[llvm] [IndVars] Support decrementing counters in LFTR (PR #225619)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 23 00:30:31 PDT 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
Author: Pengcheng Wang (wangpc-pp)
<details>
<summary>Changes</summary>
Allow LFTR to use integer induction variables with a step of -1.
Canonicalize analyzable countdown loop exits to equality comparisons.
Keep decrementing pointer induction variables unsupported.
Assisted-by: TRAE CLI (GPT-5.6)
---
Patch is 28.66 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/225619.diff
9 Files Affected:
- (modified) llvm/lib/Transforms/Scalar/IndVarSimplify.cpp (+21-13)
- (modified) llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll (+1-1)
- (modified) llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll (+2-2)
- (modified) llvm/test/Transforms/IndVarSimplify/X86/pr24356.ll (+2-2)
- (modified) llvm/test/Transforms/IndVarSimplify/drop-exact.ll (+2-6)
- (modified) llvm/test/Transforms/IndVarSimplify/eliminate-comparison.ll (+5-4)
- (modified) llvm/test/Transforms/IndVarSimplify/lftr.ll (+57-3)
- (modified) llvm/test/Transforms/IndVarSimplify/promote-iv-to-eliminate-casts.ll (+8-5)
- (modified) llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll (+24-15)
``````````diff
diff --git a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
index 4b1fb39ce9932d..975bba9fbe17c5 100644
--- a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
+++ b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
@@ -907,11 +907,15 @@ static bool hasConcreteDef(Value *V) {
return hasConcreteDefImpl(V, Visited, 0);
}
-/// Return true if the given phi is a "counter" in L. A counter is an
-/// add recurance (of integer or pointer type) with an arbitrary start, and a
-/// step of 1. Note that L must have exactly one latch.
-static bool isLoopCounter(PHINode* Phi, Loop *L,
- ScalarEvolution *SE) {
+static bool isUnitStride(const SCEV *Step, Type *Ty) {
+ return Step->isOne() || (Ty->isIntegerTy() && Step->isAllOnesValue());
+}
+
+/// Return true if the given phi is a "counter" in L. A counter is an add
+/// recurrence with an arbitrary start and a unit step. Integer counters may
+/// count up or down, while pointer counters may only count up. Note that L
+/// must have exactly one latch.
+static bool isLoopCounter(PHINode *Phi, Loop *L, ScalarEvolution *SE) {
assert(Phi->getParent() == L->getHeader());
assert(L->getLoopLatch());
@@ -919,7 +923,10 @@ static bool isLoopCounter(PHINode* Phi, Loop *L,
return false;
const SCEV *S = SE->getSCEV(Phi);
- if (!match(S, m_scev_AffineAddRec(m_SCEV(), m_scev_One(), m_SpecificLoop(L))))
+ const SCEV *Step;
+ if (!match(S,
+ m_scev_AffineAddRec(m_SCEV(), m_SCEV(Step), m_SpecificLoop(L))) ||
+ !isUnitStride(Step, Phi->getType()))
return false;
int LatchIdx = Phi->getBasicBlockIndex(L->getLoopLatch());
@@ -928,7 +935,7 @@ static bool isLoopCounter(PHINode* Phi, Loop *L,
isa<SCEVAddRecExpr>(SE->getSCEV(IncV)));
}
-/// Search the loop header for a loop counter (anadd rec w/step of one)
+/// Search the loop header for a loop counter (an add rec with a unit step)
/// suitable for use by LFTR. If multiple counters are available, select the
/// "best" one based profitable heuristics.
///
@@ -1021,7 +1028,8 @@ static Value *genLoopLimit(PHINode *IndVar, BasicBlock *ExitingBB,
assert(isLoopCounter(IndVar, L, SE));
assert(ExitCount->getType()->isIntegerTy() && "exit count must be integer");
const SCEVAddRecExpr *AR = cast<SCEVAddRecExpr>(SE->getSCEV(IndVar));
- assert(AR->getStepRecurrence(*SE)->isOne() && "only handles unit stride");
+ assert(isUnitStride(AR->getStepRecurrence(*SE), IndVar->getType()) &&
+ "only handles unit stride");
// For integer IVs, truncate the IV before computing the limit unless we
// know apriori that the limit must be a constant when evaluated in the
@@ -1052,7 +1060,7 @@ static Value *genLoopLimit(PHINode *IndVar, BasicBlock *ExitingBB,
}
/// This method rewrites the exit condition of the loop to be a canonical !=
-/// comparison against the incremented loop induction variable. This pass is
+/// comparison against the updated loop induction variable. This pass is
/// able to rewrite the exit tests of any loop where the SCEV analysis can
/// determine a loop-invariant trip count of the loop, which is actually a much
/// broader range than just linear tests.
@@ -1065,13 +1073,13 @@ linearFunctionTestReplace(Loop *L, BasicBlock *ExitingBB,
Instruction * const IncVar =
cast<Instruction>(IndVar->getIncomingValueForBlock(L->getLoopLatch()));
- // Initialize CmpIndVar to the preincremented IV.
+ // Initialize CmpIndVar to the pre-update IV.
Value *CmpIndVar = IndVar;
bool UsePostInc = false;
// If the exiting block is the same as the backedge block, we prefer to
- // compare against the post-incremented value, otherwise we must compare
- // against the preincremented value.
+ // compare against the post-update value, otherwise we must compare
+ // against the pre-update value.
if (ExitingBB == L->getLoopLatch()) {
// For pointer IVs, we chose to not strip inbounds which requires us not
// to add a potentially UB introducing use. We need to either a) show
@@ -1096,7 +1104,7 @@ linearFunctionTestReplace(Loop *L, BasicBlock *ExitingBB,
IndVar->getType()->isPointerTy() &&
"genLoopLimit missed a cast");
- // It may be necessary to drop nowrap flags on the incrementing instruction
+ // It may be necessary to drop nowrap flags on the IV update instruction
// if either LFTR moves from a pre-inc check to a post-inc check (in which
// case the increment might have previously been poison on the last iteration
// only) or if LFTR switches to a different IV that was previously dynamically
diff --git a/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll b/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll
index d4498baf0577a0..6016a4526612a2 100644
--- a/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll
+++ b/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll
@@ -237,7 +237,7 @@ define i32 @test4(i32 %a) {
; CHECK-NEXT: [[CONV3:%.*]] = trunc i32 [[OR]] to i8
; CHECK-NEXT: [[CALL:%.*]] = call i32 @fn1(i8 signext [[CONV3]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i32 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[CMP:%.*]] = icmp samesign ugt i32 [[INDVARS_IV_NEXT]], 242
+; CHECK-NEXT: [[CMP:%.*]] = icmp ne i32 [[INDVARS_IV_NEXT]], 242
; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]
; CHECK: for.end:
; CHECK-NEXT: ret i32 0
diff --git a/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll b/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll
index 9b9bc68ba7ad89..19469658293b7f 100644
--- a/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll
+++ b/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll
@@ -199,7 +199,7 @@ define void @f_ssub(ptr nocapture %a) {
; CHECK-NEXT: tail call void @llvm.trap(), !nosanitize [[META0]]
; CHECK-NEXT: unreachable, !nosanitize [[META0]]
; CHECK: [[CONT]]:
-; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i64 [[INDVARS_IV_NEXT]], -1
+; CHECK-NEXT: [[CMP:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], -1
; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP]]
;
entry:
@@ -291,7 +291,7 @@ define void @f_usub(ptr nocapture %a) {
; CHECK-NEXT: tail call void @llvm.trap(), !nosanitize [[META0]]
; CHECK-NEXT: unreachable, !nosanitize [[META0]]
; CHECK: [[CONT]]:
-; CHECK-NEXT: [[CMP:%.*]] = icmp samesign ugt i64 [[INDVARS_IV_NEXT]], 0
+; CHECK-NEXT: [[CMP:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 0
; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP]]
;
entry:
diff --git a/llvm/test/Transforms/IndVarSimplify/X86/pr24356.ll b/llvm/test/Transforms/IndVarSimplify/X86/pr24356.ll
index f2d938f6452d38..0cf6d39cfcaeca 100644
--- a/llvm/test/Transforms/IndVarSimplify/X86/pr24356.ll
+++ b/llvm/test/Transforms/IndVarSimplify/X86/pr24356.ll
@@ -14,8 +14,8 @@ bb:
bb4.preheader: ; preds = %bb, %bb16
; CHECK-LABEL: bb4.preheader:
%b.03 = phi i8 [ 0, %bb ], [ %tmp17, %bb16 ]
-; CHECK: %tmp9 = icmp ugt i8 %b.03, 1
-; CHECK-NOT: %tmp9 = icmp ugt i8 0, 1
+; CHECK: %exitcond = icmp eq i8 %b.03, -1
+; CHECK-NOT: %exitcond = icmp eq i8 0, -1
%tmp9 = icmp ugt i8 %b.03, 1
br i1 %tmp9, label %bb4.preheader.bb18.loopexit.split_crit_edge, label %bb4.preheader.bb4.preheader.split_crit_edge
diff --git a/llvm/test/Transforms/IndVarSimplify/drop-exact.ll b/llvm/test/Transforms/IndVarSimplify/drop-exact.ll
index 90129628496cc5..f778aa53d32e57 100644
--- a/llvm/test/Transforms/IndVarSimplify/drop-exact.ll
+++ b/llvm/test/Transforms/IndVarSimplify/drop-exact.ll
@@ -13,7 +13,6 @@ define void @drop_exact(ptr %p, ptr %p1) {
; CHECK-NEXT: ret void
; CHECK: bb12:
; CHECK-NEXT: [[TMP13:%.*]] = phi i32 [ -47436, [[BB:%.*]] ], [ [[TMP15:%.*]], [[BB12]] ]
-; CHECK-NEXT: [[TMP14:%.*]] = phi i32 [ 0, [[BB]] ], [ [[TMP42:%.*]], [[BB12]] ]
; CHECK-NEXT: [[TMP15]] = add nsw i32 [[TMP13]], -1
; CHECK-NEXT: [[TMP16:%.*]] = shl nsw i32 [[TMP15]], 1
; CHECK-NEXT: [[TMP17:%.*]] = sub nsw i32 42831, [[TMP16]]
@@ -23,8 +22,7 @@ define void @drop_exact(ptr %p, ptr %p1) {
; CHECK-NEXT: store i32 [[TMP22]], ptr [[P:%.*]], align 4
; CHECK-NEXT: [[TMP26:%.*]] = zext i32 [[TMP20]] to i64
; CHECK-NEXT: store i64 [[TMP26]], ptr [[P1:%.*]], align 4
-; CHECK-NEXT: [[TMP42]] = add nuw nsw i32 [[TMP14]], 1
-; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP42]], 719
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP15]], -48155
; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB7:%.*]], label [[BB12]]
;
bb:
@@ -60,7 +58,6 @@ define void @dont_drop_exact(ptr %p, ptr %p1) {
; CHECK-NEXT: ret void
; CHECK: bb12:
; CHECK-NEXT: [[TMP13:%.*]] = phi i32 [ -47436, [[BB:%.*]] ], [ [[TMP15:%.*]], [[BB12]] ]
-; CHECK-NEXT: [[TMP14:%.*]] = phi i32 [ 0, [[BB]] ], [ [[TMP42:%.*]], [[BB12]] ]
; CHECK-NEXT: [[TMP15]] = add nsw i32 [[TMP13]], -1
; CHECK-NEXT: [[TMP16:%.*]] = shl nsw i32 [[TMP15]], 1
; CHECK-NEXT: [[TMP17:%.*]] = sub nsw i32 42831, [[TMP16]]
@@ -70,8 +67,7 @@ define void @dont_drop_exact(ptr %p, ptr %p1) {
; CHECK-NEXT: store i32 [[TMP22]], ptr [[P:%.*]], align 4
; CHECK-NEXT: [[TMP26:%.*]] = zext i32 [[TMP20]] to i64
; CHECK-NEXT: store i64 [[TMP26]], ptr [[P1:%.*]], align 4
-; CHECK-NEXT: [[TMP42]] = add nuw nsw i32 [[TMP14]], 1
-; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP42]], 719
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP15]], -48155
; CHECK-NEXT: br i1 [[EXITCOND]], label [[BB7:%.*]], label [[BB12]]
;
bb:
diff --git a/llvm/test/Transforms/IndVarSimplify/eliminate-comparison.ll b/llvm/test/Transforms/IndVarSimplify/eliminate-comparison.ll
index 08f9856ac603d6..d7bd92705d49a6 100644
--- a/llvm/test/Transforms/IndVarSimplify/eliminate-comparison.ll
+++ b/llvm/test/Transforms/IndVarSimplify/eliminate-comparison.ll
@@ -68,6 +68,7 @@ return:
define i32 @_ZNK4llvm5APInt3ultERKS0_(i32 %tmp2.i1, ptr %tmp65, ptr %tmp73, ptr %tmp82, ptr %tmp90) {
; CHECK-LABEL: @_ZNK4llvm5APInt3ultERKS0_(
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[SMIN:%.*]] = call i32 @llvm.smin.i32(i32 [[TMP2_I1:%.*]], i32 -1)
; CHECK-NEXT: br label [[BB18:%.*]]
; CHECK: bb13:
; CHECK-NEXT: [[TMP66:%.*]] = load ptr, ptr [[TMP65:%.*]], align 4
@@ -88,11 +89,11 @@ define i32 @_ZNK4llvm5APInt3ultERKS0_(i32 %tmp2.i1, ptr %tmp65, ptr %tmp73, ptr
; CHECK-NEXT: [[TMP95:%.*]] = icmp ult i64 [[TMP86]], [[TMP94]]
; CHECK-NEXT: br i1 [[TMP95]], label [[BB20_LOOPEXIT]], label [[BB17:%.*]]
; CHECK: bb17:
-; CHECK-NEXT: [[TMP97:%.*]] = add nsw i32 [[I]], -1
+; CHECK-NEXT: [[TMP97:%.*]] = add i32 [[I]], -1
; CHECK-NEXT: br label [[BB18]]
; CHECK: bb18:
-; CHECK-NEXT: [[I]] = phi i32 [ [[TMP2_I1:%.*]], [[ENTRY:%.*]] ], [ [[TMP97]], [[BB17]] ]
-; CHECK-NEXT: [[TMP99:%.*]] = icmp sgt i32 [[I]], -1
+; CHECK-NEXT: [[I]] = phi i32 [ [[TMP2_I1]], [[ENTRY:%.*]] ], [ [[TMP97]], [[BB17]] ]
+; CHECK-NEXT: [[TMP99:%.*]] = icmp ne i32 [[I]], [[SMIN]]
; CHECK-NEXT: br i1 [[TMP99]], label [[BB13:%.*]], label [[BB20_LOOPEXIT]]
; CHECK: bb20.loopexit:
; CHECK-NEXT: [[TMP_0_PH:%.*]] = phi i32 [ 0, [[BB18]] ], [ 1, [[BB15]] ], [ 0, [[BB13]] ]
@@ -917,7 +918,7 @@ define void @func_24(ptr %init.ptr) {
; CHECK-NEXT: br i1 true, label [[BE]], label [[LEAVE_LOOPEXIT:%.*]]
; CHECK: be:
; CHECK-NEXT: call void @side_effect()
-; CHECK-NEXT: [[BE_COND:%.*]] = icmp sgt i32 [[IV_DEC]], 4
+; CHECK-NEXT: [[BE_COND:%.*]] = icmp ne i32 [[IV_DEC]], 4
; CHECK-NEXT: br i1 [[BE_COND]], label [[LOOP]], label [[LEAVE_LOOPEXIT]]
; CHECK: leave.loopexit:
; CHECK-NEXT: br label [[LEAVE]]
diff --git a/llvm/test/Transforms/IndVarSimplify/lftr.ll b/llvm/test/Transforms/IndVarSimplify/lftr.ll
index 5ee62ba357ab6f..a55b6381db2865 100644
--- a/llvm/test/Transforms/IndVarSimplify/lftr.ll
+++ b/llvm/test/Transforms/IndVarSimplify/lftr.ll
@@ -34,7 +34,7 @@ loopexit:
ret i32 %i
}
-; TODO: we should be able to convert the subtract into a post-decrement check
+;; Convert a pre-decrement check on the latch into a post-decrement check
define i32 @pre_to_post_sub() {
; CHECK-LABEL: @pre_to_post_sub(
; CHECK-NEXT: entry:
@@ -43,8 +43,8 @@ define i32 @pre_to_post_sub() {
; CHECK-NEXT: [[I:%.*]] = phi i32 [ 1000, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[I_NEXT]] = sub nsw i32 [[I]], 1
; CHECK-NEXT: store i32 [[I]], ptr @A, align 4
-; CHECK-NEXT: [[C:%.*]] = icmp samesign ugt i32 [[I]], 0
-; CHECK-NEXT: br i1 [[C]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i32 [[I_NEXT]], -1
+; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[LOOPEXIT:%.*]]
; CHECK: loopexit:
; CHECK-NEXT: ret i32 0
;
@@ -62,6 +62,60 @@ loopexit:
ret i32 %i
}
+;; Do not rewrite non-unit stride loops.
+define void @non_unit_stride() {
+; CHECK-LABEL: @non_unit_stride(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: [[I:%.*]] = phi i32 [ 1000, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[I_NEXT]] = sub nsw i32 [[I]], 2
+; CHECK-NEXT: store i32 [[I]], ptr @A, align 4
+; CHECK-NEXT: [[C:%.*]] = icmp samesign ugt i32 [[I]], 1
+; CHECK-NEXT: br i1 [[C]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK: loopexit:
+; CHECK-NEXT: ret void
+;
+entry:
+ br label %loop
+
+loop:
+ %i = phi i32 [ 1000, %entry ], [ %i.next, %loop ]
+ %i.next = sub i32 %i, 2
+ store i32 %i, ptr @A
+ %c = icmp ugt i32 %i, 1
+ br i1 %c, label %loop, label %loopexit
+
+loopexit:
+ ret void
+}
+
+;; Do not use a decrementing pointer IV as the LFTR loop counter.
+define void @decrementing_pointer(ptr %start) {
+; CHECK-LABEL: @decrementing_pointer(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: [[P:%.*]] = phi ptr [ [[START:%.*]], [[ENTRY:%.*]] ], [ [[P_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[P_NEXT]] = getelementptr i8, ptr [[P]], i64 -1
+; CHECK-NEXT: [[C:%.*]] = icmp ugt ptr [[P]], null
+; CHECK-NEXT: br i1 [[C]], label [[LOOP]], label [[LOOPEXIT:%.*]]
+; CHECK: loopexit:
+; CHECK-NEXT: ret void
+;
+entry:
+ br label %loop
+
+loop:
+ %p = phi ptr [ %start, %entry ], [ %p.next, %loop ]
+ %p.next = getelementptr i8, ptr %p, i64 -1
+ %c = icmp ugt ptr %p, null
+ br i1 %c, label %loop, label %loopexit
+
+loopexit:
+ ret void
+}
+
; LFTR should eliminate the need for the computation of i*i completely. It
diff --git a/llvm/test/Transforms/IndVarSimplify/promote-iv-to-eliminate-casts.ll b/llvm/test/Transforms/IndVarSimplify/promote-iv-to-eliminate-casts.ll
index 3f0ada281b1e34..b0c89df6b0a4a4 100644
--- a/llvm/test/Transforms/IndVarSimplify/promote-iv-to-eliminate-casts.ll
+++ b/llvm/test/Transforms/IndVarSimplify/promote-iv-to-eliminate-casts.ll
@@ -197,7 +197,8 @@ define void @promote_latch_condition_decrementing_loop_01(ptr %p, ptr %a) {
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
+; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], -1
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
;
@@ -241,7 +242,7 @@ define void @promote_latch_condition_decrementing_loop_02(ptr %p, ptr %a) {
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], -1
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
;
@@ -285,7 +286,7 @@ define void @promote_latch_condition_decrementing_loop_03(ptr %p, ptr %a) {
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], -1
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
;
@@ -337,7 +338,8 @@ define void @promote_latch_condition_decrementing_loop_04(ptr %p, ptr %a, i1 %co
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
+; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], -1
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
;
@@ -399,7 +401,8 @@ define void @promote_latch_condition_decrementing_loop_05(ptr %p, ptr %a, i1 %co
; CHECK-NEXT: [[EL:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store atomic i32 0, ptr [[EL]] unordered, align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp slt i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
+; CHECK-NEXT: [[LOOPCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], -1
; CHECK-NEXT: br i1 [[LOOPCOND]], label [[LOOPEXIT_LOOPEXIT:%.*]], label [[LOOP]]
;
diff --git a/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll b/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll
index d24442287cf4d7..0ef95d1414d174 100644
--- a/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll
+++ b/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll
@@ -20,7 +20,7 @@ define void @zext_postinc_constant_start(ptr %A) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp samesign ugt i64 [[INDVARS_IV_NEXT]], 6
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT:%.*]]
; CHECK: exit:
; CHECK-NEXT: ret void
@@ -51,7 +51,7 @@ define void @zext_preinc_constant_start(ptr %A) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp samesign ugt i64 [[INDVARS_IV]], 6
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 5
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT:%.*]]
; CHECK: exit:
; CHECK-NEXT: ret void
@@ -119,13 +119,16 @@ define void @zext_preinc(ptr %A, i32 %start) {
; CHECK-NEXT: br i1 [[NONPOS]], label [[EXIT:%.*]], label [[FOR_BODY_PREHEADER:%.*]]
; CHECK: for.body.preheader:
; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[START]] to i64
+; CHECK-NEXT: [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[START]], i32 6)
+; CHECK-NEXT: [[TMP1:%.*]] = add nsw i32 [[UMIN]], -1
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/225619
More information about the llvm-commits
mailing list