[flang-commits] [flang] [flang][OpenMP] Implement copyin for pointers and allocatables. (PR #107425)

[David Truby via flang-commits]

https://github.com/DavidTruby updated https://github.com/llvm/llvm-project/pull/107425

>From ee837f61369fdfd1322df53fd47b7cd4da3946f0 Mon Sep 17 00:00:00 2001
From: David Truby <david.truby at arm.com>
Date: Thu, 5 Sep 2024 14:22:58 +0100
Subject: [PATCH] [flang][OpenMP] Implement copyin for pointers and
 allocatables.

The copyin clause currently forbids pointer and allocatable variables,
which are allowed by the OpenMP 1.1 and 3.0 specifications respectively.

Since #106559 it is sufficient to remove the TODO check to get correct
behaviour.
---
 flang/lib/Lower/OpenMP/ClauseProcessor.cpp |  4 +-
 flang/test/Lower/OpenMP/copyin.f90         | 59 ++++++++++++++++++++++
 2 files changed, 60 insertions(+), 3 deletions(-)

diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 6dee31ddb6963d..3f54234b176e3f 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -619,9 +619,7 @@ bool ClauseProcessor::processCopyin() const {
               checkAndCopyHostAssociateVar(&*mem, &insPt);
             break;
           }
-          if (semantics::IsAllocatableOrObjectPointer(&sym->GetUltimate()))
-            TODO(converter.getCurrentLocation(),
-                 "pointer or allocatable variables in Copyin clause");
+
           assert(sym->has<semantics::HostAssocDetails>() &&
                  "No host-association found");
           checkAndCopyHostAssociateVar(sym);
diff --git a/flang/test/Lower/OpenMP/copyin.f90 b/flang/test/Lower/OpenMP/copyin.f90
index 4023987a841b82..c4c69657f45928 100644
--- a/flang/test/Lower/OpenMP/copyin.f90
+++ b/flang/test/Lower/OpenMP/copyin.f90
@@ -356,3 +356,62 @@ subroutine common_2()
      end do
   !$omp end parallel do
 end subroutine
+
+! CHECK-LABEL:   func.func @_QPpointer() {
+! CHECK:           %[[VAL_0:.*]] = fir.address_of(@_QFpointerEp) : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
+! CHECK:           %[[VAL_1:.*]]:2 = hlfir.declare %[[VAL_0]] {fortran_attrs = #fir.var_attrs<pointer>, uniq_name = "_QFpointerEp"} : (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>)
+! CHECK:           %[[VAL_2:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>> -> !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
+! CHECK:           %[[VAL_3:.*]]:2 = hlfir.declare %[[VAL_2]] {fortran_attrs = #fir.var_attrs<pointer>, uniq_name = "_QFpointerEp"} : (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>)
+! CHECK:           omp.parallel {
+! CHECK:             %[[VAL_4:.*]] = fir.alloca !fir.box<!fir.heap<!fir.array<?xi32>>> {pinned}
+! CHECK:             %[[VAL_5:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>> -> !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
+! CHECK:             %[[VAL_6:.*]]:2 = hlfir.declare %[[VAL_5]] {fortran_attrs = #fir.var_attrs<pointer>, uniq_name = "_QFpointerEp"} : (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>)
+! CHECK:             %[[VAL_7:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
+! CHECK:             fir.store %[[VAL_7]] to %[[VAL_6]]#0 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
+! CHECK:             omp.barrier
+! CHECK:             %[[VAL_8:.*]] = fir.load %[[VAL_6]]#0 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
+! CHECK:             %[[VAL_9:.*]]:2 = hlfir.copy_in %[[VAL_8]] to %[[VAL_4]] : (!fir.box<!fir.ptr<!fir.array<?xi32>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.box<!fir.ptr<!fir.array<?xi32>>>, i1)
+! CHECK:             %[[VAL_10:.*]] = fir.box_addr %[[VAL_9]]#0 : (!fir.box<!fir.ptr<!fir.array<?xi32>>>) -> !fir.ptr<!fir.array<?xi32>>
+! CHECK:             %[[VAL_11:.*]] = fir.convert %[[VAL_10]] : (!fir.ptr<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
+! CHECK:             fir.call @_QPsub7(%[[VAL_11]]) fastmath<contract> : (!fir.ref<!fir.array<?xi32>>) -> ()
+! CHECK:             hlfir.copy_out %[[VAL_4]], %[[VAL_9]]#1 to %[[VAL_8]] : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, i1, !fir.box<!fir.ptr<!fir.array<?xi32>>>) -> ()
+! CHECK:             omp.terminator
+! CHECK:           }
+! CHECK:           return
+! CHECK:         }
+subroutine pointer()
+  integer, pointer, save :: p(:)
+  !$omp threadprivate(p)
+
+  !$omp parallel copyin(p)
+  call sub7(p)
+  !$omp end parallel
+end subroutine
+
+! CHECK-LABEL:   func.func @_QPallocatable() {
+! CHECK:           %[[VAL_0:.*]] = fir.address_of(@_QFallocatableEp) : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK:           %[[VAL_1:.*]]:2 = hlfir.declare %[[VAL_0]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFallocatableEp"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
+! CHECK:           %[[VAL_2:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>> -> !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK:           %[[VAL_3:.*]]:2 = hlfir.declare %[[VAL_2]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFallocatableEp"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
+! CHECK:           omp.parallel {
+! CHECK:             %[[VAL_4:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>> -> !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK:             %[[VAL_5:.*]]:2 = hlfir.declare %[[VAL_4]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFallocatableEp"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
+! CHECK:             %[[VAL_6:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK:             hlfir.assign %[[VAL_6]] to %[[VAL_5]]#0 realloc : !fir.box<!fir.heap<!fir.array<?xi32>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK:             omp.barrier
+! CHECK:             %[[VAL_7:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
+! CHECK:             %[[VAL_8:.*]] = fir.box_addr %[[VAL_7]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
+! CHECK:             %[[VAL_9:.*]] = fir.convert %[[VAL_8]] : (!fir.heap<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
+! CHECK:             fir.call @_QPsub8(%[[VAL_9]]) fastmath<contract> : (!fir.ref<!fir.array<?xi32>>) -> ()
+! CHECK:             omp.terminator
+! CHECK:           }
+! CHECK:           return
+! CHECK:         }
+subroutine allocatable()
+  integer, allocatable, save :: p(:)
+  !$omp threadprivate(p)
+
+  !$omp parallel copyin(p)
+  call sub8(p)
+  !$omp end parallel
+end subroutine