diff mbox

[committed] FIx up lastprivate handling in doacross loops

Message ID 20160531134930.GA28550@tucnak.redhat.com
State New
Headers show

Commit Message

Jakub Jelinek May 31, 2016, 1:49 p.m. UTC 
Hi!

Here is the promissed fix for lastprivate.  We just need to do in the FE
what gimplifier does when using a private counter because the normal counter
is addressable - because we do the similar thing in the FE.

Tested on x86_64-linux, committed to gomp-4_5-branch.

2016-05-31  Jakub Jelinek  <jakub@redhat.com>

	* gimplify.c (gimplify_scan_omp_clauses) <case OMP_CLAUSE_DEPEND>: Add
	missing break at the end of OMP_CLAUSE_DEPEND_SINK case.

	* trans-openmp.c (gfc_trans_omp_do): For doacross loop, put into
	OMP_CLAUSE_LASTPRIVATE_STMT dovar = count * step + from or
	dovar = (count + 1) * step + from instead of dovar = dovar + step.

	* testsuite/libgomp.fortran/doacross3.f90: New test.


	Jakub
diff mbox

Patch

--- gcc/gimplify.c.jj	2016-05-27 18:32:16.000000000 +0200
+++ gcc/gimplify.c	2016-05-31 15:06:06.504711812 +0200
@@ -7188,6 +7188,7 @@  gimplify_scan_omp_clauses (tree *list_p,
 				   pre_p, NULL, is_gimple_val, fb_rvalue);
 		  deps = TREE_CHAIN (deps);
 		}
+	      break;
 	    }
 	  else if (OMP_CLAUSE_DEPEND_KIND (c) == OMP_CLAUSE_DEPEND_SOURCE)
 	    break;
--- gcc/fortran/trans-openmp.c.jj	2016-05-30 18:36:42.000000000 +0200
+++ gcc/fortran/trans-openmp.c	2016-05-31 14:39:50.693158672 +0200
@@ -3637,9 +3637,24 @@  gfc_trans_omp_do (gfc_code *code, gfc_ex
 		 OMP_CLAUSE_LASTPRIVATE_STMT, otherwise the copied dovar
 		 will have the value on entry of the last loop, rather
 		 than value after iterator increment.  */
-	      tmp = gfc_evaluate_now (step, pblock);
-	      tmp = fold_build2_loc (input_location, PLUS_EXPR, type, dovar,
-				     tmp);
+	      if (clauses->orderedc)
+		{
+		  if (clauses->collapse <= 1 || i >= clauses->collapse)
+		    tmp = count;
+		  else
+		    tmp = fold_build2_loc (input_location, PLUS_EXPR,
+					   type, count, build_one_cst (type));
+		  tmp = fold_build2_loc (input_location, MULT_EXPR, type,
+					 tmp, step);
+		  tmp = fold_build2_loc (input_location, PLUS_EXPR, type,
+					 from, tmp);
+		}
+	      else
+		{
+		  tmp = gfc_evaluate_now (step, pblock);
+		  tmp = fold_build2_loc (input_location, PLUS_EXPR, type,
+					 dovar, tmp);
+		}
 	      tmp = fold_build2_loc (input_location, MODIFY_EXPR, type,
 				     dovar, tmp);
 	      for (c = omp_clauses; c ; c = OMP_CLAUSE_CHAIN (c))
--- libgomp/testsuite/libgomp.fortran/doacross3.f90.jj	2016-05-31 15:03:08.520012394 +0200
+++ libgomp/testsuite/libgomp.fortran/doacross3.f90	2016-05-31 15:01:34.000000000 +0200
@@ -0,0 +1,261 @@ 
+! { dg-do run }
+
+  integer, parameter :: N = 256
+  integer, save :: a(N), b(N / 16, 8, 4), c(N / 32, 8, 8), g(N/16,8,6)
+  integer, save, volatile :: d, e
+  integer(kind=8), save, volatile :: f
+  integer(kind=8) :: i
+  integer :: j, k, l, m
+  integer :: m1, m2, m3, m4, m5, m6, m7, m8
+  integer :: m9, m10, m11, m12, m13, m14, m15, m16
+  d = 0
+  e = 0
+  f = 0
+  !$omp parallel private (l) shared(k)
+    !$omp do schedule(guided, 3) ordered(1)
+    do i = 2, N + f, f + 1
+      !$omp atomic write
+      a(i) = 1
+      !$omp ordered depend ( sink : i - 1 )
+      if (i.gt.2) then
+        !$omp atomic read
+        l = a(i - 1)
+        if (l.lt.2) call abort
+      end if
+      !$omp atomic write
+      a(i) = 2
+      if (i.lt.N) then
+        !$omp atomic read
+        l = a(i + 1)
+        if (l.eq.3) call abort
+      end if
+      !$omp ordered depend(source)
+      !$omp atomic write
+      a(i) = 3
+    end do
+    !$omp end do nowait
+    !$omp do schedule(guided) ordered ( 3 )
+    do i = 4, N / 16 - 1 + f, 1 + f
+      do j = 1, 8, d + 2
+        do k = 2, 4, 1 + d
+          !$omp atomic write
+          b(i, j, k) = 1
+          !$omp ordered depend(sink:i,j-2,k-1) &
+          !$omp& depend(sink: i - 2, j - 2, k + 1)
+          !$omp ordered depend(sink:i-3,j+2,k-2)
+          if (j.gt.2.and.k.gt.2) then
+            !$omp atomic read
+            l = b(i,j-2,k-1)
+            if (l.lt.2) call abort
+          end if
+          !$omp atomic write
+          b(i,j,k) = 2
+          if (i.gt.5.and.j.gt.2.and.k.lt.4) then
+            !$omp atomic read
+            l = b(i-2,j-2, k+1)
+            if (l.lt.2) call abort
+          end if
+          if (i.gt.6.and.j.le.N/16-3.and.k.eq.4) then
+            !$omp atomic read
+            l = b( i - 3, j+2, k-2)
+            if (l.lt.2) call abort
+          end if
+          !$omp ordered depend(source)
+          !$omp atomic write
+          b(i, j, k) = 3
+        end do
+      end do
+    end do
+    !$omp end do nowait
+    !$omp do schedule(guided, 15) collapse(2) ordered(13)
+    do i = 3, N / 32 + f, d + 1
+      do j = 8, 3, d - 1
+        do k = 7, 1, d - 2
+          do m1 = 4, 4, d + 1
+          do m2 = 4, 4, 1 + d
+          do m3 = 4, 4, d + 1
+          do m4 = 4, 4, 1 + d
+          do m5 = 4, 4, d + 1
+          do m6 = 4, 4, 1 + d
+          do m7 = 4, 4, d + 1
+          do m8 = 4, 4, 1 + d
+          do m9 = 4, 4
+          do m10 = 4, 4, d + 1
+          do m11 = 4, 4, 1 + d
+          do m12 = 4, 4, d + 1
+          do m13 = 4, 4
+          do m14 = 4, 4, 1 + d
+          do m15 = 4, 4, d + 1
+          do m16 = 4, 4, 1 + d
+            !$omp atomic write
+            c(i, j, k) = 1
+            !$omp ordered depend(sink: i, j, k + 2, m1, m2, m3, m4, &
+            !$omp & m5, m6, m7, m8, m9, m10) &
+            !$omp depend(sink: i - 2, j + 1, k - 4, m1,m2,m3,m4,m5, &
+            !$omp & m6,m7,m8,m9,m10) depend ( sink : i-1,j-2,k-2, &
+            !$omp& m1,m2,m3,m4 , m5, m6,m7,m8,m9,m10 )
+            if (k.le.5) then
+              !$omp atomic read
+              l = c(i, j, k + 2)
+              if (l.lt.2) call abort
+            end if
+            !$omp atomic write
+            c(i, j, k) = 2
+            if (i.ge.5.and.j.lt.8.and.k.ge.5) then
+              !$omp atomic read
+              l = c(i - 2, j + 1, k - 4)
+              if (l.lt.2) call abort
+            end if
+            if (i.ge.4.and.j.ge.5.and.k.ge.3) then
+              !$omp atomic read
+              l = c(i - 1, j - 2, k - 2)
+              if (l.lt.2) call abort
+            end if
+            !$omp ordered depend ( source )
+            !$omp atomic write
+            c(i,j,k)=3
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+          end do
+        end do
+      end do
+    end do
+    !$omp do schedule(guided, 5) ordered(3)
+    do j = 1, N / 16 - 1, d + 1
+      do k = 1, 7, 2 + d
+        do i = 4, 6 + f, f + 1
+          !$omp atomic write
+          g(j, k, i) = 1
+          !$omp ordered depend(sink: j, k-2,i-1) &
+          !$omp& depend(sink: j - 2, k - 2, i + 1)
+          !$omp ordered depend(sink:j-3,k+2,i-2)
+          if (k.gt.2.and.i.gt.4) then
+            !$omp atomic read
+            l = g(j,k-2,i-1)
+            if (l.lt.2) call abort
+          end if
+          !$omp atomic write
+          g(j,k,i) = 2
+          if (j.gt.2.and.k.gt.2.and.i.lt.6) then
+            !$omp atomic read
+            l = g(j-2,k-2, i+1)
+            if (l.lt.2) call abort
+          end if
+          if (j.gt.3.and.k.le.N/16-3.and.i.eq.6) then
+            !$omp atomic read
+            l = g( j - 3, k+2, i-2)
+            if (l.lt.2) call abort
+          end if
+          !$omp ordered depend(source)
+          !$omp atomic write
+          g(j, k, i) = 3
+        end do
+      end do
+    end do
+    !$omp end do nowait
+    !$omp do collapse(2) ordered(4) lastprivate (i,j,k)
+    do i = 2, f + 2, 1 + f
+      do j = d + 1, 0, d - 1
+        do k = 0, d - 1, d + 1
+          do l = 0, d + 1, 1 + d
+            !$omp ordered depend(source)
+            !$omp ordered depend(sink: i-2,j+2,k-2,l)
+            if (e.eq.0) call abort
+          end do
+        end do
+      end do
+    end do
+    !$omp single
+    if (i.ne.3.or.j.ne.-1.or.k.ne.0) call abort
+    i = 8; j = 9; k = 10
+    !$omp end single
+    !$omp do ordered(4) collapse(2) lastprivate (i, j, k, m)
+    do i = 2, f + 2, 1 + f
+      do j = d + 1, 0, d - 1
+        do k = 0, d + 1, 1 + d
+          do m = 0, d-1, d+1
+            !$omp ordered depend(source)
+            !$omp ordered depend(sink: i - 2, j + 2, k - 2, m)
+            call abort
+          end do
+        end do
+      end do
+    end do
+    !$omp single
+    if (i.ne.3.or.j.ne.-1.or.k.ne.2.or.m.ne.0) call abort
+    !$omp end single
+    !$omp do collapse(2) ordered(4) lastprivate (i,j,k)
+    do i = 2, f + 2, 1 + f
+      do j = d, 1, d -1
+        do k = 0, d + 1, 1 + d
+          do l = 0, d + 3, d + 1
+            !$omp ordered depend(source)
+            !$omp ordered depend(sink: i-2,j+2,k-2,l)
+            if (e.eq.0) call abort
+          end do
+        end do
+      end do
+    end do
+    !$omp end do nowait
+    !$omp single
+    if (a(1) .ne. 0) call abort
+    !$omp end single nowait
+    !$omp do
+    do i = 2, N
+      if (a(i) .ne. 3) call abort
+    end do
+    !$omp end do nowait
+    !$omp do collapse(2) private(k)
+    do i = 1, N / 16
+      do j = 1, 8
+        do k = 1, 4
+          if (i.ge.4.and.i.lt.N/16.and.iand(j,1).ne.0.and.k.ge.2) then
+            if (b(i,j,k).ne.3) call abort
+          else
+            if (b(i,j,k).ne.0) call abort
+          end if
+        end do
+      end do
+    end do
+    !$omp end do nowait
+    !$omp do collapse(3)
+    do i = 1, N / 32
+      do j = 1, 8
+        do k = 1, 4
+          if (i.ge.3.and.j.ge.3.and.iand(k,1).ne.0) then
+            if (c(i,j,k).ne.3) call abort
+          else
+            if (c(i,j,k).ne.0) call abort
+          end if
+        end do
+      end do
+    end do
+    !$omp end do nowait
+    !$omp do collapse(2) private(k)
+    do i = 1, N / 16
+      do j = 1, 8
+        do k = 1, 6
+          if (i.lt.N/16.and.iand(j,1).ne.0.and.k.ge.4) then
+            if (g(i,j,k).ne.3) call abort
+          else
+            if (g(i,j,k).ne.0) call abort
+          end if
+        end do
+      end do
+    end do
+    !$omp end do nowait
+  !$omp end parallel
+end