re PR fortran/90539 (481.wrf slowdown by 25% on Intel Kaby with -Ofast…

re PR fortran/90539 (481.wrf slowdown by 25% on Intel Kaby with -Ofast -march=native starting with r271377)

2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

	PR fortran/90539
	* trans-expr.c (gfc_conv_subref_array_arg): If the size of the
	expression can be determined to be one, treat it as contiguous.
	Set likelyhood of presence of an actual argument according to
	PRED_FORTRAN_ABSENT_DUMMY and likelyhood of being contiguous
	according to PRED_FORTRAN_CONTIGUOUS.

2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

	PR fortran/90539
	* predict.def (PRED_FORTRAN_CONTIGUOUS): New predictor.

2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>

	PR fortran/90539
	* gfortran.dg/internal_pack_24.f90: New test.

From-SVN: r271844

gcc/ChangeLog

+2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>
+
+	PR fortran/90539
+	* predict.def (PRED_FORTRAN_CONTIGUOUS): New predictor.
+
 2019-06-01  Martin Sebor  <msebor@redhat.com>
 
 	PR middle-end/90694

gcc/fortran/ChangeLog

+2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>
+
+	PR fortran/90539
+	* trans-expr.c (gfc_conv_subref_array_arg): If the size of the
+	expression can be determined to be one, treat it as contiguous.
+	Set likelyhood of presence of an actual argument according to
+	PRED_FORTRAN_ABSENT_DUMMY and likelyhood of being contiguous
+	according to PRED_FORTRAN_CONTIGUOUS.
+
 2019-05-30  Thomas Koenig  <tkoenig@gcc.gnu.org>
 
-    * gfc-internals.texi (Translating to GENERIC): New chapter.
+	* gfc-internals.texi (Translating to GENERIC): New chapter.
 
 2019-05-30  Marek Polacek  <polacek@redhat.com>
 

gcc/fortran/trans-expr.c

@@ -4922,15 +4922,35 @@ class_array_fcn:
 	  gfc_se cont_se, array_se;
 	  stmtblock_t if_block, else_block;
 	  tree if_stmt, else_stmt;
+	  mpz_t size;
+	  bool size_set;
 
 	  cont_var = gfc_create_var (boolean_type_node, "contiguous");
 
-	  /* cont_var = is_contiguous (expr); .  */
-	  gfc_init_se (&cont_se, parmse);
-	  gfc_conv_is_contiguous_expr (&cont_se, expr);
-	  gfc_add_block_to_block (&se->pre, &(&cont_se)->pre);
-	  gfc_add_modify (&se->pre, cont_var, cont_se.expr);
-	  gfc_add_block_to_block (&se->pre, &(&cont_se)->post);
+	  /* If the size is known to be one at compile-time, set
+	     cont_var to true unconditionally.  This may look
+	     inelegant, but we're only doing this during
+	     optimization, so the statements will be optimized away,
+	     and this saves complexity here.  */
+
+	  size_set = gfc_array_size (expr, &size);
+	  if (size_set && mpz_cmp_ui (size, 1) == 0)
+	    {
+	      gfc_add_modify (&se->pre, cont_var,
+			      build_one_cst (boolean_type_node));
+	    }
+	  else
+	    {
+	      /* cont_var = is_contiguous (expr); .  */
+	      gfc_init_se (&cont_se, parmse);
+	      gfc_conv_is_contiguous_expr (&cont_se, expr);
+	      gfc_add_block_to_block (&se->pre, &(&cont_se)->pre);
+	      gfc_add_modify (&se->pre, cont_var, cont_se.expr);
+	      gfc_add_block_to_block (&se->pre, &(&cont_se)->post);
+	    }
+
+	  if (size_set)
+	    mpz_clear (size);
 
 	  /* arrayse->expr = descriptor of a.  */
 	  gfc_init_se (&array_se, se);
@@ -4953,7 +4973,9 @@ class_array_fcn:
 
 	  /* And put the above into an if statement.  */
 	  pre_stmts = fold_build3_loc (input_location, COND_EXPR, void_type_node,
-				      cont_var, if_stmt, else_stmt);
+				       gfc_likely (cont_var,
+						   PRED_FORTRAN_CONTIGUOUS),
+				       if_stmt, else_stmt);
 	}
       else
 	{
@@ -4976,11 +4998,11 @@ class_array_fcn:
 	  gfc_add_modify (&else_block, pointer, build_int_cst (type, 0));
 	  else_stmt = gfc_finish_block (&else_block);
 
-	  tmp = fold_build3_loc (input_location, COND_EXPR, void_type_node, present_var,
+	  tmp = fold_build3_loc (input_location, COND_EXPR, void_type_node,
+				 gfc_likely (present_var,
+					     PRED_FORTRAN_ABSENT_DUMMY),
 				 pre_stmts, else_stmt);
 	  gfc_add_expr_to_block (&se->pre, tmp);
-
-
 	}
       else
 	gfc_add_expr_to_block (&se->pre, pre_stmts);
@@ -4995,9 +5017,16 @@ class_array_fcn:
 	  tmp = fold_build2_loc (input_location, EQ_EXPR, boolean_type_node,
 				 cont_var,
 				 build_zero_cst (boolean_type_node));
+	  tmp = gfc_unlikely (tmp, PRED_FORTRAN_CONTIGUOUS);
+
 	  if (pass_optional)
-	    post_cond = fold_build2_loc (input_location, TRUTH_ANDIF_EXPR,
-					 boolean_type_node, present_var, tmp);
+	    {
+	      tree present_likely = gfc_likely (present_var,
+						PRED_FORTRAN_ABSENT_DUMMY);
+	      post_cond = fold_build2_loc (input_location, TRUTH_ANDIF_EXPR,
+					   boolean_type_node, present_likely,
+					   tmp);
+	    }
 	  else
 	    post_cond = tmp;
 	}

gcc/predict.def

@@ -229,3 +229,10 @@ DEF_PREDICTOR (PRED_FORTRAN_ABSENT_DUMMY, "Fortran absent dummy", \
    to be very likely.  */
 DEF_PREDICTOR (PRED_FORTRAN_LOOP_PREHEADER, "Fortran loop preheader", \
 	       HITRATE (99), 0)
+
+/* Fortran assumed size arrays can be non-contiguous, so they need
+   to be repacked.  */
+
+DEF_PREDICTOR (PRED_FORTRAN_CONTIGUOUS, "Fortran contiguous", \
+	       HITRATE (75), 0)
+	

gcc/testsuite/ChangeLog

+2019-06-02  Thomas Koenig  <tkoenig@gcc.gnu.org>
+
+	PR fortran/90539
+	* gfortran.dg/internal_pack_24.f90: New test.
+
 2019-06-01  Iain Sandoe  <iain@sandoe.co.uk>
 
 	PR target/90698

gcc/testsuite/gfortran.dg/internal_pack_24.f90

+! { dg-do run }
+! { dg-additional-options "-O -fdump-tree-optimized" }
+module y
+  implicit none
+contains
+  subroutine foo(a,b,c,d,e,f)
+    real, dimension(1), intent(inout) :: a, b, c, d, e, f
+    if (any([a,b,c,d,e,f] /= [1,2,3,4,5,6])) stop 1
+    a = -a
+    b = -b
+    c = -c
+    d = -d
+    e = -e
+    f = -f
+  end subroutine foo
+end module y
+module x
+  use y
+  implicit none
+contains
+  subroutine bar(a)
+    real, dimension(:) :: a
+    integer :: n1, n3, n5
+    n1 = 1
+    n3 = 3
+    n5 = 5
+    call foo(a(n1:n1), a(n1+1:n1+1), a(n3:n3), a(n3+1:n3+1), a(n5:n5), a(n5+1:n5+1))
+  end subroutine bar
+end module x
+
+program main
+  use x
+  real, dimension(6) :: a,b
+  b = [1,2,3,4,5,6]
+  a = b
+  call bar(a)
+  if (any(a /= -b)) stop 2
+end program main
+! { dg-final { scan-tree-dump-not "contiguous" "optimized" } }