Skip to content

[LV] Ensure getScaledReductions only matches extends inside the loop #148264

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Jul 15, 2025
+262 −1
Merged

[LV] Ensure getScaledReductions only matches extends inside the loop #148264

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e40436c
[LV] Ensure getScaledReductions only matches extends inside the loop
david-arm Jul 11, 2025
58356cc
Address review comments
david-arm Jul 14, 2025
edc5d91
Address review comments
david-arm Jul 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8112,14 +8112,19 @@ bool VPRecipeBuilder::getScaledReductions(
std::optional<unsigned> BinOpc;
Type *ExtOpTypes[2] = {nullptr};

auto CollectExtInfo = [&Exts,
auto CollectExtInfo = [this, &Exts,
&ExtOpTypes](SmallVectorImpl<Value *> &Ops) -> bool {
unsigned I = 0;
for (Value *OpI : Ops) {
Value *ExtOp;
if (!match(OpI, m_ZExtOrSExt(m_Value(ExtOp))))
return false;
Exts[I] = cast<Instruction>(OpI);

// TODO: We should be able to support live-ins.
if (!CM.TheLoop->contains(Exts[I]))
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a reason it is difficult/not possible to support live-in operands?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe it is possible, but given we're close to the branch point I thought it might be better to disable it first and look at adding support for it afterwards. What do you think?

I think the problem lies in the cost model (specifically VPPartialReduction::computeCost) and some assumptions made about the ordering of PHI operands.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, TODO seems fine for now, although it would ideally be fixed at the source.

return false;

ExtOpTypes[I] = ExtOp->getType();
I++;
}
Expand Down
256 changes: 256 additions & 0 deletions llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -911,6 +911,262 @@ for.exit: ; preds = %for.body
ret i32 %add
}


define i32 @add_of_zext_outside_loop(i32 %a, ptr noalias %b, i8 %c, i32 %d) #0 {
; CHECK-INTERLEAVE1-LABEL: define i32 @add_of_zext_outside_loop(
; CHECK-INTERLEAVE1-SAME: i32 [[A:%.*]], ptr noalias [[B:%.*]], i8 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR0]] {
; CHECK-INTERLEAVE1-NEXT: entry:
; CHECK-INTERLEAVE1-NEXT: [[CONV1:%.*]] = zext i8 [[C]] to i32
; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = sub i32 1024, [[D]]
; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 16
; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-INTERLEAVE1: vector.ph:
; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 16
; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = insertelement <16 x i32> zeroinitializer, i32 [[A]], i32 0
; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32> poison, i32 [[CONV1]], i64 0
; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer
; CHECK-INTERLEAVE1-NEXT: br label [[FOR_BODY:%.*]]
; CHECK-INTERLEAVE1: vector.body:
; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ [[TMP2]], [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[FOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i32 0
; CHECK-INTERLEAVE1-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP4]], align 1
; CHECK-INTERLEAVE1-NEXT: [[TMP5]] = add <16 x i32> [[VEC_PHI]], [[BROADCAST_SPLAT]]
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
; CHECK-INTERLEAVE1: middle.block:
; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP5]])
; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-INTERLEAVE1: scalar.ph:
;
; CHECK-INTERLEAVED-LABEL: define i32 @add_of_zext_outside_loop(
; CHECK-INTERLEAVED-SAME: i32 [[A:%.*]], ptr noalias [[B:%.*]], i8 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR0]] {
; CHECK-INTERLEAVED-NEXT: entry:
; CHECK-INTERLEAVED-NEXT: [[CONV1:%.*]] = zext i8 [[C]] to i32
; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = sub i32 1024, [[D]]
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = icmp ult i32 [[TMP2]], 32
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP11]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-INTERLEAVED: vector.ph:
; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP2]], 32
; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP2]], [[N_MOD_VF]]
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = insertelement <16 x i32> zeroinitializer, i32 [[A]], i32 0
; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i32> poison, i32 [[CONV1]], i64 0
; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer
; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-INTERLEAVED: vector.body:
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ [[TMP10]], [[VECTOR_PH]] ], [ [[TMP6:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP7:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[VEC_PHI1]]
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i32 0
; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i32 16
; CHECK-INTERLEAVED-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP4]], align 1
; CHECK-INTERLEAVED-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP5]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP6]] = add <16 x i32> [[VEC_PHI]], [[BROADCAST_SPLAT]]
; CHECK-INTERLEAVED-NEXT: [[TMP7]] = add <16 x i32> [[VEC_PHI2]], [[BROADCAST_SPLAT]]
; CHECK-INTERLEAVED-NEXT: [[TMP21]] = add nuw i32 [[VEC_PHI1]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i32 [[TMP21]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP7]], [[TMP6]]
; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX]])
; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-INTERLEAVED: scalar.ph:
;
; CHECK-MAXBW-LABEL: define i32 @add_of_zext_outside_loop(
; CHECK-MAXBW-SAME: i32 [[A:%.*]], ptr noalias [[B:%.*]], i8 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR0]] {
; CHECK-MAXBW-NEXT: entry:
; CHECK-MAXBW-NEXT: [[CONV1:%.*]] = zext i8 [[C]] to i32
; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = sub i32 1024, [[D]]
; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw i32 [[TMP1]], 16
; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], [[TMP2]]
; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = mul nuw i32 [[TMP3]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], [[TMP4]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = mul nuw i32 [[TMP5]], 16
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = insertelement <vscale x 16 x i32> zeroinitializer, i32 [[A]], i32 0
; CHECK-MAXBW-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[CONV1]], i64 0
; CHECK-MAXBW-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
; CHECK-MAXBW-NEXT: br label [[FOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 16 x i32> [ [[TMP8]], [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[FOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[INDEX]]
; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[TMP9]], i32 0
; CHECK-MAXBW-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP10]], align 1
; CHECK-MAXBW-NEXT: [[TMP11]] = add <vscale x 16 x i32> [[VEC_PHI]], [[BROADCAST_SPLAT]]
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP6]]
; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i32 @llvm.vector.reduce.add.nxv16i32(<vscale x 16 x i32> [[TMP11]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-MAXBW: scalar.ph:
;
entry:
%conv1 = zext i8 %c to i32
br label %for.body

for.body:
%iv = phi i32 [ %d, %entry ], [ %iv.next, %for.body ]
%rdx = phi i32 [ %a, %entry ], [ %rdx.next, %for.body ]
%arrayidx = getelementptr inbounds i8, ptr %b, i32 %iv
store i8 0, ptr %arrayidx, align 1
%rdx.next = add nsw i32 %rdx, %conv1
%iv.next = add nuw nsw i32 %iv, 1
%cmp = icmp eq i32 %iv.next, 1024
br i1 %cmp, label %exit, label %for.body

exit:
%add.lcssa = phi i32 [ %rdx.next, %for.body ]
ret i32 %add.lcssa
}

define i32 @add_of_loop_invariant_zext(i32 %a, ptr %b, i8 %c, i32 %d) #0 {
; CHECK-INTERLEAVE1-LABEL: define i32 @add_of_loop_invariant_zext(
; CHECK-INTERLEAVE1-SAME: i32 [[A:%.*]], ptr [[B:%.*]], i8 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR0]] {
; CHECK-INTERLEAVE1-NEXT: entry:
; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = sub i32 1024, [[D]]
; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 16
; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-INTERLEAVE1: vector.ph:
; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], 16
; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[C]], i64 0
; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer
; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = insertelement <16 x i32> zeroinitializer, i32 [[A]], i32 0
; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
; CHECK-INTERLEAVE1-NEXT: br label [[FOR_BODY:%.*]]
; CHECK-INTERLEAVE1: vector.body:
; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ [[TMP2]], [[VECTOR_PH]] ], [ [[TMP6:%.*]], [[FOR_BODY]] ]
; CHECK-INTERLEAVE1-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[INDEX]]
; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 0
; CHECK-INTERLEAVE1-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP5]], align 1
; CHECK-INTERLEAVE1-NEXT: [[TMP6]] = add <16 x i32> [[VEC_PHI]], [[TMP3]]
; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 16
; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
; CHECK-INTERLEAVE1: middle.block:
; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP6]])
; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-INTERLEAVE1: scalar.ph:
;
; CHECK-INTERLEAVED-LABEL: define i32 @add_of_loop_invariant_zext(
; CHECK-INTERLEAVED-SAME: i32 [[A:%.*]], ptr [[B:%.*]], i8 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR0]] {
; CHECK-INTERLEAVED-NEXT: entry:
; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = sub i32 1024, [[D]]
; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = icmp ult i32 [[TMP2]], 32
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP11]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-INTERLEAVED: vector.ph:
; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP2]], 32
; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP2]], [[N_MOD_VF]]
; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[C]], i64 0
; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer
; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = insertelement <16 x i32> zeroinitializer, i32 [[A]], i32 0
; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[BROADCAST_SPLAT]] to <16 x i32>
; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-INTERLEAVED: vector.body:
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ [[TMP12]], [[VECTOR_PH]] ], [ [[TMP7:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ]
; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[VEC_PHI1]]
; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 0
; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 16
; CHECK-INTERLEAVED-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP5]], align 1
; CHECK-INTERLEAVED-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP6]], align 1
; CHECK-INTERLEAVED-NEXT: [[TMP7]] = add <16 x i32> [[VEC_PHI]], [[TMP3]]
; CHECK-INTERLEAVED-NEXT: [[TMP8]] = add <16 x i32> [[VEC_PHI2]], [[TMP3]]
; CHECK-INTERLEAVED-NEXT: [[TMP22]] = add nuw i32 [[VEC_PHI1]], 32
; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = icmp eq i32 [[TMP22]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
; CHECK-INTERLEAVED: middle.block:
; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP8]], [[TMP7]]
; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX]])
; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]
; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-INTERLEAVED: scalar.ph:
;
; CHECK-MAXBW-LABEL: define i32 @add_of_loop_invariant_zext(
; CHECK-MAXBW-SAME: i32 [[A:%.*]], ptr [[B:%.*]], i8 [[C:%.*]], i32 [[D:%.*]]) #[[ATTR0]] {
; CHECK-MAXBW-NEXT: entry:
; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = sub i32 1024, [[D]]
; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw i32 [[TMP1]], 16
; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], [[TMP2]]
; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK-MAXBW: vector.ph:
; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = mul nuw i32 [[TMP3]], 16
; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[TMP0]], [[TMP4]]
; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i32 [[TMP0]], [[N_MOD_VF]]
; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = mul nuw i32 [[TMP5]], 16
; CHECK-MAXBW-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i8> poison, i8 [[C]], i64 0
; CHECK-MAXBW-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i8> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i8> poison, <vscale x 16 x i32> zeroinitializer
; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = add i32 [[D]], [[N_VEC]]
; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = insertelement <vscale x 4 x i32> zeroinitializer, i32 [[A]], i32 0
; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = zext <vscale x 16 x i8> [[BROADCAST_SPLAT]] to <vscale x 16 x i32>
; CHECK-MAXBW-NEXT: br label [[FOR_BODY:%.*]]
; CHECK-MAXBW: vector.body:
; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 4 x i32> [ [[TMP8]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[FOR_BODY]] ]
; CHECK-MAXBW-NEXT: [[OFFSET_IDX:%.*]] = add i32 [[D]], [[INDEX]]
; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[OFFSET_IDX]]
; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[TMP10]], i32 0
; CHECK-MAXBW-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP11]], align 1
; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <vscale x 4 x i32> @llvm.experimental.vector.partial.reduce.add.nxv4i32.nxv16i32(<vscale x 4 x i32> [[VEC_PHI]], <vscale x 16 x i32> [[TMP9]])
; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP6]]
; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
; CHECK-MAXBW: middle.block:
; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[PARTIAL_REDUCE]])
; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK-MAXBW: scalar.ph:
;
entry:
br label %for.body

for.body:
%iv = phi i32 [ %d, %entry ], [ %iv.next, %for.body ]
%rdx = phi i32 [ %a, %entry ], [ %rdx.next, %for.body ]
%arrayidx = getelementptr inbounds i8, ptr %b, i32 %iv
store i8 0, ptr %arrayidx, align 1
%conv1 = zext i8 %c to i32
%rdx.next = add nsw i32 %rdx, %conv1
%iv.next = add nuw nsw i32 %iv, 1
%cmp = icmp eq i32 %iv.next, 1024
br i1 %cmp, label %exit, label %for.body

exit:
%add.lcssa = phi i32 [ %rdx.next, %for.body ]
ret i32 %add.lcssa
}


!0 = distinct !{!0, !1}
!1 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}
attributes #0 = { vscale_range(1,16) "target-features"="+sve" }
Expand Down