104 lines · plain
1! RUN: bbc -emit-hlfir -fopenmp -o - %s 2>&1 | FileCheck %s --check-prefix=CPU2! RUN: %flang_fc1 -emit-hlfir -fopenmp -o - %s 2>&1 | FileCheck %s --check-prefix=CPU3 4! RUN: bbc -emit-hlfir -fopenmp -fopenmp-is-target-device -fopenmp-is-gpu -o - %s 2>&1 | FileCheck %s --check-prefix=GPU5! RUN: %if amdgpu-registered-target %{ %flang_fc1 -triple amdgcn-amd-amdhsa -emit-hlfir -fopenmp -fopenmp-is-target-device -o - %s 2>&1 | \6! RUN: FileCheck %s --check-prefix=GPU %}7 8program reduce9integer, dimension(3) :: i = 010 11!$omp parallel reduction(+:i)12i(1) = 113i(2) = 214i(3) = 315!$omp end parallel16 17print *,i18end program19 20! CPU-LABEL: omp.declare_reduction @add_reduction_byref_box_3xi32 : !fir.ref<!fir.box<!fir.array<3xi32>>> attributes {byref_element_type = !fir.array<3xi32>} alloc {21! CPU: %[[VAL_8:.*]] = fir.alloca !fir.box<!fir.array<3xi32>>22! CPU: omp.yield(%[[VAL_8]] : !fir.ref<!fir.box<!fir.array<3xi32>>>)23! CPU-LABEL: } init {24! CPU: ^bb0(%[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.array<3xi32>>>, %[[ALLOC:.*]]: !fir.ref<!fir.box<!fir.array<3xi32>>>):25! CPU: %[[VAL_2:.*]] = arith.constant 0 : i3226! CPU: %[[VAL_3:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3xi32>>>27! CPU: %[[VAL_4:.*]] = arith.constant 3 : index28! CPU: %[[VAL_5:.*]] = fir.shape %[[VAL_4]] : (index) -> !fir.shape<1>29! CPU: %[[VAL_1:.*]] = fir.allocmem !fir.array<3xi32> {bindc_name = ".tmp", uniq_name = ""}30! CPU: %[[VAL_6:.*]]:2 = hlfir.declare %[[VAL_1]](%[[VAL_5]]) {uniq_name = ".tmp"} : (!fir.heap<!fir.array<3xi32>>,31!fir.shape<1>) -> (!fir.heap<!fir.array<3xi32>>, !fir.heap<!fir.array<3xi32>>)32! CPU: %[[C0:.*]] = arith.constant 0 : index33! CPU: %[[DIMS:.*]]:3 = fir.box_dims %[[VAL_3]], %[[C0]] : (!fir.box<!fir.array<3xi32>>, index) -> (index, index, index)34! CPU: %[[SHIFT:.*]] = fir.shape_shift %[[DIMS]]#0, %[[DIMS]]#1 : (index, index) -> !fir.shapeshift<1>35! CPU: %[[VAL_7:.*]] = fir.embox %[[VAL_6]]#0(%[[SHIFT]]) : (!fir.heap<!fir.array<3xi32>>, !fir.shapeshift<1>) -> !fir.box<!fir.array<3xi32>>36! CPU: hlfir.assign %[[VAL_2]] to %[[VAL_7]] : i32, !fir.box<!fir.array<3xi32>>37! CPU: fir.store %[[VAL_7]] to %[[ALLOC]] : !fir.ref<!fir.box<!fir.array<3xi32>>>38! CPU: omp.yield(%[[ALLOC]] : !fir.ref<!fir.box<!fir.array<3xi32>>>)39! CPU: } combiner {40! CPU: ^bb0(%[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.array<3xi32>>>, %[[VAL_1:.*]]: !fir.ref<!fir.box<!fir.array<3xi32>>>):41! CPU: %[[VAL_2:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3xi32>>>42! CPU: %[[VAL_3:.*]] = fir.load %[[VAL_1]] : !fir.ref<!fir.box<!fir.array<3xi32>>>43! CPU: %[[C1:.*]] = arith.constant 1 : index44! CPU: %[[C3:.*]] = arith.constant 3 : index45! CPU: %[[SHAPE_SHIFT:.*]] = fir.shape_shift %[[C1]], %[[C3]] : (index, index) -> !fir.shapeshift<1>46! CPU: %[[C1_0:.*]] = arith.constant 1 : index47! CPU: fir.do_loop %[[VAL_8:.*]] = %[[C1_0]] to %[[C3]] step %[[C1_0]] unordered {48! CPU: %[[VAL_9:.*]] = fir.array_coor %[[VAL_2]](%[[SHAPE_SHIFT]]) %[[VAL_8]] : (!fir.box<!fir.array<3xi32>>, !fir.shapeshift<1>, index) -> !fir.ref<i32>49! CPU: %[[VAL_10:.*]] = fir.array_coor %[[VAL_3]](%[[SHAPE_SHIFT]]) %[[VAL_8]] : (!fir.box<!fir.array<3xi32>>, !fir.shapeshift<1>, index) -> !fir.ref<i32>50! CPU: %[[VAL_11:.*]] = fir.load %[[VAL_9]] : !fir.ref<i32>51! CPU: %[[VAL_12:.*]] = fir.load %[[VAL_10]] : !fir.ref<i32>52! CPU: %[[VAL_13:.*]] = arith.addi %[[VAL_11]], %[[VAL_12]] : i3253! CPU: fir.store %[[VAL_13]] to %[[VAL_9]] : !fir.ref<i32>54! CPU: }55! CPU: omp.yield(%[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3xi32>>>)56! CPU: } cleanup {57! CPU: ^bb0(%[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.array<3xi32>>>):58! CPU: %[[VAL_1:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3xi32>>>59! CPU: %[[VAL_2:.*]] = fir.box_addr %[[VAL_1]] : (!fir.box<!fir.array<3xi32>>) -> !fir.ref<!fir.array<3xi32>>60! CPU: %[[VAL_3:.*]] = fir.convert %[[VAL_2]] : (!fir.ref<!fir.array<3xi32>>) -> i6461! CPU: %[[VAL_4:.*]] = arith.constant 0 : i6462! CPU: %[[VAL_5:.*]] = arith.cmpi ne, %[[VAL_3]], %[[VAL_4]] : i6463! CPU: fir.if %[[VAL_5]] {64! CPU: %[[VAL_6:.*]] = fir.convert %[[VAL_2]] : (!fir.ref<!fir.array<3xi32>>) -> !fir.heap<!fir.array<3xi32>>65! CPU: fir.freemem %[[VAL_6]] : !fir.heap<!fir.array<3xi32>>66! CPU: }67! CPU: omp.yield68! CPU: }69 70! CPU-LABEL: func.func @_QQmain()71! CPU: %[[VAL_0:.*]] = fir.address_of(@_QFEi) : !fir.ref<!fir.array<3xi32>>72! CPU: %[[VAL_1:.*]] = arith.constant 3 : index73! CPU: %[[VAL_2:.*]] = fir.shape %[[VAL_1]] : (index) -> !fir.shape<1>74! CPU: %[[VAL_3:.*]]:2 = hlfir.declare %[[VAL_0]](%[[VAL_2]]) {uniq_name = "_QFEi"} : (!fir.ref<!fir.array<3xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<3xi32>>, !fir.ref<!fir.array<3xi32>>)75! CPU: %[[VAL_4:.*]] = fir.embox %[[VAL_3]]#0(%[[VAL_2]]) : (!fir.ref<!fir.array<3xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<3xi32>>76! CPU: %[[VAL_5:.*]] = fir.alloca !fir.box<!fir.array<3xi32>>77! CPU: fir.store %[[VAL_4]] to %[[VAL_5]] : !fir.ref<!fir.box<!fir.array<3xi32>>>78! CPU: omp.parallel reduction(byref @add_reduction_byref_box_3xi32 %[[VAL_5]] -> %[[VAL_6:.*]] : !fir.ref<!fir.box<!fir.array<3xi32>>>) {79! CPU: %[[VAL_7:.*]]:2 = hlfir.declare %[[VAL_6]] {uniq_name = "_QFEi"} : (!fir.ref<!fir.box<!fir.array<3xi32>>>) -> (!fir.ref<!fir.box<!fir.array<3xi32>>>, !fir.ref<!fir.box<!fir.array<3xi32>>>)80! CPU: %[[VAL_8:.*]] = arith.constant 1 : i3281! CPU: %[[VAL_9:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<!fir.box<!fir.array<3xi32>>>82! CPU: %[[VAL_10:.*]] = arith.constant 1 : index83! CPU: %[[VAL_11:.*]] = hlfir.designate %[[VAL_9]] (%[[VAL_10]]) : (!fir.box<!fir.array<3xi32>>, index) -> !fir.ref<i32>84! CPU: hlfir.assign %[[VAL_8]] to %[[VAL_11]] : i32, !fir.ref<i32>85! CPU: %[[VAL_12:.*]] = arith.constant 2 : i3286! CPU: %[[VAL_13:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<!fir.box<!fir.array<3xi32>>>87! CPU: %[[VAL_14:.*]] = arith.constant 2 : index88! CPU: %[[VAL_15:.*]] = hlfir.designate %[[VAL_13]] (%[[VAL_14]]) : (!fir.box<!fir.array<3xi32>>, index) -> !fir.ref<i32>89! CPU: hlfir.assign %[[VAL_12]] to %[[VAL_15]] : i32, !fir.ref<i32>90! CPU: %[[VAL_16:.*]] = arith.constant 3 : i3291! CPU: %[[VAL_17:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<!fir.box<!fir.array<3xi32>>>92! CPU: %[[VAL_18:.*]] = arith.constant 3 : index93! CPU: %[[VAL_19:.*]] = hlfir.designate %[[VAL_17]] (%[[VAL_18]]) : (!fir.box<!fir.array<3xi32>>, index) -> !fir.ref<i32>94! CPU: hlfir.assign %[[VAL_16]] to %[[VAL_19]] : i32, !fir.ref<i32>95! CPU: omp.terminator96! CPU: }97 98! GPU: omp.declare_reduction {{.*}} alloc {99! GPU: } init {100! GPU-NOT: fir.allocmem {{.*}} {bindc_name = ".tmp", {{.*}}}101! GPU: fir.alloca {{.*}} {bindc_name = ".tmp"}102! GPU: } combiner {103! GPU: }104