[llvm] [VPlan] Fix miscompile when widening induction with loop-variant step (PR #206010)

Florian Hahn via llvm-commits llvm-commits at lists.llvm.org
Mon Aug 10 00:05:35 PDT 2026


================
@@ -0,0 +1,265 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes=loop-vectorize -S -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Loop with i8 induction using a variant step (zext of comparison).
+; InductionDescriptor classifies %iv as IK_IntInduction with step=1,
+; but the VPlan backedge step is loop-variant (zext i1 %cmp to i8).
+; Without the fix, ReplaceExtractsWithExitingIVValue replaces the iv
+; exit value with 'trunc i32 %n.vec to i8' (the vector iteration count)
+; instead of the actual last iv value, producing an off-by-one result.
+define { i32, i32 } @variant_step_induction(i8 %a, i32 %b) {
+; CHECK-LABEL: define { i32, i32 } @variant_step_induction(
+; CHECK-SAME: i8 [[A:%.*]], i32 [[B:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = zext i8 [[A]] to i32
+; CHECK-NEXT:    [[TMP1:%.*]] = add nuw nsw i32 [[TMP0]], 1
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP1]], 8
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[TMP1]], 8
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP1]], [[N_MOD_VF]]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i8> poison, i8 [[A]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i8> [[BROADCAST_SPLATINSERT]], <4 x i8> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i32> poison, i32 [[B]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT1]], <4 x i32> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i8> [ <i8 0, i8 1, i8 2, i8 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP6:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI3:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP7:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[STEP_ADD:%.*]] = add <4 x i8> [[VEC_IND]], splat (i8 4)
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ult <4 x i8> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <4 x i8> [[STEP_ADD]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP4:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[BROADCAST_SPLAT2]], <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = select <4 x i1> [[TMP3]], <4 x i32> [[BROADCAST_SPLAT2]], <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP6]] = add <4 x i32> [[TMP4]], [[VEC_PHI]]
+; CHECK-NEXT:    [[TMP7]] = add <4 x i32> [[TMP5]], [[VEC_PHI3]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i8> [[STEP_ADD]], splat (i8 4)
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP9:%.*]] = zext <4 x i1> [[TMP3]] to <4 x i8>
+; CHECK-NEXT:    [[TMP10:%.*]] = add <4 x i8> [[STEP_ADD]], [[TMP9]]
+; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <4 x i32> [[TMP7]], [[TMP6]]
+; CHECK-NEXT:    [[TMP11:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
+; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <4 x i8> [[TMP10]], i64 3
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP1]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
+; CHECK:       [[SCALAR_PH]]:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i8 [ [[TMP12]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP11]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[RDX:%.*]] = phi i32 [ [[BC_MERGE_RDX]], %[[SCALAR_PH]] ], [ [[RDX_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i8 [[IV]], [[A]]
+; CHECK-NEXT:    [[ADD:%.*]] = select i1 [[CMP]], i32 [[B]], i32 0
+; CHECK-NEXT:    [[RDX_NEXT]] = add nsw i32 [[ADD]], [[RDX]]
+; CHECK-NEXT:    [[INC:%.*]] = zext i1 [[CMP]] to i8
+; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], [[INC]]
+; CHECK-NEXT:    br i1 [[CMP]], label %[[LOOP]], label %[[EXIT]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    [[RDX_NEXT_LCSSA:%.*]] = phi i32 [ [[RDX_NEXT]], %[[LOOP]] ], [ [[TMP11]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    [[IV_NEXT_LCSSA:%.*]] = phi i8 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP12]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    [[CONV:%.*]] = zext i8 [[IV_NEXT_LCSSA]] to i32
+; CHECK-NEXT:    [[R0:%.*]] = insertvalue { i32, i32 } poison, i32 [[CONV]], 0
+; CHECK-NEXT:    [[R1:%.*]] = insertvalue { i32, i32 } [[R0]], i32 [[RDX_NEXT_LCSSA]], 1
+; CHECK-NEXT:    ret { i32, i32 } [[R1]]
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i8 [ 0, %entry ], [ %iv.next, %loop ]
+  %rdx = phi i32 [ 0, %entry ], [ %rdx.next, %loop ]
+  %cmp = icmp ult i8 %iv, %a
+  %add = select i1 %cmp, i32 %b, i32 0
+  %rdx.next = add nsw i32 %add, %rdx
+  %inc = zext i1 %cmp to i8
+  %iv.next = add i8 %iv, %inc
+  br i1 %cmp, label %loop, label %exit
+
+exit:
+  %conv = zext i8 %iv.next to i32
+  %r0 = insertvalue { i32, i32 } poison, i32 %conv, 0
+  %r1 = insertvalue { i32, i32 } %r0, i32 %rdx.next, 1
+  ret { i32, i32 } %r1
----------------
fhahn wrote:

simpler to add the values and return an i32?

https://github.com/llvm/llvm-project/pull/206010


More information about the llvm-commits mailing list