mlir/test/Target/LLVMIR/openmp-taskloop-context-alloca.mlir

// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s

// Check converting code where there are operations other than omp.taskloop.wrapper
// inside of omp.taskloop.context.

// The following is a common pattern generated by flang when accessing
// mutable boxes (e.g. ALLOCATABLE) inside of a taskloop

omp.private {type = private} @_QFtest_taskloopEi_private_i32 : i32

llvm.func @_QPtest_taskloop(%arg0: !llvm.ptr) {
  %0 = llvm.mlir.constant(48 : i32) : i32
  %1 = llvm.mlir.constant(1 : i32) : i32
  %2 = llvm.mlir.constant(100 : i32) : i32
  %3 = llvm.mlir.constant(1 : i64) : i64
  %4 = llvm.alloca %3 x i32 : (i64) -> !llvm.ptr
  omp.taskloop.context private(@_QFtest_taskloopEi_private_i32 %4 -> %arg1 : !llvm.ptr) {
    // test where this alloca ends up
    %5 = llvm.alloca %1 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr
    omp.taskloop.wrapper {
      omp.loop_nest (%arg2) : i32 = (%1) to (%2) inclusive step (%1) {
        llvm.store %arg2, %arg1 : i32, !llvm.ptr
        "llvm.intr.memcpy"(%5, %arg0, %0) <{arg_attrs = [{llvm.align = 8 : i64}, {llvm.align = 8 : i64}, {}], isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
        %6 = llvm.getelementptr %5[0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>
        %7 = llvm.load %6 : !llvm.ptr -> !llvm.ptr
        llvm.call @_QPdo_something(%7) {fastmathFlags = #llvm.fastmath<contract>} : (!llvm.ptr) -> ()
        omp.yield
      }
    }
    omp.terminator
  } {omp.combined}
  llvm.return
}

llvm.func @_QPdo_something(!llvm.ptr) attributes {sym_visibility = "private"}

// Outlined task function:
// CHECK-LABEL: define internal void @_QPtest_taskloop..omp_par(
// CHECK:       taskloop.alloca:
//                ...
// CHECK:         br label %[[VAL_31:taskloop.body]]

// CHECK:       taskloop.body:                                    ; preds = %[[VAL_32:.*]]
// CHECK-NEXT:    br label %[[VAL_33:omp.taskloop.context.region]]

// CHECK:       omp.taskloop.context.region:                      ; preds = %[[VAL_31]]
// This will be folded into the entry block by later optimization passes:
// CHECK-NEXT:    %[[VAL_34:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, align 8
// CHECK-NEXT:    br label %[[VAL_35:omp.taskloop.wrapper.region]]