237 lines · plain
1! REQUIRES: openmp_runtime2 3! This test checks lowering of atomic and atomic update constructs4! RUN: bbc %openmp_flags -fopenmp-version=50 -emit-hlfir %s -o - | FileCheck %s5! RUN: %flang_fc1 -emit-hlfir %openmp_flags -fopenmp-version=50 %s -o - | FileCheck %s6 7program OmpAtomicUpdate8 use omp_lib9!CHECK: %[[VAL_A:.*]] = fir.alloca !fir.box<!fir.ptr<i32>> {bindc_name = "a", uniq_name = "_QFEa"}10!CHECK: %[[ZERO:.*]] = fir.zero_bits !fir.ptr<i32>11!CHECK: %[[EMBOX:.*]] = fir.embox %[[ZERO]] : (!fir.ptr<i32>) -> !fir.box<!fir.ptr<i32>>12!CHECK: fir.store %[[EMBOX]] to %[[VAL_A]] : !fir.ref<!fir.box<!fir.ptr<i32>>>13!CHECK: %[[VAL_A_DECLARE:.*]]:2 = hlfir.declare %[[VAL_A]] {{.*}}14!CHECK: %[[VAL_B:.*]] = fir.alloca !fir.box<!fir.ptr<i32>> {bindc_name = "b", uniq_name = "_QFEb"}15!CHECK: %[[ZERO:.*]] = fir.zero_bits !fir.ptr<i32>16!CHECK: %[[EMBOX:.*]] = fir.embox %[[ZERO]] : (!fir.ptr<i32>) -> !fir.box<!fir.ptr<i32>>17!CHECK: fir.store %[[EMBOX]] to %[[VAL_B]] : !fir.ref<!fir.box<!fir.ptr<i32>>>18!CHECK: %[[VAL_B_DECLARE:.*]]:2 = hlfir.declare %[[VAL_B]] {{.*}}19!CHECK: %[[VAL_C_ADDRESS:.*]] = fir.address_of(@_QFEc) : !fir.ref<i32>20!CHECK: %[[VAL_C_DECLARE:.*]]:2 = hlfir.declare %[[VAL_C_ADDRESS]] {{.*}}21!CHECK: %[[VAL_D_ADDRESS:.*]] = fir.address_of(@_QFEd) : !fir.ref<i32>22!CHECK: %[[VAL_D_DECLARE:.*]]:2 = hlfir.declare %[[VAL_D_ADDRESS]] {{.}}23!CHECK: %[[VAL_G_ADDRESS:.*]] = fir.alloca complex<f32> {bindc_name = "g", uniq_name = "_QFEg"}24!CHECK: %[[VAL_G_DECLARE:.*]]:2 = hlfir.declare %[[VAL_G_ADDRESS]] {uniq_name = "_QFEg"} : (!fir.ref<complex<f32>>) -> (!fir.ref<complex<f32>>, !fir.ref<complex<f32>>)25!CHECK: %[[VAL_i1_ALLOCA:.*]] = fir.alloca i8 {bindc_name = "i1", uniq_name = "_QFEi1"}26!CHECK: %[[VAL_i1_DECLARE:.*]]:2 = hlfir.declare %[[VAL_i1_ALLOCA]] {{.*}}27!CHECK: %[[VAL_c5:.*]] = arith.constant 5 : index28!CHECK: %[[VAL_K_ALLOCA:.*]] = fir.alloca !fir.array<5xi32> {bindc_name = "k", uniq_name = "_QFEk"}29!CHECK: %[[VAL_K_SHAPED:.*]] = fir.shape %[[VAL_c5]] : (index) -> !fir.shape<1>30!CHECK: %[[VAL_K_DECLARE:.*]]:2 = hlfir.declare %[[VAL_K_ALLOCA]](%[[VAL_K_SHAPED]]) {{.*}}31 32!CHECK: %[[VAL_W_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "w", uniq_name = "_QFEw"}33!CHECK: %[[VAL_W_DECLARE:.*]]:2 = hlfir.declare %[[VAL_W_ALLOCA]] {uniq_name = "_QFEw"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)34!CHECK: %[[VAL_X_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFEx"}35!CHECK: %[[VAL_X_DECLARE:.*]]:2 = hlfir.declare %[[VAL_X_ALLOCA]] {uniq_name = "_QFEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)36!CHECK: %[[VAL_Y_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "y", uniq_name = "_QFEy"}37!CHECK: %[[VAL_Y_DECLARE:.*]]:2 = hlfir.declare %[[VAL_Y_ALLOCA]] {uniq_name = "_QFEy"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)38!CHECK: %[[VAL_Z_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "z", uniq_name = "_QFEz"}39!CHECK: %[[VAL_Z_DECLARE:.*]]:2 = hlfir.declare %[[VAL_Z_ALLOCA]] {uniq_name = "_QFEz"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)40 integer :: w, x, y, z41 integer, pointer :: a, b42 integer, target :: c, d43 integer(1) :: i144 integer, dimension(5) :: k45 complex :: g46 47!CHECK: %[[EMBOX:.*]] = fir.embox %[[VAL_C_DECLARE]]#0 : (!fir.ref<i32>) -> !fir.box<!fir.ptr<i32>>48!CHECK: fir.store %[[EMBOX]] to %[[VAL_A_DECLARE]]#0 : !fir.ref<!fir.box<!fir.ptr<i32>>>49!CHECK: %[[EMBOX:.*]] = fir.embox %[[VAL_D_DECLARE]]#0 : (!fir.ref<i32>) -> !fir.box<!fir.ptr<i32>>50!CHECK: fir.store %[[EMBOX]] to %[[VAL_B_DECLARE]]#0 : !fir.ref<!fir.box<!fir.ptr<i32>>>51 a=>c52 b=>d53 54!CHECK: %[[VAL_c3:.*]] = arith.constant 3 : index55!CHECK: %[[VAL_K_DESIGNATE:.*]] = hlfir.designate %[[VAL_K_DECLARE]]#0 (%[[VAL_c3]]) : (!fir.ref<!fir.array<5xi32>>, index) -> !fir.ref<i32>56!CHECK: %[[LOADED_Z:.*]] = fir.load %[[VAL_Z_DECLARE]]#0 : !fir.ref<i32>57!CHECK: omp.atomic.update %[[VAL_K_DESIGNATE]] : !fir.ref<i32> {58!CHECK: ^bb0(%[[ARG:.*]]: i32):59!CHECK: %[[TEMP:.*]] = arith.muli %[[LOADED_Z]], %[[ARG]] : i3260!CHECK: omp.yield(%[[TEMP]] : i32)61!CHECK: }62 !$omp atomic update63 k(3) = z * k(3)64 65!CHECK: %[[VAL_A_LOADED:.*]] = fir.load %[[VAL_A_DECLARE]]#0 : !fir.ref<!fir.box<!fir.ptr<i32>>>66!CHECK: %[[VAL_A_BOX_ADDR:.*]] = fir.box_addr %[[VAL_A_LOADED]] : (!fir.box<!fir.ptr<i32>>) -> !fir.ptr<i32>67!CHECK: %[[VAL_B_LOADED:.*]] = fir.load %[[VAL_B_DECLARE]]#0 : !fir.ref<!fir.box<!fir.ptr<i32>>>68!CHECK: %[[VAL_B_BOX_ADDR:.*]] = fir.box_addr %[[VAL_B_LOADED]] : (!fir.box<!fir.ptr<i32>>) -> !fir.ptr<i32>69!CHECK: %[[VAL_B:.*]] = fir.load %[[VAL_B_BOX_ADDR]] : !fir.ptr<i32>70!CHECK: omp.atomic.update %[[VAL_A_BOX_ADDR]] : !fir.ptr<i32> {71!CHECK: ^bb0(%[[ARG:.*]]: i32):72!CHECK: %[[TEMP:.*]] = arith.addi %[[ARG]], %[[VAL_B]] : i3273!CHECK: omp.yield(%[[TEMP]] : i32)74!CHECK: }75 !$omp atomic update76 a = a + b 77 78!CHECK: %[[VAL_c1:.*]] = arith.constant 1 : i3279!CHECK: omp.atomic.update %[[VAL_Y_DECLARE]]#0 : !fir.ref<i32> {80!CHECK: ^bb0(%[[ARG:.*]]: i32):81!CHECK: %[[TEMP:.*]] = arith.addi %[[ARG]], %[[VAL_c1]] : i3282!CHECK: omp.yield(%[[TEMP]] : i32)83!CHECK: }84 !$omp atomic 85 y = y + 186 87!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref<i32>88!CHECK: omp.atomic.update %[[VAL_Z_DECLARE]]#0 : !fir.ref<i32> {89!CHECK: ^bb0(%[[ARG:.*]]: i32):90!CHECK: %[[TEMP:.*]] = arith.muli %[[VAL_X_LOADED]], %[[ARG]] : i3291!CHECK: omp.yield(%[[TEMP]] : i32)92!CHECK: }93 !$omp atomic update94 z = x * z 95 96!CHECK: %[[VAL_c1:.*]] = arith.constant 1 : i3297!CHECK: omp.atomic.update hint(uncontended) memory_order(relaxed) %[[VAL_X_DECLARE]]#0 : !fir.ref<i32> {98!CHECK: ^bb0(%[[ARG:.*]]: i32):99!CHECK: %[[TEMP:.*]] = arith.subi %[[ARG]], %[[VAL_c1]] : i32100!CHECK: omp.yield(%[[TEMP]] : i32)101!CHECK: }102 !$omp atomic relaxed update hint(omp_sync_hint_uncontended)103 x = x - 1104 105!CHECK: %[[VAL_C_LOADED:.*]] = fir.load %[[VAL_C_DECLARE]]#0 : !fir.ref<i32>106!CHECK: %[[VAL_D_LOADED:.*]] = fir.load %[[VAL_D_DECLARE]]#0 : !fir.ref<i32>107!CHECK: omp.atomic.update memory_order(relaxed) %[[VAL_Y_DECLARE]]#0 : !fir.ref<i32> {108!CHECK: ^bb0(%[[ARG:.*]]: i32):109!CHECK: {{.*}} = arith.cmpi sgt, %[[ARG]], {{.*}} : i32110!CHECK: %[[TEMP:.*]] = arith.select {{.*}} : i32111!CHECK: omp.yield(%[[TEMP]] : i32)112!CHECK: }113 !$omp atomic update relaxed 114 y = max(y, c, d)115 116!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref<i32>117!CHECK: omp.atomic.update hint(contended) memory_order(relaxed) %[[VAL_Z_DECLARE]]#0 : !fir.ref<i32> {118!CHECK: ^bb0(%[[ARG:.*]]: i32):119!CHECK: %[[TEMP:.*]] = arith.addi %[[ARG]], %[[VAL_X_LOADED]] : i32120!CHECK: omp.yield(%[[TEMP]] : i32)121!CHECK: }122 !$omp atomic relaxed hint(omp_sync_hint_contended)123 z = z + x124 125!CHECK: %[[VAL_c10:.*]] = arith.constant 10 : i32126!CHECK: omp.atomic.update hint(contended) memory_order(release) %[[VAL_Z_DECLARE]]#0 : !fir.ref<i32> {127!CHECK: ^bb0(%[[ARG:.*]]: i32):128!CHECK: %[[TEMP:.*]] = arith.muli %[[VAL_c10]], %[[ARG]] : i32129!CHECK: omp.yield(%[[TEMP]] : i32)130!CHECK: }131 !$omp atomic release update hint(omp_lock_hint_contended)132 z = z * 10133 134!CHECK: %[[VAL_Z_LOADED:.*]] = fir.load %[[VAL_Z_DECLARE]]#0 : !fir.ref<i32>135!CHECK: omp.atomic.update hint(speculative) memory_order(release) %[[VAL_X_DECLARE]]#0 : !fir.ref<i32> {136!CHECK: ^bb0(%[[ARG:.*]]: i32):137!CHECK: %[[TEMP:.*]] = arith.divsi %[[ARG]], %[[VAL_Z_LOADED]] : i32138!CHECK: omp.yield(%[[TEMP]] : i32)139!CHECK: }140 !$omp atomic hint(omp_lock_hint_speculative) update release141 x = x / z142 143!CHECK: %[[VAL_c1:.*]] = arith.constant 1 : i32144!CHECK: omp.atomic.update %[[VAL_i1_DECLARE]]#0 : !fir.ref<i8> {145!CHECK: ^bb0(%[[ARG:.*]]: i8):146!CHECK: %[[CONVERT:.*]] = fir.convert %[[ARG]] : (i8) -> i32147!CHECK: %[[ADD:.*]] = arith.addi %[[CONVERT]], %[[VAL_c1]] : i32148!CHECK: %[[TEMP:.*]] = fir.convert %[[ADD]] : (i32) -> i8149!CHECK: omp.yield(%[[TEMP]] : i8)150!CHECK: }151 !$omp atomic152 i1 = i1 + 1153 !$omp end atomic154 155!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref<i32>156!CHECK: omp.atomic.update %[[VAL_Y_DECLARE]]#0 : !fir.ref<i32> {157!CHECK: ^bb0(%[[ARG_Y:.*]]: i32):158!CHECK: %[[Y_UPDATE_VAL:.*]] = arith.andi %[[VAL_X_LOADED]], %[[ARG_Y]] : i32159!CHECK: omp.yield(%[[Y_UPDATE_VAL]] : i32)160!CHECK: }161 !$omp atomic update162 y = iand(x,y)163 164!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref<i32>165!CHECK: omp.atomic.update %[[VAL_Y_DECLARE]]#0 : !fir.ref<i32> {166!CHECK: ^bb0(%[[ARG_Y:.*]]: i32):167!CHECK: %[[Y_UPDATE_VAL:.*]] = arith.xori %[[VAL_X_LOADED]], %[[ARG_Y]] : i32168!CHECK: omp.yield(%[[Y_UPDATE_VAL]] : i32)169!CHECK: }170 !$omp atomic update171 y = ieor(x,y)172 173!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref<i32>174!CHECK: %[[VAL_Y_LOADED:.*]] = fir.load %[[VAL_Y_DECLARE]]#0 : !fir.ref<i32>175!CHECK: %[[VAL_Z_LOADED:.*]] = fir.load %[[VAL_Z_DECLARE]]#0 : !fir.ref<i32>176!CHECK: omp.atomic.update %[[VAL_W_DECLARE]]#0 : !fir.ref<i32> {177!CHECK: ^bb0(%[[ARG_W:.*]]: i32):178!CHECK: %[[W_CMP:.*]] = arith.cmpi sgt, %[[ARG_W]], {{.*}} : i32179!CHECK: %[[WXYZ_MAX:.*]] = arith.select %[[W_CMP]], %[[ARG_W]], {{.*}} : i32180!CHECK: omp.yield(%[[WXYZ_MAX]] : i32)181!CHECK: }182 !$omp atomic update183 w = max(w,x,y,z)184 185!CHECK: %[[IMP_DO:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<?xi32> {186!CHECK: ^bb0(%{{.*}}: index):187! [...]188!CHECK: %[[ADD_I1:.*]] = arith.addi {{.*}} : i32189!CHECK: hlfir.yield_element %[[ADD_I1]] : i32190!CHECK: }191! [...]192!CHECK: %[[SUM:.*]] = hlfir.sum %[[IMP_DO]]193!CHECK: omp.atomic.update %[[VAL_X_DECLARE]]#0 : !fir.ref<i32> {194!CHECK: ^bb0(%[[ARG0:.*]]: i32):195!CHECK: %[[ADD_I2:.*]] = arith.addi %[[ARG0]], %[[SUM]] : i32196!CHECK: omp.yield(%[[ADD_I2]] : i32)197!CHECK: }198 !$omp atomic update199 x = x + sum([ (y+2, y=1, z) ])200 201!CHECK: %[[LOAD:.*]] = fir.load %[[VAL_G_DECLARE]]#0 : !fir.ref<complex<f32>>202!CHECK: omp.atomic.update %[[VAL_W_DECLARE]]#0 : !fir.ref<i32> {203!CHECK: ^bb0(%[[ARG:.*]]: i32):204!CHECK: %[[CVT:.*]] = fir.convert %[[ARG]] : (i32) -> f32205!CHECK: %[[CST:.*]] = arith.constant 0.000000e+00 : f32206!CHECK: %[[UNDEF:.*]] = fir.undefined complex<f32>207!CHECK: %[[IDX1:.*]] = fir.insert_value %[[UNDEF]], %[[CVT]], [0 : index] : (complex<f32>, f32) -> complex<f32>208!CHECK: %[[IDX2:.*]] = fir.insert_value %[[IDX1]], %[[CST]], [1 : index] : (complex<f32>, f32) -> complex<f32>209!CHECK: %[[ADD:.*]] = fir.addc %[[IDX2]], %[[LOAD]] {fastmath = #arith.fastmath<contract>} : complex<f32>210!CHECK: %[[EXT:.*]] = fir.extract_value %[[ADD]], [0 : index] : (complex<f32>) -> f32211!CHECK: %[[RESULT:.*]] = fir.convert %[[EXT]] : (f32) -> i32212!CHECK: omp.yield(%[[RESULT]] : i32)213 !$omp atomic update214 w = w + g 215end program OmpAtomicUpdate216 217! Check that the clean-ups associated with the function call218! are generated after the omp.atomic.update operation:219! CHECK-LABEL: func.func @_QPfunc_call_cleanup(220subroutine func_call_cleanup(v, vv)221 integer v, vv222 223! CHECK: %[[VAL_6:.*]]:3 = hlfir.associate %{{.*}} {adapt.valuebyref} : (i32) -> (!fir.ref<i32>, !fir.ref<i32>, i1)224! CHECK: %[[VAL_7:.*]] = fir.call @_QPfunc(%[[VAL_6]]#0) fastmath<contract> : (!fir.ref<i32>) -> f32225! CHECK: omp.atomic.update %{{.*}} : !fir.ref<i32> {226! CHECK: ^bb0(%[[VAL_8:.*]]: i32):227! CHECK: %[[VAL_9:.*]] = fir.convert %[[VAL_8]] : (i32) -> f32228! CHECK: %[[VAL_10:.*]] = arith.addf %[[VAL_9]], %[[VAL_7]] fastmath<contract> : f32229! CHECK: %[[VAL_11:.*]] = fir.convert %[[VAL_10]] : (f32) -> i32230! CHECK: omp.yield(%[[VAL_11]] : i32)231! CHECK: }232! CHECK: hlfir.end_associate %[[VAL_6]]#1, %[[VAL_6]]#2 : !fir.ref<i32>, i1233 !$omp atomic update234 v = v + func(vv + 1)235 !$omp end atomic236end subroutine func_call_cleanup237