[SCEV] Allow udiv canonicalization of potentially-wrapping AddRecs #169576
+50
−16
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
llvmbot
added
llvm:analysis
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Florian Hahn (fhahn) ChangesExtend the {X,+,N}/C => {(X - X%N),+,N}/C canonicalization to handle AddRecs that may wrap, when X < N <= C and both N,C are powers of 2. The alignment and power-of-2 properties ensure division results remain equivalent for all offsets [(X - X%N), X). Alive2 Proof: https://alive2.llvm.org/ce/z/VGUZo3 Fixes #168709 Full diff: https://github.com/llvm/llvm-project/pull/169576.diff 3 Files Affected:
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index a31f17b1936d6..e6be22a8e4148 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -3492,17 +3492,27 @@ const SCEV *ScalarEvolution::getUDivExpr(const SCEV *LHS,
/// {X,+,N}/C => {Y,+,N}/C where Y=X-(X%N). Safe when C%N=0.
// We can currently only fold X%N if X is constant.
const SCEVConstant *StartC = dyn_cast<SCEVConstant>(AR->getStart());
- if (StartC && !DivInt.urem(StepInt) &&
- getZeroExtendExpr(AR, ExtTy) ==
- getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
- getZeroExtendExpr(Step, ExtTy),
- AR->getLoop(), SCEV::FlagAnyWrap)) {
+ if (StartC && !DivInt.urem(StepInt)) {
const APInt &StartInt = StartC->getAPInt();
const APInt &StartRem = StartInt.urem(StepInt);
- if (StartRem != 0) {
- const SCEV *NewLHS =
- getAddRecExpr(getConstant(StartInt - StartRem), Step,
- AR->getLoop(), SCEV::FlagNW);
+ bool NoWrap =
+ getZeroExtendExpr(AR, ExtTy) ==
+ getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
+ getZeroExtendExpr(Step, ExtTy), AR->getLoop(),
+ SCEV::FlagAnyWrap);
+
+ // With X < N <= C and both N, C as powers-of-2, the transformation
+ // {X,+,N}/C => {(X - X%N),+,N}/C preserves division results even
+ // if wrapping occurs, as the division results remain equivalent for
+ // all offsets in [[(X - X%N), X).
+ bool CanFoldWithWrap = StepInt.isStrictlyPositive() &&
+ StartInt.ult(StepInt) && // X < N
+ StepInt.sle(DivInt) && // N <= C
+ StepInt.isPowerOf2() && DivInt.isPowerOf2();
+ if (StartRem != 0 && (NoWrap || CanFoldWithWrap)) {
+ const SCEV *NewLHS = getAddRecExpr(
+ getConstant(StartInt - StartRem), Step, AR->getLoop(),
+ NoWrap ? SCEV::FlagNW : SCEV::FlagAnyWrap);
if (LHS != NewLHS) {
LHS = NewLHS;
diff --git a/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll b/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll
index 0a6ef0dad4569..bb0744f8c4550 100644
--- a/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll
+++ b/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll
@@ -13,7 +13,7 @@ define void @test_step2_div4(i64 %n) {
; CHECK-NEXT: %iv.1 = add i64 %iv, 1
; CHECK-NEXT: --> {1,+,2}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.1 = udiv i64 %iv.1, 4
-; CHECK-NEXT: --> ({1,+,2}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,2}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.2 = add i64 %iv, 2
; CHECK-NEXT: --> {2,+,2}<%loop> U: [0,-1) S: [-9223372036854775808,9223372036854775807) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.2 = udiv i64 %iv.2, 4
@@ -114,15 +114,15 @@ define void @test_step4_div4(i64 %n) {
; CHECK-NEXT: %iv.1 = add i64 %iv, 1
; CHECK-NEXT: --> {1,+,4}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.1 = udiv i64 %iv.1, 4
-; CHECK-NEXT: --> ({1,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.2 = add i64 %iv, 2
; CHECK-NEXT: --> {2,+,4}<%loop> U: [0,-1) S: [-9223372036854775808,9223372036854775807) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.2 = udiv i64 %iv.2, 4
-; CHECK-NEXT: --> ({2,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.3 = add i64 %iv, 3
; CHECK-NEXT: --> {3,+,4}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.3 = udiv i64 %iv.3, 4
-; CHECK-NEXT: --> ({3,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.4 = add i64 %iv, 4
; CHECK-NEXT: --> {4,+,4}<%loop> U: [0,-3) S: [-9223372036854775808,9223372036854775805) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.4 = udiv i64 %iv.4, 4
diff --git a/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll b/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll
index 166875dd55aae..b612bfb88198e 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll
@@ -5,7 +5,7 @@
; CHECK: LV: Found an estimated cost of 1 for VF 1 For instruction: %shift = ashr i32 %val, %k
; CHECK: Cost of 2 for VF 2: WIDEN ir<%shift> = ashr ir<%val>, ir<%k>
; CHECK: Cost of 2 for VF 4: WIDEN ir<%shift> = ashr ir<%val>, ir<%k>
-define void @foo(ptr nocapture %p, i32 %k) local_unnamed_addr #0 {
+define void @foo(ptr nocapture %p, i32 %k) local_unnamed_addr {
entry:
br label %body
@@ -21,5 +21,31 @@ body:
exit:
ret void
+}
+
+; CHECK: 'shift_and_masked_load_store'
+; CHECK: Cost of 1 for VF 2: CLONE ir<%shifted> = lshr vp<{{.+}}>, ir<2>
+; CHECK: Cost of 1 for VF 4: CLONE ir<%shifted> = lshr vp<{{.+}}>, ir<2>
+; CHECK: Cost of 4 for VF 8: WIDEN ir<%shifted> = lshr ir<%iv>, ir<2>
+define void @shift_and_masked_load_store(i64 %trip.count) #0 {
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %shifted = lshr i64 %iv, 2
+ %masked.idx = and i64 %shifted, 1
+ %load.ptr = getelementptr i16, ptr poison, i64 %masked.idx
+ %val = load i16, ptr %load.ptr, align 2
+ %store.idx = shl nuw i64 %iv, 2
+ %store.ptr = getelementptr i8, ptr poison, i64 %store.idx
+ store i16 %val, ptr %store.ptr, align 2
+ %iv.next = add i64 %iv, 1
+ %cmp = icmp eq i64 %iv, %trip.count
+ br i1 %cmp, label %exit, label %loop
+exit:
+ ret void
}
+
+attributes #0 = { "target-features"="+avx2" "tune-cpu"="alderlake" }
|
Member
|
@llvm/pr-subscribers-llvm-analysis Author: Florian Hahn (fhahn) ChangesExtend the {X,+,N}/C => {(X - X%N),+,N}/C canonicalization to handle AddRecs that may wrap, when X < N <= C and both N,C are powers of 2. The alignment and power-of-2 properties ensure division results remain equivalent for all offsets [(X - X%N), X). Alive2 Proof: https://alive2.llvm.org/ce/z/VGUZo3 Fixes #168709 Full diff: https://github.com/llvm/llvm-project/pull/169576.diff 3 Files Affected:
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index a31f17b1936d6..e6be22a8e4148 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -3492,17 +3492,27 @@ const SCEV *ScalarEvolution::getUDivExpr(const SCEV *LHS,
/// {X,+,N}/C => {Y,+,N}/C where Y=X-(X%N). Safe when C%N=0.
// We can currently only fold X%N if X is constant.
const SCEVConstant *StartC = dyn_cast<SCEVConstant>(AR->getStart());
- if (StartC && !DivInt.urem(StepInt) &&
- getZeroExtendExpr(AR, ExtTy) ==
- getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
- getZeroExtendExpr(Step, ExtTy),
- AR->getLoop(), SCEV::FlagAnyWrap)) {
+ if (StartC && !DivInt.urem(StepInt)) {
const APInt &StartInt = StartC->getAPInt();
const APInt &StartRem = StartInt.urem(StepInt);
- if (StartRem != 0) {
- const SCEV *NewLHS =
- getAddRecExpr(getConstant(StartInt - StartRem), Step,
- AR->getLoop(), SCEV::FlagNW);
+ bool NoWrap =
+ getZeroExtendExpr(AR, ExtTy) ==
+ getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
+ getZeroExtendExpr(Step, ExtTy), AR->getLoop(),
+ SCEV::FlagAnyWrap);
+
+ // With X < N <= C and both N, C as powers-of-2, the transformation
+ // {X,+,N}/C => {(X - X%N),+,N}/C preserves division results even
+ // if wrapping occurs, as the division results remain equivalent for
+ // all offsets in [[(X - X%N), X).
+ bool CanFoldWithWrap = StepInt.isStrictlyPositive() &&
+ StartInt.ult(StepInt) && // X < N
+ StepInt.sle(DivInt) && // N <= C
+ StepInt.isPowerOf2() && DivInt.isPowerOf2();
+ if (StartRem != 0 && (NoWrap || CanFoldWithWrap)) {
+ const SCEV *NewLHS = getAddRecExpr(
+ getConstant(StartInt - StartRem), Step, AR->getLoop(),
+ NoWrap ? SCEV::FlagNW : SCEV::FlagAnyWrap);
if (LHS != NewLHS) {
LHS = NewLHS;
diff --git a/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll b/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll
index 0a6ef0dad4569..bb0744f8c4550 100644
--- a/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll
+++ b/llvm/test/Analysis/ScalarEvolution/addrec-may-wrap-udiv-canonicalize.ll
@@ -13,7 +13,7 @@ define void @test_step2_div4(i64 %n) {
; CHECK-NEXT: %iv.1 = add i64 %iv, 1
; CHECK-NEXT: --> {1,+,2}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.1 = udiv i64 %iv.1, 4
-; CHECK-NEXT: --> ({1,+,2}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,2}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.2 = add i64 %iv, 2
; CHECK-NEXT: --> {2,+,2}<%loop> U: [0,-1) S: [-9223372036854775808,9223372036854775807) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.2 = udiv i64 %iv.2, 4
@@ -114,15 +114,15 @@ define void @test_step4_div4(i64 %n) {
; CHECK-NEXT: %iv.1 = add i64 %iv, 1
; CHECK-NEXT: --> {1,+,4}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.1 = udiv i64 %iv.1, 4
-; CHECK-NEXT: --> ({1,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.2 = add i64 %iv, 2
; CHECK-NEXT: --> {2,+,4}<%loop> U: [0,-1) S: [-9223372036854775808,9223372036854775807) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.2 = udiv i64 %iv.2, 4
-; CHECK-NEXT: --> ({2,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.3 = add i64 %iv, 3
; CHECK-NEXT: --> {3,+,4}<%loop> U: full-set S: full-set Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.3 = udiv i64 %iv.3, 4
-; CHECK-NEXT: --> ({3,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
+; CHECK-NEXT: --> ({0,+,4}<%loop> /u 4) U: [0,4611686018427387904) S: [0,4611686018427387904) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %iv.4 = add i64 %iv, 4
; CHECK-NEXT: --> {4,+,4}<%loop> U: [0,-3) S: [-9223372036854775808,9223372036854775805) Exits: <<Unknown>> LoopDispositions: { %loop: Computable }
; CHECK-NEXT: %div.4 = udiv i64 %iv.4, 4
diff --git a/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll b/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll
index 166875dd55aae..b612bfb88198e 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/uniformshift.ll
@@ -5,7 +5,7 @@
; CHECK: LV: Found an estimated cost of 1 for VF 1 For instruction: %shift = ashr i32 %val, %k
; CHECK: Cost of 2 for VF 2: WIDEN ir<%shift> = ashr ir<%val>, ir<%k>
; CHECK: Cost of 2 for VF 4: WIDEN ir<%shift> = ashr ir<%val>, ir<%k>
-define void @foo(ptr nocapture %p, i32 %k) local_unnamed_addr #0 {
+define void @foo(ptr nocapture %p, i32 %k) local_unnamed_addr {
entry:
br label %body
@@ -21,5 +21,31 @@ body:
exit:
ret void
+}
+
+; CHECK: 'shift_and_masked_load_store'
+; CHECK: Cost of 1 for VF 2: CLONE ir<%shifted> = lshr vp<{{.+}}>, ir<2>
+; CHECK: Cost of 1 for VF 4: CLONE ir<%shifted> = lshr vp<{{.+}}>, ir<2>
+; CHECK: Cost of 4 for VF 8: WIDEN ir<%shifted> = lshr ir<%iv>, ir<2>
+define void @shift_and_masked_load_store(i64 %trip.count) #0 {
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %shifted = lshr i64 %iv, 2
+ %masked.idx = and i64 %shifted, 1
+ %load.ptr = getelementptr i16, ptr poison, i64 %masked.idx
+ %val = load i16, ptr %load.ptr, align 2
+ %store.idx = shl nuw i64 %iv, 2
+ %store.ptr = getelementptr i8, ptr poison, i64 %store.idx
+ store i16 %val, ptr %store.ptr, align 2
+ %iv.next = add i64 %iv, 1
+ %cmp = icmp eq i64 %iv, %trip.count
+ br i1 %cmp, label %exit, label %loop
+exit:
+ ret void
}
+
+attributes #0 = { "target-features"="+avx2" "tune-cpu"="alderlake" }
|
fhahn
force-pushed
the
scev-udiv-canon
branch
from
34a6be4 to
3e2f0d2
Compare
nikic
reviewed
Dec 1, 2025
| // all offsets in [[(X - X%N), X). | ||
| bool CanFoldWithWrap = StepInt.isStrictlyPositive() && | ||
| StartInt.ult(StepInt) && // X < N | ||
| StepInt.sle(DivInt) && // N <= C |
Contributor
There was a problem hiding this comment.
Can we make this ule and omit the isStrictlyPositive? https://alive2.llvm.org/ce/z/xZuj6q
Contributor
Author
There was a problem hiding this comment.
I think so, updated thanks! I couldn't find any additional cases that can be handled with the relaxed checks
fhahn
force-pushed
the
scev-udiv-canon
branch
from
3e2f0d2 to
43d04f7
Compare
Contributor
|
I don't think your proof is right, because it has a I think this is the right proof: https://alive2.llvm.org/ce/z/iu2tav I'm modelling the addrec as This shows that the |
Extend the {X,+,N}/C => {(X - X%N),+,N}/C canonicalization to handle
AddRecs that may wrap, when X < N <= C and both N,C are powers of 2.
The alignment and power-of-2 properties ensure division results remain
equivalent for all offsets [(X - X%N), X).
Alive2 Proof: https://alive2.llvm.org/ce/z/VGUZo3
Fixes llvm#168709
fhahn
force-pushed
the
scev-udiv-canon
branch
from
43d04f7 to
e596057
Compare
Contributor
Author
Yep that sounds right, thanks! Updated the PR. There have bee 2 improvements, which should be verified via https://alive2.llvm.org/ce/z/sV38FC (hopefully not over-reduced) |
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Dec 2, 2025
… AddRecs (#169576)
Extend the {X,+,N}/C => {(X - X%N),+,N}/C canonicalization to handle
AddRecs that may wrap, when X < N <= C and both N,C are powers of 2. The
alignment and power-of-2 properties ensure division results remain
equivalent for all offsets [(X - X%N), X).
Alive2 Proof: https://alive2.llvm.org/ce/z/iu2tav
Fixes llvm/llvm-project#168709
PR: llvm/llvm-project#169576
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Extend the {X,+,N}/C => {(X - X%N),+,N}/C canonicalization to handle AddRecs that may wrap, when X < N <= C and both N,C are powers of 2. The alignment and power-of-2 properties ensure division results remain equivalent for all offsets [(X - X%N), X).
Alive2 Proof: https://alive2.llvm.org/ce/z/iu2tav
Fixes #168709