[flang-commits] [flang] 222a8a1 - [flang][hlfir] Enable WHERE scheduling in LowerHLFIROrderedAssignments
Jean Perier via flang-commits
flang-commits at lists.llvm.org
Mon May 22 03:50:17 PDT 2023
Author: Jean Perier
Date: 2023-05-22T12:49:33+02:00
New Revision: 222a8a1b2da16dde3df7a9d8bda54c7fb0730b72
URL: https://github.com/llvm/llvm-project/commit/222a8a1b2da16dde3df7a9d8bda54c7fb0730b72
DIFF: https://github.com/llvm/llvm-project/commit/222a8a1b2da16dde3df7a9d8bda54c7fb0730b72.diff
LOG: [flang][hlfir] Enable WHERE scheduling in LowerHLFIROrderedAssignments
Nothing special is needed, other than adding the logging code for where
masks and to plug the pattern. This tests mainly adds test.
Note that some of the justifications to create temps shows some lacks of
side effect interface on operations (like hlfir.transpose), or on some
transparent llvm intrinsic calls (llvm.stacksave/restore).
I think we should as much as possible try to improve this on the ops
generate code rather than special casing it here.
Differential Revision: https://reviews.llvm.org/D150581
Added:
flang/test/HLFIR/order_assignments/where-fusing-scheduling.f90
flang/test/HLFIR/order_assignments/where-scheduling.f90
Modified:
flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIROrderedAssignments.cpp
flang/lib/Optimizer/HLFIR/Transforms/ScheduleOrderedAssignments.cpp
Removed:
################################################################################
diff --git a/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIROrderedAssignments.cpp b/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIROrderedAssignments.cpp
index 38c4efdb835d8..5cea4d743841b 100644
--- a/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIROrderedAssignments.cpp
+++ b/flang/lib/Optimizer/HLFIR/Transforms/LowerHLFIROrderedAssignments.cpp
@@ -371,8 +371,9 @@ class WhereOpConversion : public mlir::OpRewritePattern<hlfir::WhereOp> {
mlir::LogicalResult
matchAndRewrite(hlfir::WhereOp whereOp,
mlir::PatternRewriter &rewriter) const override {
- TODO(whereOp.getLoc(), "WHERE construct or statement in HLFIR");
- return mlir::failure();
+ auto root = mlir::cast<hlfir::OrderedAssignmentTreeOpInterface>(
+ whereOp.getOperation());
+ return ::rewrite(root, tryFusingAssignments, rewriter);
}
const bool tryFusingAssignments;
};
diff --git a/flang/lib/Optimizer/HLFIR/Transforms/ScheduleOrderedAssignments.cpp b/flang/lib/Optimizer/HLFIR/Transforms/ScheduleOrderedAssignments.cpp
index d8c909d8387b0..d9f0bc1abab4e 100644
--- a/flang/lib/Optimizer/HLFIR/Transforms/ScheduleOrderedAssignments.cpp
+++ b/flang/lib/Optimizer/HLFIR/Transforms/ScheduleOrderedAssignments.cpp
@@ -568,6 +568,12 @@ static llvm::raw_ostream &printRegionId(llvm::raw_ostream &os,
os << "rhs";
else if (&assign.getLhsRegion() == &yieldRegion)
os << "lhs";
+ } else if (auto where = mlir::dyn_cast<hlfir::WhereOp>(parent)) {
+ if (&where.getMaskRegion() == &yieldRegion)
+ os << "mask";
+ } else if (auto elseWhereOp = mlir::dyn_cast<hlfir::ElseWhereOp>(parent)) {
+ if (&elseWhereOp.getMaskRegion() == &yieldRegion)
+ os << "mask";
} else {
os << "unknown";
}
diff --git a/flang/test/HLFIR/order_assignments/where-fusing-scheduling.f90 b/flang/test/HLFIR/order_assignments/where-fusing-scheduling.f90
new file mode 100644
index 0000000000000..08d4092b49aef
--- /dev/null
+++ b/flang/test/HLFIR/order_assignments/where-fusing-scheduling.f90
@@ -0,0 +1,41 @@
+! Test scheduling of WHERE in lower-hlfir-ordered-assignments pass
+! when fusing is enabled or disabled.
+
+!RUN: bbc -hlfir -o - -pass-pipeline="builtin.module(lower-hlfir-ordered-assignments{fuse-assignments=false})" --debug-only=flang-ordered-assignment -flang-dbg-order-assignment-schedule-only %s 2>&1 | FileCheck %s --check-prefix NOFUSE
+
+!RUN: bbc -hlfir -o - -pass-pipeline="builtin.module(lower-hlfir-ordered-assignments{fuse-assignments=true})" --debug-only=flang-ordered-assignment -flang-dbg-order-assignment-schedule-only %s 2>&1 | FileCheck %s --check-prefix FUSE
+
+!REQUIRES: asserts
+
+subroutine fusable(x, y, mask)
+ real :: x(:), y(:)
+ logical :: mask(:)
+ where (mask)
+ x = 41.
+ y = 42.
+ end where
+end subroutine
+
+subroutine unfusable(x, y, mask)
+ real :: x(:), y(:)
+ logical :: mask(:)
+ where (mask)
+ x(1:10) = y
+ y = x(10:1:-1)
+ end where
+end subroutine
+
+!NOFUSE-LABEL: ------------ scheduling where in _QPfusable ------------
+!NOFUSE-NEXT: run 1 evaluate: where/region_assign1
+!NOFUSE-NEXT: run 2 evaluate: where/region_assign2
+!NOFUSE-LABEL: ------------ scheduling where in _QPunfusable ------------
+!NOFUSE-NEXT: run 1 evaluate: where/region_assign1
+!NOFUSE-NEXT: run 2 evaluate: where/region_assign2
+
+!FUSE-LABEL: ------------ scheduling where in _QPfusable ------------
+!FUSE-NEXT: run 1 evaluate: where/region_assign1
+!FUSE-NEXT: run 1 evaluate: where/region_assign2
+!FUSE-LABEL: ------------ scheduling where in _QPunfusable ------------
+!FUSE-NEXT: run 1 evaluate: where/region_assign1
+!FUSE-NEXT: conflict: R/W: <block argument> of type '!fir.box<!fir.array<?xf32>>' at index: 1 W:<block argument> of type '!fir.box<!fir.array<?xf32>>' at index: 1
+!FUSE-NEXT: run 2 evaluate: where/region_assign2
diff --git a/flang/test/HLFIR/order_assignments/where-scheduling.f90 b/flang/test/HLFIR/order_assignments/where-scheduling.f90
new file mode 100644
index 0000000000000..89a3cadbd6971
--- /dev/null
+++ b/flang/test/HLFIR/order_assignments/where-scheduling.f90
@@ -0,0 +1,128 @@
+! Test scheduling of WHERE in lower-hlfir-ordered-assignments pass.
+
+! RUN: bbc -hlfir -o - -pass-pipeline="builtin.module(lower-hlfir-ordered-assignments)" --debug-only=flang-ordered-assignment -flang-dbg-order-assignment-schedule-only %s 2>&1 | FileCheck %s
+! REQUIRES: asserts
+
+subroutine no_conflict(x, y)
+ real :: x(:), y(:)
+ where (y.gt.0) x = y
+end subroutine
+
+subroutine fake_conflict(x, y)
+ ! The conflict here could be avoided because the read and write are
+ ! aligned, so there would not be any read after write at the element
+ ! level, but this will require a bit more work to detect this (like
+ ! comparing the hlfir.designate operations).
+ real :: x(:), y(:)
+ where (x.gt.y) x = y
+end subroutine
+
+subroutine only_once(x, y, z)
+ interface
+ impure function call_me_only_once()
+ logical :: call_me_only_once(10)
+ end function
+ end interface
+ real :: x(:), y(:), z(:)
+ where (call_me_only_once())
+ x = y
+ z = y
+ end where
+end subroutine
+
+subroutine rhs_lhs_conflict(x, y)
+ real :: x(:, :), y(:, :)
+ where (y.gt.0.) x = transpose(x)
+end subroutine
+
+subroutine where_construct_no_conflict(x, y, z, mask1, mask2)
+ real :: x(:), y(:), z(:)
+ logical :: mask1(:), mask2(:)
+ where (mask1)
+ x = y
+ elsewhere (mask2)
+ z = y
+ end where
+end subroutine
+
+subroutine where_construct_conflict(x, y)
+ real :: x(:, :), y(:, :)
+ where (y.gt.0.)
+ x = y
+ elsewhere (x.gt.0)
+ y = x
+ end where
+end subroutine
+
+subroutine where_construct_conflict_2(x, y)
+ real :: x(:, :), y(:, :)
+ where (x.gt.0.)
+ x = y
+ elsewhere (y.gt.0)
+ y = x
+ end where
+end subroutine
+
+subroutine where_vector_subscript_conflict_1(x, vec1)
+ real :: x(10)
+ integer :: vec1(10)
+ where (x(vec1).lt.0.) x = 42.
+end subroutine
+
+subroutine where_vector_subscript_conflict_2(x, vec1)
+ integer :: x(10)
+ real :: y(10)
+ where (y(x).lt.0.) x = 0
+end subroutine
+
+subroutine where_in_forall_conflict(x)
+ real :: x(:, :)
+ forall (i = 1:10)
+ where (x(i, :).gt.0) x(:, i) = x(i, :)
+ end forall
+end subroutine
+
+!CHECK-LABEL: ------------ scheduling where in _QPno_conflict ------------
+!CHECK-NEXT: run 1 evaluate: where/region_assign1
+!CHECK-LABEL: ------------ scheduling where in _QPfake_conflict ------------
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.box<!fir.array<?xf32>>' at index: 0 W:<block argument> of type '!fir.box<!fir.array<?xf32>>' at index: 0
+!CHECK-NEXT: run 1 save : where/mask
+!CHECK-NEXT: run 2 evaluate: where/region_assign1
+!CHECK-LABEL: ------------ scheduling where in _QPonly_once ------------
+!CHECK-NEXT: unknown effect: %9 = fir.call @llvm.stacksave() fastmath<contract> : () -> !fir.ref<i8>
+!CHECK-NEXT: run 1 save (w): where/mask
+!CHECK-NEXT: run 2 evaluate: where/region_assign1
+!CHECK-NEXT: run 3 evaluate: where/region_assign2
+!CHECK-LABEL: ------------ scheduling where in _QPrhs_lhs_conflict ------------
+!CHECK-NEXT: unknown effect: %2 = hlfir.transpose %0#0 : (!fir.box<!fir.array<?x?xf32>>) -> !hlfir.expr<?x?xf32>
+!CHECK-NEXT: run 1 save (w): where/region_assign1/rhs
+!CHECK-NEXT: run 2 evaluate: where/region_assign1
+!CHECK-LABEL: ------------ scheduling where in _QPwhere_construct_no_conflict ------------
+!CHECK-NEXT: run 1 evaluate: where/region_assign1
+!CHECK-NEXT: run 2 evaluate: where/elsewhere1/region_assign1
+!CHECK-LABEL: ------------ scheduling where in _QPwhere_construct_conflict ------------
+!CHECK-NEXT: run 1 evaluate: where/region_assign1
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 1 W:<block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 1
+!CHECK-NEXT: run 2 save : where/mask
+!CHECK-NEXT: run 3 evaluate: where/elsewhere1/region_assign1
+!CHECK-LABEL: ------------ scheduling where in _QPwhere_construct_conflict_2 ------------
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 0 W:<block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 0
+!CHECK-NEXT: run 1 save : where/mask
+!CHECK-NEXT: run 2 evaluate: where/region_assign1
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 1 W:<block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 1
+!CHECK-NEXT: run 3 save : where/elsewhere1/mask
+!CHECK-NEXT: run 4 evaluate: where/elsewhere1/region_assign1
+!CHECK-LABEL: ------------ scheduling where in _QPwhere_vector_subscript_conflict_1 ------------
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.ref<!fir.array<10xf32>>' at index: 0 W:<block argument> of type '!fir.ref<!fir.array<10xf32>>' at index: 0
+!CHECK-NEXT: run 1 save : where/mask
+!CHECK-NEXT: run 2 evaluate: where/region_assign1
+!CHECK-LABEL: ------------ scheduling where in _QPwhere_vector_subscript_conflict_2 ------------
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.ref<!fir.array<10xi32>>' at index: 0 W:<block argument> of type '!fir.ref<!fir.array<10xi32>>' at index: 0
+!CHECK-NEXT: run 1 save : where/mask
+!CHECK-NEXT: run 2 evaluate: where/region_assign1
+!CHECK-LABEL: ------------ scheduling forall in _QPwhere_in_forall_conflict ------------
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 0 W:<block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 0
+!CHECK-NEXT: run 1 save : forall/where1/mask
+!CHECK-NEXT: conflict: R/W: <block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 0 W:<block argument> of type '!fir.box<!fir.array<?x?xf32>>' at index: 0
+!CHECK-NEXT: run 1 save : forall/where1/region_assign1/rhs
+!CHECK-NEXT: run 2 evaluate: forall/where1/region_assign1
More information about the flang-commits
mailing list