Skip to content

[LoopInterchange] Forbid interchange when the dependency is confused #78533

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 2 commits into from
Closed

[LoopInterchange] Forbid interchange when the dependency is confused #78533

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7f25fae
Add pr47259.ll
ShivaChen Jan 11, 2024
7734472
[LoopInterchange] Forbid interchange when the dependency is confused
ShivaChen Jan 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions llvm/lib/Transforms/Scalar/LoopInterchange.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,6 +123,13 @@ static bool populateDependencyMatrix(CharMatrix &DepMatrix, unsigned Level,
// Track Output, Flow, and Anti dependencies.
if (auto D = DI->depends(Src, Dst, true)) {
assert(D->isOrdered() && "Expected an output, flow or anti dep.");

if (D->isConfused()) {
LLVM_DEBUG(dbgs() << " Confused dependency between:\n"
<< " " << *Src << "\n"
<< " " << *Dst << "\n");
return false;
}
// If the direction vector is negative, normalize it to
// make it non-negative.
if (D->normalize(SE))
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/Transforms/LoopInterchange/lcssa.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -180,7 +180,7 @@ for.end16: ; preds = %for.exit
; REMARK: Interchanged
; REMARK-NEXT: lcssa_05

define void @lcssa_05(ptr %ptr) {
define void @lcssa_05(ptr noalias %ptr) {
entry:
br label %outer.header

Expand Down Expand Up @@ -225,7 +225,7 @@ for.end16: ; preds = %for.exit
; REMARK: UnsupportedExitPHI
; REMARK-NEXT: lcssa_06

define void @lcssa_06(ptr %ptr, ptr %ptr1) {
define void @lcssa_06(ptr %ptr, ptr noalias %ptr1) {
Copy link
Contributor

@artagnon artagnon Sep 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are all these other noalias changes required as part of the patch? If so, I can recommend marking both arguments noalias, and mentioning why in the commit message (eg. if these tests were faulty to begin with).

entry:
br label %outer.header

Expand Down
2 changes: 1 addition & 1 deletion llvm/test/Transforms/LoopInterchange/pr43326-ideal-access-pattern.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@
; REMARKS-NEXT: Name: Interchanged
; REMARKS-NEXT: Function: pr43326-triply-nested

define void @pr43326-triply-nested(ptr %e, ptr %f) {
define void @pr43326-triply-nested(ptr noalias %e, ptr %f) {
entry:
br label %for.outermost.header

Expand Down
55 changes: 55 additions & 0 deletions llvm/test/Transforms/LoopInterchange/pr47259.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
; REQUIRES: asserts
; RUN: opt < %s -passes=loop-interchange -cache-line-size=64 -verify-dom-info -verify-loop-info \
; RUN: -S -debug 2>&1 | FileCheck %s
Comment on lines +2 to +3
Copy link
Contributor

@artagnon artagnon Sep 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

-disable-output -debug-only=loop-interchange? Also, is -cache-line-size=64 required for the test?


;; Test that a confused dependency in loop should prevent interchange in
;; loops i and j.
;;
;; void test_deps() {
;; for (int i = 0; i <= 3; i++)
;; for (int j = 0; j <= 3; j++) {
;; *f ^= 0x1000;
;; c[j][i] = *f;
;; }
;; }


; CHECK: Confused dependency between:
; CHECK: store i32 %xor, ptr %arrayidx6, align 4
; CHECK: %1 = load i32, ptr %0, align 4
; CHECK-NOT: Loops interchanged.

@a = global i32 0, align 4
@f = global ptr @a, align 8
@c = global [4 x [4 x i32]] zeroinitializer, align 8
Comment on lines +22 to +24
Copy link
Contributor

@artagnon artagnon Sep 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do these need to be globals? Can they not be passed as arguments to the function?


define void @test_deps() {
Copy link
Contributor

@artagnon artagnon Sep 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would recommend naming this function confused_deps, as the "test" part is already implied.

entry:
%0 = load ptr, ptr @f, align 8
br label %for.cond1.preheader

; Loop:
for.cond1.preheader: ; preds = %entry, %for.cond.cleanup3
%indvars.iv16 = phi i64 [ 0, %entry ], [ %indvars.iv.next17, %for.cond.cleanup3 ]
br label %for.body4

for.cond.cleanup3: ; preds = %for.body4
%indvars.iv.next17 = add nuw nsw i64 %indvars.iv16, 1
%exitcond18 = icmp ne i64 %indvars.iv.next17, 4
br i1 %exitcond18, label %for.cond1.preheader, label %for.cond.cleanup

for.body4: ; preds = %for.cond1.preheader, %for.body4
%indvars.iv = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next, %for.body4 ]
%1 = load i32, ptr %0, align 4
%xor = xor i32 %1, 4096
store i32 %xor, ptr %0, align 4
%arrayidx6 = getelementptr inbounds [4 x [4 x i32]], ptr @c, i64 0, i64 %indvars.iv, i64 %indvars.iv16
store i32 %xor, ptr %arrayidx6, align 4
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp ne i64 %indvars.iv.next, 4
br i1 %exitcond, label %for.body4, label %for.cond.cleanup3
Comment on lines +42 to +50
Copy link
Contributor

@artagnon artagnon Sep 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It would be nice if the test were cleaned up a bit. Example: s/indvars.iv/iv.inner/, s/for.body.4/inner.loop/, s/for.cond1.preheader/outer.loop/?


; Exit blocks
for.cond.cleanup: ; preds = %for.cond.cleanup3
ret void
}
6 changes: 3 additions & 3 deletions llvm/test/Transforms/LoopInterchange/reductions-across-inner-and-outer-loop.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,7 +156,7 @@ for1.loopexit: ; preds = %for1.inc
; REMARKS-NEXT: Name: UnsupportedPHIOuter
; REMARKS-NEXT: Function: test4

define i64 @test4(ptr %Arr, ptr %dst) {
define i64 @test4(ptr %Arr, ptr noalias %dst) {
entry:
%gep.dst = getelementptr inbounds i64, ptr %dst, i64 42
br label %for1.header
Expand Down Expand Up @@ -234,7 +234,7 @@ for1.loopexit: ; preds = %for1.inc
; REMARKS-NEXT: Name: Interchanged
; REMARKS-NEXT: Function: test5

define float @test5(ptr %Arr, ptr %Arr2) {
define float @test5(ptr %Arr, ptr noalias %Arr2) {
entry:
br label %outer.header

Expand Down Expand Up @@ -274,7 +274,7 @@ for.exit: ; preds = %outer.inc
; REMARKS-NEXT: Name: UnsupportedPHIOuter
; REMARKS-NEXT: Function: test6

define float @test6(ptr %Arr, ptr %Arr2) {
define float @test6(ptr %Arr, ptr noalias %Arr2) {
entry:
br label %outer.header

Expand Down
20 changes: 5 additions & 15 deletions llvm/test/Transforms/LoopInterchange/vector-gep-operand.ll
View file
Open in desktop
Copy link
Contributor

@artagnon artagnon Sep 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Was this test faulty?

Original file line number Diff line number Diff line change
Expand Up @@ -6,35 +6,25 @@
define void @test(ptr noalias %src, ptr %dst) {
; CHECK-LABEL: @test(
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[INNER_PREHEADER:%.*]]
; CHECK: outer.header.preheader:
; CHECK-NEXT: br label [[OUTER_HEADER:%.*]]
; CHECK: outer.header:
; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[I_NEXT:%.*]], [[OUTER_LATCH:%.*]] ], [ 0, [[OUTER_HEADER_PREHEADER:%.*]] ]
; CHECK-NEXT: br label [[INNER_SPLIT1:%.*]]
; CHECK: inner.preheader:
; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[I_NEXT:%.*]], [[OUTER_LATCH:%.*]] ], [ 0, [[ENTRY:%.*]] ]
; CHECK-NEXT: br label [[INNER:%.*]]
; CHECK: inner:
; CHECK-NEXT: [[J:%.*]] = phi i64 [ [[TMP0:%.*]], [[INNER_SPLIT:%.*]] ], [ 0, [[INNER_PREHEADER]] ]
; CHECK-NEXT: br label [[OUTER_HEADER_PREHEADER]]
; CHECK: inner.split1:
; CHECK-NEXT: [[J:%.*]] = phi i64 [ 0, [[OUTER_HEADER]] ], [ [[J_NEXT:%.*]], [[INNER]] ]
; CHECK-NEXT: [[SRC_GEP:%.*]] = getelementptr inbounds [256 x float], ptr [[SRC:%.*]], <2 x i64> <i64 0, i64 1>, i64 [[J]]
; CHECK-NEXT: [[SRC_0:%.*]] = extractelement <2 x ptr> [[SRC_GEP]], i32 0
; CHECK-NEXT: [[LV_0:%.*]] = load float, ptr [[SRC_0]], align 4
; CHECK-NEXT: [[ADD_0:%.*]] = fadd float [[LV_0]], 1.000000e+00
; CHECK-NEXT: [[DST_GEP:%.*]] = getelementptr inbounds float, ptr [[DST:%.*]], i64 [[J]]
; CHECK-NEXT: store float [[ADD_0]], ptr [[DST_GEP]], align 4
; CHECK-NEXT: [[J_NEXT:%.*]] = add nuw nsw i64 [[J]], 1
; CHECK-NEXT: [[J_NEXT]] = add nuw nsw i64 [[J]], 1
; CHECK-NEXT: [[INNER_EXITCOND:%.*]] = icmp eq i64 [[J_NEXT]], 100
; CHECK-NEXT: br label [[OUTER_LATCH]]
; CHECK: inner.split:
; CHECK-NEXT: [[TMP0]] = add nuw nsw i64 [[J]], 1
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[TMP0]], 100
; CHECK-NEXT: br i1 [[TMP1]], label [[EXIT:%.*]], label [[INNER]]
; CHECK-NEXT: br i1 [[INNER_EXITCOND]], label [[OUTER_LATCH]], label [[INNER]]
; CHECK: outer.latch:
; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i32 [[I]], 1
; CHECK-NEXT: [[OUTER_EXITCOND:%.*]] = icmp eq i32 [[I_NEXT]], 100
; CHECK-NEXT: br i1 [[OUTER_EXITCOND]], label [[INNER_SPLIT]], label [[OUTER_HEADER]]
; CHECK-NEXT: br i1 [[OUTER_EXITCOND]], label [[EXIT:%.*]], label [[OUTER_HEADER]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
Expand Down