[flang] Postpone hlfir.end_associate generation for calls. #138786
+126
−11
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
If we generate hlfir.end_associate at the end of the statement,
we get easier optimizable HLFIR, because there are no compiler
generated operations with side-effects in between the call
and the consumers. This allows more hlfir.eval_in_mem to reuse
the LHS instead of allocating temporary buffer.
I do not think the same can be done for hlfir.copy_out always, e.g.:
```
subroutine test2(x)
interface
function array_func2(x,y)
real:: x(*), array_func2(10), y
end function array_func2
end interface
real :: x(:)
x = array_func2(x, 1.0)
end subroutine test2
```
If we postpone the copy-out until after the assignment, then
the result may be wrong.
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Slava Zakharin (vzakhari) ChangesIf we generate hlfir.end_associate at the end of the statement, I do not think the same can be done for hlfir.copy_out always, e.g.: If we postpone the copy-out until after the assignment, then Full diff: https://github.com/llvm/llvm-project/pull/138786.diff 2 Files Affected:
diff --git a/flang/lib/Lower/ConvertCall.cpp b/flang/lib/Lower/ConvertCall.cpp
index a5b85e25b1af0..d37d51f6ec634 100644
--- a/flang/lib/Lower/ConvertCall.cpp
+++ b/flang/lib/Lower/ConvertCall.cpp
@@ -960,9 +960,26 @@ struct CallCleanUp {
mlir::Value tempVar;
mlir::Value mustFree;
};
- void genCleanUp(mlir::Location loc, fir::FirOpBuilder &builder) {
- Fortran::common::visit([&](auto &c) { c.genCleanUp(loc, builder); },
+
+ /// Generate clean-up code.
+ /// If \p postponeAssociates is true, the ExprAssociate clean-up
+ /// is not generated, and instead the corresponding CallCleanUp
+ /// object is returned as the result.
+ std::optional<CallCleanUp> genCleanUp(mlir::Location loc,
+ fir::FirOpBuilder &builder,
+ bool postponeAssociates) {
+ std::optional<CallCleanUp> postponed;
+ Fortran::common::visit(Fortran::common::visitors{
+ [&](CopyIn &c) { c.genCleanUp(loc, builder); },
+ [&](ExprAssociate &c) {
+ if (postponeAssociates)
+ postponed = CallCleanUp{c};
+ else
+ c.genCleanUp(loc, builder);
+ },
+ },
cleanUp);
+ return postponed;
}
std::variant<CopyIn, ExprAssociate> cleanUp;
};
@@ -1729,10 +1746,23 @@ genUserCall(Fortran::lower::PreparedActualArguments &loweredActuals,
caller, callSiteType, callContext.resultType,
callContext.isElementalProcWithArrayArgs());
- /// Clean-up associations and copy-in.
- for (auto cleanUp : callCleanUps)
- cleanUp.genCleanUp(loc, builder);
-
+ // Clean-up associations and copy-in.
+ // The association clean-ups are postponed to the end of the statement
+ // lowering. The copy-in clean-ups may be delayed as well,
+ // but they are done immediately after the call currently.
+ llvm::SmallVector<CallCleanUp> associateCleanups;
+ for (auto cleanUp : callCleanUps) {
+ auto postponed =
+ cleanUp.genCleanUp(loc, builder, /*postponeAssociates=*/true);
+ if (postponed)
+ associateCleanups.push_back(*postponed);
+ }
+
+ fir::FirOpBuilder *bldr = &builder;
+ callContext.stmtCtx.attachCleanup([=]() {
+ for (auto cleanUp : associateCleanups)
+ (void)cleanUp.genCleanUp(loc, *bldr, /*postponeAssociates=*/false);
+ });
if (auto *entity = std::get_if<hlfir::EntityWithAttributes>(&loweredResult))
return *entity;
diff --git a/flang/test/Lower/HLFIR/call-postponed-associate.f90 b/flang/test/Lower/HLFIR/call-postponed-associate.f90
new file mode 100644
index 0000000000000..18df62b44324b
--- /dev/null
+++ b/flang/test/Lower/HLFIR/call-postponed-associate.f90
@@ -0,0 +1,85 @@
+! RUN: bbc -emit-hlfir -o - %s -I nowhere | FileCheck %s
+
+subroutine test1
+ interface
+ function array_func1(x)
+ real:: x, array_func1(10)
+ end function array_func1
+ end interface
+ real :: x(10)
+ x = array_func1(1.0)
+end subroutine test1
+! CHECK-LABEL: func.func @_QPtest1() {
+! CHECK: %[[VAL_5:.*]] = arith.constant 1.000000e+00 : f32
+! CHECK: %[[VAL_6:.*]]:3 = hlfir.associate %[[VAL_5]] {adapt.valuebyref} : (f32) -> (!fir.ref<f32>, !fir.ref<f32>, i1)
+! CHECK: %[[VAL_17:.*]] = hlfir.eval_in_mem shape %{{.*}} : (!fir.shape<1>) -> !hlfir.expr<10xf32> {
+! CHECK: fir.call @_QParray_func1
+! CHECK: fir.save_result
+! CHECK: }
+! CHECK: hlfir.assign %[[VAL_17]] to %{{.*}} : !hlfir.expr<10xf32>, !fir.ref<!fir.array<10xf32>>
+! CHECK: hlfir.end_associate %[[VAL_6]]#1, %[[VAL_6]]#2 : !fir.ref<f32>, i1
+
+subroutine test2(x)
+ interface
+ function array_func2(x,y)
+ real:: x(*), array_func2(10), y
+ end function array_func2
+ end interface
+ real :: x(:)
+ x = array_func2(x, 1.0)
+end subroutine test2
+! CHECK-LABEL: func.func @_QPtest2(
+! CHECK: %[[VAL_3:.*]] = arith.constant 1.000000e+00 : f32
+! CHECK: %[[VAL_4:.*]]:2 = hlfir.copy_in %{{.*}} to %{{.*}} : (!fir.box<!fir.array<?xf32>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> (!fir.box<!fir.array<?xf32>>, i1)
+! CHECK: %[[VAL_5:.*]] = fir.box_addr %[[VAL_4]]#0 : (!fir.box<!fir.array<?xf32>>) -> !fir.ref<!fir.array<?xf32>>
+! CHECK: %[[VAL_6:.*]]:3 = hlfir.associate %[[VAL_3]] {adapt.valuebyref} : (f32) -> (!fir.ref<f32>, !fir.ref<f32>, i1)
+! CHECK: %[[VAL_17:.*]] = hlfir.eval_in_mem shape %{{.*}} : (!fir.shape<1>) -> !hlfir.expr<10xf32> {
+! CHECK: ^bb0(%[[VAL_18:.*]]: !fir.ref<!fir.array<10xf32>>):
+! CHECK: %[[VAL_19:.*]] = fir.call @_QParray_func2(%[[VAL_5]], %[[VAL_6]]#0) fastmath<contract> : (!fir.ref<!fir.array<?xf32>>, !fir.ref<f32>) -> !fir.array<10xf32>
+! CHECK: fir.save_result %[[VAL_19]] to %[[VAL_18]](%{{.*}}) : !fir.array<10xf32>, !fir.ref<!fir.array<10xf32>>, !fir.shape<1>
+! CHECK: }
+! CHECK: hlfir.copy_out %{{.*}}, %[[VAL_4]]#1 to %{{.*}} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>, i1, !fir.box<!fir.array<?xf32>>) -> ()
+! CHECK: hlfir.assign %[[VAL_17]] to %{{.*}} : !hlfir.expr<10xf32>, !fir.box<!fir.array<?xf32>>
+! CHECK: hlfir.end_associate %[[VAL_6]]#1, %[[VAL_6]]#2 : !fir.ref<f32>, i1
+! CHECK: hlfir.destroy %[[VAL_17]] : !hlfir.expr<10xf32>
+
+subroutine test3(x)
+ interface
+ function array_func3(x)
+ real :: x, array_func3(10)
+ end function array_func3
+ end interface
+ logical :: x
+ if (any(array_func3(1.0).le.array_func3(2.0))) x = .true.
+end subroutine test3
+! CHECK-LABEL: func.func @_QPtest3(
+! CHECK: %[[VAL_2:.*]] = arith.constant 1.000000e+00 : f32
+! CHECK: %[[VAL_3:.*]]:3 = hlfir.associate %[[VAL_2]] {adapt.valuebyref} : (f32) -> (!fir.ref<f32>, !fir.ref<f32>, i1)
+! CHECK: %[[VAL_14:.*]] = hlfir.eval_in_mem shape %{{.*}} : (!fir.shape<1>) -> !hlfir.expr<10xf32> {
+! CHECK: ^bb0(%[[VAL_15:.*]]: !fir.ref<!fir.array<10xf32>>):
+! CHECK: %[[VAL_16:.*]] = fir.call @_QParray_func3(%[[VAL_3]]#0) fastmath<contract> : (!fir.ref<f32>) -> !fir.array<10xf32>
+! CHECK: fir.save_result %[[VAL_16]] to %[[VAL_15]](%{{.*}}) : !fir.array<10xf32>, !fir.ref<!fir.array<10xf32>>, !fir.shape<1>
+! CHECK: }
+! CHECK: %[[VAL_17:.*]] = arith.constant 2.000000e+00 : f32
+! CHECK: %[[VAL_18:.*]]:3 = hlfir.associate %[[VAL_17]] {adapt.valuebyref} : (f32) -> (!fir.ref<f32>, !fir.ref<f32>, i1)
+! CHECK: %[[VAL_29:.*]] = hlfir.eval_in_mem shape %{{.*}} : (!fir.shape<1>) -> !hlfir.expr<10xf32> {
+! CHECK: ^bb0(%[[VAL_30:.*]]: !fir.ref<!fir.array<10xf32>>):
+! CHECK: %[[VAL_31:.*]] = fir.call @_QParray_func3(%[[VAL_18]]#0) fastmath<contract> : (!fir.ref<f32>) -> !fir.array<10xf32>
+! CHECK: fir.save_result %[[VAL_31]] to %[[VAL_30]](%{{.*}}) : !fir.array<10xf32>, !fir.ref<!fir.array<10xf32>>, !fir.shape<1>
+! CHECK: }
+! CHECK: %[[VAL_32:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<?x!fir.logical<4>> {
+! CHECK: ^bb0(%[[VAL_33:.*]]: index):
+! CHECK: %[[VAL_34:.*]] = hlfir.apply %[[VAL_14]], %[[VAL_33]] : (!hlfir.expr<10xf32>, index) -> f32
+! CHECK: %[[VAL_35:.*]] = hlfir.apply %[[VAL_29]], %[[VAL_33]] : (!hlfir.expr<10xf32>, index) -> f32
+! CHECK: %[[VAL_36:.*]] = arith.cmpf ole, %[[VAL_34]], %[[VAL_35]] fastmath<contract> : f32
+! CHECK: %[[VAL_37:.*]] = fir.convert %[[VAL_36]] : (i1) -> !fir.logical<4>
+! CHECK: hlfir.yield_element %[[VAL_37]] : !fir.logical<4>
+! CHECK: }
+! CHECK: %[[VAL_38:.*]] = hlfir.any %[[VAL_32]] : (!hlfir.expr<?x!fir.logical<4>>) -> !fir.logical<4>
+! CHECK: hlfir.destroy %[[VAL_32]] : !hlfir.expr<?x!fir.logical<4>>
+! CHECK: hlfir.end_associate %[[VAL_18]]#1, %[[VAL_18]]#2 : !fir.ref<f32>, i1
+! CHECK: hlfir.destroy %[[VAL_29]] : !hlfir.expr<10xf32>
+! CHECK: hlfir.end_associate %[[VAL_3]]#1, %[[VAL_3]]#2 : !fir.ref<f32>, i1
+! CHECK: hlfir.destroy %[[VAL_14]] : !hlfir.expr<10xf32>
+! CHECK: %[[VAL_39:.*]] = fir.convert %[[VAL_38]] : (!fir.logical<4>) -> i1
+! CHECK: fir.if %[[VAL_39]] {
|
jeanPerier
approved these changes
May 7, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
If we generate hlfir.end_associate at the end of the statement,
we get easier optimizable HLFIR, because there are no compiler
generated operations with side-effects in between the call
and the consumers. This allows more hlfir.eval_in_mem to reuse
the LHS instead of allocating temporary buffer.
I do not think the same can be done for hlfir.copy_out always, e.g.:
If we postpone the copy-out until after the assignment, then
the result may be wrong.