[Reland][SCEV] teach isImpliedViaOperations about samesign #133711
Merged
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
Use CmpPredicate::getMatching in isImpliedCondBalancedTypes to pass samesign information to isImpliedViaOperations, and teach it to call CmpPredicate::getPreferredSignedPredicate, effectively making it optimize with samesign information.
llvmbot
added
llvm:analysis
|
@llvm/pr-subscribers-llvm-transforms Author: Yingwei Zheng (dtcxzyw) ChangesThis patch relands #124270. The root cause is that we incorrectly preserve the samesign flag after truncating operands of an icmp: Patch is 21.07 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/133711.diff 5 Files Affected:
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 600a061d4435e..32e669a400c13 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -11804,8 +11804,10 @@ bool ScalarEvolution::isImpliedCond(CmpPredicate Pred, const SCEV *LHS,
MaxValue)) {
const SCEV *TruncFoundLHS = getTruncateExpr(FoundLHS, NarrowType);
const SCEV *TruncFoundRHS = getTruncateExpr(FoundRHS, NarrowType);
- if (isImpliedCondBalancedTypes(Pred, LHS, RHS, FoundPred, TruncFoundLHS,
- TruncFoundRHS, CtxI))
+ // We cannot preserve samesign after truncation.
+ if (isImpliedCondBalancedTypes(
+ Pred, LHS, RHS, static_cast<ICmpInst::Predicate>(FoundPred),
+ TruncFoundLHS, TruncFoundRHS, CtxI))
return true;
}
}
@@ -11862,15 +11864,13 @@ bool ScalarEvolution::isImpliedCondBalancedTypes(
}
// Check whether the found predicate is the same as the desired predicate.
- // FIXME: use CmpPredicate::getMatching here.
- if (FoundPred == static_cast<CmpInst::Predicate>(Pred))
- return isImpliedCondOperands(Pred, LHS, RHS, FoundLHS, FoundRHS, CtxI);
+ if (auto P = CmpPredicate::getMatching(FoundPred, Pred))
+ return isImpliedCondOperands(*P, LHS, RHS, FoundLHS, FoundRHS, CtxI);
// Check whether swapping the found predicate makes it the same as the
// desired predicate.
- // FIXME: use CmpPredicate::getMatching here.
- if (ICmpInst::getSwappedCmpPredicate(FoundPred) ==
- static_cast<CmpInst::Predicate>(Pred)) {
+ if (auto P = CmpPredicate::getMatching(
+ ICmpInst::getSwappedCmpPredicate(FoundPred), Pred)) {
// We can write the implication
// 0. LHS Pred RHS <- FoundLHS SwapPred FoundRHS
// using one of the following ways:
@@ -11881,22 +11881,23 @@ bool ScalarEvolution::isImpliedCondBalancedTypes(
// Forms 1. and 2. require swapping the operands of one condition. Don't
// do this if it would break canonical constant/addrec ordering.
if (!isa<SCEVConstant>(RHS) && !isa<SCEVAddRecExpr>(LHS))
- return isImpliedCondOperands(FoundPred, RHS, LHS, FoundLHS, FoundRHS,
- CtxI);
+ return isImpliedCondOperands(ICmpInst::getSwappedCmpPredicate(*P), RHS,
+ LHS, FoundLHS, FoundRHS, CtxI);
if (!isa<SCEVConstant>(FoundRHS) && !isa<SCEVAddRecExpr>(FoundLHS))
- return isImpliedCondOperands(Pred, LHS, RHS, FoundRHS, FoundLHS, CtxI);
+ return isImpliedCondOperands(*P, LHS, RHS, FoundRHS, FoundLHS, CtxI);
// There's no clear preference between forms 3. and 4., try both. Avoid
// forming getNotSCEV of pointer values as the resulting subtract is
// not legal.
if (!LHS->getType()->isPointerTy() && !RHS->getType()->isPointerTy() &&
- isImpliedCondOperands(FoundPred, getNotSCEV(LHS), getNotSCEV(RHS),
- FoundLHS, FoundRHS, CtxI))
+ isImpliedCondOperands(ICmpInst::getSwappedCmpPredicate(*P),
+ getNotSCEV(LHS), getNotSCEV(RHS), FoundLHS,
+ FoundRHS, CtxI))
return true;
if (!FoundLHS->getType()->isPointerTy() &&
!FoundRHS->getType()->isPointerTy() &&
- isImpliedCondOperands(Pred, LHS, RHS, getNotSCEV(FoundLHS),
+ isImpliedCondOperands(*P, LHS, RHS, getNotSCEV(FoundLHS),
getNotSCEV(FoundRHS), CtxI))
return true;
@@ -12572,14 +12573,16 @@ bool ScalarEvolution::isImpliedViaOperations(CmpPredicate Pred, const SCEV *LHS,
return false;
// We only want to work with GT comparison so far.
- if (Pred == ICmpInst::ICMP_ULT || Pred == ICmpInst::ICMP_SLT) {
+ if (ICmpInst::isLT(Pred)) {
Pred = ICmpInst::getSwappedCmpPredicate(Pred);
std::swap(LHS, RHS);
std::swap(FoundLHS, FoundRHS);
}
+ CmpInst::Predicate P = Pred.getPreferredSignedPredicate();
+
// For unsigned, try to reduce it to corresponding signed comparison.
- if (Pred == ICmpInst::ICMP_UGT)
+ if (P == ICmpInst::ICMP_UGT)
// We can replace unsigned predicate with its signed counterpart if all
// involved values are non-negative.
// TODO: We could have better support for unsigned.
@@ -12592,10 +12595,10 @@ bool ScalarEvolution::isImpliedViaOperations(CmpPredicate Pred, const SCEV *LHS,
FoundRHS) &&
isImpliedCondOperands(ICmpInst::ICMP_SGT, RHS, MinusOne, FoundLHS,
FoundRHS))
- Pred = ICmpInst::ICMP_SGT;
+ P = ICmpInst::ICMP_SGT;
}
- if (Pred != ICmpInst::ICMP_SGT)
+ if (P != ICmpInst::ICMP_SGT)
return false;
auto GetOpFromSExt = [&](const SCEV *S) {
diff --git a/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll b/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll
index 93c6bc08af2a0..4d569cc69fa2b 100644
--- a/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll
+++ b/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll
@@ -5,9 +5,9 @@
define i32 @exit_count_samesign(i32 %iter.count, ptr %ptr) {
; CHECK-LABEL: 'exit_count_samesign'
; CHECK-NEXT: Determining loop execution counts for: @exit_count_samesign
-; CHECK-NEXT: Loop %inner.loop: backedge-taken count is (-1 + (1 smax {(-1 + %iter.count)<nsw>,+,-1}<nsw><%outer.loop>))<nsw>
+; CHECK-NEXT: Loop %inner.loop: backedge-taken count is {(-2 + %iter.count),+,-1}<nw><%outer.loop>
; CHECK-NEXT: Loop %inner.loop: constant max backedge-taken count is i32 2147483646
-; CHECK-NEXT: Loop %inner.loop: symbolic max backedge-taken count is (-1 + (1 smax {(-1 + %iter.count)<nsw>,+,-1}<nsw><%outer.loop>))<nsw>
+; CHECK-NEXT: Loop %inner.loop: symbolic max backedge-taken count is {(-2 + %iter.count),+,-1}<nw><%outer.loop>
; CHECK-NEXT: Loop %inner.loop: Trip multiple is 1
; CHECK-NEXT: Loop %outer.loop: <multiple exits> Unpredictable backedge-taken count.
; CHECK-NEXT: Loop %outer.loop: Unpredictable constant max backedge-taken count.
diff --git a/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll b/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll
index a1d30406095ec..d83301243ef30 100644
--- a/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll
+++ b/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll
@@ -2,12 +2,10 @@
; RUN: opt < %s -disable-output -passes="print<scalar-evolution>" \
; RUN: -scalar-evolution-classify-expressions=0 2>&1 | FileCheck %s
-declare void @llvm.experimental.guard(i1, ...)
-
-define void @test_1(i32 %n) nounwind {
-; Prove that (n > 1) ===> (n / 2 > 0).
-; CHECK-LABEL: 'test_1'
-; CHECK-NEXT: Determining loop execution counts for: @test_1
+define void @implied1(i32 %n) {
+; Prove that (n s> 1) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied1'
+; CHECK-NEXT: Determining loop execution counts for: @implied1
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + %n.div.2)<nsw>
@@ -29,10 +27,35 @@ exit:
ret void
}
-define void @test_1neg(i32 %n) nounwind {
-; Prove that (n > 0) =\=> (n / 2 > 0).
-; CHECK-LABEL: 'test_1neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_1neg
+define void @implied1_samesign(i32 %n) {
+; Prove that (n > 1) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied1_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied1_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign ugt i32 %n, 1
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sgt i32 %n.div.2, %indvar.next
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied1_neg(i32 %n) {
+; Prove that (n s> 0) =\=> (n / 2 s> 0).
+; CHECK-LABEL: 'implied1_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied1_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
@@ -54,10 +77,10 @@ exit:
ret void
}
-define void @test_2(i32 %n) nounwind {
-; Prove that (n >= 2) ===> (n / 2 > 0).
-; CHECK-LABEL: 'test_2'
-; CHECK-NEXT: Determining loop execution counts for: @test_2
+define void @implied2(i32 %n) {
+; Prove that (n s>= 2) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied2'
+; CHECK-NEXT: Determining loop execution counts for: @implied2
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + %n.div.2)<nsw>
@@ -79,10 +102,35 @@ exit:
ret void
}
-define void @test_2neg(i32 %n) nounwind {
-; Prove that (n >= 1) =\=> (n / 2 > 0).
-; CHECK-LABEL: 'test_2neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_2neg
+define void @implied2_samesign(i32 %n) {
+; Prove that (n >= 2) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied2_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied2_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign uge i32 %n, 2
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sgt i32 %n.div.2, %indvar.next
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied2_neg(i32 %n) {
+; Prove that (n s>= 1) =\=> (n / 2 s> 0).
+; CHECK-LABEL: 'implied2_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied2_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
@@ -104,10 +152,10 @@ exit:
ret void
}
-define void @test_3(i32 %n) nounwind {
-; Prove that (n > -2) ===> (n / 2 >= 0).
-; CHECK-LABEL: 'test_3'
-; CHECK-NEXT: Determining loop execution counts for: @test_3
+define void @implied3(i32 %n) {
+; Prove that (n s> -2) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied3'
+; CHECK-NEXT: Determining loop execution counts for: @implied3
; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
@@ -129,10 +177,35 @@ exit:
ret void
}
-define void @test_3neg(i32 %n) nounwind {
+define void @implied3_samesign(i32 %n) {
+; Prove that (n > -2) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied3_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied3_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign ugt i32 %n, -2
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sge i32 %n.div.2, %indvar
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied3_neg(i32 %n) {
; Prove that (n > -3) =\=> (n / 2 >= 0).
-; CHECK-LABEL: 'test_3neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_3neg
+; CHECK-LABEL: 'implied3_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied3_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
@@ -154,10 +227,10 @@ exit:
ret void
}
-define void @test_4(i32 %n) nounwind {
-; Prove that (n >= -1) ===> (n / 2 >= 0).
-; CHECK-LABEL: 'test_4'
-; CHECK-NEXT: Determining loop execution counts for: @test_4
+define void @implied4(i32 %n) {
+; Prove that (n s>= -1) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied4'
+; CHECK-NEXT: Determining loop execution counts for: @implied4
; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
@@ -179,10 +252,35 @@ exit:
ret void
}
-define void @test_4neg(i32 %n) nounwind {
-; Prove that (n >= -2) =\=> (n / 2 >= 0).
-; CHECK-LABEL: 'test_4neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_4neg
+define void @implied4_samesign(i32 %n) {
+; Prove that (n >= -1) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied4_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied4_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign uge i32 %n, -1
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sge i32 %n.div.2, %indvar
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied4_neg(i32 %n) {
+; Prove that (n s>= -2) =\=> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied4_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied4_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
diff --git a/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll b/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll
index 1207f47c5e3c9..c4e26c98ed24a 100644
--- a/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll
+++ b/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll
@@ -68,28 +68,32 @@ define i32 @iv_zext_zext_gt_slt(i32 %iter.count, ptr %ptr) {
; CHECK-LABEL: define i32 @iv_zext_zext_gt_slt(
; CHECK-SAME: i32 [[ITER_COUNT:%.*]], ptr [[PTR:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
-; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[ITER_COUNT]] to i64
+; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[ITER_COUNT]], -1
; CHECK-NEXT: br label %[[OUTER_LOOP:.*]]
; CHECK: [[PH_LOOPEXIT:.*]]:
; CHECK-NEXT: br label %[[PH:.*]]
; CHECK: [[PH]]:
+; CHECK-NEXT: [[INDVARS_IV_NEXT3:%.*]] = add i32 [[INDVARS_IV1:%.*]], -1
; CHECK-NEXT: br label %[[OUTER_LOOP]]
; CHECK: [[OUTER_LOOP]]:
-; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[PH]] ], [ [[TMP0]], %[[ENTRY]] ]
-; CHECK-NEXT: [[INDVARS_IV_NEXT2]] = add nsw i64 [[INDVARS_IV1]], -1
+; CHECK-NEXT: [[INDVARS_IV1]] = phi i32 [ [[INDVARS_IV_NEXT3]], %[[PH]] ], [ [[TMP0]], %[[ENTRY]] ]
+; CHECK-NEXT: [[IV_OUTER:%.*]] = phi i32 [ [[IV_OUTER_1:%.*]], %[[PH]] ], [ [[ITER_COUNT]], %[[ENTRY]] ]
+; CHECK-NEXT: [[IV_OUTER_1]] = add nsw i32 [[IV_OUTER]], -1
+; CHECK-NEXT: [[INDVARS_IV_NEXT2:%.*]] = zext nneg i32 [[IV_OUTER_1]] to i64
; CHECK-NEXT: [[GEP_OUTER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
; CHECK-NEXT: store i8 0, ptr [[GEP_OUTER]], align 1
-; CHECK-NEXT: [[EXIT_COND_OUTER:%.*]] = icmp samesign ugt i64 [[INDVARS_IV1]], 1
+; CHECK-NEXT: [[EXIT_COND_OUTER:%.*]] = icmp samesign ugt i32 [[IV_OUTER]], 1
; CHECK-NEXT: br i1 [[EXIT_COND_OUTER]], label %[[INNER_LOOP_PREHEADER:.*]], label %[[PH]]
; CHECK: [[INNER_LOOP_PREHEADER]]:
+; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[INDVARS_IV1]] to i64
; CHECK-NEXT: br label %[[INNER_LOOP:.*]]
; CHECK: [[INNER_LOOP]]:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[INNER_LOOP_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[INNER_LOOP]] ]
; CHECK-NEXT: [[GEP_INNER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store i8 0, ptr [[GEP_INNER]], align 1
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; CHECK-NEXT: [[EXIT_COND_INNER:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], [[INDVARS_IV_NEXT2]]
-; CHECK-NEXT: br i1 [[EXIT_COND_INNER]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
; CHECK: [[EXIT:.*:]]
; CHECK-NEXT: ret i32 0
;
@@ -424,28 +428,32 @@ define i32 @iv_sext_sext_gt_slt(i32 %iter.count, ptr %ptr) {
; CHECK-LABEL: define i32 @iv_sext_sext_gt_slt(
; CHECK-SAME: i32 [[ITER_COUNT:%.*]], ptr [[PTR:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP1:%.*]] = add nsw i32 [[ITER_COUNT]], -1
; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[ITER_COUNT]] to i64
; CHECK-NEXT: br label %[[OUTER_LOOP:.*]]
; CHECK: [[PH_LOOPEXIT:.*]]:
; CHECK-NEXT: br label %[[PH:.*]]
; CHECK: [[PH]]:
+; CHECK-NEXT: [[INDVARS_IV_NEXT3:%.*]] = add i32 [[INDVARS_IV2:%.*]], -1
; CHECK-NEXT: br label %[[OUTER_LOOP]]
; CHECK: [[OUTER_LOOP]]:
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[PH]] ], [ [[TMP0]], %[[ENTRY]] ]
+; CHECK-NEXT: [[INDVARS_IV2]] = phi i32 [ [[INDVARS_IV_NEXT3]], %[[PH]] ], [ [[TMP1]], %[[ENTRY]] ]
; CHECK-NEXT: [[INDVARS_IV_NEXT2]] = add nsw i64 [[INDVARS_IV1]], -1
; CHECK-NEXT: [[GEP_OUTER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
; CHECK-NEXT: store i8 0, ptr [[GEP_OUTER]], align 1
; CHECK-NEXT: [[EXIT_COND_OUTER:%.*]] = icmp samesign ugt i64 [[INDVARS_IV1]], 1
; CHECK-NEXT: br i1 [[EXIT_COND_OUTER]], label %[[INNER_LOOP_PREHEADER:.*]], label %[[PH]]
; CHECK: [[INNER_LOOP_PREHEADER]]:
+; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[INDVARS_IV2]] to i64
; CHECK-NEXT: br label %[[INNER_LOOP:.*]]
; CHECK: [[INNER_LOOP]]:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[INNER_LOOP_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[INNER_LOOP]] ]
; CHECK-NEXT: [[GEP_INNER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store i8 0, ptr [[GEP_INNER]], align 1
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; CHECK-NEXT: [[EXIT_COND_INNER:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], [[INDVARS_IV_NEXT2]]
-; CHECK-NEXT: br i1 [[EXIT_COND_INNER]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
; CHECK: [[EXIT:.*:]]
; CHECK-NEXT: ret i32 0
;
diff --git a/llvm/test/Transforms/IndVarSimplify/pr126409.ll b/llvm/test/Transforms/IndVarSimplify/pr126409.ll
new file mode 100644
index 0000000000000..a491ce81878de
--- /dev/null
+++ b/llvm/test/Transforms/IndVarSi...
[truncated]
|
|
@llvm/pr-subscribers-llvm-analysis Author: Yingwei Zheng (dtcxzyw) ChangesThis patch relands #124270. The root cause is that we incorrectly preserve the samesign flag after truncating operands of an icmp: Patch is 21.07 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/133711.diff 5 Files Affected:
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 600a061d4435e..32e669a400c13 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -11804,8 +11804,10 @@ bool ScalarEvolution::isImpliedCond(CmpPredicate Pred, const SCEV *LHS,
MaxValue)) {
const SCEV *TruncFoundLHS = getTruncateExpr(FoundLHS, NarrowType);
const SCEV *TruncFoundRHS = getTruncateExpr(FoundRHS, NarrowType);
- if (isImpliedCondBalancedTypes(Pred, LHS, RHS, FoundPred, TruncFoundLHS,
- TruncFoundRHS, CtxI))
+ // We cannot preserve samesign after truncation.
+ if (isImpliedCondBalancedTypes(
+ Pred, LHS, RHS, static_cast<ICmpInst::Predicate>(FoundPred),
+ TruncFoundLHS, TruncFoundRHS, CtxI))
return true;
}
}
@@ -11862,15 +11864,13 @@ bool ScalarEvolution::isImpliedCondBalancedTypes(
}
// Check whether the found predicate is the same as the desired predicate.
- // FIXME: use CmpPredicate::getMatching here.
- if (FoundPred == static_cast<CmpInst::Predicate>(Pred))
- return isImpliedCondOperands(Pred, LHS, RHS, FoundLHS, FoundRHS, CtxI);
+ if (auto P = CmpPredicate::getMatching(FoundPred, Pred))
+ return isImpliedCondOperands(*P, LHS, RHS, FoundLHS, FoundRHS, CtxI);
// Check whether swapping the found predicate makes it the same as the
// desired predicate.
- // FIXME: use CmpPredicate::getMatching here.
- if (ICmpInst::getSwappedCmpPredicate(FoundPred) ==
- static_cast<CmpInst::Predicate>(Pred)) {
+ if (auto P = CmpPredicate::getMatching(
+ ICmpInst::getSwappedCmpPredicate(FoundPred), Pred)) {
// We can write the implication
// 0. LHS Pred RHS <- FoundLHS SwapPred FoundRHS
// using one of the following ways:
@@ -11881,22 +11881,23 @@ bool ScalarEvolution::isImpliedCondBalancedTypes(
// Forms 1. and 2. require swapping the operands of one condition. Don't
// do this if it would break canonical constant/addrec ordering.
if (!isa<SCEVConstant>(RHS) && !isa<SCEVAddRecExpr>(LHS))
- return isImpliedCondOperands(FoundPred, RHS, LHS, FoundLHS, FoundRHS,
- CtxI);
+ return isImpliedCondOperands(ICmpInst::getSwappedCmpPredicate(*P), RHS,
+ LHS, FoundLHS, FoundRHS, CtxI);
if (!isa<SCEVConstant>(FoundRHS) && !isa<SCEVAddRecExpr>(FoundLHS))
- return isImpliedCondOperands(Pred, LHS, RHS, FoundRHS, FoundLHS, CtxI);
+ return isImpliedCondOperands(*P, LHS, RHS, FoundRHS, FoundLHS, CtxI);
// There's no clear preference between forms 3. and 4., try both. Avoid
// forming getNotSCEV of pointer values as the resulting subtract is
// not legal.
if (!LHS->getType()->isPointerTy() && !RHS->getType()->isPointerTy() &&
- isImpliedCondOperands(FoundPred, getNotSCEV(LHS), getNotSCEV(RHS),
- FoundLHS, FoundRHS, CtxI))
+ isImpliedCondOperands(ICmpInst::getSwappedCmpPredicate(*P),
+ getNotSCEV(LHS), getNotSCEV(RHS), FoundLHS,
+ FoundRHS, CtxI))
return true;
if (!FoundLHS->getType()->isPointerTy() &&
!FoundRHS->getType()->isPointerTy() &&
- isImpliedCondOperands(Pred, LHS, RHS, getNotSCEV(FoundLHS),
+ isImpliedCondOperands(*P, LHS, RHS, getNotSCEV(FoundLHS),
getNotSCEV(FoundRHS), CtxI))
return true;
@@ -12572,14 +12573,16 @@ bool ScalarEvolution::isImpliedViaOperations(CmpPredicate Pred, const SCEV *LHS,
return false;
// We only want to work with GT comparison so far.
- if (Pred == ICmpInst::ICMP_ULT || Pred == ICmpInst::ICMP_SLT) {
+ if (ICmpInst::isLT(Pred)) {
Pred = ICmpInst::getSwappedCmpPredicate(Pred);
std::swap(LHS, RHS);
std::swap(FoundLHS, FoundRHS);
}
+ CmpInst::Predicate P = Pred.getPreferredSignedPredicate();
+
// For unsigned, try to reduce it to corresponding signed comparison.
- if (Pred == ICmpInst::ICMP_UGT)
+ if (P == ICmpInst::ICMP_UGT)
// We can replace unsigned predicate with its signed counterpart if all
// involved values are non-negative.
// TODO: We could have better support for unsigned.
@@ -12592,10 +12595,10 @@ bool ScalarEvolution::isImpliedViaOperations(CmpPredicate Pred, const SCEV *LHS,
FoundRHS) &&
isImpliedCondOperands(ICmpInst::ICMP_SGT, RHS, MinusOne, FoundLHS,
FoundRHS))
- Pred = ICmpInst::ICMP_SGT;
+ P = ICmpInst::ICMP_SGT;
}
- if (Pred != ICmpInst::ICMP_SGT)
+ if (P != ICmpInst::ICMP_SGT)
return false;
auto GetOpFromSExt = [&](const SCEV *S) {
diff --git a/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll b/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll
index 93c6bc08af2a0..4d569cc69fa2b 100644
--- a/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll
+++ b/llvm/test/Analysis/ScalarEvolution/exit-count-samesign.ll
@@ -5,9 +5,9 @@
define i32 @exit_count_samesign(i32 %iter.count, ptr %ptr) {
; CHECK-LABEL: 'exit_count_samesign'
; CHECK-NEXT: Determining loop execution counts for: @exit_count_samesign
-; CHECK-NEXT: Loop %inner.loop: backedge-taken count is (-1 + (1 smax {(-1 + %iter.count)<nsw>,+,-1}<nsw><%outer.loop>))<nsw>
+; CHECK-NEXT: Loop %inner.loop: backedge-taken count is {(-2 + %iter.count),+,-1}<nw><%outer.loop>
; CHECK-NEXT: Loop %inner.loop: constant max backedge-taken count is i32 2147483646
-; CHECK-NEXT: Loop %inner.loop: symbolic max backedge-taken count is (-1 + (1 smax {(-1 + %iter.count)<nsw>,+,-1}<nsw><%outer.loop>))<nsw>
+; CHECK-NEXT: Loop %inner.loop: symbolic max backedge-taken count is {(-2 + %iter.count),+,-1}<nw><%outer.loop>
; CHECK-NEXT: Loop %inner.loop: Trip multiple is 1
; CHECK-NEXT: Loop %outer.loop: <multiple exits> Unpredictable backedge-taken count.
; CHECK-NEXT: Loop %outer.loop: Unpredictable constant max backedge-taken count.
diff --git a/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll b/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll
index a1d30406095ec..d83301243ef30 100644
--- a/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll
+++ b/llvm/test/Analysis/ScalarEvolution/implied-via-division.ll
@@ -2,12 +2,10 @@
; RUN: opt < %s -disable-output -passes="print<scalar-evolution>" \
; RUN: -scalar-evolution-classify-expressions=0 2>&1 | FileCheck %s
-declare void @llvm.experimental.guard(i1, ...)
-
-define void @test_1(i32 %n) nounwind {
-; Prove that (n > 1) ===> (n / 2 > 0).
-; CHECK-LABEL: 'test_1'
-; CHECK-NEXT: Determining loop execution counts for: @test_1
+define void @implied1(i32 %n) {
+; Prove that (n s> 1) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied1'
+; CHECK-NEXT: Determining loop execution counts for: @implied1
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + %n.div.2)<nsw>
@@ -29,10 +27,35 @@ exit:
ret void
}
-define void @test_1neg(i32 %n) nounwind {
-; Prove that (n > 0) =\=> (n / 2 > 0).
-; CHECK-LABEL: 'test_1neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_1neg
+define void @implied1_samesign(i32 %n) {
+; Prove that (n > 1) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied1_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied1_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign ugt i32 %n, 1
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sgt i32 %n.div.2, %indvar.next
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied1_neg(i32 %n) {
+; Prove that (n s> 0) =\=> (n / 2 s> 0).
+; CHECK-LABEL: 'implied1_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied1_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
@@ -54,10 +77,10 @@ exit:
ret void
}
-define void @test_2(i32 %n) nounwind {
-; Prove that (n >= 2) ===> (n / 2 > 0).
-; CHECK-LABEL: 'test_2'
-; CHECK-NEXT: Determining loop execution counts for: @test_2
+define void @implied2(i32 %n) {
+; Prove that (n s>= 2) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied2'
+; CHECK-NEXT: Determining loop execution counts for: @implied2
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + %n.div.2)<nsw>
@@ -79,10 +102,35 @@ exit:
ret void
}
-define void @test_2neg(i32 %n) nounwind {
-; Prove that (n >= 1) =\=> (n / 2 > 0).
-; CHECK-LABEL: 'test_2neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_2neg
+define void @implied2_samesign(i32 %n) {
+; Prove that (n >= 2) ===> (n / 2 s> 0).
+; CHECK-LABEL: 'implied2_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied2_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign uge i32 %n, 2
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sgt i32 %n.div.2, %indvar.next
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied2_neg(i32 %n) {
+; Prove that (n s>= 1) =\=> (n / 2 s> 0).
+; CHECK-LABEL: 'implied2_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied2_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741822
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (-1 + (1 smax %n.div.2))<nsw>
@@ -104,10 +152,10 @@ exit:
ret void
}
-define void @test_3(i32 %n) nounwind {
-; Prove that (n > -2) ===> (n / 2 >= 0).
-; CHECK-LABEL: 'test_3'
-; CHECK-NEXT: Determining loop execution counts for: @test_3
+define void @implied3(i32 %n) {
+; Prove that (n s> -2) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied3'
+; CHECK-NEXT: Determining loop execution counts for: @implied3
; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
@@ -129,10 +177,35 @@ exit:
ret void
}
-define void @test_3neg(i32 %n) nounwind {
+define void @implied3_samesign(i32 %n) {
+; Prove that (n > -2) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied3_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied3_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign ugt i32 %n, -2
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sge i32 %n.div.2, %indvar
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied3_neg(i32 %n) {
; Prove that (n > -3) =\=> (n / 2 >= 0).
-; CHECK-LABEL: 'test_3neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_3neg
+; CHECK-LABEL: 'implied3_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied3_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
@@ -154,10 +227,10 @@ exit:
ret void
}
-define void @test_4(i32 %n) nounwind {
-; Prove that (n >= -1) ===> (n / 2 >= 0).
-; CHECK-LABEL: 'test_4'
-; CHECK-NEXT: Determining loop execution counts for: @test_4
+define void @implied4(i32 %n) {
+; Prove that (n s>= -1) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied4'
+; CHECK-NEXT: Determining loop execution counts for: @implied4
; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
@@ -179,10 +252,35 @@ exit:
ret void
}
-define void @test_4neg(i32 %n) nounwind {
-; Prove that (n >= -2) =\=> (n / 2 >= 0).
-; CHECK-LABEL: 'test_4neg'
-; CHECK-NEXT: Determining loop execution counts for: @test_4neg
+define void @implied4_samesign(i32 %n) {
+; Prove that (n >= -1) ===> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied4_samesign'
+; CHECK-NEXT: Determining loop execution counts for: @implied4_samesign
+; CHECK-NEXT: Loop %header: backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1
+; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (1 + %n.div.2)<nsw>
+; CHECK-NEXT: Loop %header: Trip multiple is 1
+;
+entry:
+ %cmp1 = icmp samesign uge i32 %n, -1
+ %n.div.2 = sdiv i32 %n, 2
+ call void @llvm.assume(i1 %cmp1)
+ br label %header
+
+header:
+ %indvar = phi i32 [ %indvar.next, %header ], [ 0, %entry ]
+ %indvar.next = add i32 %indvar, 1
+ %exitcond = icmp sge i32 %n.div.2, %indvar
+ br i1 %exitcond, label %header, label %exit
+
+exit:
+ ret void
+}
+
+define void @implied4_neg(i32 %n) {
+; Prove that (n s>= -2) =\=> (n / 2 s>= 0).
+; CHECK-LABEL: 'implied4_neg'
+; CHECK-NEXT: Determining loop execution counts for: @implied4_neg
; CHECK-NEXT: Loop %header: backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
; CHECK-NEXT: Loop %header: constant max backedge-taken count is i32 1073741824
; CHECK-NEXT: Loop %header: symbolic max backedge-taken count is (0 smax (1 + %n.div.2)<nsw>)
diff --git a/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll b/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll
index 1207f47c5e3c9..c4e26c98ed24a 100644
--- a/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll
+++ b/llvm/test/Transforms/IndVarSimplify/iv-ext-samesign.ll
@@ -68,28 +68,32 @@ define i32 @iv_zext_zext_gt_slt(i32 %iter.count, ptr %ptr) {
; CHECK-LABEL: define i32 @iv_zext_zext_gt_slt(
; CHECK-SAME: i32 [[ITER_COUNT:%.*]], ptr [[PTR:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
-; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[ITER_COUNT]] to i64
+; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[ITER_COUNT]], -1
; CHECK-NEXT: br label %[[OUTER_LOOP:.*]]
; CHECK: [[PH_LOOPEXIT:.*]]:
; CHECK-NEXT: br label %[[PH:.*]]
; CHECK: [[PH]]:
+; CHECK-NEXT: [[INDVARS_IV_NEXT3:%.*]] = add i32 [[INDVARS_IV1:%.*]], -1
; CHECK-NEXT: br label %[[OUTER_LOOP]]
; CHECK: [[OUTER_LOOP]]:
-; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[PH]] ], [ [[TMP0]], %[[ENTRY]] ]
-; CHECK-NEXT: [[INDVARS_IV_NEXT2]] = add nsw i64 [[INDVARS_IV1]], -1
+; CHECK-NEXT: [[INDVARS_IV1]] = phi i32 [ [[INDVARS_IV_NEXT3]], %[[PH]] ], [ [[TMP0]], %[[ENTRY]] ]
+; CHECK-NEXT: [[IV_OUTER:%.*]] = phi i32 [ [[IV_OUTER_1:%.*]], %[[PH]] ], [ [[ITER_COUNT]], %[[ENTRY]] ]
+; CHECK-NEXT: [[IV_OUTER_1]] = add nsw i32 [[IV_OUTER]], -1
+; CHECK-NEXT: [[INDVARS_IV_NEXT2:%.*]] = zext nneg i32 [[IV_OUTER_1]] to i64
; CHECK-NEXT: [[GEP_OUTER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
; CHECK-NEXT: store i8 0, ptr [[GEP_OUTER]], align 1
-; CHECK-NEXT: [[EXIT_COND_OUTER:%.*]] = icmp samesign ugt i64 [[INDVARS_IV1]], 1
+; CHECK-NEXT: [[EXIT_COND_OUTER:%.*]] = icmp samesign ugt i32 [[IV_OUTER]], 1
; CHECK-NEXT: br i1 [[EXIT_COND_OUTER]], label %[[INNER_LOOP_PREHEADER:.*]], label %[[PH]]
; CHECK: [[INNER_LOOP_PREHEADER]]:
+; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[INDVARS_IV1]] to i64
; CHECK-NEXT: br label %[[INNER_LOOP:.*]]
; CHECK: [[INNER_LOOP]]:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[INNER_LOOP_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[INNER_LOOP]] ]
; CHECK-NEXT: [[GEP_INNER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store i8 0, ptr [[GEP_INNER]], align 1
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; CHECK-NEXT: [[EXIT_COND_INNER:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], [[INDVARS_IV_NEXT2]]
-; CHECK-NEXT: br i1 [[EXIT_COND_INNER]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
; CHECK: [[EXIT:.*:]]
; CHECK-NEXT: ret i32 0
;
@@ -424,28 +428,32 @@ define i32 @iv_sext_sext_gt_slt(i32 %iter.count, ptr %ptr) {
; CHECK-LABEL: define i32 @iv_sext_sext_gt_slt(
; CHECK-SAME: i32 [[ITER_COUNT:%.*]], ptr [[PTR:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP1:%.*]] = add nsw i32 [[ITER_COUNT]], -1
; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[ITER_COUNT]] to i64
; CHECK-NEXT: br label %[[OUTER_LOOP:.*]]
; CHECK: [[PH_LOOPEXIT:.*]]:
; CHECK-NEXT: br label %[[PH:.*]]
; CHECK: [[PH]]:
+; CHECK-NEXT: [[INDVARS_IV_NEXT3:%.*]] = add i32 [[INDVARS_IV2:%.*]], -1
; CHECK-NEXT: br label %[[OUTER_LOOP]]
; CHECK: [[OUTER_LOOP]]:
; CHECK-NEXT: [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[PH]] ], [ [[TMP0]], %[[ENTRY]] ]
+; CHECK-NEXT: [[INDVARS_IV2]] = phi i32 [ [[INDVARS_IV_NEXT3]], %[[PH]] ], [ [[TMP1]], %[[ENTRY]] ]
; CHECK-NEXT: [[INDVARS_IV_NEXT2]] = add nsw i64 [[INDVARS_IV1]], -1
; CHECK-NEXT: [[GEP_OUTER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
; CHECK-NEXT: store i8 0, ptr [[GEP_OUTER]], align 1
; CHECK-NEXT: [[EXIT_COND_OUTER:%.*]] = icmp samesign ugt i64 [[INDVARS_IV1]], 1
; CHECK-NEXT: br i1 [[EXIT_COND_OUTER]], label %[[INNER_LOOP_PREHEADER:.*]], label %[[PH]]
; CHECK: [[INNER_LOOP_PREHEADER]]:
+; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[INDVARS_IV2]] to i64
; CHECK-NEXT: br label %[[INNER_LOOP:.*]]
; CHECK: [[INNER_LOOP]]:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[INNER_LOOP_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[INNER_LOOP]] ]
; CHECK-NEXT: [[GEP_INNER:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store i8 0, ptr [[GEP_INNER]], align 1
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; CHECK-NEXT: [[EXIT_COND_INNER:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], [[INDVARS_IV_NEXT2]]
-; CHECK-NEXT: br i1 [[EXIT_COND_INNER]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label %[[INNER_LOOP]], label %[[PH_LOOPEXIT]]
; CHECK: [[EXIT:.*:]]
; CHECK-NEXT: ret i32 0
;
diff --git a/llvm/test/Transforms/IndVarSimplify/pr126409.ll b/llvm/test/Transforms/IndVarSimplify/pr126409.ll
new file mode 100644
index 0000000000000..a491ce81878de
--- /dev/null
+++ b/llvm/test/Transforms/IndVarSi...
[truncated]
|
artagnon
reviewed
Mar 31, 2025
There was a problem hiding this comment.
Thanks for following up on this! This is quite a non-trivial fix, and the inter-diff seems obviously correct, although I can't be certain that there are other pending issues: I think you created some kind of fuzz-service for LLVM; if possible, can we fuzz the patch pre-landing?
Csmith generated 2 million tests, and all the tests passed. Is it good enough to merge? |
nikic
approved these changes
Apr 2, 2025
| TruncFoundRHS, CtxI)) | ||
| // We cannot preserve samesign after truncation. | ||
| if (isImpliedCondBalancedTypes( | ||
| Pred, LHS, RHS, static_cast<ICmpInst::Predicate>(FoundPred), |
There was a problem hiding this comment.
I kind of wonder whether we should have some more explicit API for this like FoundPred.withoutSameSign(). Doing this by going through a static_cast is a bit subtle...
There was a problem hiding this comment.
I will fix it in a follow up patch.
artagnon
approved these changes
Apr 2, 2025
dtcxzyw
added a commit
that referenced
this pull request
Apr 2, 2025
Address review comment #133711 (comment)
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.
This patch relands #124270.
Closes #126409.
The root cause is that we incorrectly preserve the samesign flag after truncating operands of an icmp:
https://alive2.llvm.org/ce/z/4NE9gS