Skip to content

Commit e3464e2

Browse files
kaviya2510kaviya2510
committed
[Flang][OpenMP] Fix copyin allocatable lowering to MLIR
1 parent b7b9ccf commit e3464e2

File tree

2 files changed

+65
-8
lines changed

2 files changed

+65
-8
lines changed

‎flang/lib/Lower/Bridge.cpp‎

Lines changed: 7 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1292,10 +1292,15 @@ class FirConverter : public Fortran::lower::AbstractConverter {
12921292
} else if (isAllocatable &&
12931293
(flags.test(Fortran::semantics::Symbol::Flag::OmpFirstPrivate) ||
12941294
flags.test(Fortran::semantics::Symbol::Flag::OmpCopyIn))) {
1295-
// For firstprivate and copyin allocatable variables, RHS must be copied
1295+
// For firstprivate allocatable variables, RHS must be copied
12961296
// only when LHS is allocated.
1297+
1298+
// For copyin allocatable variables, RHS must be copied to lhs
1299+
// only when rhs is allocated.
1300+
hlfir::Entity entity =
1301+
flags.test(Fortran::semantics::Symbol::Flag::OmpCopyIn) ? rhs : lhs;
12971302
hlfir::Entity temp =
1298-
hlfir::derefPointersAndAllocatables(loc, *builder, lhs);
1303+
hlfir::derefPointersAndAllocatables(loc, *builder, entity);
12991304
mlir::Value addr = hlfir::genVariableRawAddress(loc, *builder, temp);
13001305
mlir::Value isAllocated = builder->genIsNotNullAddr(loc, addr);
13011306
builder->genIfThen(loc, isAllocated)

‎flang/test/Lower/OpenMP/copyin.f90‎

Lines changed: 58 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -391,12 +391,19 @@ subroutine pointer()
391391
! 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>>>>
392392
! 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>>>>)
393393
! CHECK: %[[VAL_6:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
394-
! 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>>>>
394+
! CHECK: %[[VAL_7:.*]] = fir.box_addr %[[VAL_6]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
395+
! CHECK: %[[VAL_8:.*]] = fir.convert %[[VAL_7]] : (!fir.heap<!fir.array<?xi32>>) -> i64
396+
! CHECK: %[[C0_I64:.*]] = arith.constant 0 : i64
397+
! CHECK: %[[VAL_9:.*]] = arith.cmpi ne, %[[VAL_8]], %c0_i64 : i64
398+
! CHECK: fir.if %9 {
399+
! CHECK: %[[VAL_10:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
400+
! 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>>>>
401+
! CHECK: }
395402
! CHECK: omp.barrier
396-
! CHECK: %[[VAL_7:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
397-
! CHECK: %[[VAL_8:.*]] = fir.box_addr %[[VAL_7]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
398-
! CHECK: %[[VAL_9:.*]] = fir.convert %[[VAL_8]] : (!fir.heap<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
399-
! CHECK: fir.call @_QPsub8(%[[VAL_9]]) fastmath<contract> : (!fir.ref<!fir.array<?xi32>>) -> ()
403+
! CHECK: %[[VAL_11:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>
404+
! CHECK: %[[VAL_12:.*]] = fir.box_addr %[[VAL_11]] : (!fir.box<!fir.heap<!fir.array<?xi32>>>) -> !fir.heap<!fir.array<?xi32>>
405+
! CHECK: %[[VAL_13:.*]] = fir.convert %[[VAL_12]] : (!fir.heap<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
406+
! CHECK: fir.call @_QPsub8(%[[VAL_13]]) fastmath<contract> : (!fir.ref<!fir.array<?xi32>>) -> ()
400407
! CHECK: omp.terminator
401408
! CHECK: }
402409
! CHECK: return
@@ -418,7 +425,7 @@ subroutine allocatable()
418425
! CHECK: omp.parallel {
419426
! CHECK: %[[VAL_4:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>> -> !fir.ref<!fir.box<!fir.heap<i32>>>
420427
! 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>>>)
421-
! CHECK: %[[VAL_6:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
428+
! CHECK: %[[VAL_6:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
422429
! CHECK: %[[VAL_7:.*]] = fir.box_addr %[[VAL_6]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
423430
! CHECK: %[[VAL_8:.*]] = fir.convert %[[VAL_7]] : (!fir.heap<i32>) -> i64
424431
! CHECK: %[[VAL_9:.*]] = arith.constant 0 : i64
@@ -444,3 +451,48 @@ subroutine allocatable2()
444451
a = 1
445452
!$omp end parallel
446453
end subroutine
454+
455+
! CHECK: func.func @_QPallocatable3() {
456+
! CHECK: %[[VAL_0:.*]] = fir.address_of(@_QFallocatable3Ea) : !fir.ref<!fir.box<!fir.heap<i32>>>
457+
! 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>>>)
458+
! CHECK: %[[VAL_2:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>> -> !fir.ref<!fir.box<!fir.heap<i32>>>
459+
! 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>>>)
460+
! CHECK: %[[VAL_4:.*]] = fir.allocmem i32 {fir.must_be_heap = true, uniq_name = "_QFallocatable3Ea.alloc"}
461+
! CHECK: %[[VAL_5:.*]] = fir.embox %[[VAL_4]] : (!fir.heap<i32>) -> !fir.box<!fir.heap<i32>>
462+
! CHECK: fir.store %[[VAL_5]] to %[[VAL_3]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>>
463+
! CHECK: %[[C10_I32:.*]] = arith.constant 10 : i32
464+
! CHECK: hlfir.assign %[[C10_I32]] to %[[VAL_3]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
465+
! CHECK: omp.parallel {
466+
! CHECK: %[[VAL_6:.*]] = omp.threadprivate %[[VAL_1]]#1 : !fir.ref<!fir.box<!fir.heap<i32>>> -> !fir.ref<!fir.box<!fir.heap<i32>>>
467+
! 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>>>)
468+
! CHECK: %[[VAL_8:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
469+
! CHECK: %[[VAL_9:.*]] = fir.box_addr %[[VAL_8]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
470+
! CHECK: %[[VAL_10:.*]] = fir.convert %[[VAL_9]] : (!fir.heap<i32>) -> i64
471+
! CHECK: %[[C10_I64:.*]] = arith.constant 0 : i64
472+
! CHECK: %[[VAL_11:.*]] = arith.cmpi ne, %[[VAL_10]], %[[C10_I64]] : i64
473+
! CHECK: fir.if %[[VAL_11]] {
474+
! CHECK: %[[VAL_16:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
475+
! CHECK: %[[VAL_17:.*]] = fir.box_addr %[[VAL_16]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
476+
! CHECK: %[[VAL_18:.*]] = fir.load %[[VAL_17]] : !fir.heap<i32>
477+
! CHECK: hlfir.assign %[[VAL_18]] to %[[VAL_7]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
478+
! CHECK: }
479+
! CHECK: omp.barrier
480+
! CHECK: %[[VAL_12:.*]] = fir.load %7#0 : !fir.ref<!fir.box<!fir.heap<i32>>>
481+
! CHECK: %[[VAL_13:.*]] = fir.box_addr %[[VAL_12]] : (!fir.box<!fir.heap<i32>>) -> !fir.heap<i32>
482+
! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_13]] : !fir.heap<i32>
483+
! CHECK: %[[C1_I32:.*]] = arith.constant 1 : i32
484+
! CHECK: %[[VAL_15:.*]]= arith.addi %[[VAL_14]], %[[C1_I32]] : i32
485+
! CHECK: hlfir.assign %[[VAL_15]]to %[[VAL_7]]#0 realloc : i32, !fir.ref<!fir.box<!fir.heap<i32>>>
486+
! CHECK: omp.terminator
487+
! CHECK: }
488+
! CHECK: return
489+
! CHECK: }
490+
subroutine allocatable3()
491+
integer, allocatable, save :: a
492+
!$omp threadprivate(a)
493+
allocate(a)
494+
a = 10
495+
!$omp parallel copyin(a)
496+
a = a + 1
497+
!$omp end parallel
498+
end subroutine

0 commit comments

Comments
 (0)