Index: llvm/lib/Transforms/Utils/LoopUtils.cpp =================================================================== --- llvm/lib/Transforms/Utils/LoopUtils.cpp +++ llvm/lib/Transforms/Utils/LoopUtils.cpp @@ -1535,7 +1535,7 @@ // Use this type for pointer arithmetic. Type *PtrArithTy = Type::getInt8PtrTy(Ctx, AS); - if (SE->isLoopInvariant(Sc, TheLoop)) { + if (SE->isLoopInvariant(Sc, TheLoop) && (CG->Low == CG->High)) { LLVM_DEBUG(dbgs() << "LAA: Adding RT check for a loop invariant ptr:" << *Ptr << "\n"); // Ptr could be in the loop body. If so, expand a new one at the correct Index: llvm/test/Transforms/LoopVectorize/pr50686.ll =================================================================== --- llvm/test/Transforms/LoopVectorize/pr50686.ll +++ llvm/test/Transforms/LoopVectorize/pr50686.ll @@ -1,32 +1,23 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py ; RUN: opt -loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S < %s | FileCheck %s -@k = common dso_local local_unnamed_addr global i32 0, align 4 - +; %p2 checking group has multiple invariant members with different addresses. define void @m(i32* nocapture %p, i32* nocapture %p2, i32 %q) { ; CHECK-LABEL: @m( ; CHECK-NEXT: entry: ; CHECK-NEXT: [[P1:%.*]] = bitcast i32* [[P:%.*]] to i8* -; CHECK-NEXT: [[I:%.*]] = load i32, i32* @k, align 4 -; CHECK-NEXT: [[CMP32:%.*]] = icmp slt i32 [[I]], [[Q:%.*]] -; CHECK-NEXT: br i1 [[CMP32]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND2_PREHEADER:%.*]] -; CHECK: for.body.preheader: -; CHECK-NEXT: br label [[FOR_BODY:%.*]] -; CHECK: for.cond2.preheader.loopexit: -; CHECK-NEXT: br label [[FOR_COND2_PREHEADER]] +; CHECK-NEXT: [[P23:%.*]] = bitcast i32* [[P2:%.*]] to i8* +; CHECK-NEXT: br label [[FOR_COND2_PREHEADER:%.*]] ; CHECK: for.cond2.preheader: -; CHECK-NEXT: [[ARRAYIDX9_1:%.*]] = getelementptr inbounds i32, i32* [[P2:%.*]], i64 1 +; CHECK-NEXT: [[ARRAYIDX9_1:%.*]] = getelementptr inbounds i32, i32* [[P2]], i64 1 ; CHECK-NEXT: [[ARRAYIDX9_2:%.*]] = getelementptr inbounds i32, i32* [[P2]], i64 2 +; CHECK-NEXT: [[ARRAYIDX9_24:%.*]] = bitcast i32* [[ARRAYIDX9_2]] to i8* ; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] ; CHECK: vector.memcheck: ; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[P]], i64 63 ; CHECK-NEXT: [[SCEVGEP2:%.*]] = bitcast i32* [[SCEVGEP]] to i8* -; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i32, i32* [[P2]], i64 2 -; CHECK-NEXT: [[SCEVGEP34:%.*]] = bitcast i32* [[SCEVGEP3]] to i8* -; CHECK-NEXT: [[UGLYGEP:%.*]] = getelementptr i8, i8* [[SCEVGEP34]], i64 1 -; CHECK-NEXT: [[BC:%.*]] = bitcast i32* [[ARRAYIDX9_2]] to i8* -; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[P1]], [[UGLYGEP]] -; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[BC]], [[SCEVGEP2]] +; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[P1]], [[ARRAYIDX9_24]] +; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[P23]], [[SCEVGEP2]] ; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] ; CHECK-NEXT: [[MEMCHECK_CONFLICT:%.*]] = and i1 [[FOUND_CONFLICT]], true ; CHECK-NEXT: br i1 [[MEMCHECK_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] @@ -60,16 +51,6 @@ ; CHECK: scalar.ph: ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 60, [[MIDDLE_BLOCK]] ], [ 0, [[FOR_COND2_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ] ; CHECK-NEXT: br label [[FOR_COND5_PREHEADER:%.*]] -; CHECK: for.body: -; CHECK-NEXT: [[I1:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[I]], [[FOR_BODY_PREHEADER]] ] -; CHECK-NEXT: [[IDXPROM:%.*]] = sext i32 [[I1]] to i64 -; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[P2]], i64 [[IDXPROM]] -; CHECK-NEXT: store i32 2, i32* [[ARRAYIDX]], align 4 -; CHECK-NEXT: [[I2:%.*]] = load i32, i32* @k, align 4 -; CHECK-NEXT: [[INC]] = add nsw i32 [[I2]], 1 -; CHECK-NEXT: store i32 [[INC]], i32* @k, align 4 -; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[INC]], [[Q]] -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND2_PREHEADER_LOOPEXIT:%.*]] ; CHECK: for.cond5.preheader: ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_COND5_PREHEADER]] ] ; CHECK-NEXT: [[I3:%.*]] = load i32, i32* [[P2]], align 4 @@ -87,31 +68,105 @@ ; CHECK-NEXT: ret void ; entry: - %i = load i32, i32* @k, align 4 - %cmp32 = icmp slt i32 %i, %q - br i1 %cmp32, label %for.body.preheader, label %for.cond2.preheader - -for.body.preheader: ; preds = %entry - br label %for.body - -for.cond2.preheader.loopexit: ; preds = %for.body br label %for.cond2.preheader -for.cond2.preheader: ; preds = %for.cond2.preheader.loopexit, %entry +for.cond2.preheader: ; preds = %entry %arrayidx9.1 = getelementptr inbounds i32, i32* %p2, i64 1 %arrayidx9.2 = getelementptr inbounds i32, i32* %p2, i64 2 br label %for.cond5.preheader -for.body: ; preds = %for.body, %for.body.preheader - %i1 = phi i32 [ %inc, %for.body ], [ %i, %for.body.preheader ] - %idxprom = sext i32 %i1 to i64 - %arrayidx = getelementptr inbounds i32, i32* %p2, i64 %idxprom - store i32 2, i32* %arrayidx, align 4 - %i2 = load i32, i32* @k, align 4 - %inc = add nsw i32 %i2, 1 - store i32 %inc, i32* @k, align 4 - %cmp = icmp slt i32 %inc, %q - br i1 %cmp, label %for.body, label %for.cond2.preheader.loopexit +for.cond5.preheader: ; preds = %for.cond5.preheader, %for.cond2.preheader + %indvars.iv = phi i64 [ 0, %for.cond2.preheader ], [ %indvars.iv.next, %for.cond5.preheader ] + %i3 = load i32, i32* %p2, align 4 + %sub = sub nsw i32 0, %i3 + %i4 = load i32, i32* %arrayidx9.1, align 4 + %sub.1 = sub nsw i32 %sub, %i4 + %i5 = load i32, i32* %arrayidx9.2, align 4 + %sub.2 = sub nsw i32 %sub.1, %i5 + %arrayidx14 = getelementptr inbounds i32, i32* %p, i64 %indvars.iv + store i32 %sub.2, i32* %arrayidx14, align 4 + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 + %exitcond = icmp eq i64 %indvars.iv.next, 63 + br i1 %exitcond, label %for.end17, label %for.cond5.preheader + +for.end17: ; preds = %for.cond5.preheader + ret void +} + +; %p2 checking group has multiple invariant members with the same addresses. +define void @multiple_member_cg(i32* nocapture %p, i32* nocapture %p2, i32 %q) { +; CHECK-LABEL: @multiple_member_cg( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[P1:%.*]] = bitcast i32* [[P:%.*]] to i8* +; CHECK-NEXT: br label [[FOR_COND2_PREHEADER:%.*]] +; CHECK: for.cond2.preheader: +; CHECK-NEXT: [[ARRAYIDX9_1:%.*]] = getelementptr inbounds i32, i32* [[P2:%.*]], i64 0 +; CHECK-NEXT: [[ARRAYIDX9_13:%.*]] = bitcast i32* [[ARRAYIDX9_1]] to i8* +; CHECK-NEXT: [[ARRAYIDX9_2:%.*]] = getelementptr inbounds i32, i32* [[P2]], i64 0 +; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] +; CHECK: vector.memcheck: +; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i32, i32* [[P]], i64 63 +; CHECK-NEXT: [[SCEVGEP2:%.*]] = bitcast i32* [[SCEVGEP]] to i8* +; CHECK-NEXT: [[UGLYGEP:%.*]] = getelementptr i8, i8* [[ARRAYIDX9_13]], i64 1 +; CHECK-NEXT: [[BC:%.*]] = bitcast i32* [[ARRAYIDX9_2]] to i8* +; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult i8* [[P1]], [[UGLYGEP]] +; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult i8* [[BC]], [[SCEVGEP2]] +; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] +; CHECK-NEXT: [[MEMCHECK_CONFLICT:%.*]] = and i1 [[FOUND_CONFLICT]], true +; CHECK-NEXT: br i1 [[MEMCHECK_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] +; CHECK: vector.ph: +; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] +; CHECK: vector.body: +; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] +; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 +; CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[P2]], align 4, !alias.scope !8 +; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[TMP1]], i32 0 +; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer +; CHECK-NEXT: [[TMP2:%.*]] = sub nsw <4 x i32> zeroinitializer, [[BROADCAST_SPLAT]] +; CHECK-NEXT: [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX9_1]], align 4, !alias.scope !8 +; CHECK-NEXT: [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <4 x i32> poison, i32 [[TMP3]], i32 0 +; CHECK-NEXT: [[BROADCAST_SPLAT5:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT4]], <4 x i32> poison, <4 x i32> zeroinitializer +; CHECK-NEXT: [[TMP4:%.*]] = sub nsw <4 x i32> [[TMP2]], [[BROADCAST_SPLAT5]] +; CHECK-NEXT: [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX9_2]], align 4, !alias.scope !8 +; CHECK-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <4 x i32> poison, i32 [[TMP5]], i32 0 +; CHECK-NEXT: [[BROADCAST_SPLAT7:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT6]], <4 x i32> poison, <4 x i32> zeroinitializer +; CHECK-NEXT: [[TMP6:%.*]] = sub nsw <4 x i32> [[TMP4]], [[BROADCAST_SPLAT7]] +; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 [[TMP0]] +; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[TMP7]], i32 0 +; CHECK-NEXT: [[TMP9:%.*]] = bitcast i32* [[TMP8]] to <4 x i32>* +; CHECK-NEXT: store <4 x i32> [[TMP6]], <4 x i32>* [[TMP9]], align 4, !alias.scope !11, !noalias !8 +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 +; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 60 +; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] +; CHECK: middle.block: +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 63, 60 +; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END17:%.*]], label [[SCALAR_PH]] +; CHECK: scalar.ph: +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 60, [[MIDDLE_BLOCK]] ], [ 0, [[FOR_COND2_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ] +; CHECK-NEXT: br label [[FOR_COND5_PREHEADER:%.*]] +; CHECK: for.cond5.preheader: +; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_COND5_PREHEADER]] ] +; CHECK-NEXT: [[I3:%.*]] = load i32, i32* [[P2]], align 4 +; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 0, [[I3]] +; CHECK-NEXT: [[I4:%.*]] = load i32, i32* [[ARRAYIDX9_1]], align 4 +; CHECK-NEXT: [[SUB_1:%.*]] = sub nsw i32 [[SUB]], [[I4]] +; CHECK-NEXT: [[I5:%.*]] = load i32, i32* [[ARRAYIDX9_2]], align 4 +; CHECK-NEXT: [[SUB_2:%.*]] = sub nsw i32 [[SUB_1]], [[I5]] +; CHECK-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds i32, i32* [[P]], i64 [[INDVARS_IV]] +; CHECK-NEXT: store i32 [[SUB_2]], i32* [[ARRAYIDX14]], align 4 +; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 +; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 63 +; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END17]], label [[FOR_COND5_PREHEADER]], !llvm.loop [[LOOP14:![0-9]+]] +; CHECK: for.end17: +; CHECK-NEXT: ret void +; +entry: + br label %for.cond2.preheader + +for.cond2.preheader: ; preds = %entry + %arrayidx9.1 = getelementptr inbounds i32, i32* %p2, i64 0 + %arrayidx9.2 = getelementptr inbounds i32, i32* %p2, i64 0 + br label %for.cond5.preheader for.cond5.preheader: ; preds = %for.cond5.preheader, %for.cond2.preheader %indvars.iv = phi i64 [ 0, %for.cond2.preheader ], [ %indvars.iv.next, %for.cond5.preheader ]