
> On Mon, Mar 15, 2010 at 05:38, Richard Guenther
> <richard.guenther@gmail.com> wrote:
> > Unconditionally adding two passes will just slow down compile.
> 
> The patch from Ira would not apply: here is an updated patch,
> that anyways would execute conditionally based on running
> the graphite pass.
> 
> Sebastian

> From edcd395abce6e7f59b4496794cc6da5f9b3fb3db Mon Sep 17 00:00:00 2001
> From: Sebastian Pop <sebpop@gmail.com>
> Date: Mon, 15 Mar 2010 03:25:42 -0500
> Subject: [PATCH 1/2] Add pass_copy_prop after Graphite.
> 
> ---
>  gcc/passes.c |    1 +
>  1 files changed, 1 insertions(+), 0 deletions(-)
> 
> diff --git a/gcc/passes.c b/gcc/passes.c
> index 9d0c55a..1ac8694 100644
> --- a/gcc/passes.c
> +++ b/gcc/passes.c
> @@ -897,6 +897,7 @@ 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);
>  	    }
> -- 
> 1.6.3.3
> 

Sebastian,
    Does this patch resolve the missed vectorizations in
any of the proposed testcases with -fgraphite-identity
(such as that proposed in http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43359#c4)?
                    Jack

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Mon, Mar 15, 2010 at 12:21, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> Sebastian,
>    Does this patch resolve the missed vectorizations in
> any of the proposed testcases with -fgraphite-identity
> (such as that proposed in http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43359#c4)?

I have no idea yet, I will have to dig into the code and see
why the vectorization cannot handle the code after graphite.
I will give it a look in the next days.

Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Toon Moene]

[ Somewhat belatedly, as I was in Oslo for the work-for-hire
   experience (meteorology) last week ]

Tobias Burnus wrote:

> On 03/15/2010 04:12 PM, Jack Howarth wrote:

>> On Mon, Mar 15, 2010 at 03:11:21PM +0100, Richard Guenther wrote:
>>> Updating the default x86 arch does bring you nothing (well, if
>>> you're not clueless in case you shouldn't build GCC yourself).
>>    I mentioned that because of the 2% improvement in the Polyhedron
>> 2005 benchmarks which coincided with the x86 arch changes which
>> enabled -msse2 across the board...
>>
>> http://gcc.gnu.org/ml/gcc/2010-03/msg00075.html
> 
> I somehow doubt that the changing of the default has anything to do with
> it. I could well imagine that Toon uses -march=native or similar.

I changed to using --with-arch-64=native --with-tune-64=native instead 
of --with-arch-64=core2 --with-tune-64=core2 after the default was 
changed.  Before that the --with-{blah}=native wasn't accepted.

This of course (I hope :-) affected the build of the various run-time 
libraries and hence the performance of code ...

At least, that was what *I* thought had brought me 2 % performance 
improvement.

But as Richard Guenther writes: Really figuring out what 
enables/inhibits performance improvements is *hard work*, and has to be 
balanced against other hard work (like fixing P1 bugs :-)

[ As an aside: the representative of Intel on the Fortran
   Standardization Committee was aghast that the Polyhedron suite
   "suddenly" started to be important in evaluating the performance of
   compiler-generated code, because that meant they had to track *yet
   another* suite of programs for performance regressions.

   You know, I agreed ... ]

Cheers,

-- 
Toon Moene - e-mail: toon@moene.org - phone: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
At home: http://moene.org/~toon/
Progress of GNU Fortran: http://gcc.gnu.org/gcc-4.5/changes.html

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Tue, Mar 16, 2010 at 09:38, Michael Matz <matz@suse.de> wrote:
> Hi,
>
> On Tue, 16 Mar 2010, Sebastian Pop wrote:
>
>> > In Toon's testcase from
>> > http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00515.html vectorization fails
>> > because scev analysis failure: "evolution of offset is not affine." It
>> > probably gets confused by
>> >
>> >  # graphite_IV.22_66 = PHI <0(4), graphite_IV.22_67(6)>
>> >  var.23_77 = (character(kind=4)) graphite_IV.22_66;
>> >  D.1214_68 = (integer(kind=4)) var.23_77;
>> >  i_69 = D.1214_68 + 1;
>> >  D.1192_70 = (integer(kind=8)) i_69;
>> >  D.1193_71 = D.1192_70 + -1;
>> >  D.1194_72 = (*a_39(D))[D.1193_71];
>> >
>> >  instead (without Graphite)
>> >
>> >  # i_1 = PHI <1(3), i_48(5)>
>> >  D.1192_37 = (integer(kind=8)) i_1;
>> >  D.1193_38 = D.1192_37 + -1;
>> >  D.1194_40 = (*a_39(D))[D.1193_38];
>> >
>>
>> Looks like the scev analysis fails because of too many casts
>> on the way, otherwise I see no particular difficulty in the code
>> generated by Graphite.
>
> Both sequences aren't equivalent, so my guess would be that the
> transformation done by graphite actually is buggy or at least suboptimal.
> The difference lies in the overflow behaviour.  Without graphite, i_1 has
> undefined overflow, as D.1192_37 too, hence we can look through the
> conversion.  With graphite var.23_77 (of unsigned type) has defined
> overflow semantics, hence we can't look through that conversion.
>

Right, the new induction variable that we create does not maintain the
signedness of the original IV.  This is due to IV canonicalization
that systematically creates unsigned IVs.  The usual example of why we
need to generate an unsigned IV is due to the fact that with signed
IVs we can generate a loop that rolls for MAX iterations, with MAX the
maximal value of the unsigned type:

  char i;
  for (i = -127; i < 128; i++)
    S;

This loop runs for 255 iterations, and thus in the IV canonical form,
that is one IV per loop, IV->base = 0, and IV->step = 1, the type of
the IV should be unsigned char.

The fix that I am proposing to preserve the signedness of the original
IVs, is to remove the need for the IV canonicalization.  I am
preparing a set of patches for doing that.

Sebastian

Re: [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

Re: [patch] Merge to trunk from Graphite branch

[Jack Howarth]

On Tue, Mar 16, 2010 at 06:57:07PM -0500, Sebastian Pop wrote:
> On Tue, Mar 16, 2010 at 17:26, Dominique Dhumieres <dominiq@lps.ens.fr> 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?
> 
> On trunk 157499 I also see the exact same thing:
> 
> $ grep "LOOP VECTORIZED" before.txt | wc
>      24      96    1848
> 
> $ grep "LOOP VECTORIZED" after.txt | wc  # changes to canonicalize_loop_ivs
>      26     104    2001
> 
> $ grep "LOOP VECTORIZED" after1.txt | wc # add pass_copy_prop
>      33     132    2533
> 
> Also on the Graphite branch I tried to disable all the loop transforms
> that could have added more loops, but I get the same thing:
> 
> $ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
> -fgraphite-identity -fno-loop-block -fno-loop-interchange
> -fno-loop-strip-mine -fno-graphite gas_dyn.f90 &> after1.txt
> $ grep "LOOP VECTORIZED" after1.txt | wc
>      40     160    3065
> 
> $ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
> -fno-graphite-identity -fno-loop-block -fno-loop-interchange
> -fno-loop-strip-mine -fno-graphite
> ~/gcc/bench/pb05/lin/source/gas_dyn.f90 &> after2.txt
> $ grep "LOOP VECTORIZED" after2.txt | wc
>      41     164    3142
> 
> I will look at the diff between the Graphite branch and trunk to see
> why we have this difference.

Or remerge trunk back into graphite branch and see if the difference
starts to appear there as well.

> 
> Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Tue, Mar 16, 2010 at 17:26, Dominique Dhumieres <dominiq@lps.ens.fr> 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?

On trunk 157499 I also see the exact same thing:

$ grep "LOOP VECTORIZED" before.txt | wc
     24      96    1848

$ grep "LOOP VECTORIZED" after.txt | wc  # changes to canonicalize_loop_ivs
     26     104    2001

$ grep "LOOP VECTORIZED" after1.txt | wc # add pass_copy_prop
     33     132    2533

Also on the Graphite branch I tried to disable all the loop transforms
that could have added more loops, but I get the same thing:

$ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
-fgraphite-identity -fno-loop-block -fno-loop-interchange
-fno-loop-strip-mine -fno-graphite gas_dyn.f90 &> after1.txt
$ grep "LOOP VECTORIZED" after1.txt | wc
     40     160    3065

$ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
-fno-graphite-identity -fno-loop-block -fno-loop-interchange
-fno-loop-strip-mine -fno-graphite
~/gcc/bench/pb05/lin/source/gas_dyn.f90 &> after2.txt
$ grep "LOOP VECTORIZED" after2.txt | wc
     41     164    3142

I will look at the diff between the Graphite branch and trunk to see
why we have this difference.

Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Dominique Dhumieres]

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

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 594 bytes --]

On Tue, Mar 16, 2010 at 14:29, Dominique Dhumieres <dominiq@lps.ens.fr> wrote:
> With your patch (with the previous 2 additional passes from previous
> attempt and lim) I have only 33 vectorized loops instead of 41 without
> -fgraphite*. The timing is going from ~10.8s to 6.5s compared to a
> reference time of 4.7s. Still investigating.
>

I got 40 loops vectorized with Graphite:
./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2 -fgraphite-identity
gas_dyn.f90
and 41 loops vectorized without Graphite.

I was using the attached patch on top of the Graphite branch in my tests.

Sebastian

[-- Attachment #2: 0001-Add-pass_copy_prop-after-Graphite.patch --]
[-- Type: text/x-diff, Size: 682 bytes --]

From edcd395abce6e7f59b4496794cc6da5f9b3fb3db Mon Sep 17 00:00:00 2001
From: Sebastian Pop <sebpop@gmail.com>
Date: Mon, 15 Mar 2010 03:25:42 -0500
Subject: [PATCH 1/3] Add pass_copy_prop after Graphite.

---
 gcc/passes.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/gcc/passes.c b/gcc/passes.c
index 9d0c55a..1ac8694 100644
--- a/gcc/passes.c
+++ b/gcc/passes.c
@@ -897,6 +897,7 @@ 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);
 	    }
-- 
1.6.3.3

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Tue, Mar 16, 2010 at 13:40, Sebastian Pop <sebpop@gmail.com> wrote:
> With the attached patch we now get:
>
> $ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
> -fgraphite-identity gas_dyn.f90 &> before.txt
> $ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
> -fgraphite-identity gas_dyn.f90 &> after.txt
>
> $ grep "LOOP VECTORIZED" before.txt | wc
>     24      96    1848
> $ grep "LOOP VECTORIZED" after.txt | wc
>     40     160    3065
>
> I will test this patch on amd64-linux.
> Ok for trunk if it passes bootstrap and test?
>

Here is the diff for make check RUNTESTFLAGS=vect.exp
without and with the patch in the Graphite branch, where
Graphite is enabled in -O2 and above:

$ diff -u before.txt after.txt | grep "^-FAIL" | wc
     35     307    3190

After the patch we still have quite some regressions for the vectorizer:
$ cat after.txt | grep "^FAIL" | wc
    210    1734   18794

Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Dominique Dhumieres]

With your patch (with the previous 2 additional passes from previous
attempt and lim) I have only 33 vectorized loops instead of 41 without
-fgraphite*. The timing is going from ~10.8s to 6.5s compared to a
reference time of 4.7s. Still investigating.

Thanks for the interest;-)

Dominique

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 1818 bytes --]

On Tue, Mar 16, 2010 at 12:59, Sebastian Pop <sebpop@gmail.com> wrote:
> On Tue, Mar 16, 2010 at 08:24, Ira Rosen <IRAR@il.ibm.com> wrote:
>> In gas_dyn vectorization fails because Graphite inserts statements in latch
>> blocks, e.g. (line 409),
>>
>> <bb 12>:
>>  # ivtmp.2047_274 = PHI <ivtmp.2047_275(13), 0(11)>
>>  D.7535_276 = (integer(kind=8)) ivtmp.2047_274;
>>  S.26_217 = D.7535_276 + 1;
>>  D.4310_112 = S.26_217 + -1;
>>  (*gamma_113(D))[D.4310_112] = D.4292_73;
>>  if (ivtmp.2047_274 < D.7533_233)
>>    goto <bb 13>;
>>  else
>>    goto <bb 14>;
>>
>> <bb 13>:
>>  ivtmp.2047_275 = ivtmp.2047_274 + 1;  <============= this statement
>>  goto <bb 12>;
>>
>
> This is due to the fact that we call
>
>  loop->single_iv = canonicalize_loop_ivs (loop, &nit);
>
> This function was used only after the vectorizer in parloops:
>
> $ grep canonicalize_loop_ivs *.[ch]
> graphite-sese-to-poly.c:2897:  loop->single_iv = canonicalize_loop_ivs
> (loop, &nit);
> tree-flow.h:697:tree canonicalize_loop_ivs (struct loop *, tree *);
> tree-parloops.c:1629:  canonicalize_loop_ivs (loop, &nit);
> tree-ssa-loop-manip.c:1188:canonicalize_loop_ivs (struct loop *loop, tree *nit)
>
> but now it is used in Graphite this should be cleaned up to not
> generate the iv bumping in the loop->latch.
>

With the attached patch we now get:

$ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
-fgraphite-identity gas_dyn.f90 &> before.txt
$ ./f951 -O3 -ffast-math -ftree-vectorizer-verbose=2
-fgraphite-identity gas_dyn.f90 &> after.txt

$ grep "LOOP VECTORIZED" before.txt | wc
     24      96    1848
$ grep "LOOP VECTORIZED" after.txt | wc
     40     160    3065

I will test this patch on amd64-linux.
Ok for trunk if it passes bootstrap and test?

Thanks,
Sebastian

[-- Attachment #2: 0001-canonicalize_loop_ivs-should-add-the-IV-bump-in-loop.patch --]
[-- Type: text/x-diff, Size: 1540 bytes --]

From 0be713a0cc67d73e55f88c8c07e55a22c3ae62ee Mon Sep 17 00:00:00 2001
From: Sebastian Pop <sebpop@gmail.com>
Date: Tue, 16 Mar 2010 13:39:16 -0500
Subject: [PATCH] canonicalize_loop_ivs should add the IV bump in loop->header.

---
 gcc/graphite-sese-to-poly.c |    2 +-
 gcc/tree-ssa-loop-manip.c   |    4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index b0bc348..a016285 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -2886,7 +2886,7 @@ graphite_loop_normal_form (loop_p loop)
 
   bool known_niter = number_of_iterations_exit (loop, exit, &niter, false);
 
-  /* At this point we should know the number of iterations,  */
+  /* At this point we should know the number of iterations.  */
   gcc_assert (known_niter);
 
   nit = force_gimple_operand (unshare_expr (niter.niter), &stmts, true,
diff --git a/gcc/tree-ssa-loop-manip.c b/gcc/tree-ssa-loop-manip.c
index 7c54c87..9d5d697 100644
--- a/gcc/tree-ssa-loop-manip.c
+++ b/gcc/tree-ssa-loop-manip.c
@@ -1215,9 +1215,9 @@ canonicalize_loop_ivs (struct loop *loop, tree *nit)
 	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);
 
-- 
1.6.3.3

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Tue, Mar 16, 2010 at 08:24, Ira Rosen <IRAR@il.ibm.com> wrote:
> In gas_dyn vectorization fails because Graphite inserts statements in latch
> blocks, e.g. (line 409),
>
> <bb 12>:
>  # ivtmp.2047_274 = PHI <ivtmp.2047_275(13), 0(11)>
>  D.7535_276 = (integer(kind=8)) ivtmp.2047_274;
>  S.26_217 = D.7535_276 + 1;
>  D.4310_112 = S.26_217 + -1;
>  (*gamma_113(D))[D.4310_112] = D.4292_73;
>  if (ivtmp.2047_274 < D.7533_233)
>    goto <bb 13>;
>  else
>    goto <bb 14>;
>
> <bb 13>:
>  ivtmp.2047_275 = ivtmp.2047_274 + 1;  <============= this statement
>  goto <bb 12>;
>

This is due to the fact that we call

  loop->single_iv = canonicalize_loop_ivs (loop, &nit);

This function was used only after the vectorizer in parloops:

$ grep canonicalize_loop_ivs *.[ch]
graphite-sese-to-poly.c:2897:  loop->single_iv = canonicalize_loop_ivs
(loop, &nit);
tree-flow.h:697:tree canonicalize_loop_ivs (struct loop *, tree *);
tree-parloops.c:1629:  canonicalize_loop_ivs (loop, &nit);
tree-ssa-loop-manip.c:1188:canonicalize_loop_ivs (struct loop *loop, tree *nit)

but now it is used in Graphite this should be cleaned up to not
generate the iv bumping in the loop->latch.

Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Michael Matz]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1388 bytes --]

Hi,

On Tue, 16 Mar 2010, Sebastian Pop wrote:

> > In Toon's testcase from
> > http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00515.html vectorization fails
> > because scev analysis failure: "evolution of offset is not affine." It
> > probably gets confused by
> >
> >  # graphite_IV.22_66 = PHI <0(4), graphite_IV.22_67(6)>
> >  var.23_77 = (character(kind=4)) graphite_IV.22_66;
> >  D.1214_68 = (integer(kind=4)) var.23_77;
> >  i_69 = D.1214_68 + 1;
> >  D.1192_70 = (integer(kind=8)) i_69;
> >  D.1193_71 = D.1192_70 + -1;
> >  D.1194_72 = (*a_39(D))[D.1193_71];
> >
> >  instead (without Graphite)
> >
> >  # i_1 = PHI <1(3), i_48(5)>
> >  D.1192_37 = (integer(kind=8)) i_1;
> >  D.1193_38 = D.1192_37 + -1;
> >  D.1194_40 = (*a_39(D))[D.1193_38];
> >
> 
> Looks like the scev analysis fails because of too many casts
> on the way, otherwise I see no particular difficulty in the code
> generated by Graphite.

Both sequences aren't equivalent, so my guess would be that the 
transformation done by graphite actually is buggy or at least suboptimal.  
The difference lies in the overflow behaviour.  Without graphite, i_1 has 
undefined overflow, as D.1192_37 too, hence we can look through the 
conversion.  With graphite var.23_77 (of unsigned type) has defined 
overflow semantics, hence we can't look through that conversion.


Ciao,
Michael.

Re: [patch] Merge to trunk from Graphite branch

[Ira Rosen]

Sebastian Pop <sebpop@gmail.com> wrote on 16/03/2010 03:41:22 PM:

>
> This is how the IV canonicalization is inserting the IV bumping.
> I will try to put this stmt before the exit condition in the header bb.
>
> Why is the vectorizer so picky about the latch bb to be empty?
> Probably we should fix the vectorizer for this particular case?

I think most of the answers can be found in this old thread
http://gcc.gnu.org/ml/gcc-patches/2005-11/msg01978.html.

Ira

Re: [patch] Merge to trunk from Graphite branch

[Sebastian Pop]

On Tue, Mar 16, 2010 at 08:24, Ira Rosen <IRAR@il.ibm.com> wrote:
>
>
> dominiq@lps.ens.fr (Dominique Dhumieres) wrote on 16/03/2010 12:07:42 AM:
>
>> The missed vectorization seems related to the loop index being unknown:
>> if I replace 'do i = 1, n' with 'do i = 1, nl' with nl set to
>> some given value (say 100) the loop vectorize with -fgraphite-identity.
>>

Ok, I should look at the computation of niter on the code generated
by Graphite to see why it fails.

>
> In gas_dyn vectorization fails because Graphite inserts statements in latch
> blocks, e.g. (line 409),
>
> <bb 12>:
>  # ivtmp.2047_274 = PHI <ivtmp.2047_275(13), 0(11)>
>  D.7535_276 = (integer(kind=8)) ivtmp.2047_274;
>  S.26_217 = D.7535_276 + 1;
>  D.4310_112 = S.26_217 + -1;
>  (*gamma_113(D))[D.4310_112] = D.4292_73;
>  if (ivtmp.2047_274 < D.7533_233)
>    goto <bb 13>;
>  else
>    goto <bb 14>;
>
> <bb 13>:
>  ivtmp.2047_275 = ivtmp.2047_274 + 1;  <============= this statement
>  goto <bb 12>;
>

This is how the IV canonicalization is inserting the IV bumping.
I will try to put this stmt before the exit condition in the header bb.

Why is the vectorizer so picky about the latch bb to be empty?
Probably we should fix the vectorizer for this particular case?

> In Toon's testcase from
> http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00515.html vectorization fails
> because scev analysis failure: "evolution of offset is not affine." It
> probably gets confused by
>
>  # graphite_IV.22_66 = PHI <0(4), graphite_IV.22_67(6)>
>  var.23_77 = (character(kind=4)) graphite_IV.22_66;
>  D.1214_68 = (integer(kind=4)) var.23_77;
>  i_69 = D.1214_68 + 1;
>  D.1192_70 = (integer(kind=8)) i_69;
>  D.1193_71 = D.1192_70 + -1;
>  D.1194_72 = (*a_39(D))[D.1193_71];
>
>  instead (without Graphite)
>
>  # i_1 = PHI <1(3), i_48(5)>
>  D.1192_37 = (integer(kind=8)) i_1;
>  D.1193_38 = D.1192_37 + -1;
>  D.1194_40 = (*a_39(D))[D.1193_38];
>

Looks like the scev analysis fails because of too many casts
on the way, otherwise I see no particular difficulty in the code
generated by Graphite.  I will give a look at this as well.

Thanks for the detailed analysis of the missed vectorization cases,
Sebastian

Re: [patch] Merge to trunk from Graphite branch

[Ira Rosen]

dominiq@lps.ens.fr (Dominique Dhumieres) wrote on 16/03/2010 12:07:42 AM:

> The missed vectorization seems related to the loop index being unknown:
> if I replace 'do i = 1, n' with 'do i = 1, nl' with nl set to
> some given value (say 100) the loop vectorize with -fgraphite-identity.
>

In gas_dyn vectorization fails because Graphite inserts statements in latch
blocks, e.g. (line 409),

<bb 12>:
  # ivtmp.2047_274 = PHI <ivtmp.2047_275(13), 0(11)>
  D.7535_276 = (integer(kind=8)) ivtmp.2047_274;
  S.26_217 = D.7535_276 + 1;
  D.4310_112 = S.26_217 + -1;
  (*gamma_113(D))[D.4310_112] = D.4292_73;
  if (ivtmp.2047_274 < D.7533_233)
    goto <bb 13>;
  else
    goto <bb 14>;

<bb 13>:
  ivtmp.2047_275 = ivtmp.2047_274 + 1;  <============= this statement
  goto <bb 12>;




In Toon's testcase from
http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00515.html vectorization fails
because scev analysis failure: "evolution of offset is not affine." It
probably gets confused by

  # graphite_IV.22_66 = PHI <0(4), graphite_IV.22_67(6)>
  var.23_77 = (character(kind=4)) graphite_IV.22_66;
  D.1214_68 = (integer(kind=4)) var.23_77;
  i_69 = D.1214_68 + 1;
  D.1192_70 = (integer(kind=8)) i_69;
  D.1193_71 = D.1192_70 + -1;
  D.1194_72 = (*a_39(D))[D.1193_71];

 instead (without Graphite)

  # i_1 = PHI <1(3), i_48(5)>
  D.1192_37 = (integer(kind=8)) i_1;
  D.1193_38 = D.1192_37 + -1;
  D.1194_40 = (*a_39(D))[D.1193_38];


Ira

> Dominique

Re: [patch] Merge to trunk from Graphite branch

[Dominique Dhumieres]

The missed vectorization seems related to the loop index being unknown:
if I replace 'do i = 1, n' with 'do i = 1, nl' with nl set to
some given value (say 100) the loop vectorize with -fgraphite-identity.

Dominique

Re: [patch] Merge to trunk from Graphite branch

[Dominique Dhumieres]

> The patch from Ira would not apply: here is an updated patch,
> that anyways would execute conditionally based on running
> the graphite pass.

The patch does not fix the missed vectorization with -fgraphite-identity.

Even if this problem is not release critical it should be P1 on the graphite
agenda: there is no point to fine tune loops if it breaks the vectorization.
On x86 the potential speed-up of vectorization is 4 for floats and 2 for 
doubles.

Dominique

[patch] Merge to trunk from Graphite branch

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 201 bytes --]

Hi,

I am merging to trunk the last changes from the Graphite branch
These should fix some of the last P1 bugs related to Graphite.

Sebastian Pop
--
AMD / Open Source Compiler Engineering / GNU Tools

[-- Attachment #2: 0001-Add-constraints-on-the-type-of-parameters-to-the-sco.patch --]
[-- Type: text/x-diff, Size: 1451 bytes --]

From 940163e15b52aead6cc08b8c011d94102705ec95 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Tue, 23 Feb 2010 13:31:26 +0000
Subject: [PATCH 01/15] Add constraints on the type of parameters to the scop context.

2010-02-23  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-sese-to-poly.c (add_param_constraints): Enabled: remove
	early return.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157002 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite      |    5 +++++
 gcc/graphite-sese-to-poly.c |    3 ---
 2 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 1771284..a3dbd5f 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,8 @@
+2010-02-23  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (add_param_constraints): Enabled: remove
+	early return.
+
 2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
 
 	PR middle-end/43083
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index d4889b0..3ee431f 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -1501,9 +1501,6 @@ add_param_constraints (scop_p scop, ppl_Polyhedron_t context, graphite_dim_t p)
   tree type = TREE_TYPE (parameter);
   tree lb, ub;
 
-  /* Disabled until we fix CPU2006.  */
-  return;
-
   if (!INTEGRAL_TYPE_P (type))
     return;
 
-- 
1.6.3.3


[-- Attachment #3: 0002-Cleanup.patch --]
[-- Type: text/x-diff, Size: 2574 bytes --]

From ce550713b29740b1621e37d499cc55b2629bf567 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Tue, 23 Feb 2010 13:31:32 +0000
Subject: [PATCH 02/15] Cleanup.

2010-02-23  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-clast-to-gimple.c (find_cloog_iv_in_expr): Simplify
	and clean up the logic.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157003 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite         |    5 +++++
 gcc/graphite-clast-to-gimple.c |   30 +++++++++++++-----------------
 2 files changed, 18 insertions(+), 17 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index a3dbd5f..e3e6ecd 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,10 @@
 2010-02-23  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* graphite-clast-to-gimple.c (find_cloog_iv_in_expr): Simplify
+	and clean up the logic.
+
+2010-02-23  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-sese-to-poly.c (add_param_constraints): Enabled: remove
 	early return.
 
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 424f5c6..3e82075 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -997,34 +997,30 @@ static const char *
 find_cloog_iv_in_expr (struct clast_expr *expr)
 {
   struct clast_term *term = (struct clast_term *) expr;
-
-  if (expr->type == expr_term
-      && !term->var)
-    return NULL;
+  struct clast_reduction *red;
+  int i;
 
   if (expr->type == expr_term)
     return term->var;
 
-  if (expr->type == expr_red)
-    {
-      int i;
-      struct clast_reduction *red = (struct clast_reduction *) expr;
+  if (expr->type != expr_red)
+    return NULL;
 
-      for (i = 0; i < red->n; i++)
-	{
-	  const char *res = find_cloog_iv_in_expr ((red)->elts[i]);
+  red = (struct clast_reduction *) expr;
+  for (i = 0; i < red->n; i++)
+    {
+      const char *res = find_cloog_iv_in_expr (red->elts[i]);
 
-	  if (res)
-	    return res;
-	}
+      if (res)
+	return res;
     }
 
   return NULL;
 }
 
-/* Build for a clast_user_stmt USER_STMT a map between the CLAST
-   induction variables and the corresponding GCC old induction
-   variables.  This information is stored on each GRAPHITE_BB.  */
+/* Build for USER_STMT a map between the CLAST induction variables and
+   the corresponding GCC old induction variables.  This information is
+   stored on each GRAPHITE_BB.  */
 
 static void
 compute_cloog_iv_types_1 (poly_bb_p pbb, struct clast_user_stmt *user_stmt)
-- 
1.6.3.3


[-- Attachment #4: 0003-Fix-PR42326-handle-default-definitions.patch --]
[-- Type: text/x-diff, Size: 2432 bytes --]

From cd3223f902edff054f9d8f394324e874353b6e5f Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Tue, 2 Mar 2010 11:51:21 +0000
Subject: [PATCH 03/15] Fix PR42326: handle default definitions.

2010-03-02  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/42326
	* sese.c (name_defined_in_loop_p): Return false for default
	definitions.

	* gcc.dg/graphite/pr42326.c: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157162 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                  |    8 ++++++++
 gcc/sese.c                              |    5 ++---
 gcc/testsuite/gcc.dg/graphite/pr42326.c |   20 ++++++++++++++++++++
 3 files changed, 30 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/graphite/pr42326.c

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index e3e6ecd..5406717 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,11 @@
+2010-03-02  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/42326
+	* sese.c (name_defined_in_loop_p): Return false for default
+	definitions.
+
+	* gcc.dg/graphite/pr42326.c: New.
+
 2010-02-23  Sebastian Pop  <sebastian.pop@amd.com>
 
 	* graphite-clast-to-gimple.c (find_cloog_iv_in_expr): Simplify
diff --git a/gcc/sese.c b/gcc/sese.c
index d7a9faa..545b1c6 100644
--- a/gcc/sese.c
+++ b/gcc/sese.c
@@ -1106,9 +1106,8 @@ get_false_edge_from_guard_bb (basic_block bb)
 static bool
 name_defined_in_loop_p (tree name, loop_p loop)
 {
-  gimple stmt = SSA_NAME_DEF_STMT (name);
-
-  return (gimple_bb (stmt)->loop_father == loop);
+  return !SSA_NAME_IS_DEFAULT_DEF (name)
+    && gimple_bb (SSA_NAME_DEF_STMT (name))->loop_father == loop;
 }
 
 /* Returns true when EXPR contains SSA_NAMEs defined in LOOP.  */
diff --git a/gcc/testsuite/gcc.dg/graphite/pr42326.c b/gcc/testsuite/gcc.dg/graphite/pr42326.c
new file mode 100644
index 0000000..de5d56e
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/pr42326.c
@@ -0,0 +1,20 @@
+/* { dg-options "-O1 -floop-parallelize-all" } */
+
+double lagrange(const double x[],
+                const double y[],
+                long n,
+                double xval)
+{
+  long i, j;
+  double yval = 0.;
+
+  for( i=0; i < n; i++ )
+    {
+      double l = 1.;
+      for( j=0; j < n; j++ )
+	if( i != j )
+	  l *= (xval-x[j])/(x[i]-x[j]);
+      yval += y[i]*l;
+    }
+  return yval;
+}
-- 
1.6.3.3


[-- Attachment #5: 0004-New-function-combine_context_id_scat.patch --]
[-- Type: text/x-diff, Size: 3829 bytes --]

From c88ab668784be9616db77719263c66eae41773b5 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Thu, 4 Mar 2010 12:31:18 +0000
Subject: [PATCH 04/15] New function combine_context_id_scat.

2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-poly.h (struct poly_scattering): Add layout documentation.
	(struct poly_bb): Same.
	(combine_context_id_scat): New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157219 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite |    6 +++++
 gcc/graphite-poly.h    |   53 ++++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 57 insertions(+), 2 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 5406717..3665132 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,9 @@
+2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-poly.h (struct poly_scattering): Add layout documentation.
+	(struct poly_bb): Same.
+	(combine_context_id_scat): New.
+
 2010-03-02  Sebastian Pop  <sebastian.pop@amd.com>
 
 	PR middle-end/42326
diff --git a/gcc/graphite-poly.h b/gcc/graphite-poly.h
index 0a8204e..b586699 100644
--- a/gcc/graphite-poly.h
+++ b/gcc/graphite-poly.h
@@ -263,7 +263,9 @@ typedef struct poly_scattering *poly_scattering_p;
 
 struct poly_scattering
 {
-  /* The scattering function containing the transformations.  */
+  /* The scattering function containing the transformations: the
+     layout of this polyhedron is: T|I|G with T the transform
+     scattering, I the iteration domain, G the context parameters.  */
   ppl_Polyhedron_t scattering;
 
   /* The number of local variables.  */
@@ -283,7 +285,9 @@ struct poly_bb
   /* Pointer to the SCOP containing this PBB.  */
   scop_p scop;
 
-  /* The iteration domain of this bb.
+  /* The iteration domain of this bb.  The layout of this polyhedron
+     is I|G with I the iteration domain, G the context parameters.
+
      Example:
 
      for (i = a - 7*b + 8; i <= 3*a + 13*b + 20; i++)
@@ -1467,4 +1471,49 @@ restore_scattering (scop_p scop)
   restore_lst_schedule (scop);
 }
 
+/* For a given PBB, add to RES the scop context, the iteration domain,
+   the original scattering when ORIGINAL_P is true, otherwise add the
+   transformed scattering.  */
+
+static inline void
+combine_context_id_scat (ppl_Pointset_Powerset_C_Polyhedron_t *res,
+			 poly_bb_p pbb, bool original_p)
+{
+  ppl_Pointset_Powerset_C_Polyhedron_t context;
+  ppl_Pointset_Powerset_C_Polyhedron_t id;
+
+  ppl_new_Pointset_Powerset_C_Polyhedron_from_C_Polyhedron
+    (res, original_p ?
+     PBB_ORIGINAL_SCATTERING (pbb) : PBB_TRANSFORMED_SCATTERING (pbb));
+
+  ppl_new_Pointset_Powerset_C_Polyhedron_from_Pointset_Powerset_C_Polyhedron
+    (&context, SCOP_CONTEXT (PBB_SCOP (pbb)));
+
+  ppl_new_Pointset_Powerset_C_Polyhedron_from_Pointset_Powerset_C_Polyhedron
+    (&id, PBB_DOMAIN (pbb));
+
+  /* Extend the context and the iteration domain to the dimension of
+     the scattering: T|I|G.  */
+  {
+    ppl_dimension_type gdim, tdim, idim;
+
+    ppl_Pointset_Powerset_C_Polyhedron_space_dimension (*res, &tdim);
+    ppl_Pointset_Powerset_C_Polyhedron_space_dimension (context, &gdim);
+    ppl_Pointset_Powerset_C_Polyhedron_space_dimension (id, &idim);
+
+    if (tdim > gdim)
+      ppl_insert_dimensions_pointset (context, 0, tdim - gdim);
+
+    if (tdim > idim)
+      ppl_insert_dimensions_pointset (id, 0, tdim - idim);
+  }
+
+  /* Add the context and the iteration domain to the result.  */
+  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (*res, context);
+  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (*res, id);
+
+  ppl_delete_Pointset_Powerset_C_Polyhedron (context);
+  ppl_delete_Pointset_Powerset_C_Polyhedron (id);
+}
+
 #endif
-- 
1.6.3.3


[-- Attachment #6: 0005-Cleanup-data-dep-polyhedron-construction.patch --]
[-- Type: text/x-diff, Size: 6393 bytes --]

From 02d4a49457b6fabd698e32d4c489eb89930a1a47 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Thu, 4 Mar 2010 12:31:24 +0000
Subject: [PATCH 05/15] Cleanup data dep polyhedron construction.

2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-dependences.c (map_into_dep_poly): Removed.
	(dependence_polyhedron_1): Use combine_context_id_scat.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157220 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite     |    5 +++
 gcc/graphite-dependences.c |   60 ++++++++-----------------------------------
 2 files changed, 16 insertions(+), 49 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 3665132..c190671 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,10 @@
 2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* graphite-dependences.c (map_into_dep_poly): Removed.
+	(dependence_polyhedron_1): Use combine_context_id_scat.
+
+2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-poly.h (struct poly_scattering): Add layout documentation.
 	(struct poly_bb): Same.
 	(combine_context_id_scat): New.
diff --git a/gcc/graphite-dependences.c b/gcc/graphite-dependences.c
index bd83e15..f9d9daa 100644
--- a/gcc/graphite-dependences.c
+++ b/gcc/graphite-dependences.c
@@ -269,28 +269,6 @@ poly_drs_may_alias_p (poly_dr_p pdr1, poly_dr_p pdr2)
   return !empty_p;
 }
 
-/* Returns a polyhedron of dimension DIM.
-
-   Maps the dimensions [0, ..., cut - 1] of polyhedron P to OFFSET
-   and the dimensions [cut, ..., nb_dim] to DIM - GDIM.  */
-
-static ppl_Pointset_Powerset_C_Polyhedron_t
-map_into_dep_poly (graphite_dim_t dim, graphite_dim_t gdim,
-		   ppl_Pointset_Powerset_C_Polyhedron_t p,
-		   graphite_dim_t cut,
-		   graphite_dim_t offset)
-{
-  ppl_Pointset_Powerset_C_Polyhedron_t res;
-
-  ppl_new_Pointset_Powerset_C_Polyhedron_from_Pointset_Powerset_C_Polyhedron
-    (&res, p);
-  ppl_insert_dimensions_pointset (res, 0, offset);
-  ppl_insert_dimensions_pointset (res, offset + cut,
-				  dim - offset - cut - gdim);
-
-  return res;
-}
-
 /* Swap [cut0, ..., cut1] to the end of DR: "a CUT0 b CUT1 c" is
    transformed into "a CUT0 c CUT1' b"
 
@@ -486,41 +464,33 @@ dependence_polyhedron_1 (poly_dr_p pdr1, poly_dr_p pdr2,
   poly_bb_p pbb1 = PDR_PBB (pdr1);
   poly_bb_p pbb2 = PDR_PBB (pdr2);
   scop_p scop = PBB_SCOP (pbb1);
-  ppl_Pointset_Powerset_C_Polyhedron_t d1 = PBB_DOMAIN (pbb1);
-  ppl_Pointset_Powerset_C_Polyhedron_t d2 = PBB_DOMAIN (pbb2);
   graphite_dim_t tdim1 = original_scattering_p ?
     pbb_nb_scattering_orig (pbb1) : pbb_nb_scattering_transform (pbb1);
   graphite_dim_t tdim2 = original_scattering_p ?
     pbb_nb_scattering_orig (pbb2) : pbb_nb_scattering_transform (pbb2);
-  ppl_Polyhedron_t scat1 = original_scattering_p ?
-    PBB_ORIGINAL_SCATTERING (pbb1) : PBB_TRANSFORMED_SCATTERING (pbb1);
-  ppl_Polyhedron_t scat2 = original_scattering_p ?
-    PBB_ORIGINAL_SCATTERING (pbb2) : PBB_TRANSFORMED_SCATTERING (pbb2);
   graphite_dim_t ddim1 = pbb_dim_iter_domain (pbb1);
   graphite_dim_t ddim2 = pbb_dim_iter_domain (pbb2);
   graphite_dim_t sdim1 = PDR_NB_SUBSCRIPTS (pdr1) + 1;
+  graphite_dim_t sdim2 = PDR_NB_SUBSCRIPTS (pdr2) + 1;
   graphite_dim_t gdim = scop_nb_params (scop);
   graphite_dim_t dim1 = pdr_dim (pdr1);
   graphite_dim_t dim2 = pdr_dim (pdr2);
   graphite_dim_t dim = tdim1 + tdim2 + dim1 + dim2 - gdim;
   ppl_Pointset_Powerset_C_Polyhedron_t res;
-  ppl_Pointset_Powerset_C_Polyhedron_t id1, id2, isc1, isc2, idr1, idr2;
+  ppl_Pointset_Powerset_C_Polyhedron_t idr1, idr2;
   ppl_Pointset_Powerset_C_Polyhedron_t sc1, sc2, dreq;
-  ppl_Pointset_Powerset_C_Polyhedron_t context;
 
   gcc_assert (PBB_SCOP (pbb1) == PBB_SCOP (pbb2));
 
-  ppl_new_Pointset_Powerset_C_Polyhedron_from_Pointset_Powerset_C_Polyhedron
-    (&context, SCOP_CONTEXT (scop));
-  ppl_insert_dimensions_pointset (context, 0, dim - gdim);
+  combine_context_id_scat (&sc1, pbb1, original_scattering_p);
+  combine_context_id_scat (&sc2, pbb2, original_scattering_p);
 
-  ppl_new_Pointset_Powerset_C_Polyhedron_from_C_Polyhedron (&sc1, scat1);
-  ppl_new_Pointset_Powerset_C_Polyhedron_from_C_Polyhedron (&sc2, scat2);
+  ppl_insert_dimensions_pointset (sc1, tdim1 + ddim1,
+				  tdim2 + ddim2 + sdim1 + sdim2);
 
-  id1 = map_into_dep_poly (dim, gdim, d1, ddim1, tdim1);
-  id2 = map_into_dep_poly (dim, gdim, d2, ddim2, tdim1 + ddim1 + tdim2);
-  isc1 = map_into_dep_poly (dim, gdim, sc1, ddim1 + tdim1, 0);
-  isc2 = map_into_dep_poly (dim, gdim, sc2, ddim2 + tdim2, tdim1 + ddim1);
+  ppl_insert_dimensions_pointset (sc2, 0, tdim1 + ddim1);
+  ppl_insert_dimensions_pointset (sc2, tdim1 + ddim1 + tdim2 + ddim2,
+				  sdim1 + sdim2);
 
   idr1 = map_dr_into_dep_poly (dim, PDR_ACCESSES (pdr1), ddim1, ddim1 + gdim,
 			       tdim1, tdim2 + ddim2);
@@ -531,21 +501,13 @@ dependence_polyhedron_1 (poly_dr_p pdr1, poly_dr_p pdr2,
   dreq = dr_equality_constraints (dim, tdim1 + ddim1 + tdim2 + ddim2, sdim1);
 
   ppl_new_Pointset_Powerset_C_Polyhedron_from_space_dimension (&res, dim, 0);
-  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, context);
-  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, id1);
-  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, id2);
-  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, isc1);
-  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, isc2);
+  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, sc1);
+  ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, sc2);
   ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, idr1);
   ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, idr2);
   ppl_Pointset_Powerset_C_Polyhedron_intersection_assign (res, dreq);
-  ppl_delete_Pointset_Powerset_C_Polyhedron (context);
-  ppl_delete_Pointset_Powerset_C_Polyhedron (id1);
-  ppl_delete_Pointset_Powerset_C_Polyhedron (id2);
   ppl_delete_Pointset_Powerset_C_Polyhedron (sc1);
   ppl_delete_Pointset_Powerset_C_Polyhedron (sc2);
-  ppl_delete_Pointset_Powerset_C_Polyhedron (isc1);
-  ppl_delete_Pointset_Powerset_C_Polyhedron (isc2);
   ppl_delete_Pointset_Powerset_C_Polyhedron (idr1);
   ppl_delete_Pointset_Powerset_C_Polyhedron (idr2);
   ppl_delete_Pointset_Powerset_C_Polyhedron (dreq);
-- 
1.6.3.3


[-- Attachment #7: 0006-Cleanup.patch --]
[-- Type: text/x-diff, Size: 2894 bytes --]

From 80f59eb05e36c6dfea5182a7b0fd80ed3f6c6d14 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Thu, 4 Mar 2010 12:31:31 +0000
Subject: [PATCH 06/15] Cleanup.

2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-ppl.c (ppl_min_for_le_polyhedron): Renamed
	ppl_min_for_le_pointset.  Use ppl_Pointset_Powerset_C_Polyhedron_minimize.
	* graphite-ppl.h (ppl_min_for_le_polyhedron): Update declaration.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157221 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite |    6 ++++++
 gcc/graphite-ppl.c     |    8 ++++----
 gcc/graphite-ppl.h     |    4 ++--
 3 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index c190671..e5d5653 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,11 @@
 2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* graphite-ppl.c (ppl_min_for_le_polyhedron): Renamed
+	ppl_min_for_le_pointset.  Use ppl_Pointset_Powerset_C_Polyhedron_minimize.
+	* graphite-ppl.h (ppl_min_for_le_polyhedron): Update declaration.
+
+2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-dependences.c (map_into_dep_poly): Removed.
 	(dependence_polyhedron_1): Use combine_context_id_scat.
 
diff --git a/gcc/graphite-ppl.c b/gcc/graphite-ppl.c
index 0b76335..c80cf1b 100644
--- a/gcc/graphite-ppl.c
+++ b/gcc/graphite-ppl.c
@@ -673,18 +673,18 @@ ppl_max_for_le_pointset (ppl_Pointset_Powerset_C_Polyhedron_t ps,
    polyhedron POL.  */
 
 void
-ppl_min_for_le_polyhedron (ppl_Polyhedron_t pol,
-			   ppl_Linear_Expression_t le, Value res)
+ppl_min_for_le_pointset (ppl_Pointset_Powerset_C_Polyhedron_t ps,
+			 ppl_Linear_Expression_t le, Value res)
 {
   ppl_Coefficient_t num, denom;
   Value dv, nv;
-  int maximum, err;
+  int minimum, err;
 
   value_init (nv);
   value_init (dv);
   ppl_new_Coefficient (&num);
   ppl_new_Coefficient (&denom);
-  err = ppl_Polyhedron_minimize (pol, le, num, denom, &maximum);
+  err = ppl_Pointset_Powerset_C_Polyhedron_minimize (ps, le, num, denom, &minimum);
 
   if (err > 0)
     {
diff --git a/gcc/graphite-ppl.h b/gcc/graphite-ppl.h
index 488ad6f..5a3ec6a 100644
--- a/gcc/graphite-ppl.h
+++ b/gcc/graphite-ppl.h
@@ -48,8 +48,8 @@ void ppl_set_inhomogeneous_gmp (ppl_Linear_Expression_t, Value);
 void ppl_set_coef_gmp (ppl_Linear_Expression_t, ppl_dimension_type, Value);
 void ppl_max_for_le_pointset (ppl_Pointset_Powerset_C_Polyhedron_t,
                               ppl_Linear_Expression_t, Value);
-void ppl_min_for_le_polyhedron (ppl_Polyhedron_t, ppl_Linear_Expression_t,
-				Value);
+void ppl_min_for_le_pointset (ppl_Pointset_Powerset_C_Polyhedron_t,
+			      ppl_Linear_Expression_t, Value);
 ppl_Constraint_t ppl_build_relation (int, int, int, int,
 				     enum ppl_enum_Constraint_Type);
 
-- 
1.6.3.3


[-- Attachment #8: 0007-Fold-convert-SSA_NAMEs-in-the-scalar-expander.patch --]
[-- Type: text/x-diff, Size: 3701 bytes --]

From 9187b4b6f838d67c624f3f2aff60597ab6714c25 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Thu, 4 Mar 2010 12:31:37 +0000
Subject: [PATCH 07/15] Fold convert SSA_NAMEs in the scalar expander.

2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>

	* sese.c (expand_scalar_variables_ssa_name): Add new argument for type.
	Call fold_convert on all the returned values.
	(expand_scalar_variables_expr): Pass to expand_scalar_variables_ssa_name
	the type of the resulting expression.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157222 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite |    7 +++++++
 gcc/sese.c             |   14 +++++++-------
 2 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index e5d5653..25e0c51 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,12 @@
 2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* sese.c (expand_scalar_variables_ssa_name): Add new argument for type.
+	Call fold_convert on all the returned values.
+	(expand_scalar_variables_expr): Pass to expand_scalar_variables_ssa_name
+	the type of the resulting expression.
+
+2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-ppl.c (ppl_min_for_le_polyhedron): Renamed
 	ppl_min_for_le_pointset.  Use ppl_Pointset_Powerset_C_Polyhedron_minimize.
 	* graphite-ppl.h (ppl_min_for_le_polyhedron): Update declaration.
diff --git a/gcc/sese.c b/gcc/sese.c
index 545b1c6..975eb36 100644
--- a/gcc/sese.c
+++ b/gcc/sese.c
@@ -778,7 +778,7 @@ expand_scalar_variables_call (gimple stmt, basic_block bb, sese region,
    to translate the names of induction variables.  */
 
 static tree
-expand_scalar_variables_ssa_name (tree op0, basic_block bb,
+expand_scalar_variables_ssa_name (tree type, tree op0, basic_block bb,
 				  sese region, htab_t map,
 				  gimple_stmt_iterator *gsi)
 {
@@ -787,7 +787,7 @@ expand_scalar_variables_ssa_name (tree op0, basic_block bb,
 
   if (is_parameter (region, op0)
       || is_iv (op0))
-    return get_rename (map, op0);
+    return fold_convert (type, get_rename (map, op0));
 
   def_stmt = SSA_NAME_DEF_STMT (op0);
 
@@ -796,7 +796,7 @@ expand_scalar_variables_ssa_name (tree op0, basic_block bb,
 
   if (new_op != op0
       && gimple_bb (SSA_NAME_DEF_STMT (new_op)) == bb)
-    return new_op;
+    return fold_convert (type, new_op);
 
   if (gimple_bb (def_stmt) == bb)
     {
@@ -804,13 +804,13 @@ expand_scalar_variables_ssa_name (tree op0, basic_block bb,
 	 we do not need to create a new expression for it, we
 	 only need to ensure its operands are expanded.  */
       expand_scalar_variables_stmt (def_stmt, bb, region, map, gsi);
-      return new_op;
+      return fold_convert (type, new_op);
     }
   else
     {
       if (!gimple_bb (def_stmt)
 	  || !bb_in_sese_p (gimple_bb (def_stmt), region))
-	return new_op;
+	return fold_convert (type, new_op);
 
       switch (gimple_code (def_stmt))
 	{
@@ -871,7 +871,7 @@ expand_scalar_variables_expr (tree type, tree op0, enum tree_code code,
 	  {
 	    tree old_name = TREE_OPERAND (op0, 0);
 	    tree expr = expand_scalar_variables_ssa_name
-	      (old_name, bb, region, map, gsi);
+	      (type, old_name, bb, region, map, gsi);
 
 	    if (TREE_CODE (expr) != SSA_NAME
 		&& is_gimple_reg (old_name))
@@ -938,7 +938,7 @@ expand_scalar_variables_expr (tree type, tree op0, enum tree_code code,
     }
 
   if (code == SSA_NAME)
-    return expand_scalar_variables_ssa_name (op0, bb, region, map, gsi);
+    return expand_scalar_variables_ssa_name (type, op0, bb, region, map, gsi);
 
   if (code == ADDR_EXPR)
     {
-- 
1.6.3.3


[-- Attachment #9: 0008-Do-not-short-cut-code-generation-with-gloog_error.patch --]
[-- Type: text/x-diff, Size: 1584 bytes --]

From d14310f6f43a37395bd94bed693c098814ed5ad3 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Thu, 4 Mar 2010 12:31:43 +0000
Subject: [PATCH 08/15] Do not short-cut code generation with gloog_error.

2010-03-04  Tobias Grosser  <grosser@fim.uni-passau.de>

	* graphite-clast-to-gimple.c (translate_clast): Do not short-cut
	code generation with gloog_error.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157223 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite         |    5 +++++
 gcc/graphite-clast-to-gimple.c |    2 +-
 2 files changed, 6 insertions(+), 1 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 25e0c51..ff6533a 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,8 @@
+2010-03-04  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	* graphite-clast-to-gimple.c (translate_clast): Do not short-cut
+	code generation with gloog_error.
+
 2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
 
 	* sese.c (expand_scalar_variables_ssa_name): Add new argument for type.
diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index 3e82075..fd631a4 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -952,7 +952,7 @@ translate_clast (sese region, loop_p context_loop, struct clast_stmt *stmt,
 		 htab_t newivs_index, htab_t bb_pbb_mapping, int level,
 		 htab_t params_index)
 {
-  if (!stmt || gloog_error)
+  if (!stmt)
     return next_e;
 
   if (CLAST_STMT_IS_A (stmt, stmt_root))
-- 
1.6.3.3


[-- Attachment #10: 0009-Fix-type-problems-in-loop-ivs.patch --]
[-- Type: text/x-diff, Size: 9502 bytes --]

From ac92c9f41ecba26c95e7aca21e5bbc416981575e Mon Sep 17 00:00:00 2001
From: grosser <grosser@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Fri, 5 Mar 2010 14:30:16 +0000
Subject: [PATCH 09/15] Fix type problems in loop ivs.

Fix pr42644.
Fix pr42130 (dealII).

2010-03-03  Tobias Grosser  <grosser@fim.uni-passau.de>

	* gcc/graphite-clast-to-gimple.c (clast_to_gcc_expression): Also
	handle conversions from pointer to integers.
	(gcc_type_for_cloog_iv): Choose the smalles signed integer as an
	induction variable, to be able to work with code generated by
	CLooG.
	* gcc/graphite-sese-to-poly.c (scop_ivs_can_be_represented): New.
	(build_poly_scop): Bail out if we cannot codegen a loop.
	* gcc/testsuite/gcc.dg/graphite/id-18.c: New.
	* gcc/testsuite/gcc.dg/graphite/run-id-pr42644.c: New.
	* libgomp/testsuite/libgomp.graphite/force-parallel-1.c: Adjust.
	* libgomp/testsuite/libgomp.graphite/force-parallel-2.c: Adjust.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157241 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/graphite-clast-to-gimple.c                     |   67 +++++++++++++++++--
 gcc/graphite-sese-to-poly.c                        |   36 +++++++++++
 gcc/testsuite/gcc.dg/graphite/id-18.c              |    7 ++
 gcc/testsuite/gcc.dg/graphite/run-id-pr42644.c     |   32 +++++++++
 .../testsuite/libgomp.graphite/force-parallel-1.c  |    2 +-
 .../testsuite/libgomp.graphite/force-parallel-2.c  |    2 +-
 6 files changed, 137 insertions(+), 9 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/graphite/id-18.c
 create mode 100644 gcc/testsuite/gcc.dg/graphite/run-id-pr42644.c

diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
index fd631a4..64ddbb8 100644
--- a/gcc/graphite-clast-to-gimple.c
+++ b/gcc/graphite-clast-to-gimple.c
@@ -282,14 +282,24 @@ clast_to_gcc_expression (tree type, struct clast_expr *e,
 	      {
 		tree name = clast_name_to_gcc (t->var, region, newivs,
 					       newivs_index, params_index);
-		return fold_convert (type, name);
+
+		if (POINTER_TYPE_P (TREE_TYPE (name)) != POINTER_TYPE_P (type))
+		  name = fold_convert (sizetype, name);
+
+		name = fold_convert (type, name);
+		return name;
 	      }
 
 	    else if (value_mone_p (t->val))
 	      {
 		tree name = clast_name_to_gcc (t->var, region, newivs,
 					       newivs_index, params_index);
+
+		if (POINTER_TYPE_P (TREE_TYPE (name)) != POINTER_TYPE_P (type))
+		  name = fold_convert (sizetype, name);
+
 		name = fold_convert (type, name);
+
 		return fold_build1 (NEGATE_EXPR, type, name);
 	      }
 	    else
@@ -297,7 +307,12 @@ clast_to_gcc_expression (tree type, struct clast_expr *e,
 		tree name = clast_name_to_gcc (t->var, region, newivs,
 					       newivs_index, params_index);
 		tree cst = gmp_cst_to_tree (type, t->val);
+
+		if (POINTER_TYPE_P (TREE_TYPE (name)) != POINTER_TYPE_P (type))
+		  name = fold_convert (sizetype, name);
+
 		name = fold_convert (type, name);
+
 		if (!POINTER_TYPE_P (type))
 		  return fold_build2 (MULT_EXPR, type, cst, name);
 
@@ -532,9 +547,16 @@ clast_get_body_of_loop (struct clast_stmt *stmt)
   gcc_unreachable ();
 }
 
-/* Given a CLOOG_IV, returns the type that it should have in GCC land.
-   If the information is not available, i.e. in the case one of the
-   transforms created the loop, just return integer_type_node.  */
+/* Given a CLOOG_IV, return the type that CLOOG_IV should have in GCC
+   land.  The selected type is big enough to include the original loop
+   iteration variable, but signed to work with the subtractions CLooG
+   may have introduced.  If such a type is not available, we fail.
+
+   TODO: Do not always return long_long, but the smallest possible
+   type, that still holds the original type.
+
+   TODO: Get the types using CLooG instead.  This enables further
+   optimizations, but needs CLooG support.  */
 
 static tree
 gcc_type_for_cloog_iv (const char *cloog_iv, gimple_bb_p gbb)
@@ -546,9 +568,40 @@ gcc_type_for_cloog_iv (const char *cloog_iv, gimple_bb_p gbb)
   slot = htab_find_slot (GBB_CLOOG_IV_TYPES (gbb), &tmp, NO_INSERT);
 
   if (slot && *slot)
-    return ((ivtype_map_elt) *slot)->type;
+    {
+      tree type = ((ivtype_map_elt) *slot)->type;
+      int type_precision = TYPE_PRECISION (type);
+
+      /* Find the smallest signed type possible.  */
+      if (!TYPE_UNSIGNED (type))
+	{
+	  if (type_precision <= TYPE_PRECISION (integer_type_node))
+	    return integer_type_node;
+
+	  if (type_precision <= TYPE_PRECISION (long_integer_type_node))
+	    return long_integer_type_node;
+
+	  if (type_precision <= TYPE_PRECISION (long_long_integer_type_node))
+	    return long_long_integer_type_node;
+
+	  gcc_unreachable ();
+	}
+
+      if (type_precision < TYPE_PRECISION (integer_type_node))
+	return integer_type_node;
+
+      if (type_precision < TYPE_PRECISION (long_integer_type_node))
+	return long_integer_type_node;
+
+      if (type_precision < TYPE_PRECISION (long_long_integer_type_node))
+	return long_long_integer_type_node;
+
+      /* There is no signed type available, that is large enough to hold the
+	 original value.  */
+      gcc_unreachable ();
+    }
 
-  return integer_type_node;
+  return long_long_integer_type_node;
 }
 
 /* Returns the induction variable for the loop that gets translated to
@@ -1046,7 +1099,7 @@ compute_cloog_iv_types_1 (poly_bb_p pbb, struct clast_user_stmt *user_stmt)
       if (slot && !*slot)
 	{
 	  tree oldiv = pbb_to_depth_to_oldiv (pbb, index);
-	  tree type = oldiv ? TREE_TYPE (oldiv) : integer_type_node;
+	  tree type = TREE_TYPE (oldiv);
 	  *slot = new_ivtype_map_elt (tmp.cloog_iv, type);
 	}
     }
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 3ee431f..279a905 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -2893,6 +2893,38 @@ scop_canonicalize_loops (scop_p scop)
       graphite_loop_normal_form (loop);
 }
 
+/* Can all ivs be represented by a signed integer?
+   As CLooG might generate negative values in its expressions, signed loop ivs
+   are required in the backend. */
+static bool
+scop_ivs_can_be_represented (scop_p scop)
+{
+  loop_iterator li;
+  loop_p loop;
+
+  FOR_EACH_LOOP (li, loop, 0)
+    {
+      tree type;
+      int precision;
+
+      if (!loop_in_sese_p (loop, SCOP_REGION (scop)))
+	continue;
+
+      if (!loop->single_iv)
+	continue;
+
+      type = TREE_TYPE(loop->single_iv);
+      precision = TYPE_PRECISION (type);
+
+      if (TYPE_UNSIGNED (type)
+	  && precision >= TYPE_PRECISION (long_long_integer_type_node))
+	return false;
+    }
+
+  return true;
+}
+
+
 /* Builds the polyhedral representation for a SESE region.  */
 
 bool
@@ -2915,6 +2947,10 @@ build_poly_scop (scop_p scop)
     return false;
 
   scop_canonicalize_loops (scop);
+
+  if (!scop_ivs_can_be_represented (scop))
+    return false;
+
   build_sese_loop_nests (region);
   build_sese_conditions (region);
   find_scop_parameters (scop);
diff --git a/gcc/testsuite/gcc.dg/graphite/id-18.c b/gcc/testsuite/gcc.dg/graphite/id-18.c
new file mode 100644
index 0000000..77628fa
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/id-18.c
@@ -0,0 +1,7 @@
+long do_hash (const char * lo, const char * hi)
+{
+	int val = 0;
+	for (; lo < hi; ++lo)
+		val = *lo;
+	return val;
+}
diff --git a/gcc/testsuite/gcc.dg/graphite/run-id-pr42644.c b/gcc/testsuite/gcc.dg/graphite/run-id-pr42644.c
new file mode 100644
index 0000000..2e90064
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/run-id-pr42644.c
@@ -0,0 +1,32 @@
+/* Testcase extracted from test 183.equake in SPEC CPU2000.  */
+double Ke[2], ds[2];
+
+void foo(double Ke[2], int i, double ds[],  int column)
+{
+  double tt, ts;
+  int j;
+
+  for (j = 0; j < 2; j++)
+    {
+      ++column;
+      ts = ds[i];
+      if (i == j)
+	tt = 123;
+      else
+	tt = 0;
+      Ke[column] = Ke[column] + ts + tt;
+    }
+}
+
+int
+main ()
+{
+  int i, j;
+
+  ds[0] = 1.0;
+  ds[1] = 1.0;
+
+  foo(Ke, 0, ds, -1);
+
+  return (int) Ke[0] != 124;
+}
diff --git a/libgomp/testsuite/libgomp.graphite/force-parallel-1.c b/libgomp/testsuite/libgomp.graphite/force-parallel-1.c
index 7f043d8..1ad4b41 100644
--- a/libgomp/testsuite/libgomp.graphite/force-parallel-1.c
+++ b/libgomp/testsuite/libgomp.graphite/force-parallel-1.c
@@ -23,7 +23,7 @@ int main(void)
 }
 
 /* Check that parallel code generation part make the right answer.  */
-/* { dg-final { scan-tree-dump-times "1 loops carried no dependency" 2 "graphite" } } */
+/* { dg-final { scan-tree-dump-times "1 loops carried no dependency" 1 "graphite" } } */
 /* { dg-final { cleanup-tree-dump "graphite" } } */
 /* { dg-final { scan-tree-dump-times "loopfn" 5 "optimized" } } */
 /* { dg-final { cleanup-tree-dump "parloops" } } */
diff --git a/libgomp/testsuite/libgomp.graphite/force-parallel-2.c b/libgomp/testsuite/libgomp.graphite/force-parallel-2.c
index 03d8236..1ce0feb 100644
--- a/libgomp/testsuite/libgomp.graphite/force-parallel-2.c
+++ b/libgomp/testsuite/libgomp.graphite/force-parallel-2.c
@@ -23,7 +23,7 @@ int main(void)
 }
 
 /* Check that parallel code generation part make the right answer.  */
-/* { dg-final { scan-tree-dump-times "2 loops carried no dependency" 2 "graphite" } } */
+/* { dg-final { scan-tree-dump-times "2 loops carried no dependency" 1 "graphite" } } */
 /* { dg-final { cleanup-tree-dump "graphite" } } */
 /* { dg-final { scan-tree-dump-times "loopfn" 5 "optimized" } } */
 /* { dg-final { cleanup-tree-dump "parloops" } } */
-- 
1.6.3.3


[-- Attachment #11: 0010-Add-forgotten-ChangeLog-entries.patch --]
[-- Type: text/x-diff, Size: 3147 bytes --]

From 7a2af21d7cdfd9935f1a2edc5697fd1753a5dec2 Mon Sep 17 00:00:00 2001
From: grosser <grosser@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Fri, 5 Mar 2010 14:51:16 +0000
Subject: [PATCH 10/15] Add forgotten ChangeLog entries.

Eric pointed out that my ChangeLog was incorrect. As I forgot to
commit it at all, here the corrected ChangeLogs for the last commit.

gcc/
2010-03-05  Tobias Grosser  <grosser@fim.uni-passau.de>
	    Ramakrishna Upadrasta <Ramakrishna.Upadrasta@inria.fr>

	* graphite-clast-to-gimple.c (clast_to_gcc_expression): Also
	handle conversions from pointer to integers.
	(gcc_type_for_cloog_iv): Choose the smalles signed integer as an
	induction variable, to be able to work with code generated by
	CLooG.
	* graphite-sese-to-poly.c (scop_ivs_can_be_represented): New.
	(build_poly_scop): Bail out if we cannot codegen a loop.

gcc/testsuite/
2010-03-05  Tobias Grosser  <grosser@fim.uni-passau.de>

	* gcc.dg/graphite/id-18.c: New.
	* gcc.dg/graphite/run-id-pr42644.c: New.

libgomp/
2010-03-05  Tobias Grosser  <grosser@fim.uni-passau.de>

	* testsuite/libgomp.graphite/force-parallel-1.c: Adjust.
	* testsuite/libgomp.graphite/force-parallel-2.c: Adjust.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157242 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite           |   10 ++++++++++
 gcc/testsuite/ChangeLog.graphite |    5 +++++
 libgomp/ChangeLog.graphite       |    5 +++++
 3 files changed, 20 insertions(+), 0 deletions(-)
 create mode 100644 gcc/testsuite/ChangeLog.graphite

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index ff6533a..8790e87 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,13 @@
+2010-03-05  Tobias Grosser  <grosser@fim.uni-passau.de>
+	    Ramakrishna Upadrasta <Ramakrishna.Upadrasta@inria.fr>
+
+	* graphite-clast-to-gimple.c (clast_to_gcc_expression): Also
+	handle conversions from pointer to integers.
+	(gcc_type_for_cloog_iv): Choose the smalles signed integer as an
+	induction variable, to be able to work with code generated by CLooG.
+	* graphite-sese-to-poly.c (scop_ivs_can_be_represented): New.
+	(build_poly_scop): Bail out if we cannot codegen a loop.
+
 2010-03-04  Tobias Grosser  <grosser@fim.uni-passau.de>
 
 	* graphite-clast-to-gimple.c (translate_clast): Do not short-cut
diff --git a/gcc/testsuite/ChangeLog.graphite b/gcc/testsuite/ChangeLog.graphite
new file mode 100644
index 0000000..3d70811
--- /dev/null
+++ b/gcc/testsuite/ChangeLog.graphite
@@ -0,0 +1,5 @@
+2010-03-05  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	* gcc.dg/graphite/id-18.c: New.
+	* gcc.dg/graphite/run-id-pr42644.c: New.
+
diff --git a/libgomp/ChangeLog.graphite b/libgomp/ChangeLog.graphite
index ee2cddb..74049a9 100644
--- a/libgomp/ChangeLog.graphite
+++ b/libgomp/ChangeLog.graphite
@@ -1,3 +1,8 @@
+2010-03-05  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	* testsuite/libgomp.graphite/force-parallel-1.c: Adjust.
+	* testsuite/libgomp.graphite/force-parallel-2.c: Adjust.
+
 2010-02-07  Sebastian Pop  <sebastian.pop@amd.com>
 
 	* testsuite/libgomp.graphite/force-parallel-5.c: Un-XFAILed.
-- 
1.6.3.3


[-- Attachment #12: 0011-Add-testcase-from-PR43065.patch --]
[-- Type: text/x-diff, Size: 1613 bytes --]

From aef3bb9b5cd9f69fc4052792f4618f4b185b69e0 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Fri, 5 Mar 2010 23:04:17 +0000
Subject: [PATCH 11/15] Add testcase from PR43065.

2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/43065
	* gcc.dg/graphite/run-id-3.c: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157247 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                   |    5 +++++
 gcc/testsuite/gcc.dg/graphite/run-id-3.c |   24 ++++++++++++++++++++++++
 2 files changed, 29 insertions(+), 0 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/graphite/run-id-3.c

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 8790e87..4d2853c 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,3 +1,8 @@
+2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43065
+	* gcc.dg/graphite/run-id-3.c: New.
+
 2010-03-05  Tobias Grosser  <grosser@fim.uni-passau.de>
 	    Ramakrishna Upadrasta <Ramakrishna.Upadrasta@inria.fr>
 
diff --git a/gcc/testsuite/gcc.dg/graphite/run-id-3.c b/gcc/testsuite/gcc.dg/graphite/run-id-3.c
new file mode 100644
index 0000000..e708ba0
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/run-id-3.c
@@ -0,0 +1,24 @@
+extern void abort (void);
+
+__attribute__ ((noinline)) int
+foo (int *zzz, unsigned int kk)
+{
+  int a, b, d;
+
+  a = b = 0;
+  for (d = 0; d < 1000; d++)
+    {
+      if (kk != 0)
+        b = *zzz;
+    }
+
+  return b;
+}
+
+int
+main (void)
+{
+  if (foo (0, 0) != 0)
+    abort();
+  return 0;
+}
-- 
1.6.3.3


[-- Attachment #13: 0012-Fix-PR43065-Insert-bounds-on-pointer-type-parameters.patch --]
[-- Type: text/x-diff, Size: 2983 bytes --]

From 926b0ec36b29d975ac956ac0da50050a490661e9 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Fri, 5 Mar 2010 23:48:43 +0000
Subject: [PATCH 12/15] Fix PR43065: Insert bounds on pointer type parameters.

2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
	    Reza Yazdani  <reza.yazdani@amd.com>

	PR middle-end/43065
	* graphite-sese-to-poly.c (add_param_constraints): Insert bounds
	on pointer type parameters.

	* gcc.dg/graphite/run-id-4.c: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157248 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                   |   11 ++++++++++-
 gcc/graphite-sese-to-poly.c              |   18 ++++++++++++------
 gcc/testsuite/gcc.dg/graphite/run-id-4.c |   28 ++++++++++++++++++++++++++++
 3 files changed, 50 insertions(+), 7 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/graphite/run-id-4.c

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 4d2853c..c3d8220 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,4 +1,13 @@
-2010-03-04  Sebastian Pop  <sebastian.pop@amd.com>
+2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
+	    Reza Yazdani  <reza.yazdani@amd.com>
+
+	PR middle-end/43065
+	* graphite-sese-to-poly.c (add_param_constraints): Insert bounds
+	on pointer type parameters.
+
+	* gcc.dg/graphite/run-id-4.c: New.
+
+2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
 
 	PR middle-end/43065
 	* gcc.dg/graphite/run-id-3.c: New.
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 279a905..8933072 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -1499,13 +1499,19 @@ add_param_constraints (scop_p scop, ppl_Polyhedron_t context, graphite_dim_t p)
   ppl_Linear_Expression_t le;
   tree parameter = VEC_index (tree, SESE_PARAMS (SCOP_REGION (scop)), p);
   tree type = TREE_TYPE (parameter);
-  tree lb, ub;
+  tree lb = NULL_TREE;
+  tree ub = NULL_TREE;
 
-  if (!INTEGRAL_TYPE_P (type))
-    return;
-
-  lb = TYPE_MIN_VALUE (type);
-  ub = TYPE_MAX_VALUE (type);
+  if (INTEGRAL_TYPE_P (type))
+    {
+      lb = TYPE_MIN_VALUE (type);
+      ub = TYPE_MAX_VALUE (type);
+    }
+  else if (POINTER_TYPE_P (type))
+    {
+      lb = TYPE_MIN_VALUE (unsigned_type_node);
+      ub = TYPE_MAX_VALUE (unsigned_type_node);
+    }
 
   if (lb)
     {
diff --git a/gcc/testsuite/gcc.dg/graphite/run-id-4.c b/gcc/testsuite/gcc.dg/graphite/run-id-4.c
new file mode 100644
index 0000000..143a449
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/run-id-4.c
@@ -0,0 +1,28 @@
+/* PR rtl-optimization/24899 */
+
+extern void abort (void);
+
+__attribute__ ((noinline)) int
+foo (int x, int y, int *z)
+{
+  int a, b, c, d;
+
+  a = b = 0;
+  for (d = 0; d < y; d++)
+    {
+      if (z)
+	b = d * *z;
+      for (c = 0; c < x; c++)
+	a += b;
+    }
+
+  return a;
+}
+
+int
+main (void)
+{
+  if (foo (3, 2, 0) != 0)
+    abort ();
+  return 0;
+}
-- 
1.6.3.3


[-- Attachment #14: 0013-Use-sizetype-instead-of-unsigned_type_node.patch --]
[-- Type: text/x-diff, Size: 1591 bytes --]

From 4cd5894e9ce9ecfdc5ff689c03bee8331cee2b51 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Fri, 5 Mar 2010 23:54:14 +0000
Subject: [PATCH 13/15] Use sizetype instead of unsigned_type_node.

2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-sese-to-poly.c (add_param_constraints): Use sizetype
	instead of unsigned_type_node.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157249 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite      |    5 +++++
 gcc/graphite-sese-to-poly.c |    4 ++--
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index c3d8220..57176ad 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,4 +1,9 @@
 2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (add_param_constraints): Use sizetype
+	instead of unsigned_type_node.
+
+2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
 	    Reza Yazdani  <reza.yazdani@amd.com>
 
 	PR middle-end/43065
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 8933072..99d83d6 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -1509,8 +1509,8 @@ add_param_constraints (scop_p scop, ppl_Polyhedron_t context, graphite_dim_t p)
     }
   else if (POINTER_TYPE_P (type))
     {
-      lb = TYPE_MIN_VALUE (unsigned_type_node);
-      ub = TYPE_MAX_VALUE (unsigned_type_node);
+      lb = TYPE_MIN_VALUE (sizetype);
+      ub = TYPE_MAX_VALUE (sizetype);
     }
 
   if (lb)
-- 
1.6.3.3


[-- Attachment #15: 0014-Use-lower-upper-_bound_in_type.patch --]
[-- Type: text/x-diff, Size: 1997 bytes --]

From dc8bd59bfbc6e9bb7baf8537be6cd413f7ed6e05 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Sat, 6 Mar 2010 07:23:30 +0000
Subject: [PATCH 14/15] Use {lower,upper}_bound_in_type.

2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>

	* graphite-sese-to-poly.c (add_param_constraints): Use
	lower_bound_in_type and upper_bound_in_type.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@157255 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite      |    5 +++++
 gcc/graphite-sese-to-poly.c |   19 +++++++++----------
 2 files changed, 14 insertions(+), 10 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 57176ad..9dcdcda 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,10 @@
 2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
 
+	* graphite-sese-to-poly.c (add_param_constraints): Use
+	lower_bound_in_type and upper_bound_in_type.
+
+2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
+
 	* graphite-sese-to-poly.c (add_param_constraints): Use sizetype
 	instead of unsigned_type_node.
 
diff --git a/gcc/graphite-sese-to-poly.c b/gcc/graphite-sese-to-poly.c
index 99d83d6..11bddf8 100644
--- a/gcc/graphite-sese-to-poly.c
+++ b/gcc/graphite-sese-to-poly.c
@@ -1502,16 +1502,15 @@ add_param_constraints (scop_p scop, ppl_Polyhedron_t context, graphite_dim_t p)
   tree lb = NULL_TREE;
   tree ub = NULL_TREE;
 
-  if (INTEGRAL_TYPE_P (type))
-    {
-      lb = TYPE_MIN_VALUE (type);
-      ub = TYPE_MAX_VALUE (type);
-    }
-  else if (POINTER_TYPE_P (type))
-    {
-      lb = TYPE_MIN_VALUE (sizetype);
-      ub = TYPE_MAX_VALUE (sizetype);
-    }
+  if (POINTER_TYPE_P (type) || !TYPE_MIN_VALUE (type))
+    lb = lower_bound_in_type (type, type);
+  else
+    lb = TYPE_MIN_VALUE (type);
+
+  if (POINTER_TYPE_P (type) || !TYPE_MAX_VALUE (type))
+    ub = upper_bound_in_type (type, type);
+  else
+    ub = TYPE_MAX_VALUE (type);
 
   if (lb)
     {
-- 
1.6.3.3


[-- Attachment #16: 0015-Add-ChangeLog-entries.patch --]
[-- Type: text/x-diff, Size: 4501 bytes --]

From f3ec517e276e87858a62f5ef84fb42ec6b092feb Mon Sep 17 00:00:00 2001
From: Sebastian Pop <sebpop@gmail.com>
Date: Mon, 8 Mar 2010 11:38:36 -0600
Subject: [PATCH 15/15] Add ChangeLog entries.

---
 gcc/ChangeLog           |   74 +++++++++++++++++++++++++++++++++++++++++++++++
 gcc/testsuite/ChangeLog |   23 ++++++++++++++
 libgomp/ChangeLog       |    7 ++++
 3 files changed, 104 insertions(+), 0 deletions(-)

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 67086f5..7eae746 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,77 @@
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (add_param_constraints): Use
+	lower_bound_in_type and upper_bound_in_type.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (add_param_constraints): Use sizetype
+	instead of unsigned_type_node.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+	    Reza Yazdani  <reza.yazdani@amd.com>
+
+	PR middle-end/43065
+	* graphite-sese-to-poly.c (add_param_constraints): Insert bounds
+	on pointer type parameters.
+
+2010-03-08  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	PR middle-end/42644
+	PR middle-end/42130
+	* graphite-clast-to-gimple.c (clast_to_gcc_expression): Also
+	handle conversions from pointer to integers.
+	(gcc_type_for_cloog_iv): Choose the smalles signed integer as an
+	induction variable, to be able to work with code generated by
+	CLooG.
+	* graphite-sese-to-poly.c (scop_ivs_can_be_represented): New.
+	(build_poly_scop): Bail out if we cannot codegen a loop.
+
+2010-03-08  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	* graphite-clast-to-gimple.c (translate_clast): Do not short-cut
+	code generation with gloog_error.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* sese.c (expand_scalar_variables_ssa_name): Add new argument for type.
+	Call fold_convert on all the returned values.
+	(expand_scalar_variables_expr): Pass to expand_scalar_variables_ssa_name
+	the type of the resulting expression.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-ppl.c (ppl_min_for_le_polyhedron): Renamed
+	ppl_min_for_le_pointset.  Use ppl_Pointset_Powerset_C_Polyhedron_minimize.
+	* graphite-ppl.h (ppl_min_for_le_polyhedron): Update declaration.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-dependences.c (map_into_dep_poly): Removed.
+	(dependence_polyhedron_1): Use combine_context_id_scat.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-poly.h (struct poly_scattering): Add layout documentation.
+	(struct poly_bb): Same.
+	(combine_context_id_scat): New.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/42326
+	* sese.c (name_defined_in_loop_p): Return false for default
+	definitions.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-clast-to-gimple.c (find_cloog_iv_in_expr): Simplify
+	and clean up the logic.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	* graphite-sese-to-poly.c (add_param_constraints): Enabled: remove
+	early return.
+
 2010-03-08  Jakub Jelinek  <jakub@redhat.com>
 
 	* var-tracking.c (remove_cselib_value_chains): Define only for
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 197d695..b9ade14 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,26 @@
+2010-03-05  Sebastian Pop  <sebastian.pop@amd.com>
+	    Reza Yazdani  <reza.yazdani@amd.com>
+
+	PR middle-end/43065
+	* gcc.dg/graphite/run-id-4.c: New.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43065
+	* gcc.dg/graphite/run-id-3.c: New.
+
+2010-03-08  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	PR middle-end/42644
+	PR middle-end/42130
+	* gcc.dg/graphite/id-18.c: New.
+	* gcc.dg/graphite/run-id-pr42644.c: New.
+
+2010-03-08  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/42326
+	* gcc.dg/graphite/pr42326.c: New.
+
 2010-03-08  Richard Guenther  <rguenther@suse.de>
 
 	PR tree-optimization/43269
diff --git a/libgomp/ChangeLog b/libgomp/ChangeLog
index dcd532f..420f354 100644
--- a/libgomp/ChangeLog
+++ b/libgomp/ChangeLog
@@ -1,3 +1,10 @@
+2010-03-08  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	PR middle-end/42644
+	PR middle-end/42130
+	* testsuite/libgomp.graphite/force-parallel-1.c: Adjust.
+	* testsuite/libgomp.graphite/force-parallel-2.c: Adjust.
+
 2010-01-29  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>
 
 	* testsuite/libgomp.c++/task-1.C: Renamed err to e.
-- 
1.6.3.3

[patch] Merge to trunk from Graphite branch

[Sebastian Pop]

[-- Attachment #1: Type: text/plain, Size: 137 bytes --]

Hi,

I committed to trunk the attached patches to fix 4 PRs.
These patches passed regstrap and the SPEC 2006 benchmark tests.

Sebastian

[-- Attachment #2: 0001-Fix-PR43026-handle-COMPONENT_REFs-in-expand-scalar-e.patch --]
[-- Type: text/x-diff, Size: 3143 bytes --]

From 986be805e7cad0595bd395db50681db55bfd497f Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Mon, 22 Feb 2010 13:10:49 +0000
Subject: [PATCH 1/7] Fix PR43026: handle COMPONENT_REFs in expand scalar expressions.

2010-02-11  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/43026
	* sese.c (expand_scalar_variables_expr): Handle COMPONENT_REF.

	* g++.dg/graphite/pr43026.C: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@156956 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                  |    7 ++++
 gcc/sese.c                              |    3 ++
 gcc/testsuite/g++.dg/graphite/pr43026.C |   51 +++++++++++++++++++++++++++++++
 3 files changed, 61 insertions(+), 0 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/graphite/pr43026.C

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 4bb6dc2..cb969c5 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,12 @@
 2010-02-11  Sebastian Pop  <sebastian.pop@amd.com>
 
+	PR middle-end/43026
+	* sese.c (expand_scalar_variables_expr): Handle COMPONENT_REF.
+
+	* g++.dg/graphite/pr43026.C: New.
+
+2010-02-11  Sebastian Pop  <sebastian.pop@amd.com>
+
 	PR middle-end/43012
 	* gcc.dg/graphite/pr43012.c: New.
 
diff --git a/gcc/sese.c b/gcc/sese.c
index 6fb4065..ebf9154 100644
--- a/gcc/sese.c
+++ b/gcc/sese.c
@@ -897,6 +897,9 @@ expand_scalar_variables_expr (tree type, tree op0, enum tree_code code,
 	    return build4 (ARRAY_REF, type, base, subscript, op02, op03);
 	  }
 
+	case COMPONENT_REF:
+	  return op0;
+
 	default:
 	  /* The above cases should catch everything.  */
 	  gcc_unreachable ();
diff --git a/gcc/testsuite/g++.dg/graphite/pr43026.C b/gcc/testsuite/g++.dg/graphite/pr43026.C
new file mode 100644
index 0000000..0e33f10
--- /dev/null
+++ b/gcc/testsuite/g++.dg/graphite/pr43026.C
@@ -0,0 +1,51 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fgraphite-identity -m32" } */
+
+template<typename Tp > class vector { };
+
+template <int rank, int dim> class Tensor;
+
+template <int dim> class Tensor<1,dim> {
+public:
+  Tensor (const Tensor<1,dim> &);
+private:
+  double values[(dim != 0) ? (dim) : 1];
+};
+
+template <int dim>
+#ifdef NOINLINE
+// declaring this noinline prevents the ICE
+__attribute__ ((noinline))
+#endif
+Tensor<1,dim>::Tensor (const Tensor<1,dim> &p)
+{
+  for (unsigned int i = 0; i < dim; ++i)
+    values[i] = p.values[i];
+}
+
+template <int rank, int dim>
+class Tensor {
+  Tensor<rank-1,dim> subtensor[dim];
+};
+
+template <int dim> class Base {
+public:
+  const unsigned int npoints;
+  const unsigned int dofs;
+  const Tensor<2,dim> &s2d (const unsigned int fno,
+                            const unsigned int pno) const;
+  void getf2d (vector<Tensor<2,dim> >& d2) const;
+};
+
+template <int dim>
+void Base<dim>:: getf2d
+  (vector<Tensor<2,dim> > &d2) const
+{
+  unsigned int point, sf;
+
+  for (point = 0; point < npoints; ++point)
+    for (sf = 0; sf < dofs; ++sf)
+      Tensor<2,dim> tmp = s2d (sf, point);
+}
+
+template void Base<3>::getf2d (vector<Tensor<2,3> > &) const;
-- 
1.6.3.3


[-- Attachment #3: 0002-Fix-PR43140-Add-Wno-conversion-null-to-pr41305.C.patch --]
[-- Type: text/x-diff, Size: 1725 bytes --]

From be0c4357df7018479c04999b2f69a14858c2ba44 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Mon, 22 Feb 2010 14:55:08 +0000
Subject: [PATCH 2/7] Fix PR43140: Add -Wno-conversion-null to pr41305.C.
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 8bit

2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
	    Manuel López-Ibáñez  <manu@gcc.gnu.org>

	PR middle-end/43140
	* g++.dg/graphite/pr41305.C: Add -Wno-conversion-null.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@156963 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                  |    8 +++++++-
 gcc/testsuite/g++.dg/graphite/pr41305.C |    2 +-
 2 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index cb969c5..56b4694 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,4 +1,10 @@
-2010-02-11  Sebastian Pop  <sebastian.pop@amd.com>
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+	    Manuel López-Ibáñez  <manu@gcc.gnu.org>
+
+	PR middle-end/43140
+	* g++.dg/graphite/pr41305.C: Add -Wno-conversion-null.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
 
 	PR middle-end/43026
 	* sese.c (expand_scalar_variables_expr): Handle COMPONENT_REF.
diff --git a/gcc/testsuite/g++.dg/graphite/pr41305.C b/gcc/testsuite/g++.dg/graphite/pr41305.C
index f558e7c..6a30b0e 100644
--- a/gcc/testsuite/g++.dg/graphite/pr41305.C
+++ b/gcc/testsuite/g++.dg/graphite/pr41305.C
@@ -1,5 +1,5 @@
 // { dg-do compile }
-// { dg-options "-O3 -floop-interchange" }
+// { dg-options "-O3 -floop-interchange -Wno-conversion-null" }
 
 void __throw_bad_alloc ();
 
-- 
1.6.3.3


[-- Attachment #4: 0003-Fix-PR43097-rename-only-SSA_NAMEs.patch --]
[-- Type: text/x-diff, Size: 3362 bytes --]

From 0b9a70f792614cfeb710b442663c04bc6f4b5521 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Mon, 22 Feb 2010 15:13:39 +0000
Subject: [PATCH 3/7] Fix PR43097: rename only SSA_NAMEs.

2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/43140
	* sese.c (get_rename): Assert that old_name is an SSA_NAME.
	(rename_variables_in_stmt): Continue when the use is not an SSA_NAME.

	* gfortran.dg/graphite/pr43097.f: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@156964 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                       |    8 ++++++++
 gcc/sese.c                                   |   12 +++++++++---
 gcc/testsuite/gfortran.dg/graphite/pr43097.f |   25 +++++++++++++++++++++++++
 3 files changed, 42 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/gfortran.dg/graphite/pr43097.f

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index 56b4694..aae16e7 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,4 +1,12 @@
 2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43140
+	* sese.c (get_rename): Assert that old_name is an SSA_NAME.
+	(rename_variables_in_stmt): Continue when the use is not an SSA_NAME.
+
+	* gfortran.dg/graphite/pr43097.f: New.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
 	    Manuel López-Ibáñez  <manu@gcc.gnu.org>
 
 	PR middle-end/43140
diff --git a/gcc/sese.c b/gcc/sese.c
index ebf9154..d7a9faa 100644
--- a/gcc/sese.c
+++ b/gcc/sese.c
@@ -494,6 +494,7 @@ get_rename (htab_t map, tree old_name)
   struct rename_map_elt_s tmp;
   PTR *slot;
 
+  gcc_assert (TREE_CODE (old_name) == SSA_NAME);
   tmp.old_name = old_name;
   slot = htab_find_slot (map, &tmp, NO_INSERT);
 
@@ -658,14 +659,19 @@ rename_variables_in_stmt (gimple stmt, htab_t map, gimple_stmt_iterator *insert_
   FOR_EACH_SSA_USE_OPERAND (use_p, stmt, iter, SSA_OP_ALL_USES)
     {
       tree use = USE_FROM_PTR (use_p);
-      tree expr = get_rename (map, use);
-      tree type_use = TREE_TYPE (use);
-      tree type_expr = TREE_TYPE (expr);
+      tree expr, type_use, type_expr;
       gimple_seq stmts;
 
+      if (TREE_CODE (use) != SSA_NAME)
+	continue;
+
+      expr = get_rename (map, use);
       if (use == expr)
 	continue;
 
+      type_use = TREE_TYPE (use);
+      type_expr = TREE_TYPE (expr);
+
       if (type_use != type_expr
 	  || (TREE_CODE (expr) != SSA_NAME
 	      && is_gimple_reg (use)))
diff --git a/gcc/testsuite/gfortran.dg/graphite/pr43097.f b/gcc/testsuite/gfortran.dg/graphite/pr43097.f
new file mode 100644
index 0000000..4ddeed8
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/graphite/pr43097.f
@@ -0,0 +1,25 @@
+! { dg-options "-O2 -fgraphite-identity" }
+
+      subroutine foo (ldmx,ldmy,nx,ny,v)
+      implicit real*8 (a-h, o-z)
+      dimension v(5,ldmx,ldmy,*)
+      dimension tmat(5,5)
+
+      k = 2
+      do j = 2, ny-1
+         do i = 2, nx-1
+            do ip = 1, 4
+               do m = ip+1, 5
+                  v(m,i,j,k) = v(m,i,j,k) * m
+               end do
+            end do
+            do m = 5, 1, -1
+               do l = m+1, 5
+                  v(m,i,j,k) = v(l,i,j,k)
+               end do
+               v(m,i,j,k) = m
+           end do
+         end do
+      end do
+      return
+      end
-- 
1.6.3.3


[-- Attachment #5: 0004-Fix-PR-number-in-the-changelog-entry.patch --]
[-- Type: text/x-diff, Size: 829 bytes --]

From 08cc5dfadeb3c8461ff132c36ef9297f7c8679be Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Mon, 22 Feb 2010 15:17:38 +0000
Subject: [PATCH 4/7] Fix PR number in the changelog entry.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@156965 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index aae16e7..b3a820e 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,6 +1,6 @@
 2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
 
-	PR middle-end/43140
+	PR middle-end/43097
 	* sese.c (get_rename): Assert that old_name is an SSA_NAME.
 	(rename_variables_in_stmt): Continue when the use is not an SSA_NAME.
 
-- 
1.6.3.3


[-- Attachment #6: 0005-Fix-PR43083-Do-not-handle-regions-ending-with-multip.patch --]
[-- Type: text/x-diff, Size: 3946 bytes --]

From 9e199674aa67ed17bf84cb68512709d7b50a3c16 Mon Sep 17 00:00:00 2001
From: spop <spop@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Mon, 22 Feb 2010 17:40:23 +0000
Subject: [PATCH 5/7] Fix PR43083: Do not handle regions ending with multiple edges on the exit BB.

2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>

	PR middle-end/43083
	* graphite-scop-detection.c (create_single_exit_edge): Move
	the call to find_single_exit_edge to....
	(create_sese_edges): ...here.  Don't handle multiple edges
	exiting the function.
	(build_graphite_scops): Don't handle multiple edges
	exiting the function.

	* gcc.dg/graphite/pr43083.c: New.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/graphite@156970 138bc75d-0d04-0410-961f-82ee72b054a4
---
 gcc/ChangeLog.graphite                  |   12 ++++++++++++
 gcc/graphite-scop-detection.c           |   17 +++++++++++------
 gcc/testsuite/gcc.dg/graphite/pr43083.c |   14 ++++++++++++++
 3 files changed, 37 insertions(+), 6 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/graphite/pr43083.c

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index b3a820e..f1e9005 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -1,5 +1,17 @@
 2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
 
+	PR middle-end/43083
+	* graphite-scop-detection.c (create_single_exit_edge): Move
+	the call to find_single_exit_edge to....
+	(create_sese_edges): ...here.  Don't handle multiple edges
+	exiting the function.
+	(build_graphite_scops): Don't handle multiple edges
+	exiting the function.
+
+	* gcc.dg/graphite/pr43083.c: New.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
 	PR middle-end/43097
 	* sese.c (get_rename): Assert that old_name is an SSA_NAME.
 	(rename_variables_in_stmt): Continue when the use is not an SSA_NAME.
diff --git a/gcc/graphite-scop-detection.c b/gcc/graphite-scop-detection.c
index d89f0f8..5c1dbbd 100644
--- a/gcc/graphite-scop-detection.c
+++ b/gcc/graphite-scop-detection.c
@@ -935,9 +935,6 @@ create_single_exit_edge (sd_region *region)
   edge forwarder = NULL;
   basic_block exit;
 
-  if (find_single_exit_edge (region))
-    return;
-
   /* We create a forwarder bb (5) for all edges leaving this region
      (3->5, 4->5).  All other edges leading to the same bb, are moved
      to a new bb (6).  If these edges where part of another region (2->5)
@@ -1031,7 +1028,10 @@ create_sese_edges (VEC (sd_region, heap) *regions)
   mark_exit_edges (regions);
 
   for (i = 0; VEC_iterate (sd_region, regions, i, s); i++)
-    create_single_exit_edge (s);
+    /* Don't handle multiple edges exiting the function.  */
+    if (!find_single_exit_edge (s)
+	&& s->exit != EXIT_BLOCK_PTR)
+      create_single_exit_edge (s);
 
   unmark_exit_edges (regions);
 
@@ -1057,7 +1057,12 @@ build_graphite_scops (VEC (sd_region, heap) *regions,
     {
       edge entry = find_single_entry_edge (s);
       edge exit = find_single_exit_edge (s);
-      scop_p scop = new_scop (new_sese (entry, exit));
+      scop_p scop;
+
+      if (!exit)
+	continue;
+
+      scop = new_scop (new_sese (entry, exit));
       VEC_safe_push (scop_p, heap, *scops, scop);
 
       /* Are there overlapping SCoPs?  */
@@ -1323,7 +1328,7 @@ build_scops (VEC (scop_p, heap) **scops)
 
   canonicalize_loop_closed_ssa_form ();
   build_scops_1 (single_succ (ENTRY_BLOCK_PTR), ENTRY_BLOCK_PTR->loop_father,
-			      &regions, loop);
+		 &regions, loop);
   create_sese_edges (regions);
   build_graphite_scops (regions, scops);
 
diff --git a/gcc/testsuite/gcc.dg/graphite/pr43083.c b/gcc/testsuite/gcc.dg/graphite/pr43083.c
new file mode 100644
index 0000000..afb97af
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/graphite/pr43083.c
@@ -0,0 +1,14 @@
+/* { dg-options "-O3 -fgraphite-identity" } */
+
+extern void baz(void);
+
+static inline int bar(void)
+{
+  int i;
+  for (i = 0; i < 10; i++) baz();
+}
+
+int foo(void)
+{
+  if (bar() != 0) return 0;
+}
-- 
1.6.3.3


[-- Attachment #7: 0006-Add-missing-changelog-entry.patch --]
[-- Type: text/x-diff, Size: 689 bytes --]

From a886129a62af29ceb93581a2a2a17a52f5a5d195 Mon Sep 17 00:00:00 2001
From: Sebastian Pop <sebpop@gmail.com>
Date: Tue, 23 Feb 2010 06:49:39 -0600
Subject: [PATCH 6/7] Add missing changelog entry.

---
 gcc/ChangeLog.graphite |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/gcc/ChangeLog.graphite b/gcc/ChangeLog.graphite
index f1e9005..1771284 100644
--- a/gcc/ChangeLog.graphite
+++ b/gcc/ChangeLog.graphite
@@ -31,6 +31,10 @@
 
 	* g++.dg/graphite/pr43026.C: New.
 
+2010-02-13  Tobias Grosser  <grosser@fim.uni-passau.de>
+
+	* Merge from mainline (154736:156693).
+
 2010-02-11  Sebastian Pop  <sebastian.pop@amd.com>
 
 	PR middle-end/43012
-- 
1.6.3.3


[-- Attachment #8: 0007-Add-ChangeLog-entries.patch --]
[-- Type: text/x-diff, Size: 2048 bytes --]

From 347eebde254ee70a90487c0a6017243049fa83e6 Mon Sep 17 00:00:00 2001
From: Sebastian Pop <sebpop@gmail.com>
Date: Tue, 23 Feb 2010 06:57:24 -0600
Subject: [PATCH 7/7] Add ChangeLog entries.

---
 gcc/ChangeLog           |   21 +++++++++++++++++++++
 gcc/testsuite/ChangeLog |   21 +++++++++++++++++++++
 2 files changed, 42 insertions(+), 0 deletions(-)

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 195d69c..d500d8e 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,24 @@
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43083
+	* graphite-scop-detection.c (create_single_exit_edge): Move
+	the call to find_single_exit_edge to....
+	(create_sese_edges): ...here.  Don't handle multiple edges
+	exiting the function.
+	(build_graphite_scops): Don't handle multiple edges
+	exiting the function.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43097
+	* sese.c (get_rename): Assert that old_name is an SSA_NAME.
+	(rename_variables_in_stmt): Continue when the use is not an SSA_NAME.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43026
+	* sese.c (expand_scalar_variables_expr): Handle COMPONENT_REF.
+
 2010-02-22  Manuel López-Ibáñez  <manu@gcc.gnu.org>
 
 	PR c++/43126
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 7ca5442..d8bcce5 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,24 @@
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43083
+	* gcc.dg/graphite/pr43083.c: New.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43097
+	* gfortran.dg/graphite/pr43097.f: New.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+	    Manuel López-Ibáñez  <manu@gcc.gnu.org>
+
+	PR middle-end/43140
+	* g++.dg/graphite/pr41305.C: Add -Wno-conversion-null.
+
+2010-02-22  Sebastian Pop  <sebastian.pop@amd.com>
+
+	PR middle-end/43026
+	* g++.dg/graphite/pr43026.C: New.
+
 2010-02-22  Manuel López-Ibáñez  <manu@gcc.gnu.org>
 
 	PR c++/43126
-- 
1.6.3.3