[SECV] Try to push the op into ZExt: A + zext (-A + B) -> zext (B) #151227

fhahn · 2025-07-29T20:47:47Z

Try to push the constant operand into a ZExt:
A + zext (-A + B) -> zext (B), if trunc (A) + -A + B does not unsigned-wrap.

The actual code supports ZExts with arbitrary number of arguments, hence the getAddExpr in the return.

This helps SCEV reasoning in some cases, commonly when adding an offset to a zero-extended SCEV that subtracts the same offset.

Note that this is restricted to cases where we can fold away an operand of the inner Add. This is needed to avoid bad interactions with patterns when forming ZExts, which try to push to ZExt to add operands.

https://alive2.llvm.org/ce/z/q7d303

llvmbot · 2025-07-29T20:48:22Z

@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: Florian Hahn (fhahn)

Changes

Try to push the constant operand into a ZExt:
A + zext (-A + B) -> zext (B), if trunc (A) + -A + B does not unsigned-wrap.

The actual code supports ZExts with arbitrary number of arguments, hence the getAddExpr in the return.

This helps SCEV reasoning in some cases, commonly when adding an offset to a zero-extended SCEV that subtracts the same offset.

Note that this is restricted to cases where we can fold away an operand of the inner Add. This is needed to avoid bad interactions with patterns when forming ZExts, which try to push to ZExt to add operands.

https://alive2.llvm.org/ce/z/uuYGC3k

Full diff: https://github.com/llvm/llvm-project/pull/151227.diff

3 Files Affected:

(modified) llvm/lib/Analysis/ScalarEvolution.cpp (+15)
(modified) llvm/test/Transforms/IndVarSimplify/AArch64/fold-ext-add.ll (+5-5)
(modified) llvm/test/Transforms/IndVarSimplify/zext-nuw.ll (+2-4)

diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 0990a0daac80c..8e587d8cbced6 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -2682,6 +2682,21 @@ const SCEV *ScalarEvolution::getAddExpr(SmallVectorImpl<const SCEV *> &Ops,
         return getAddExpr(NewOps, PreservedFlags);
       }
     }
+
+    // Try to push the constant operand into a ZExt: A + zext (-A + B) -> zext
+    // (B), if trunc (A) + -A + B  does not unsigned-wrap.
+    if (auto *ZExt = dyn_cast<SCEVZeroExtendExpr>(Ops[1])) {
+      const SCEV *B = ZExt->getOperand(0);
+      const SCEV *NarrowA = getTruncateExpr(A, B->getType());
+      if (isa<SCEVAddExpr>(B) &&
+          NarrowA == getNegativeSCEV(cast<SCEVAddExpr>(B)->getOperand(0)) &&
+          getZeroExtendExpr(NarrowA, ZExt->getType()) == A &&
+          hasFlags(
+              StrengthenNoWrapFlags(this, scAddExpr, {NarrowA, B}, OrigFlags),
+              SCEV::FlagNUW)) {
+        return getZeroExtendExpr(getAddExpr(NarrowA, B), ZExt->getType());
+      }
+    }
   }
 
   // Canonicalize (-1 * urem X, Y) + X --> (Y * X/Y)
diff --git a/llvm/test/Transforms/IndVarSimplify/AArch64/fold-ext-add.ll b/llvm/test/Transforms/IndVarSimplify/AArch64/fold-ext-add.ll
index 48b92e905b8ce..640c910be4a82 100644
--- a/llvm/test/Transforms/IndVarSimplify/AArch64/fold-ext-add.ll
+++ b/llvm/test/Transforms/IndVarSimplify/AArch64/fold-ext-add.ll
@@ -10,21 +10,21 @@ define void @pred_mip_12(ptr %dst, ptr %src, i32 %n, i64 %offset) {
 ; CHECK-SAME: ptr [[DST:%.*]], ptr [[SRC:%.*]], i32 [[N:%.*]], i64 [[OFFSET:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*]]:
 ; CHECK-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[N]], i32 1)
+; CHECK-NEXT:    [[TMP0:%.*]] = zext nneg i32 [[SMAX]] to i64
+; CHECK-NEXT:    [[TMP1:%.*]] = mul i64 [[OFFSET]], [[TMP0]]
+; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP1]]
 ; CHECK-NEXT:    br label %[[OUTER_LOOP:.*]]
 ; CHECK:       [[OUTER_LOOP_LOOPEXIT:.*]]:
-; CHECK-NEXT:    [[PTR_IV_NEXT_LCSSA:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], %[[INNER_LOOP:.*]] ]
 ; CHECK-NEXT:    br label %[[OUTER_LOOP]]
 ; CHECK:       [[OUTER_LOOP]]:
-; CHECK-NEXT:    [[OUTER_PTR:%.*]] = phi ptr [ [[SRC]], %[[ENTRY]] ], [ [[PTR_IV_NEXT_LCSSA]], %[[OUTER_LOOP_LOOPEXIT]] ]
+; CHECK-NEXT:    [[OUTER_PTR:%.*]] = phi ptr [ [[SRC]], %[[ENTRY]] ], [ [[SCEVGEP]], %[[OUTER_LOOP_LOOPEXIT]] ]
 ; CHECK-NEXT:    [[C:%.*]] = call i1 @cond()
 ; CHECK-NEXT:    br i1 [[C]], label %[[INNER_LOOP_PREHEADER:.*]], label %[[EXIT:.*]]
 ; CHECK:       [[INNER_LOOP_PREHEADER]]:
-; CHECK-NEXT:    br label %[[INNER_LOOP]]
+; CHECK-NEXT:    br label %[[INNER_LOOP:.*]]
 ; CHECK:       [[INNER_LOOP]]:
 ; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], %[[INNER_LOOP]] ], [ 0, %[[INNER_LOOP_PREHEADER]] ]
-; CHECK-NEXT:    [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT]], %[[INNER_LOOP]] ], [ [[SRC]], %[[INNER_LOOP_PREHEADER]] ]
 ; CHECK-NEXT:    [[L:%.*]] = load i8, ptr [[OUTER_PTR]], align 1
-; CHECK-NEXT:    [[PTR_IV_NEXT]] = getelementptr i8, ptr [[PTR_IV]], i64 [[OFFSET]]
 ; CHECK-NEXT:    store i8 [[L]], ptr [[DST]], align 2
 ; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
 ; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[IV_NEXT]], [[SMAX]]
diff --git a/llvm/test/Transforms/IndVarSimplify/zext-nuw.ll b/llvm/test/Transforms/IndVarSimplify/zext-nuw.ll
index d24f9a4e40e38..17921afc5ff06 100644
--- a/llvm/test/Transforms/IndVarSimplify/zext-nuw.ll
+++ b/llvm/test/Transforms/IndVarSimplify/zext-nuw.ll
@@ -15,11 +15,9 @@ define void @_Z3fn1v() {
 ; CHECK-NEXT:    [[J_SROA_0_0_COPYLOAD:%.*]] = load i8, ptr [[X5]], align 1
 ; CHECK-NEXT:    br label [[DOTPREHEADER4_LR_PH:%.*]]
 ; CHECK:       .preheader4.lr.ph:
-; CHECK-NEXT:    [[TMP1:%.*]] = add nsw i32 [[X4]], -1
-; CHECK-NEXT:    [[TMP2:%.*]] = zext nneg i32 [[TMP1]] to i64
-; CHECK-NEXT:    [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
 ; CHECK-NEXT:    [[TMP4:%.*]] = sext i8 [[J_SROA_0_0_COPYLOAD]] to i64
-; CHECK-NEXT:    [[TMP5:%.*]] = mul i64 [[TMP3]], [[TMP4]]
+; CHECK-NEXT:    [[TMP2:%.*]] = zext nneg i32 [[X4]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = mul i64 [[TMP4]], [[TMP2]]
 ; CHECK-NEXT:    br label [[DOTPREHEADER4:%.*]]
 ; CHECK:       .preheader4:
 ; CHECK-NEXT:    [[K_09:%.*]] = phi ptr [ undef, [[DOTPREHEADER4_LR_PH]] ], [ [[X25:%.*]], [[X22:%.*]] ]

efriedma-quic · 2025-07-29T20:56:22Z

llvm/lib/Analysis/ScalarEvolution.cpp

OrigFlags is the flags of the outer add. I can't see any way it's correct to use those flags for a narrower add.

(Even if both the outer and inner additions are nsw/nuw, that isn't enough for this transform. See https://alive2.llvm.org/ce/z/DFLN-5 .)

Yep, it should not pass the original flags. Updated to pass AnyWrapFlags.

This checks adding the truncated first operand and the inner expression of the ZExt does not wrap. Updated proof: https://alive2.llvm.org/ce/z/q7d303

efriedma-quic · 2025-07-29T20:56:36Z

llvm/test/Transforms/IndVarSimplify/AArch64/fold-ext-add.ll

In addition to the indvars tests, can you add a scalar-evolution analysis test that shows the flags in a dump?

Yup, added an analysis only test, thanks!

Adds a SCEV-only tests for #151227.

Try to push the constant operand into a ZExt: A + zext (-A + B) -> zext (B), if trunc (A) + -A + B does not unsigned-wrap. The actual code supports ZExts with arbitrary number of arguments, hence the getAddExpr in the return. This helps SCEV reasoning in some cases, commonly when adding an offset to a zero-extended SCEV that subtracts the same offset. Note that this is restricted to cases where we can fold away an operand of the inner Add. This is needed to avoid bad interactions with patterns when forming ZExts, which try to push to ZExt to add operands. https://alive2.llvm.org/ce/z/uuYGC3k

Adds a SCEV-only tests for llvm/llvm-project#151227.

efriedma-quic

LGTM

…zext (B) (#151227) Try to push the constant operand into a ZExt: A + zext (-A + B) -> zext (B), if trunc (A) + -A + B does not unsigned-wrap. The actual code supports ZExts with arbitrary number of arguments, hence the getAddExpr in the return. This helps SCEV reasoning in some cases, commonly when adding an offset to a zero-extended SCEV that subtracts the same offset. Note that this is restricted to cases where we can fold away an operand of the inner Add. This is needed to avoid bad interactions with patterns when forming ZExts, which try to push to ZExt to add operands. https://alive2.llvm.org/ce/z/q7d303 PR: llvm/llvm-project#151227

nikic · 2025-07-31T09:34:52Z

llvm/lib/Analysis/ScalarEvolution.cpp

+    // (B), if trunc (A) + -A + B  does not unsigned-wrap.
+    if (auto *ZExt = dyn_cast<SCEVZeroExtendExpr>(Ops[1])) {
+      const SCEV *B = ZExt->getOperand(0);
+      const SCEV *NarrowA = getTruncateExpr(A, B->getType());


Check isa<SCEVAddExpr>(B) before computing getTruncateExpr?

Thanks, should be done in ab9b23c by introducing a new m_scec_Add() which matches and captures SCEVAddExpr with an abitrary number of operands

Follow-up to #151227 (review) to check the inner expression is an Add before calling getTruncateExpr. Adds a new matcher that just matches and captures SCEVAddExpr, to support matching a SCEVAddExpr with arbitrary number of operands.

Follow-up to llvm/llvm-project#151227 (review) to check the inner expression is an Add before calling getTruncateExpr. Adds a new matcher that just matches and captures SCEVAddExpr, to support matching a SCEVAddExpr with arbitrary number of operands.

Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to llvm#151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL

…#155300) Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to #151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL PR: #155300

…(A*C + B*C) (#155300) Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to llvm/llvm-project#151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL PR: llvm/llvm-project#155300

Adds a SCEV-only tests for llvm#151227. (cherry picked from commit c6f7fa7)

…lvm#151227) Try to push the constant operand into a ZExt: A + zext (-A + B) -> zext (B), if trunc (A) + -A + B does not unsigned-wrap. The actual code supports ZExts with arbitrary number of arguments, hence the getAddExpr in the return. This helps SCEV reasoning in some cases, commonly when adding an offset to a zero-extended SCEV that subtracts the same offset. Note that this is restricted to cases where we can fold away an operand of the inner Add. This is needed to avoid bad interactions with patterns when forming ZExts, which try to push to ZExt to add operands. https://alive2.llvm.org/ce/z/q7d303 PR: llvm#151227 (cherry picked from commit d74d841)

Follow-up to llvm#151227 (review) to check the inner expression is an Add before calling getTruncateExpr. Adds a new matcher that just matches and captures SCEVAddExpr, to support matching a SCEVAddExpr with arbitrary number of operands. (cherry picked from commit ab9b23c)

…llvm#155300) Try to push constant multiply operand into a ZExt containing an add, if possible. In general we are trying to push down ops through ZExt if possible. This is similar to llvm#151227 which did the same for additions. For now this is restricted to adds with a constant operand, which is similar to some of the logic above. This enables some additional simplifications. Alive2 Proof: https://alive2.llvm.org/ce/z/97pbSL PR: llvm#155300 (cherry-picked from 9143746)

fhahn requested review from efriedma-quic and preames July 29, 2025 20:47

fhahn requested a review from nikic as a code owner July 29, 2025 20:47

llvmbot added llvm:analysis Includes value tracking, cost tables and constant folding llvm:transforms labels Jul 29, 2025

efriedma-quic reviewed Jul 29, 2025

View reviewed changes

fhahn added a commit that referenced this pull request Jul 30, 2025

[SCEV] Add test for pushing constant add into zext.

c6f7fa7

Adds a SCEV-only tests for #151227.

fhahn added 2 commits July 30, 2025 10:05

!fixup pass FlagAnyWrap.

e834442

fhahn force-pushed the perf/scev-push-const-op-into-zext branch from df16937 to e834442 Compare July 30, 2025 09:05

llvm-sync bot pushed a commit to arm/arm-toolchain that referenced this pull request Jul 30, 2025

Automerge: [SCEV] Add test for pushing constant add into zext.

b609d15

Adds a SCEV-only tests for llvm/llvm-project#151227.

efriedma-quic approved these changes Jul 30, 2025

View reviewed changes

fhahn merged commit d74d841 into llvm:main Jul 30, 2025
9 checks passed

nikic reviewed Jul 31, 2025

View reviewed changes

fhahn deleted the perf/scev-push-const-op-into-zext branch July 31, 2025 11:48

fhahn mentioned this pull request Aug 25, 2025

[SCEV] Try to push op into ZExt: C * zext (A + B) -> zext (A*C + B*C) #155300

Merged

aemerson mentioned this pull request Aug 30, 2025

[NFC] [clangd] [Modules] remove dot in log #156207

Closed

fhahn added a commit to fhahn/llvm-project that referenced this pull request Nov 4, 2025

[SCEV] Add test for pushing constant add into zext.

9184074

Adds a SCEV-only tests for llvm#151227. (cherry picked from commit c6f7fa7)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[SECV] Try to push the op into ZExt: A + zext (-A + B) -> zext (B) #151227

[SECV] Try to push the op into ZExt: A + zext (-A + B) -> zext (B) #151227

Uh oh!

fhahn commented Jul 29, 2025 •

edited

Loading

Uh oh!

llvmbot commented Jul 29, 2025 •

edited

Loading

Uh oh!

efriedma-quic Jul 29, 2025

Uh oh!

efriedma-quic Jul 29, 2025

Uh oh!

fhahn Jul 30, 2025

Uh oh!

efriedma-quic Jul 29, 2025

Uh oh!

fhahn Jul 30, 2025

Uh oh!

efriedma-quic left a comment

Uh oh!

Uh oh!

nikic Jul 31, 2025

Uh oh!

fhahn Jul 31, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

[SECV] Try to push the op into ZExt: A + zext (-A + B) -> zext (B) #151227

[SECV] Try to push the op into ZExt: A + zext (-A + B) -> zext (B) #151227

Uh oh!

Conversation

fhahn commented Jul 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jul 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

efriedma-quic Jul 29, 2025

Choose a reason for hiding this comment

Uh oh!

efriedma-quic Jul 29, 2025

Choose a reason for hiding this comment

Uh oh!

fhahn Jul 30, 2025

Choose a reason for hiding this comment

Uh oh!

efriedma-quic Jul 29, 2025

Choose a reason for hiding this comment

Uh oh!

fhahn Jul 30, 2025

Choose a reason for hiding this comment

Uh oh!

efriedma-quic left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

nikic Jul 31, 2025

Choose a reason for hiding this comment

Uh oh!

fhahn Jul 31, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

fhahn commented Jul 29, 2025 •

edited

Loading

llvmbot commented Jul 29, 2025 •

edited

Loading