[flang-commits] [flang] [flang][OpenMP] Lower DO, SIMD, and DO SIMD metadirective variants (PR #218555)

[via flang-commits]

================
@@ -0,0 +1,562 @@
+! Test lowering of metadirectives with ordinary loop-associated variants.
+
+! RUN: %flang_fc1 -fopenmp -emit-hlfir -fopenmp-version=52 %s -o - | FileCheck %s
+
+! CHECK: #loop_unroll = #llvm.loop_unroll<disable = false, count = 4 : i64>
+! CHECK: #loop_annotation = #llvm.loop_annotation<unroll = #loop_unroll>
+
+! CHECK-LABEL: func.func @_QPtest_do(
+! CHECK-NOT:     omp.parallel
+! CHECK:         omp.wsloop
+! CHECK:           omp.loop_nest
+! CHECK:             hlfir.assign
+! CHECK:             omp.yield
+! CHECK:         return
+subroutine test_do(n, a)
+  integer :: n, a(n), i
+  !$omp metadirective &
+  !$omp & when(implementation={vendor(llvm)}: do) &
+  !$omp & otherwise(nothing)
+  do i = 1, n
+    a(i) = i
+  end do
+end subroutine
+
+! CHECK-LABEL: func.func @_QPtest_simd(
+! CHECK-NOT:     omp.wsloop
+! CHECK:         omp.simd linear(
+! CHECK:           omp.loop_nest
+! CHECK:             hlfir.assign
+! CHECK:             omp.yield
+! CHECK-NOT:     fir.do_loop
+! CHECK:         fir.load
+! CHECK:         hlfir.assign
+! CHECK:         return
+subroutine test_simd(n, a, after)
+  integer :: n, a(n), after, i
+  !$omp metadirective &
+  !$omp & when(implementation={vendor(llvm)}: simd) &
+  !$omp & otherwise(nothing)
+  do i = 1, n
+    a(i) = i
+  end do
+  after = i
+end subroutine
+
+! CHECK-LABEL: func.func @_QPtest_do_simd(
+! CHECK-NOT:     omp.parallel
+! CHECK:         omp.wsloop
+! CHECK:           omp.simd linear(
+! CHECK:             omp.loop_nest
+! CHECK:               hlfir.assign
+! CHECK:               omp.yield
+! CHECK:         return
+subroutine test_do_simd(n, a)
+  integer :: n, a(n), i
+  !$omp metadirective &
+  !$omp & when(implementation={vendor(llvm)}: do simd) &
+  !$omp & otherwise(nothing)
+  do i = 1, n
+    a(i) = i
+  end do
+end subroutine
+
+! CHECK-LABEL: func.func @_QPtest_begin_do(
+! CHECK:         fir.if {{.*}} {
+! CHECK:           omp.wsloop
+! CHECK:             omp.loop_nest
+! CHECK:               hlfir.assign
+! CHECK:         } else {
+! CHECK-NOT:       omp.
+! CHECK:           fir.do_loop
+! CHECK:             hlfir.assign
+! CHECK:         }
+! CHECK:         return
----------------
MattPD wrote:

The `test_begin_do` CHECK block still passes when the checked output contains a third `fir.do_loop` between the `fir.if` and the `return` for `test_begin_do`. The checks therefore do not detect a duplicate associated loop after the `fir.if`.

Could bounded `CHECK-NOT` lines exclude two kinds of operations in that range: `fir.do_loop` and OpenMP loop operations? The test would then prove that lowering emits the associated loop only once.

https://github.com/llvm/llvm-project/pull/218555