shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[flang][OpenMP] Add OutlineableOpenMPOpInterface to omp.teams (#131109)

Browse Source 
Given the following input:
```fortran
program rep_loopbind
  implicit none
  integer :: i
  real :: priv_val

  !$omp teams private(priv_val)
    !$omp distribute
    do i=1,1000
    end do
  !$omp end teams
end program
```
the `AllocaOpConversion` pattern in `FIRToLLVMLowering` would **move**
the private allocations that belong to the `teams` directive (i.e. the
allocations needed for the private copies of `priv_val` and the loop's
iteration variable) from the the `omp.teams` op to the outside scope.

This is not correct since these allocations should be eventually emitted
inside the outlined region for the `teams` directive. Without this fix,
these allocation would be emitted in the parent function (or the parent
scope whatever it is).
This commit is contained in:
Kareem Ergawy 2025-03-13 16:03:19 +01:00 committed by GitHub
parent 7661526fdf
commit b003face11
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 40 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
flang/test/Fir/omp-teams.fir Normal file
View File

@ -0,0 +1,38 @@
// RUN: fir-opt --fir-to-llvm-ir %s | FileCheck %s
// Tests that private allocations of an `omp.teams` op remain nested inside the
// owning `omp.teams` op after FIR to LLVM conversion.
func.func @foo() {
omp.teams {
%2 = fir.alloca f32 {bindc_name = "priv_val", pinned, uniq_name = "_QFEpriv_val"}
%3 = fir.alloca i32 {bindc_name = "i", pinned, uniq_name = "_QFEi"}
%c1_i32 = arith.constant 1 : i32
%c1000_i32 = arith.constant 1000 : i32
%c1_i32_0 = arith.constant 1 : i32
omp.distribute {
omp.loop_nest (%arg0) : i32 = (%c1_i32) to (%c1000_i32) inclusive step (%c1_i32_0) {
fir.store %arg0 to %3 : !fir.ref<i32>
omp.yield
}
}
omp.terminator
}
return
}
// CHECK-LABEL: llvm.func @foo() {
// CHECK: omp.teams {
// CHECK: %[[TEAMS_IV_ALLOC:.*]] = llvm.alloca %{{.*}} x i32 {bindc_name = "i", pinned}
// CHECK: %[[TEAMS_PRIV_ALLOC:.*]] = llvm.alloca %{{.*}} x f32 {bindc_name = "priv_val", pinned}
// CHECK: omp.distribute {
// CHECK: omp.loop_nest (%{{.*}}) : i32 = (%{{.*}}) to (%{{.*}}) inclusive step (%{{.*}}) {
// CHECK: llvm.store %{{.*}}, %[[TEAMS_IV_ALLOC]] : i32, !llvm.ptr
// CHECK: omp.yield
// CHECK: }
// CHECK: }
// CHECK: omp.terminator
// CHECK: }
// CHECK: llvm.return
// CHECK: }

2
flang/test/Transforms/stack-arrays-hlfir.f90
View File

@ -71,8 +71,8 @@ end subroutine omp_target_wsloop
! CHECK: omp.target {{.*}} {
! CHECK-NOT: fir.allocmem
! CHECK-NOT: fir.freemem
! CHECK: fir.alloca !fir.array<2xi64>
! CHECK: omp.teams {
! CHECK: fir.alloca !fir.array<2xi64>
! CHECK: omp.distribute {
! CHECK: omp.loop_nest {{.*}} {
! CHECK-NOT: fir.allocmem

2
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
View File

@ -236,7 +236,7 @@ def TerminatorOp : OpenMP_Op<"terminator", [Terminator, Pure]> {
// 2.7 teams Construct
//===----------------------------------------------------------------------===//
def TeamsOp : OpenMP_Op<"teams", traits = [
AttrSizedOperandSegments, RecursiveMemoryEffects
AttrSizedOperandSegments, RecursiveMemoryEffects, OutlineableOpenMPOpInterface
], clauses = [
OpenMP_AllocateClause, OpenMP_IfClause, OpenMP_NumTeamsClause,
OpenMP_PrivateClause, OpenMP_ReductionClause, OpenMP_ThreadLimitClause