Skip to content

[Flang][OpenMP] Fix copyin allocatable lowering to MLIR #122097

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kaviya2510 merged 3 commits into from
Jan 23, 2025
Merged

[Flang][OpenMP] Fix copyin allocatable lowering to MLIR #122097

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e3464e2
[Flang][OpenMP] Fix copyin allocatable lowering to MLIR
kaviya2510 Jan 8, 2025
b36716a
Fix allocation status handling: Ensure the copy inherits the unalloca…
kaviya2510 Jan 20, 2025
bc6750d
[TEST] Testcase fix
kaviya2510 Jan 22, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 24 additions & 3 deletions flang/lib/Lower/Bridge.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1290,9 +1290,30 @@ class FirConverter : public Fortran::lower::AbstractConverter {
auto loadVal = builder->create<fir::LoadOp>(loc, rhs);
builder->create<fir::StoreOp>(loc, loadVal, lhs);
} else if (isAllocatable &&
(flags.test(Fortran::semantics::Symbol::Flag::OmpFirstPrivate) ||
flags.test(Fortran::semantics::Symbol::Flag::OmpCopyIn))) {
// For firstprivate and copyin allocatable variables, RHS must be copied
flags.test(Fortran::semantics::Symbol::Flag::OmpCopyIn)) {
// For copyin allocatable variables, RHS must be copied to lhs
// only when rhs is allocated.
hlfir::Entity temp =
hlfir::derefPointersAndAllocatables(loc, *builder, rhs);
mlir::Value addr = hlfir::genVariableRawAddress(loc, *builder, temp);
mlir::Value isAllocated = builder->genIsNotNullAddr(loc, addr);
builder->genIfThenElse(loc, isAllocated)
.genThen([&]() { copyData(lhs, rhs); })
.genElse([&]() {
fir::ExtendedValue hexv = symBoxToExtendedValue(dst);
hexv.match(
[&](const fir::MutableBoxValue &new_box) -> void {
// if the allocation status of original list item is
// unallocated, unallocate the copy if it is allocated, else
// do nothing.
Fortran::lower::genDeallocateIfAllocated(*this, new_box, loc);
},
[&](const auto &) -> void {});
})
.end();
} else if (isAllocatable &&
flags.test(Fortran::semantics::Symbol::Flag::OmpFirstPrivate)) {
// For firstprivate allocatable variables, RHS must be copied
// only when LHS is allocated.
hlfir::Entity temp =
hlfir::derefPointersAndAllocatables(loc, *builder, lhs);
Expand Down
112 changes: 103 additions & 9 deletions flang/test/Lower/OpenMP/copyin.f90
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -391,12 +391,34 @@ subroutine pointer()
! 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: %[[VAL_7:.*]] = fir.box_addr %[[VAL_6]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
! CHECK: %[[VAL_8:.*]] = fir.convert %[[VAL_7]] : (!fir.heap<!fir.array<?xi32>>) -> i64
! CHECK: %[[C0_I64:.*]] = arith.constant 0 : i64
! CHECK: %[[VAL_9:.*]] = arith.cmpi ne, %[[VAL_8]], %[[C0_I64]] : i64
! CHECK: fir.if %[[VAL_9]] {
! CHECK: %[[VAL_10:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
! CHECK: hlfir.assign %[[VAL_10]] to %[[VAL_5]]#0 realloc : !fir.box<!fir.heap<!fir.array<?xi32>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
! CHECK: } else {
! CHECK: %[[VAL_10:.*]] = fir.load %[[VAL_5]]#1 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
! CHECK: %[[VAL_11:.*]] = fir.box_addr %[[VAL_10]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
! CHECK: %[[VAL_12:.*]] = fir.convert %[[VAL_11]] : (!fir.heap<!fir.array<?xi32>>) -> i64
! CHECK: %[[C0_I64_0:.*]] = arith.constant 0 : i64
! CHECK: %[[VAL_13:.*]] = arith.cmpi ne, %[[VAL_12]], %[[C0_I64_0]] : i64
! CHECK: fir.if %[[VAL_13]] {
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_5]]#1 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
! CHECK: %[[VAL_15:.*]] = fir.box_addr %[[VAL_14]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
! CHECK: fir.freemem %[[VAL_15]] : !fir.heap<!fir.array<?xi32>>
! CHECK: %[[VAL_16:.*]] = fir.zero_bits !fir.heap<!fir.array<?xi32>>
! CHECK: %[[C0:.*]] = arith.constant 0 : index
! CHECK: %[[VAL_17:.*]] = fir.shape %[[C0]] : (index) -> !fir.shape<1>
! CHECK: %[[VAL_18:.*]] = fir.embox %[[VAL_16]](%[[VAL_17]]) : (!fir.heap<!fir.array<?xi32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<?xi32>>>
! CHECK: fir.store %[[VAL_18]] to %[[VAL_5]]#1 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
! CHECK: }
! 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: %[[VAL_19:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
! CHECK: %[[VAL_20:.*]] = fir.box_addr %[[VAL_19]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
! CHECK: %[[VAL_21:.*]] = fir.convert %[[VAL_20]] : (!fir.heap<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
! CHECK: fir.call @_QPsub8(%[[VAL_21]]) fastmath<contract> : (!fir.ref<!fir.array<?xi32>>) -> ()
! CHECK: omp.terminator
! CHECK: }
! CHECK: return
Expand All @@ -418,7 +440,7 @@ subroutine allocatable()
! CHECK: omp.parallel {
! CHECK: %[[VAL_4:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>> -> !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_5:.*]]:2 = hlfir.declare %[[VAL_4]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFallocatable2Ea"} : (!fir.ref<!fir.box<!fir.heap<i32>>>) -> (!fir.ref<!fir.box<!fir.heap<i32>>>, !fir.ref<!fir.box<!fir.heap<i32>>>)
! CHECK: %[[VAL_6:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_6:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_7:.*]] = fir.box_addr %[[VAL_6]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: %[[VAL_8:.*]] = fir.convert %[[VAL_7]] : (!fir.heap<i32>) -> i64
! CHECK: %[[VAL_9:.*]] = arith.constant 0 : i64
Expand All @@ -428,10 +450,23 @@ subroutine allocatable()
! CHECK: %[[VAL_12:.*]] = fir.box_addr %[[VAL_11]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: %[[VAL_13:.*]] = fir.load %[[VAL_12]] : !fir.heap<i32>
! CHECK: hlfir.assign %[[VAL_13]] to %[[VAL_5]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: }
! CHECK: } else {
! CHECK: %[[VAL_11:.*]] = fir.load %[[VAL_5]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_15:.*]] = fir.box_addr %[[VAL_11]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: %[[VAL_16:.*]] = fir.convert %[[VAL_15]] : (!fir.heap<i32>) -> i64
! CHECK: %[[C0_I64_0:.*]] = arith.constant 0 : i64
! CHECK: %[[VAL_17:.*]] = arith.cmpi ne, %[[VAL_16]], %[[C0_I64_0]] : i64
! CHECK: fir.if %[[VAL_17]] {
! CHECK: %[[VAL_18:.*]] = fir.load %[[VAL_5]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_19:.*]] = fir.box_addr %[[VAL_18]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: fir.freemem %[[VAL_19]] : !fir.heap<i32>
! CHECK: %[[VAL_20:.*]] = fir.zero_bits !fir.heap<i32>
! CHECK: %[[VAL_21:.*]] = fir.embox %[[VAL_20]] : (!fir.heap<i32>) -> !fir.box<!fir.heap<i32>>
! CHECK: fir.store %[[VAL_21]] to %[[VAL_5]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: }
! CHECK: omp.barrier
! CHECK: %[[VAL_14:.*]] = arith.constant 1 : i32
! CHECK: hlfir.assign %[[VAL_14]] to %[[VAL_5]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_22:.*]] = arith.constant 1 : i32
! CHECK: hlfir.assign %[[VAL_22]] to %[[VAL_5]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: omp.terminator
! CHECK: }
! CHECK: return
Expand All @@ -444,3 +479,62 @@ subroutine allocatable2()
a = 1
!$omp end parallel
end subroutine

! CHECK: func.func @_QPallocatable3() {
! CHECK: %[[VAL_0:.*]] = fir.address_of(@_QFallocatable3Ea) : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_1:.*]]:2 = hlfir.declare %[[VAL_0]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFallocatable3Ea"} : (!fir.ref<!fir.box<!fir.heap<i32>>>) -> (!fir.ref<!fir.box<!fir.heap<i32>>>, !fir.ref<!fir.box<!fir.heap<i32>>>)
! CHECK: %[[VAL_2:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>> -> !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_3:.*]]:2 = hlfir.declare %[[VAL_2]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFallocatable3Ea"} : (!fir.ref<!fir.box<!fir.heap<i32>>>) -> (!fir.ref<!fir.box<!fir.heap<i32>>>, !fir.ref<!fir.box<!fir.heap<i32>>>)
! CHECK: %[[VAL_4:.*]] = fir.allocmem i32 {fir.must_be_heap = true, uniq_name = "_QFallocatable3Ea.alloc"}
! CHECK: %[[VAL_5:.*]] = fir.embox %[[VAL_4]] : (!fir.heap<i32>) -> !fir.box<!fir.heap<i32>>
! CHECK: fir.store %[[VAL_5]] to %[[VAL_3]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[C10_I32:.*]] = arith.constant 10 : i32
! CHECK: hlfir.assign %[[C10_I32]] to %[[VAL_3]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: omp.parallel {
! CHECK: %[[VAL_6:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>> -> !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_7:.*]]:2 = hlfir.declare %[[VAL_6]] {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFallocatable3Ea"} : (!fir.ref<!fir.box<!fir.heap<i32>>>) -> (!fir.ref<!fir.box<!fir.heap<i32>>>, !fir.ref<!fir.box<!fir.heap<i32>>>)
! CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_9:.*]] = fir.box_addr %[[VAL_8]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: %[[VAL_10:.*]] = fir.convert %[[VAL_9]] : (!fir.heap<i32>) -> i64
! CHECK: %[[C10_I64:.*]] = arith.constant 0 : i64
! CHECK: %[[VAL_11:.*]] = arith.cmpi ne, %[[VAL_10]], %[[C10_I64]] : i64
! CHECK: fir.if %[[VAL_11]] {
! CHECK: %[[VAL_12:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_13:.*]] = fir.box_addr %[[VAL_12]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_13]] : !fir.heap<i32>
! CHECK: hlfir.assign %[[VAL_14]] to %[[VAL_7]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: } else {
! CHECK: %[[VAL_12:.*]] = fir.load %[[VAL_7]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_15:.*]] = fir.box_addr %[[VAL_12]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: %[[VAL_16:.*]] = fir.convert %[[VAL_15]] : (!fir.heap<i32>) -> i64
! CHECK: %[[C0_I64_0:.*]] = arith.constant 0 : i64
! CHECK: %[[VAL_17:.*]] = arith.cmpi ne, %[[VAL_16]], %[[C0_I64_0]] : i64
! CHECK: fir.if %[[VAL_17]] {
! CHECK: %[[VAL_18:.*]] = fir.load %[[VAL_7]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_19:.*]] = fir.box_addr %[[VAL_18]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: fir.freemem %[[VAL_19]] : !fir.heap<i32>
! CHECK: %[[VAL_20:.*]] = fir.zero_bits !fir.heap<i32>
! CHECK: %[[VAL_21:.*]] = fir.embox %[[VAL_20]] : (!fir.heap<i32>) -> !fir.box<!fir.heap<i32>>
! CHECK: fir.store %[[VAL_21]] to %[[VAL_7]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: }
! CHECK: }
! CHECK: omp.barrier
! CHECK: %[[VAL_22:.*]] = fir.load %7#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: %[[VAL_23:.*]] = fir.box_addr %[[VAL_22]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
! CHECK: %[[VAL_24:.*]] = fir.load %[[VAL_23]] : !fir.heap<i32>
! CHECK: %[[C1_I32:.*]] = arith.constant 1 : i32
! CHECK: %[[VAL_25:.*]]= arith.addi %[[VAL_24]], %[[C1_I32]] : i32
! CHECK: hlfir.assign %[[VAL_25]]to %[[VAL_7]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
! CHECK: omp.terminator
! CHECK: }
! CHECK: return
! CHECK: }
subroutine allocatable3()
integer, allocatable, save :: a
!$omp threadprivate(a)
allocate(a)
a = 10
!$omp parallel copyin(a)
a = a + 1
!$omp end parallel
end subroutine
Loading