Skip to content

Commit 40278bb

Browse files
vzakharivzakhari
authored
[mlir][acc] Added async to data clause operations. (#97307)
As long as the data clause operations are not tightly "associated" with the compute/data operations (e.g. they can be optimized as SSA producers and made block arguments), the information about the original async() clause should be attached to the data clause operations to make it easier to generate proper runtime actions for them. This change propagates the async() information from the OpenACC data/compute constructs to the data clause operations. This change also adds the CurrentDeviceIdResource to guarantee proper ordering of the operations that read and write the current device identifier.
1 parent e89890e commit 40278bb

File tree

10 files changed

+580
-192
lines changed

10 files changed

+580
-192
lines changed

‎flang/lib/Lower/OpenACC.cpp

Lines changed: 276 additions & 135 deletions
Large diffs are not rendered by default.

‎flang/test/Lower/OpenACC/acc-data.f90

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -155,11 +155,13 @@ subroutine acc_data
155155
! CHECK: acc.data dataOperands(%{{.*}}) {
156156
! CHECK: } attributes {asyncOnly = [#acc.device_type<none>]}
157157

158-
!$acc data present(a) async(1)
158+
!$acc data copy(a) async(1)
159159
!$acc end data
160160

161-
! CHECK: acc.data async(%{{.*}} : i32) dataOperands(%{{.*}}) {
161+
! CHECK: %[[COPYIN:.*]] = acc.copyin varPtr(%{{.*}} : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async(%[[ASYNC:.*]] : i32) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_copy>, name = "a"}
162+
! CHECK: acc.data async(%[[ASYNC]] : i32) dataOperands(%[[COPYIN]] : !fir.ref<!fir.array<10x10xf32>>) {
162163
! CHECK: }{{$}}
164+
! CHECK: acc.copyout accPtr(%[[COPYIN]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async(%[[ASYNC]] : i32) to varPtr(%{{.*}} : !fir.ref<!fir.array<10x10xf32>>) {dataClause = #acc<data_clause acc_copy>, name = "a"}
163165

164166
!$acc data present(a) wait
165167
!$acc end data

‎flang/test/Lower/OpenACC/acc-enter-data.f90

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -94,7 +94,7 @@ subroutine acc_enter_data
9494
!$acc enter data create(a) async
9595
!CHECK: %[[BOUND0:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
9696
!CHECK: %[[BOUND1:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[EXTENT_C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
97-
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) -> !fir.ref<!fir.array<10x10xf32>> {name = "a", structured = false}
97+
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) -> !fir.ref<!fir.array<10x10xf32>> {asyncOnly = [#acc.device_type<none>], name = "a", structured = false}
9898
!CHECK: acc.enter_data dataOperands(%[[CREATE_A]] : !fir.ref<!fir.array<10x10xf32>>) attributes {async}
9999

100100
!$acc enter data create(a) wait
@@ -106,22 +106,22 @@ subroutine acc_enter_data
106106
!$acc enter data create(a) async wait
107107
!CHECK: %[[BOUND0:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
108108
!CHECK: %[[BOUND1:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[EXTENT_C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
109-
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) -> !fir.ref<!fir.array<10x10xf32>> {name = "a", structured = false}
109+
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) -> !fir.ref<!fir.array<10x10xf32>> {asyncOnly = [#acc.device_type<none>], name = "a", structured = false}
110110
!CHECK: acc.enter_data dataOperands(%[[CREATE_A]] : !fir.ref<!fir.array<10x10xf32>>) attributes {async, wait}
111111

112112
!$acc enter data create(a) async(1)
113+
!CHECK: %[[ASYNC1:.*]] = arith.constant 1 : i32
113114
!CHECK: %[[BOUND0:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
114115
!CHECK: %[[BOUND1:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[EXTENT_C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
115-
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) -> !fir.ref<!fir.array<10x10xf32>> {name = "a", structured = false}
116-
!CHECK: %[[ASYNC1:.*]] = arith.constant 1 : i32
116+
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) async(%[[ASYNC1]] : i32) -> !fir.ref<!fir.array<10x10xf32>> {name = "a", structured = false}
117117
!CHECK: acc.enter_data async(%[[ASYNC1]] : i32) dataOperands(%[[CREATE_A]] : !fir.ref<!fir.array<10x10xf32>>)
118118

119119
!$acc enter data create(a) async(async)
120+
!CHECK: %[[ASYNC2:.*]] = fir.load %{{.*}} : !fir.ref<i32>
120121
!CHECK: %[[BOUND0:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
121122
!CHECK: %[[BOUND1:.*]] = acc.bounds lowerbound(%{{.*}} : index) upperbound(%{{.*}} : index) extent(%[[EXTENT_C10]] : index) stride(%c1{{.*}} : index) startIdx(%{{.*}} : index)
122123

123-
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) -> !fir.ref<!fir.array<10x10xf32>> {name = "a", structured = false}
124-
!CHECK: %[[ASYNC2:.*]] = fir.load %{{.*}} : !fir.ref<i32>
124+
!CHECK: %[[CREATE_A:.*]] = acc.create varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%[[BOUND0]], %[[BOUND1]]) async(%[[ASYNC2]] : i32) -> !fir.ref<!fir.array<10x10xf32>> {name = "a", structured = false}
125125
!CHECK: acc.enter_data async(%[[ASYNC2]] : i32) dataOperands(%[[CREATE_A]] : !fir.ref<!fir.array<10x10xf32>>)
126126

127127
!$acc enter data create(a) wait(1)

‎flang/test/Lower/OpenACC/acc-exit-data.f90

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -56,32 +56,32 @@ subroutine acc_exit_data
5656
!CHECK: acc.detach accPtr(%[[DEVPTR_D]] : !fir.ptr<f32>) {name = "d", structured = false}
5757

5858
!$acc exit data delete(a) async
59-
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
59+
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {asyncOnly = [#acc.device_type<none>], dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
6060
!CHECK: acc.exit_data dataOperands(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) attributes {async}
61-
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) {name = "a", structured = false}
61+
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) {asyncOnly = [#acc.device_type<none>], name = "a", structured = false}
6262

6363
!$acc exit data delete(a) wait
6464
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
6565
!CHECK: acc.exit_data dataOperands(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) attributes {wait}
6666
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) {name = "a", structured = false}
6767

6868
!$acc exit data delete(a) async wait
69-
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
69+
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {asyncOnly = [#acc.device_type<none>], dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
7070
!CHECK: acc.exit_data dataOperands(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) attributes {async, wait}
71-
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) {name = "a", structured = false}
71+
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) {asyncOnly = [#acc.device_type<none>], name = "a", structured = false}
7272

7373
!$acc exit data delete(a) async(1)
74-
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
7574
!CHECK: %[[ASYNC1:.*]] = arith.constant 1 : i32
75+
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async(%[[ASYNC1]] : i32) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
7676
!CHECK: acc.exit_data async(%[[ASYNC1]] : i32) dataOperands(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>)
77-
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) {name = "a", structured = false}
77+
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async(%[[ASYNC1]] : i32) {name = "a", structured = false}
7878

7979

8080
!$acc exit data delete(a) async(async)
81-
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
8281
!CHECK: %[[ASYNC2:.*]] = fir.load %{{.*}} : !fir.ref<i32>
82+
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async(%[[ASYNC2]] : i32) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}
8383
!CHECK: acc.exit_data async(%[[ASYNC2]] : i32) dataOperands(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>)
84-
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) {name = "a", structured = false}
84+
!CHECK: acc.delete accPtr(%[[DEVPTR]] : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async(%[[ASYNC2]] : i32) {name = "a", structured = false}
8585

8686
!$acc exit data delete(a) wait(1)
8787
!CHECK: %[[DEVPTR:.*]] = acc.getdeviceptr varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {dataClause = #acc<data_clause acc_delete>, name = "a", structured = false}

‎flang/test/Lower/OpenACC/acc-parallel.f90

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -60,7 +60,7 @@ subroutine acc_parallel
6060
!$acc parallel async
6161
!$acc end parallel
6262

63-
! CHECK: acc.parallel {
63+
! CHECK: acc.parallel {
6464
! CHECK: acc.yield
6565
! CHECK-NEXT: } attributes {asyncOnly = [#acc.device_type<none>]}
6666

@@ -76,7 +76,7 @@ subroutine acc_parallel
7676
!$acc end parallel
7777

7878
! CHECK: [[ASYNC2:%.*]] = fir.load %{{.*}} : !fir.ref<i32>
79-
! CHECK: acc.parallel async([[ASYNC2]] : i32) {
79+
! CHECK-NEXT: acc.parallel async([[ASYNC2]] : i32) {
8080
! CHECK: acc.yield
8181
! CHECK-NEXT: }{{$}}
8282

@@ -324,13 +324,13 @@ subroutine acc_parallel
324324
! CHECK: acc.detach accPtr(%[[ATTACH_D]] : !fir.ptr<f32>) {dataClause = #acc<data_clause acc_attach>, name = "d"}
325325
! CHECK: acc.detach accPtr(%[[ATTACH_E]] : !fir.ptr<f32>) {dataClause = #acc<data_clause acc_attach>, name = "e"}
326326

327-
!$acc parallel private(a) firstprivate(b) private(c)
327+
!$acc parallel private(a) firstprivate(b) private(c) async(1)
328328
!$acc end parallel
329329

330-
! CHECK: %[[ACC_PRIVATE_A:.*]] = acc.private varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {name = "a"}
331-
! CHECK: %[[ACC_FPRIVATE_B:.*]] = acc.firstprivate varPtr(%[[DECLB]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {name = "b"}
332-
! CHECK: %[[ACC_PRIVATE_C:.*]] = acc.private varPtr(%[[DECLC]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<10x10xf32>> {name = "c"}
333-
! CHECK: acc.parallel firstprivate(@firstprivatization_section_ext10xext10_ref_10x10xf32 -> %[[ACC_FPRIVATE_B]] : !fir.ref<!fir.array<10x10xf32>>) private(@privatization_ref_10x10xf32 -> %[[ACC_PRIVATE_A]] : !fir.ref<!fir.array<10x10xf32>>, @privatization_ref_10x10xf32 -> %[[ACC_PRIVATE_C]] : !fir.ref<!fir.array<10x10xf32>>) {
330+
! CHECK: %[[ACC_PRIVATE_A:.*]] = acc.private varPtr(%[[DECLA]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async([[ASYNC3:%.*]]) -> !fir.ref<!fir.array<10x10xf32>> {name = "a"}
331+
! CHECK: %[[ACC_FPRIVATE_B:.*]] = acc.firstprivate varPtr(%[[DECLB]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async([[ASYNC3]]) -> !fir.ref<!fir.array<10x10xf32>> {name = "b"}
332+
! CHECK: %[[ACC_PRIVATE_C:.*]] = acc.private varPtr(%[[DECLC]]#0 : !fir.ref<!fir.array<10x10xf32>>) bounds(%{{.*}}, %{{.*}}) async([[ASYNC3]]) -> !fir.ref<!fir.array<10x10xf32>> {name = "c"}
333+
! CHECK: acc.parallel async([[ASYNC3]]) firstprivate(@firstprivatization_section_ext10xext10_ref_10x10xf32 -> %[[ACC_FPRIVATE_B]] : !fir.ref<!fir.array<10x10xf32>>) private(@privatization_ref_10x10xf32 -> %[[ACC_PRIVATE_A]] : !fir.ref<!fir.array<10x10xf32>>, @privatization_ref_10x10xf32 -> %[[ACC_PRIVATE_C]] : !fir.ref<!fir.array<10x10xf32>>) {
334334
! CHECK: acc.yield
335335
! CHECK-NEXT: }{{$}}
336336

‎flang/test/Lower/OpenACC/acc-serial.f90

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -60,7 +60,7 @@ subroutine acc_serial
6060
!$acc serial async
6161
!$acc end serial
6262

63-
! CHECK: acc.serial {
63+
! CHECK: acc.serial {
6464
! CHECK: acc.yield
6565
! CHECK-NEXT: } attributes {asyncOnly = [#acc.device_type<none>]}
6666

@@ -76,7 +76,7 @@ subroutine acc_serial
7676
!$acc end serial
7777

7878
! CHECK: [[ASYNC2:%.*]] = fir.load %{{.*}} : !fir.ref<i32>
79-
! CHECK: acc.serial async([[ASYNC2]] : i32) {
79+
! CHECK-NEXT: acc.serial async([[ASYNC2]] : i32) {
8080
! CHECK: acc.yield
8181
! CHECK-NEXT: }{{$}}
8282

0 commit comments

Comments
 (0)