[committed] Fix Fortran linear clause handling (PR fortran/65597)

[Jakub Jelinek]

Hi!

This patch makes sure we have OMP_CLAUSE_LINEAR_NO_COPYIN set on
iterator vars, both with implicit and explicit linear clause, so that
we don't ICE on those - the iterator vars don't need original list item
value.  The patch also fixes up the step value for the artificial count
iterator var, which always has step of one.

The patch doesn't cure linear clauses on combined do simd (or for simd)
constructs if the var in the linear clause is not iterator var and
is private in outer context - have asked on omp-lang whether it should be
rejected or what behavior it should have.

Bootstrapped/regtested on x86_64-linux and i686-linux, committed to trunk so
far (plan to backport to 4.9 eventually).

2015-03-30  Jakub Jelinek  <jakub@redhat.com>

	PR fortran/65597
	* trans-openmp.c (gfc_trans_omp_do): For !simple simd with explicit
	linear clause for the iterator set OMP_CLAUSE_LINEAR_NO_COPYIN.
	For implcitly added !simple OMP_CLAUSE_LINEAR set it too.  Use step 1
	instead of the original step on the new iterator - count.

	* testsuite/libgomp.fortran/pr65597.f90: New test.

--- gcc/fortran/trans-openmp.c.jj	2015-01-28 08:39:53.000000000 +0100
+++ gcc/fortran/trans-openmp.c	2015-03-30 16:50:27.710714334 +0200
@@ -3255,6 +3255,19 @@ gfc_trans_omp_do (gfc_code *code, gfc_ex
 	  inits.safe_push (e);
 	}
 
+      if (dovar_found == 2
+	  && op == EXEC_OMP_SIMD
+	  && collapse == 1
+	  && !simple)
+	{
+	  for (tmp = omp_clauses; tmp; tmp = OMP_CLAUSE_CHAIN (tmp))
+	    if (OMP_CLAUSE_CODE (tmp) == OMP_CLAUSE_LINEAR
+		&& OMP_CLAUSE_DECL (tmp) == dovar)
+	      {
+		OMP_CLAUSE_LINEAR_NO_COPYIN (tmp) = 1;
+		break;
+	      }
+	}
       if (!dovar_found)
 	{
 	  if (op == EXEC_OMP_SIMD)
@@ -3263,6 +3276,7 @@ gfc_trans_omp_do (gfc_code *code, gfc_ex
 		{
 		  tmp = build_omp_clause (input_location, OMP_CLAUSE_LINEAR);
 		  OMP_CLAUSE_LINEAR_STEP (tmp) = step;
+		  OMP_CLAUSE_LINEAR_NO_COPYIN (tmp) = 1;
 		}
 	      else
 		tmp = build_omp_clause (input_location, OMP_CLAUSE_LASTPRIVATE);
@@ -3330,7 +3344,7 @@ gfc_trans_omp_do (gfc_code *code, gfc_ex
 	  else if (collapse == 1)
 	    {
 	      tmp = build_omp_clause (input_location, OMP_CLAUSE_LINEAR);
-	      OMP_CLAUSE_LINEAR_STEP (tmp) = step;
+	      OMP_CLAUSE_LINEAR_STEP (tmp) = build_int_cst (type, 1);
 	      OMP_CLAUSE_LINEAR_NO_COPYIN (tmp) = 1;
 	      OMP_CLAUSE_LINEAR_NO_COPYOUT (tmp) = 1;
 	    }
--- libgomp/testsuite/libgomp.fortran/pr65597.f90.jj	2015-03-30 16:57:50.685614350 +0200
+++ libgomp/testsuite/libgomp.fortran/pr65597.f90	2015-03-30 16:57:10.000000000 +0200
@@ -0,0 +1,21 @@
+! PR fortran/65597
+! { dg-do run }
+
+  integer :: i, a(151)
+  a(:) = 0
+  !$omp do simd
+    do i = 1, 151, 31
+      a(i) = a(i) + 1
+    end do
+  !$omp do simd linear (i: 31)
+    do i = 1, 151, 31
+      a(i) = a(i) + 1
+    end do
+  do i = 1, 151
+    if (mod (i, 31) .eq. 1) then
+      if (a(i) .ne. 2) call abort
+    else
+      if (a(i) .ne. 0) call abort
+    end if
+  end do
+end

	Jakub