diff --git a/flang/test/Lower/OpenMP/atomic-capture.f90 b/flang/test/Lower/OpenMP/atomic-capture.f90 new file mode 100644 index 0000000000000..cde0281dbdc84 --- /dev/null +++ b/flang/test/Lower/OpenMP/atomic-capture.f90 @@ -0,0 +1,98 @@ +! This test checks the lowering of atomic capture + +! RUN: bbc -fopenmp -emit-hlfir %s -o - | FileCheck %s +! RUN: %flang_fc1 -emit-hlfir -fopenmp %s -o - | FileCheck %s + + + +program OmpAtomicCapture + use omp_lib + +!CHECK: %[[VAL_X_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFEx"} +!CHECK: %[[VAL_X_DECLARE:.*]]:2 = hlfir.declare %[[VAL_X_ALLOCA]] {{.*}} +!CHECK: %[[VAL_Y_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "y", uniq_name = "_QFEy"} +!CHECK: %[[VAL_Y_DECLARE:.*]]:2 = hlfir.declare %[[VAL_Y_ALLOCA]] {{.*}} + integer :: x, y + +!CHECK: %[[VAL_Y_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref +!CHECK: omp.atomic.capture hint(uncontended) { +!CHECK: omp.atomic.update %[[VAL_Y_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.muli %[[VAL_Y_LOADED]], %[[ARG]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } +!CHECK: omp.atomic.read %[[VAL_X_DECLARE]]#1 = %[[VAL_Y_DECLARE]]#1 : !fir.ref, i32 +!CHECK: } + !$omp atomic hint(omp_sync_hint_uncontended) capture + y = x * y + x = y + !$omp end atomic + +!CHECK: %[[VAL_20:.*]] = arith.constant 20 : i32 +!CHECK: %[[VAL_8:.*]] = arith.constant 8 : i32 +!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref +!CHECK: %[[SUB:.*]] = arith.subi %[[VAL_8]], %[[VAL_X_LOADED]] : i32 +!CHECK: %[[NO_REASSOC:.*]] = hlfir.no_reassoc %[[SUB]] : i32 +!CHECK: %[[ADD:.*]] = arith.addi %[[VAL_20]], %[[NO_REASSOC]] : i32 +!CHECK: omp.atomic.capture memory_order(acquire) hint(nonspeculative) { +!CHECK: omp.atomic.read %[[VAL_X_DECLARE]]#1 = %[[VAL_Y_DECLARE]]#1 : !fir.ref, i32 +!CHECK: omp.atomic.write %[[VAL_Y_DECLARE]]#1 = %[[ADD]] : !fir.ref, i32 +!CHECK: } +!CHECK: return +!CHECK: } + !$omp atomic hint(omp_lock_hint_nonspeculative) capture acquire + x = y + y = 2 * 10 + (8 - x) + !$omp end atomic +end program + + +!CHECK: func.func @_QPpointers_in_atomic_capture() { +subroutine pointers_in_atomic_capture() + +!CHECK: %[[VAL_A_ALLOCA:.*]] = fir.alloca !fir.box> {bindc_name = "a", uniq_name = "_QFpointers_in_atomic_captureEa"} +!CHECK: %[[ZERO:.*]] = fir.zero_bits !fir.ptr +!CHECK: %[[EMBOX:.*]] = fir.embox %[[ZERO]] : (!fir.ptr) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_A_ALLOCA]] : !fir.ref>> +!CHECK: %[[VAL_A_DECLARE:.*]]:2 = hlfir.declare %[[VAL_A_ALLOCA]] {{.*}} +!CHECK: %[[VAL_B_ALLOCA:.*]] = fir.alloca !fir.box> {bindc_name = "b", uniq_name = "_QFpointers_in_atomic_captureEb"} +!CHECK: %[[ZERO:.*]] = fir.zero_bits !fir.ptr +!CHECK: %[[EMBOX:.*]] = fir.embox %[[ZERO]] : (!fir.ptr) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_B_ALLOCA]] : !fir.ref>> +!CHECK: %[[VAL_B_DECLARE:.*]]:2 = hlfir.declare %[[VAL_B_ALLOCA]] {{.*}} +!CHECK: %[[VAL_C_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "c", fir.target, uniq_name = "_QFpointers_in_atomic_captureEc"} +!CHECK: %[[VAL_C_DECLARE:.*]]:2 = hlfir.declare %[[VAL_C_ALLOCA]] {{.*}} +!CHECK: %[[VAL_D_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "d", fir.target, uniq_name = "_QFpointers_in_atomic_captureEd"} +!CHECK: %[[VAL_D_DECLARE:.*]]:2 = hlfir.declare %[[VAL_D_ALLOCA]] {{.*}} + integer, pointer :: a, b + integer, target :: c, d + +!CHECK: %[[EMBOX:.*]] = fir.embox %[[VAL_C_DECLARE]]#1 : (!fir.ref) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_A_DECLARE]]#1 : !fir.ref>> +!CHECK: %[[EMBOX:.*]] = fir.embox %[[VAL_D_DECLARE]]#1 : (!fir.ref) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_B_DECLARE]]#1 : !fir.ref>> + a=>c + b=>d + +!CHECK: %[[VAL_A_LOADED:.*]] = fir.load %[[VAL_A_DECLARE]]#0 : !fir.ref>> +!CHECK: %[[VAL_A_BOX_ADDR:.*]] = fir.box_addr %[[VAL_A_LOADED]] : (!fir.box>) -> !fir.ptr +!CHECK: %[[VAL_B_LOADED:.*]] = fir.load %[[VAL_B_DECLARE]]#0 : !fir.ref>> +!CHECK: %[[VAL_B_BOX_ADDR:.*]] = fir.box_addr %[[VAL_B_LOADED]] : (!fir.box>) -> !fir.ptr +!CHECK: %[[VAL_B_LOADED_2:.*]] = fir.load %[[VAL_B_DECLARE]]#0 : !fir.ref>> +!CHECK: %[[VAL_B_BOX_ADDR_2:.*]] = fir.box_addr %[[VAL_B_LOADED_2]] : (!fir.box>) -> !fir.ptr +!CHECK: %[[VAL_B:.*]] = fir.load %[[VAL_B_BOX_ADDR_2]] : !fir.ptr +!CHECK: omp.atomic.capture { +!CHECK: omp.atomic.update %[[VAL_A_BOX_ADDR]] : !fir.ptr { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.addi %[[ARG]], %[[VAL_B]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } +!CHECK: omp.atomic.read %[[VAL_B_BOX_ADDR]] = %[[VAL_A_BOX_ADDR]] : !fir.ptr, i32 +!CHECK: } +!CHECK: return +!CHECK: } + !$omp atomic capture + a = a + b + b = a + !$omp end atomic +end subroutine diff --git a/flang/test/Lower/OpenMP/atomic-update-hlfir.f90 b/flang/test/Lower/OpenMP/atomic-update-hlfir.f90 deleted file mode 100644 index 329009ab8ef8e..0000000000000 --- a/flang/test/Lower/OpenMP/atomic-update-hlfir.f90 +++ /dev/null @@ -1,23 +0,0 @@ -! This test checks lowering of atomic and atomic update constructs with HLFIR -! RUN: bbc -hlfir -fopenmp -emit-hlfir %s -o - | FileCheck %s -! RUN: %flang_fc1 -flang-experimental-hlfir -emit-hlfir -fopenmp %s -o - | FileCheck %s - -subroutine sb - integer :: x, y - - !$omp atomic update - x = x + y -end subroutine - -!CHECK-LABEL: @_QPsb -!CHECK: %[[X_REF:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFsbEx"} -!CHECK: %[[X_DECL:.*]]:2 = hlfir.declare %[[X_REF]] {uniq_name = "_QFsbEx"} : (!fir.ref) -> (!fir.ref, !fir.ref) -!CHECK: %[[Y_REF:.*]] = fir.alloca i32 {bindc_name = "y", uniq_name = "_QFsbEy"} -!CHECK: %[[Y_DECL:.*]]:2 = hlfir.declare %[[Y_REF]] {uniq_name = "_QFsbEy"} : (!fir.ref) -> (!fir.ref, !fir.ref) -!CHECK: %[[Y_VAL:.*]] = fir.load %[[Y_DECL]]#0 : !fir.ref -!CHECK: omp.atomic.update %[[X_DECL]]#1 : !fir.ref { -!CHECK: ^bb0(%[[ARG_X:.*]]: i32): -!CHECK: %[[X_UPDATE_VAL:.*]] = arith.addi %[[ARG_X]], %[[Y_VAL]] : i32 -!CHECK: omp.yield(%[[X_UPDATE_VAL]] : i32) -!CHECK: } -!CHECK: return diff --git a/flang/test/Lower/OpenMP/atomic-update.f90 b/flang/test/Lower/OpenMP/atomic-update.f90 new file mode 100644 index 0000000000000..e6319f70f3736 --- /dev/null +++ b/flang/test/Lower/OpenMP/atomic-update.f90 @@ -0,0 +1,149 @@ +! This test checks lowering of atomic and atomic update constructs +! RUN: bbc -fopenmp -emit-hlfir %s -o - | FileCheck %s +! RUN: %flang_fc1 -emit-hlfir -fopenmp %s -o - | FileCheck %s + +program OmpAtomicUpdate + use omp_lib +!CHECK: %[[VAL_A:.*]] = fir.alloca !fir.box> {bindc_name = "a", uniq_name = "_QFEa"} +!CHECK: %[[ZERO:.*]] = fir.zero_bits !fir.ptr +!CHECK: %[[EMBOX:.*]] = fir.embox %[[ZERO]] : (!fir.ptr) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_A]] : !fir.ref>> +!CHECK: %[[VAL_A_DECLARE:.*]]:2 = hlfir.declare %[[VAL_A]] {{.*}} +!CHECK: %[[VAL_B:.*]] = fir.alloca !fir.box> {bindc_name = "b", uniq_name = "_QFEb"} +!CHECK: %[[ZERO:.*]] = fir.zero_bits !fir.ptr +!CHECK: %[[EMBOX:.*]] = fir.embox %[[ZERO]] : (!fir.ptr) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_B]] : !fir.ref>> +!CHECK: %[[VAL_B_DECLARE:.*]]:2 = hlfir.declare %[[VAL_B]] {{.*}} +!CHECK: %[[VAL_C_ADDRESS:.*]] = fir.address_of(@_QFEc) : !fir.ref +!CHECK: %[[VAL_C_DECLARE:.*]]:2 = hlfir.declare %[[VAL_C_ADDRESS]] {{.*}} +!CHECK: %[[VAL_D_ADDRESS:.*]] = fir.address_of(@_QFEd) : !fir.ref +!CHECK: %[[VAL_D_DECLARE:.*]]:2 = hlfir.declare %[[VAL_D_ADDRESS]] {{.}} +!CHECK: %[[VAL_i1_ALLOCA:.*]] = fir.alloca i8 {bindc_name = "i1", uniq_name = "_QFEi1"} +!CHECK: %[[VAL_i1_DECLARE:.*]]:2 = hlfir.declare %[[VAL_i1_ALLOCA]] {{.*}} +!CHECK: %[[VAL_c5:.*]] = arith.constant 5 : index +!CHECK: %[[VAL_K_ALLOCA:.*]] = fir.alloca !fir.array<5xi32> {bindc_name = "k", uniq_name = "_QFEk"} +!CHECK: %[[VAL_K_SHAPED:.*]] = fir.shape %[[VAL_c5]] : (index) -> !fir.shape<1> +!CHECK: %[[VAL_K_DECLARE:.*]]:2 = hlfir.declare %[[VAL_K_ALLOCA]](%[[VAL_K_SHAPED]]) {{.*}} + +!CHECK: %[[VAL_X_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFEx"} +!CHECK: %[[VAL_X_DECLARE:.*]]:2 = hlfir.declare %[[VAL_X_ALLOCA]] {uniq_name = "_QFEx"} : (!fir.ref) -> (!fir.ref, !fir.ref) +!CHECK: %[[VAL_Y_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "y", uniq_name = "_QFEy"} +!CHECK: %[[VAL_Y_DECLARE:.*]]:2 = hlfir.declare %[[VAL_Y_ALLOCA]] {uniq_name = "_QFEy"} : (!fir.ref) -> (!fir.ref, !fir.ref) +!CHECK: %[[VAL_Z_ALLOCA:.*]] = fir.alloca i32 {bindc_name = "z", uniq_name = "_QFEz"} +!CHECK: %[[VAL_Z_DECLARE:.*]]:2 = hlfir.declare %[[VAL_Z_ALLOCA]] {uniq_name = "_QFEz"} : (!fir.ref) -> (!fir.ref, !fir.ref) + integer :: x, y, z + integer, pointer :: a, b + integer, target :: c, d + integer(1) :: i1 + integer, dimension(5) :: k + +!CHECK: %[[EMBOX:.*]] = fir.embox %[[VAL_C_DECLARE]]#1 : (!fir.ref) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_A_DECLARE]]#1 : !fir.ref>> +!CHECK: %[[EMBOX:.*]] = fir.embox %[[VAL_D_DECLARE]]#1 : (!fir.ref) -> !fir.box> +!CHECK: fir.store %[[EMBOX]] to %[[VAL_B_DECLARE]]#1 : !fir.ref>> + a=>c + b=>d + +!CHECK: %[[VAL_c3:.*]] = arith.constant 3 : index +!CHECK: %[[VAL_K_DESIGNATE:.*]] = hlfir.designate %[[VAL_K_DECLARE]]#0 (%[[VAL_c3]]) : (!fir.ref>, index) -> !fir.ref +!CHECK: %[[LOADED_Z:.*]] = fir.load %[[VAL_Z_DECLARE]]#0 : !fir.ref +!CHECK: omp.atomic.update %[[VAL_K_DESIGNATE]] : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.muli %[[LOADED_Z]], %[[ARG]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic update + k(3) = z * k(3) + +!CHECK: %[[VAL_A_LOADED:.*]] = fir.load %[[VAL_A_DECLARE]]#0 : !fir.ref>> +!CHECK: %[[VAL_A_BOX_ADDR:.*]] = fir.box_addr %[[VAL_A_LOADED]] : (!fir.box>) -> !fir.ptr +!CHECK: %[[VAL_B_LOADED:.*]] = fir.load %[[VAL_B_DECLARE]]#0 : !fir.ref>> +!CHECK: %[[VAL_B_BOX_ADDR:.*]] = fir.box_addr %[[VAL_B_LOADED]] : (!fir.box>) -> !fir.ptr +!CHECK: %[[VAL_B:.*]] = fir.load %[[VAL_B_BOX_ADDR]] : !fir.ptr +!CHECK: omp.atomic.update %[[VAL_A_BOX_ADDR]] : !fir.ptr { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.addi %[[ARG]], %[[VAL_B]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic update + a = a + b + +!CHECK: %[[VAL_c1:.*]] = arith.constant 1 : i32 +!CHECK: omp.atomic.update %[[VAL_Y_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.addi %[[ARG]], %[[VAL_c1]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic + y = y + 1 + +!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref +!CHECK: omp.atomic.update %[[VAL_Z_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.muli %[[VAL_X_LOADED]], %[[ARG]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic update + z = x * z + +!CHECK: %[[VAL_c1:.*]] = arith.constant 1 : i32 +!CHECK: omp.atomic.update memory_order(relaxed) hint(uncontended) %[[VAL_X_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.subi %[[ARG]], %[[VAL_c1]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic relaxed update hint(omp_sync_hint_uncontended) + x = x - 1 + +!CHECK: omp.atomic.update memory_order(relaxed) %[[VAL_Y_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[VAL_C_LOADED:.*]] = fir.load %[[VAL_C_DECLARE]]#0 : !fir.ref +!CHECK: %[[VAL_D_LOADED:.*]] = fir.load %[[VAL_D_DECLARE]]#0 : !fir.ref +!CHECK: {{.*}} = arith.cmpi sgt, %[[ARG]], {{.*}} : i32 +!CHECK: {{.*}} = arith.select {{.*}}, %[[ARG]], {{.*}} : i32 +!CHECK: {{.*}} = arith.cmpi sgt, {{.*}} +!CHECK: %[[TEMP:.*]] = arith.select {{.*}} : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic update relaxed + y = max(y, c, d) + +!CHECK: %[[VAL_X_LOADED:.*]] = fir.load %[[VAL_X_DECLARE]]#0 : !fir.ref +!CHECK: omp.atomic.update memory_order(relaxed) hint(contended) %[[VAL_Z_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.addi %[[ARG]], %[[VAL_X_LOADED]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic relaxed hint(omp_sync_hint_contended) + z = z + x + +!CHECK: %[[VAL_c10:.*]] = arith.constant 10 : i32 +!CHECK: omp.atomic.update memory_order(release) hint(contended) %[[VAL_Z_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.muli %[[VAL_c10]], %[[ARG]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic release update hint(omp_lock_hint_contended) + z = z * 10 + +!CHECK: %[[VAL_Z_LOADED:.*]] = fir.load %[[VAL_Z_DECLARE]]#0 : !fir.ref +!CHECK: omp.atomic.update memory_order(release) hint(speculative) %[[VAL_X_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i32): +!CHECK: %[[TEMP:.*]] = arith.divsi %[[ARG]], %[[VAL_Z_LOADED]] : i32 +!CHECK: omp.yield(%[[TEMP]] : i32) +!CHECK: } + !$omp atomic hint(omp_lock_hint_speculative) update release + x = x / z + +!CHECK: %[[VAL_c1:.*]] = arith.constant 1 : i32 +!CHECK: omp.atomic.update %[[VAL_i1_DECLARE]]#1 : !fir.ref { +!CHECK: ^bb0(%[[ARG:.*]]: i8): +!CHECK: %[[CONVERT:.*]] = fir.convert %[[ARG]] : (i8) -> i32 +!CHECK: %[[ADD:.*]] = arith.addi %[[CONVERT]], %[[VAL_c1]] : i32 +!CHECK: %[[TEMP:.*]] = fir.convert %[[ADD]] : (i32) -> i8 +!CHECK: omp.yield(%[[TEMP]] : i8) +!CHECK: } + !$omp atomic + i1 = i1 + 1 + !$omp end atomic +end program OmpAtomicUpdate