[llvm] [LV] Fix MVE regression from #132190 (PR #141736)

Sam Tebbs via llvm-commits llvm-commits at lists.llvm.org
Wed Jun 11 07:00:09 PDT 2025


https://github.com/SamTebbs33 updated https://github.com/llvm/llvm-project/pull/141736

>From 6b78004592a9fc358a739ba28be5a93579a15923 Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Wed, 28 May 2025 10:23:09 +0100
Subject: [PATCH 1/8] [LV] Fix MVE regression from #132190

Register pressure was only considered if the vector bandwidth was being
maximised (chosen either by the target or user options), but #132190
inadvertently caused high pressure VFs to be pruned even when max
bandwidth wasn't enabled. This PR returns to the previous behaviour.
---
 .../Transforms/Vectorize/LoopVectorize.cpp    | 27 ++++++++++++++-----
 1 file changed, 21 insertions(+), 6 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 427c1460fcfc9..83bc5cde95abf 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -956,6 +956,10 @@ class LoopVectorizationCostModel {
     return expectedCost(UserVF).isValid();
   }
 
+  /// \return True if maximizing vector bandwidth is enabled by the target or
+  /// user options.
+  bool useMaxBandwidth(TargetTransformInfo::RegisterKind RegKind);
+
   /// \return The size (in bits) of the smallest and widest types in the code
   /// that needs to be vectorized. We ignore values that remain scalar such as
   /// 64 bit loop indices.
@@ -3918,6 +3922,14 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
   return FixedScalableVFPair::getNone();
 }
 
+bool LoopVectorizationCostModel::useMaxBandwidth(
+    TargetTransformInfo::RegisterKind RegKind) {
+  return MaximizeBandwidth || (MaximizeBandwidth.getNumOccurrences() == 0 &&
+                               (TTI.shouldMaximizeVectorBandwidth(RegKind) ||
+                                (UseWiderVFIfCallVariantsPresent &&
+                                 Legal->hasVectorCallVariants())));
+}
+
 ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
     unsigned MaxTripCount, unsigned SmallestType, unsigned WidestType,
     ElementCount MaxSafeVF, bool FoldTailByMasking) {
@@ -3983,10 +3995,7 @@ ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
       ComputeScalableMaxVF ? TargetTransformInfo::RGK_ScalableVector
                            : TargetTransformInfo::RGK_FixedWidthVector;
   ElementCount MaxVF = MaxVectorElementCount;
-  if (MaximizeBandwidth ||
-      (MaximizeBandwidth.getNumOccurrences() == 0 &&
-       (TTI.shouldMaximizeVectorBandwidth(RegKind) ||
-        (UseWiderVFIfCallVariantsPresent && Legal->hasVectorCallVariants())))) {
+  if (useMaxBandwidth(RegKind)) {
     auto MaxVectorElementCountMaxBW = ElementCount::get(
         llvm::bit_floor(WidestRegister.getKnownMinValue() / SmallestType),
         ComputeScalableMaxVF);
@@ -4349,7 +4358,10 @@ VectorizationFactor LoopVectorizationPlanner::selectVectorizationFactor() {
 
       /// Don't consider the VF if it exceeds the number of registers for the
       /// target.
-      if (RU.exceedsMaxNumRegs(TTI))
+      if (CM.useMaxBandwidth(VF.isScalable()
+                                 ? TargetTransformInfo::RGK_ScalableVector
+                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
+          RU.exceedsMaxNumRegs(TTI))
         continue;
 
       InstructionCost C = CM.expectedCost(VF);
@@ -7119,7 +7131,10 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
       InstructionCost Cost = cost(*P, VF);
       VectorizationFactor CurrentFactor(VF, Cost, ScalarCost);
 
-      if (RU.exceedsMaxNumRegs(TTI)) {
+      if (CM.useMaxBandwidth(VF.isScalable()
+                                 ? TargetTransformInfo::RGK_ScalableVector
+                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
+          RU.exceedsMaxNumRegs(TTI)) {
         LLVM_DEBUG(dbgs() << "LV(REG): Not considering vector loop of width "
                           << VF << " because it uses too many registers\n");
         continue;

>From 95543d1a90fe1f3aecdb486678c131b3b809b0cb Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 29 May 2025 13:46:02 +0100
Subject: [PATCH 2/8] Add test

---
 .../ARM/mve-reg-pressure-vmla.ll              | 147 ++++++++++++++++++
 1 file changed, 147 insertions(+)
 create mode 100644 llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll

diff --git a/llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll b/llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll
new file mode 100644
index 0000000000000..667022e0470e4
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll
@@ -0,0 +1,147 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "^scalar.ph:" --version 5
+; RUN: opt -passes=loop-vectorize < %s -S -o - | FileCheck %s
+
+source_filename = "<source>"
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "thumbv8.1m.main-unknown-none-eabihf"
+
+; Even though it has high register pressure, this example should still vectorise since the mul+add chains become VMLAs.
+
+define void @fn(i32 noundef %n, ptr %in, ptr %out) #0 {
+; CHECK-LABEL: define void @fn(
+; CHECK-SAME: i32 noundef [[N:%.*]], ptr [[IN:%.*]], ptr [[OUT:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[CMP46_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT:    br i1 [[CMP46_NOT]], [[FOR_COND_CLEANUP:label %.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK:       [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
+; CHECK:       [[VECTOR_MEMCHECK]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = mul i32 [[N]], 3
+; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr i8, ptr [[OUT]], i32 [[TMP0]]
+; CHECK-NEXT:    [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[IN]], i32 [[TMP0]]
+; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult ptr [[OUT]], [[SCEVGEP1]]
+; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult ptr [[IN]], [[SCEVGEP]]
+; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
+; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i32 [[N]], 3
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[N_RND_UP]], 4
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[N_MOD_VF]]
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[POINTER_PHI:%.*]] = phi ptr [ [[IN]], %[[VECTOR_PH]] ], [ [[PTR_IND:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[POINTER_PHI2:%.*]] = phi ptr [ [[OUT]], %[[VECTOR_PH]] ], [ [[PTR_IND3:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i32> <i32 0, i32 3, i32 6, i32 9>
+; CHECK-NEXT:    [[VECTOR_GEP4:%.*]] = getelementptr i8, ptr [[POINTER_PHI2]], <4 x i32> <i32 0, i32 3, i32 6, i32 9>
+; CHECK-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 [[INDEX]], i32 [[N]])
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw i8, <4 x ptr> [[VECTOR_GEP]], i32 1
+; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i8> @llvm.masked.gather.v4i8.v4p0(<4 x ptr> [[VECTOR_GEP]], i32 1, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i8> poison), !alias.scope [[META0:![0-9]+]]
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw i8, <4 x ptr> [[VECTOR_GEP]], i32 2
+; CHECK-NEXT:    [[WIDE_MASKED_GATHER5:%.*]] = call <4 x i8> @llvm.masked.gather.v4i8.v4p0(<4 x ptr> [[TMP1]], i32 1, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i8> poison), !alias.scope [[META0]]
+; CHECK-NEXT:    [[WIDE_MASKED_GATHER6:%.*]] = call <4 x i8> @llvm.masked.gather.v4i8.v4p0(<4 x ptr> [[TMP2]], i32 1, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i8> poison), !alias.scope [[META0]]
+; CHECK-NEXT:    [[TMP3:%.*]] = zext <4 x i8> [[WIDE_MASKED_GATHER]] to <4 x i32>
+; CHECK-NEXT:    [[TMP4:%.*]] = mul nuw nsw <4 x i32> [[TMP3]], splat (i32 19595)
+; CHECK-NEXT:    [[TMP5:%.*]] = zext <4 x i8> [[WIDE_MASKED_GATHER5]] to <4 x i32>
+; CHECK-NEXT:    [[TMP6:%.*]] = mul nuw nsw <4 x i32> [[TMP5]], splat (i32 38470)
+; CHECK-NEXT:    [[TMP7:%.*]] = zext <4 x i8> [[WIDE_MASKED_GATHER6]] to <4 x i32>
+; CHECK-NEXT:    [[TMP8:%.*]] = mul nuw nsw <4 x i32> [[TMP7]], splat (i32 7471)
+; CHECK-NEXT:    [[TMP9:%.*]] = add nuw nsw <4 x i32> [[TMP4]], splat (i32 32768)
+; CHECK-NEXT:    [[TMP10:%.*]] = add nuw nsw <4 x i32> [[TMP9]], [[TMP6]]
+; CHECK-NEXT:    [[TMP11:%.*]] = add nuw nsw <4 x i32> [[TMP10]], [[TMP8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = lshr <4 x i32> [[TMP11]], splat (i32 16)
+; CHECK-NEXT:    [[TMP13:%.*]] = trunc <4 x i32> [[TMP12]] to <4 x i8>
+; CHECK-NEXT:    [[TMP14:%.*]] = mul nuw nsw <4 x i32> [[TMP3]], splat (i32 32767)
+; CHECK-NEXT:    [[TMP15:%.*]] = mul nuw <4 x i32> [[TMP5]], splat (i32 16762097)
+; CHECK-NEXT:    [[TMP16:%.*]] = mul nuw <4 x i32> [[TMP7]], splat (i32 16759568)
+; CHECK-NEXT:    [[TMP17:%.*]] = add nuw nsw <4 x i32> [[TMP14]], splat (i32 32768)
+; CHECK-NEXT:    [[TMP18:%.*]] = add nuw <4 x i32> [[TMP17]], [[TMP15]]
+; CHECK-NEXT:    [[TMP19:%.*]] = add <4 x i32> [[TMP18]], [[TMP16]]
+; CHECK-NEXT:    [[TMP20:%.*]] = lshr <4 x i32> [[TMP19]], splat (i32 16)
+; CHECK-NEXT:    [[TMP21:%.*]] = trunc <4 x i32> [[TMP20]] to <4 x i8>
+; CHECK-NEXT:    [[TMP22:%.*]] = mul nuw nsw <4 x i32> [[TMP3]], splat (i32 13282)
+; CHECK-NEXT:    [[TMP23:%.*]] = mul nuw <4 x i32> [[TMP5]], splat (i32 16744449)
+; CHECK-NEXT:    [[TMP24:%.*]] = mul nuw nsw <4 x i32> [[TMP7]], splat (i32 19485)
+; CHECK-NEXT:    [[TMP25:%.*]] = add nuw nsw <4 x i32> [[TMP22]], splat (i32 32768)
+; CHECK-NEXT:    [[TMP26:%.*]] = add nuw <4 x i32> [[TMP25]], [[TMP23]]
+; CHECK-NEXT:    [[TMP27:%.*]] = add nuw <4 x i32> [[TMP26]], [[TMP24]]
+; CHECK-NEXT:    [[TMP28:%.*]] = lshr <4 x i32> [[TMP27]], splat (i32 16)
+; CHECK-NEXT:    [[TMP29:%.*]] = trunc <4 x i32> [[TMP28]] to <4 x i8>
+; CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds nuw i8, <4 x ptr> [[VECTOR_GEP4]], i32 1
+; CHECK-NEXT:    call void @llvm.masked.scatter.v4i8.v4p0(<4 x i8> [[TMP13]], <4 x ptr> [[VECTOR_GEP4]], i32 1, <4 x i1> [[ACTIVE_LANE_MASK]]), !alias.scope [[META3:![0-9]+]], !noalias [[META0]]
+; CHECK-NEXT:    [[TMP31:%.*]] = getelementptr inbounds nuw i8, <4 x ptr> [[VECTOR_GEP4]], i32 2
+; CHECK-NEXT:    call void @llvm.masked.scatter.v4i8.v4p0(<4 x i8> [[TMP21]], <4 x ptr> [[TMP30]], i32 1, <4 x i1> [[ACTIVE_LANE_MASK]]), !alias.scope [[META3]], !noalias [[META0]]
+; CHECK-NEXT:    call void @llvm.masked.scatter.v4i8.v4p0(<4 x i8> [[TMP29]], <4 x ptr> [[TMP31]], i32 1, <4 x i1> [[ACTIVE_LANE_MASK]]), !alias.scope [[META3]], !noalias [[META0]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
+; CHECK-NEXT:    [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i32 12
+; CHECK-NEXT:    [[PTR_IND3]] = getelementptr i8, ptr [[POINTER_PHI2]], i32 12
+; CHECK-NEXT:    [[TMP32:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP32]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    br [[FOR_COND_CLEANUP_LOOPEXIT:label %.*]]
+; CHECK:       [[SCALAR_PH]]:
+;
+entry:
+  %cmp46.not = icmp eq i32 %n, 0
+  br i1 %cmp46.not, label %for.cond.cleanup, label %for.body.preheader
+
+for.body.preheader:                               ; preds = %entry
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %for.body
+  br label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret void
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %in.addr.049 = phi ptr [ %incdec.ptr2, %for.body ], [ %in, %for.body.preheader ]
+  %out.addr.048 = phi ptr [ %incdec.ptr34, %for.body ], [ %out, %for.body.preheader ]
+  %i.047 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
+  %incdec.ptr = getelementptr inbounds nuw i8, ptr %in.addr.049, i32 1
+  %0 = load i8, ptr %in.addr.049, align 1
+  %incdec.ptr1 = getelementptr inbounds nuw i8, ptr %in.addr.049, i32 2
+  %1 = load i8, ptr %incdec.ptr, align 1
+  %incdec.ptr2 = getelementptr inbounds nuw i8, ptr %in.addr.049, i32 3
+  %2 = load i8, ptr %incdec.ptr1, align 1
+  %conv = zext i8 %0 to i32
+  %mul = mul nuw nsw i32 %conv, 19595
+  %conv3 = zext i8 %1 to i32
+  %mul4 = mul nuw nsw i32 %conv3, 38470
+  %conv5 = zext i8 %2 to i32
+  %mul6 = mul nuw nsw i32 %conv5, 7471
+  %add = add nuw nsw i32 %mul, 32768
+  %add7 = add nuw nsw i32 %add, %mul4
+  %add8 = add nuw nsw i32 %add7, %mul6
+  %shr = lshr i32 %add8, 16
+  %conv9 = trunc nuw i32 %shr to i8
+  %mul11 = mul nuw nsw i32 %conv, 32767
+  %mul13 = mul nuw i32 %conv3, 16762097
+  %mul16 = mul nuw i32 %conv5, 16759568
+  %add14 = add nuw nsw i32 %mul11, 32768
+  %add17 = add nuw i32 %add14, %mul13
+  %add18 = add i32 %add17, %mul16
+  %shr19 = lshr i32 %add18, 16
+  %conv20 = trunc i32 %shr19 to i8
+  %mul22 = mul nuw nsw i32 %conv, 13282
+  %mul24 = mul nuw i32 %conv3, 16744449
+  %mul27 = mul nuw nsw i32 %conv5, 19485
+  %add25 = add nuw nsw i32 %mul22, 32768
+  %add28 = add nuw i32 %add25, %mul24
+  %add29 = add nuw i32 %add28, %mul27
+  %shr30 = lshr i32 %add29, 16
+  %conv31 = trunc i32 %shr30 to i8
+  %incdec.ptr32 = getelementptr inbounds nuw i8, ptr %out.addr.048, i32 1
+  store i8 %conv9, ptr %out.addr.048, align 1
+  %incdec.ptr33 = getelementptr inbounds nuw i8, ptr %out.addr.048, i32 2
+  store i8 %conv20, ptr %incdec.ptr32, align 1
+  %incdec.ptr34 = getelementptr inbounds nuw i8, ptr %out.addr.048, i32 3
+  store i8 %conv31, ptr %incdec.ptr33, align 1
+  %inc = add nuw i32 %i.047, 1
+  %exitcond.not = icmp eq i32 %inc, %n
+  br i1 %exitcond.not, label %for.cond.cleanup.loopexit, label %for.body, !llvm.loop !7
+}
+
+attributes #0 = { nofree norecurse nosync nounwind memory(argmem: readwrite) "target-features"="+mve" }
+
+!7 = !{!"llvm.loop.mustprogress"}

>From 8663143b359108b980be96f75796cca02b6b4d59 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 29 May 2025 14:23:23 +0100
Subject: [PATCH 3/8] Only compute reg usage if needed

---
 .../Transforms/Vectorize/LoopVectorize.cpp    | 23 ++++++++---
 .../LoopVectorize/RISCV/reg-usage-bf16.ll     |  3 +-
 .../LoopVectorize/RISCV/reg-usage-f16.ll      |  6 +--
 .../LoopVectorize/RISCV/reg-usage.ll          | 40 ++++++++-----------
 4 files changed, 36 insertions(+), 36 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 83bc5cde95abf..b0c2f9ee46a9b 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -4350,8 +4350,13 @@ VectorizationFactor LoopVectorizationPlanner::selectVectorizationFactor() {
   for (auto &P : VPlans) {
     ArrayRef<ElementCount> VFs(P->vectorFactors().begin(),
                                P->vectorFactors().end());
-    auto RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
-    for (auto [VF, RU] : zip_equal(VFs, RUs)) {
+    SmallVector<VPRegisterUsage, 8> RUs;
+    if (CM.useMaxBandwidth(TargetTransformInfo::RGK_ScalableVector) ||
+        CM.useMaxBandwidth(TargetTransformInfo::RGK_FixedWidthVector))
+      RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
+
+    for (unsigned I = 0; I < VFs.size(); I++) {
+      ElementCount VF = VFs[I];
       // The cost for scalar VF=1 is already calculated, so ignore it.
       if (VF.isScalar())
         continue;
@@ -4361,7 +4366,7 @@ VectorizationFactor LoopVectorizationPlanner::selectVectorizationFactor() {
       if (CM.useMaxBandwidth(VF.isScalable()
                                  ? TargetTransformInfo::RGK_ScalableVector
                                  : TargetTransformInfo::RGK_FixedWidthVector) &&
-          RU.exceedsMaxNumRegs(TTI))
+          RUs[I].exceedsMaxNumRegs(TTI))
         continue;
 
       InstructionCost C = CM.expectedCost(VF);
@@ -7108,8 +7113,14 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
   for (auto &P : VPlans) {
     ArrayRef<ElementCount> VFs(P->vectorFactors().begin(),
                                P->vectorFactors().end());
-    auto RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
-    for (auto [VF, RU] : zip_equal(VFs, RUs)) {
+
+    SmallVector<VPRegisterUsage, 8> RUs;
+    if (CM.useMaxBandwidth(TargetTransformInfo::RGK_ScalableVector) ||
+        CM.useMaxBandwidth(TargetTransformInfo::RGK_FixedWidthVector))
+      RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
+
+    for (unsigned I = 0; I < VFs.size(); I++) {
+      ElementCount VF = VFs[I];
       if (VF.isScalar())
         continue;
       if (!ForceVectorization && !willGenerateVectors(*P, VF, TTI)) {
@@ -7134,7 +7145,7 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
       if (CM.useMaxBandwidth(VF.isScalable()
                                  ? TargetTransformInfo::RGK_ScalableVector
                                  : TargetTransformInfo::RGK_FixedWidthVector) &&
-          RU.exceedsMaxNumRegs(TTI)) {
+          RUs[I].exceedsMaxNumRegs(TTI)) {
         LLVM_DEBUG(dbgs() << "LV(REG): Not considering vector loop of width "
                           << VF << " because it uses too many registers\n");
         continue;
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
index cb071f989dafa..5a67b54c7a3d5 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
@@ -3,8 +3,7 @@
 
 define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture readonly %src2, i32 signext %size, ptr noalias nocapture writeonly %result) {
 ; CHECK-LABEL: add
-; CHECK:       LV(REG): VF = 8
-; CHECK-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK:  LV(REG): Found max usage: 2 item
 ; CHECK-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; CHECK-NEXT:  LV(REG): Found invariant usage: 1 item
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
index 15facfc48137b..d4909fa61b4f5 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
@@ -4,14 +4,12 @@
 
 define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture readonly %src2, i32 signext %size, ptr noalias nocapture writeonly %result) {
 ; CHECK-LABEL: add
-; ZVFH:       LV(REG): VF = 8
-; ZVFH-NEXT:  LV(REG): Found max usage: 2 item
+; ZVFH:  LV(REG): Found max usage: 2 item
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; ZVFH-NEXT:  LV(REG): Found invariant usage: 1 item
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; ZVFHMIN:       LV(REG): VF = 8
-; ZVFHMIN-NEXT:  LV(REG): Found max usage: 2 item
+; ZVFHMIN:  LV(REG): Found max usage: 2 item
 ; ZVFHMIN-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; ZVFHMIN-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; ZVFHMIN-NEXT:  LV(REG): Found invariant usage: 1 item
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
index 870f52876c5a9..318ecd5934e13 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
@@ -28,28 +28,24 @@ define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture rea
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::FPRRC, 2 registers
 ; CHECK-SCALAR-NEXT: LV(REG): Found invariant usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL1:       LV(REG): VF = 2
-; CHECK-LMUL1-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL2:       LV(REG): VF = 4
-; CHECK-LMUL2-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
+; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL4:       LV(REG): VF = 8
-; CHECK-LMUL4-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
+; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL8:       LV(REG): VF = 16
-; CHECK-LMUL8-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
+; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 16 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
 
@@ -80,21 +76,17 @@ define void @goo(ptr nocapture noundef %a, i32 noundef signext %n) {
 ; CHECK-SCALAR:      LV(REG): VF = 1
 ; CHECK-SCALAR-NEXT: LV(REG): Found max usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL1:       LV(REG): VF = 2
-; CHECK-LMUL1-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
-; CHECK-LMUL2:       LV(REG): VF = 4
-; CHECK-LMUL2-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
+; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
+; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 1 registers
+; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
-; CHECK-LMUL4:       LV(REG): VF = 8
-; CHECK-LMUL4-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
+; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
-; CHECK-LMUL8:       LV(REG): VF = 16
-; CHECK-LMUL8-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
+; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 entry:
   %cmp3 = icmp sgt i32 %n, 0

>From 07f6b77fa5b05d32d7e7b18d09eeac7786b2296a Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Fri, 30 May 2025 17:38:51 +0100
Subject: [PATCH 4/8] Interpret VPWidenPointerInductionRecipe as a scalar
 instead

---
 .../Transforms/Vectorize/LoopVectorize.cpp    | 46 ++++---------------
 .../Transforms/Vectorize/VPlanAnalysis.cpp    |  2 +-
 .../LoopVectorize/RISCV/reg-usage-bf16.ll     |  3 +-
 .../LoopVectorize/RISCV/reg-usage-f16.ll      |  6 ++-
 .../LoopVectorize/RISCV/reg-usage.ll          | 40 +++++++++-------
 5 files changed, 41 insertions(+), 56 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index b0c2f9ee46a9b..427c1460fcfc9 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -956,10 +956,6 @@ class LoopVectorizationCostModel {
     return expectedCost(UserVF).isValid();
   }
 
-  /// \return True if maximizing vector bandwidth is enabled by the target or
-  /// user options.
-  bool useMaxBandwidth(TargetTransformInfo::RegisterKind RegKind);
-
   /// \return The size (in bits) of the smallest and widest types in the code
   /// that needs to be vectorized. We ignore values that remain scalar such as
   /// 64 bit loop indices.
@@ -3922,14 +3918,6 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
   return FixedScalableVFPair::getNone();
 }
 
-bool LoopVectorizationCostModel::useMaxBandwidth(
-    TargetTransformInfo::RegisterKind RegKind) {
-  return MaximizeBandwidth || (MaximizeBandwidth.getNumOccurrences() == 0 &&
-                               (TTI.shouldMaximizeVectorBandwidth(RegKind) ||
-                                (UseWiderVFIfCallVariantsPresent &&
-                                 Legal->hasVectorCallVariants())));
-}
-
 ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
     unsigned MaxTripCount, unsigned SmallestType, unsigned WidestType,
     ElementCount MaxSafeVF, bool FoldTailByMasking) {
@@ -3995,7 +3983,10 @@ ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
       ComputeScalableMaxVF ? TargetTransformInfo::RGK_ScalableVector
                            : TargetTransformInfo::RGK_FixedWidthVector;
   ElementCount MaxVF = MaxVectorElementCount;
-  if (useMaxBandwidth(RegKind)) {
+  if (MaximizeBandwidth ||
+      (MaximizeBandwidth.getNumOccurrences() == 0 &&
+       (TTI.shouldMaximizeVectorBandwidth(RegKind) ||
+        (UseWiderVFIfCallVariantsPresent && Legal->hasVectorCallVariants())))) {
     auto MaxVectorElementCountMaxBW = ElementCount::get(
         llvm::bit_floor(WidestRegister.getKnownMinValue() / SmallestType),
         ComputeScalableMaxVF);
@@ -4350,23 +4341,15 @@ VectorizationFactor LoopVectorizationPlanner::selectVectorizationFactor() {
   for (auto &P : VPlans) {
     ArrayRef<ElementCount> VFs(P->vectorFactors().begin(),
                                P->vectorFactors().end());
-    SmallVector<VPRegisterUsage, 8> RUs;
-    if (CM.useMaxBandwidth(TargetTransformInfo::RGK_ScalableVector) ||
-        CM.useMaxBandwidth(TargetTransformInfo::RGK_FixedWidthVector))
-      RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
-
-    for (unsigned I = 0; I < VFs.size(); I++) {
-      ElementCount VF = VFs[I];
+    auto RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
+    for (auto [VF, RU] : zip_equal(VFs, RUs)) {
       // The cost for scalar VF=1 is already calculated, so ignore it.
       if (VF.isScalar())
         continue;
 
       /// Don't consider the VF if it exceeds the number of registers for the
       /// target.
-      if (CM.useMaxBandwidth(VF.isScalable()
-                                 ? TargetTransformInfo::RGK_ScalableVector
-                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
-          RUs[I].exceedsMaxNumRegs(TTI))
+      if (RU.exceedsMaxNumRegs(TTI))
         continue;
 
       InstructionCost C = CM.expectedCost(VF);
@@ -7113,14 +7096,8 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
   for (auto &P : VPlans) {
     ArrayRef<ElementCount> VFs(P->vectorFactors().begin(),
                                P->vectorFactors().end());
-
-    SmallVector<VPRegisterUsage, 8> RUs;
-    if (CM.useMaxBandwidth(TargetTransformInfo::RGK_ScalableVector) ||
-        CM.useMaxBandwidth(TargetTransformInfo::RGK_FixedWidthVector))
-      RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
-
-    for (unsigned I = 0; I < VFs.size(); I++) {
-      ElementCount VF = VFs[I];
+    auto RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
+    for (auto [VF, RU] : zip_equal(VFs, RUs)) {
       if (VF.isScalar())
         continue;
       if (!ForceVectorization && !willGenerateVectors(*P, VF, TTI)) {
@@ -7142,10 +7119,7 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
       InstructionCost Cost = cost(*P, VF);
       VectorizationFactor CurrentFactor(VF, Cost, ScalarCost);
 
-      if (CM.useMaxBandwidth(VF.isScalable()
-                                 ? TargetTransformInfo::RGK_ScalableVector
-                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
-          RUs[I].exceedsMaxNumRegs(TTI)) {
+      if (RU.exceedsMaxNumRegs(TTI)) {
         LLVM_DEBUG(dbgs() << "LV(REG): Not considering vector loop of width "
                           << VF << " because it uses too many registers\n");
         continue;
diff --git a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
index 76da5b0314a8e..23ddd74f5d2be 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
@@ -547,7 +547,7 @@ SmallVector<VPRegisterUsage, 8> llvm::calculateRegisterUsageForPlan(
 
         if (VFs[J].isScalar() ||
             isa<VPCanonicalIVPHIRecipe, VPReplicateRecipe, VPDerivedIVRecipe,
-                VPScalarIVStepsRecipe>(R) ||
+                VPScalarIVStepsRecipe, VPWidenPointerInductionRecipe>(R) ||
             (isa<VPInstruction>(R) &&
              all_of(cast<VPSingleDefRecipe>(R)->users(),
                     [&](VPUser *U) {
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
index 5a67b54c7a3d5..cb071f989dafa 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
@@ -3,7 +3,8 @@
 
 define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture readonly %src2, i32 signext %size, ptr noalias nocapture writeonly %result) {
 ; CHECK-LABEL: add
-; CHECK:  LV(REG): Found max usage: 2 item
+; CHECK:       LV(REG): VF = 8
+; CHECK-NEXT:  LV(REG): Found max usage: 2 item
 ; CHECK-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; CHECK-NEXT:  LV(REG): Found invariant usage: 1 item
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
index d4909fa61b4f5..15facfc48137b 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
@@ -4,12 +4,14 @@
 
 define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture readonly %src2, i32 signext %size, ptr noalias nocapture writeonly %result) {
 ; CHECK-LABEL: add
-; ZVFH:  LV(REG): Found max usage: 2 item
+; ZVFH:       LV(REG): VF = 8
+; ZVFH-NEXT:  LV(REG): Found max usage: 2 item
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; ZVFH-NEXT:  LV(REG): Found invariant usage: 1 item
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; ZVFHMIN:  LV(REG): Found max usage: 2 item
+; ZVFHMIN:       LV(REG): VF = 8
+; ZVFHMIN-NEXT:  LV(REG): Found max usage: 2 item
 ; ZVFHMIN-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; ZVFHMIN-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; ZVFHMIN-NEXT:  LV(REG): Found invariant usage: 1 item
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
index 318ecd5934e13..870f52876c5a9 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
@@ -28,24 +28,28 @@ define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture rea
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::FPRRC, 2 registers
 ; CHECK-SCALAR-NEXT: LV(REG): Found invariant usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1:       LV(REG): VF = 2
+; CHECK-LMUL1-NEXT:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2:       LV(REG): VF = 4
+; CHECK-LMUL2-NEXT:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
+; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4:       LV(REG): VF = 8
+; CHECK-LMUL4-NEXT:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
+; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8:       LV(REG): VF = 16
+; CHECK-LMUL8-NEXT:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 16 registers
+; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
 
@@ -76,17 +80,21 @@ define void @goo(ptr nocapture noundef %a, i32 noundef signext %n) {
 ; CHECK-SCALAR:      LV(REG): VF = 1
 ; CHECK-SCALAR-NEXT: LV(REG): Found max usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
-; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 1 registers
-; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
+; CHECK-LMUL1:       LV(REG): VF = 2
+; CHECK-LMUL1-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
+; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
+; CHECK-LMUL2:       LV(REG): VF = 4
+; CHECK-LMUL2-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
-; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
+; CHECK-LMUL4:       LV(REG): VF = 8
+; CHECK-LMUL4-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
-; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
+; CHECK-LMUL8:       LV(REG): VF = 16
+; CHECK-LMUL8-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 entry:
   %cmp3 = icmp sgt i32 %n, 0

>From 8de22c0c664e968326cd6cb8d6014864797536e0 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Wed, 4 Jun 2025 16:17:00 +0100
Subject: [PATCH 5/8] Revert "Interpret VPWidenPointerInductionRecipe as a
 scalar instead"

This reverts commit 760eb570768fa32f36dc64be73eb770bd27b691f.
---
 .../Transforms/Vectorize/LoopVectorize.cpp    | 47 +++++++++++++++----
 .../Transforms/Vectorize/VPlanAnalysis.cpp    |  2 +-
 .../LoopVectorize/RISCV/reg-usage-bf16.ll     |  3 +-
 .../LoopVectorize/RISCV/reg-usage-f16.ll      |  6 +--
 .../LoopVectorize/RISCV/reg-usage.ll          | 40 +++++++---------
 5 files changed, 57 insertions(+), 41 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 427c1460fcfc9..83e7ed57f7bc4 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -956,6 +956,10 @@ class LoopVectorizationCostModel {
     return expectedCost(UserVF).isValid();
   }
 
+  /// \return True if maximizing vector bandwidth is enabled by the target or
+  /// user options.
+  bool useMaxBandwidth(TargetTransformInfo::RegisterKind RegKind);
+
   /// \return The size (in bits) of the smallest and widest types in the code
   /// that needs to be vectorized. We ignore values that remain scalar such as
   /// 64 bit loop indices.
@@ -3918,6 +3922,14 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
   return FixedScalableVFPair::getNone();
 }
 
+bool LoopVectorizationCostModel::useMaxBandwidth(
+    TargetTransformInfo::RegisterKind RegKind) {
+  return MaximizeBandwidth || (MaximizeBandwidth.getNumOccurrences() == 0 &&
+                               (TTI.shouldMaximizeVectorBandwidth(RegKind) ||
+                                (UseWiderVFIfCallVariantsPresent &&
+                                 Legal->hasVectorCallVariants())));
+}
+
 ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
     unsigned MaxTripCount, unsigned SmallestType, unsigned WidestType,
     ElementCount MaxSafeVF, bool FoldTailByMasking) {
@@ -3983,10 +3995,7 @@ ElementCount LoopVectorizationCostModel::getMaximizedVFForTarget(
       ComputeScalableMaxVF ? TargetTransformInfo::RGK_ScalableVector
                            : TargetTransformInfo::RGK_FixedWidthVector;
   ElementCount MaxVF = MaxVectorElementCount;
-  if (MaximizeBandwidth ||
-      (MaximizeBandwidth.getNumOccurrences() == 0 &&
-       (TTI.shouldMaximizeVectorBandwidth(RegKind) ||
-        (UseWiderVFIfCallVariantsPresent && Legal->hasVectorCallVariants())))) {
+  if (useMaxBandwidth(RegKind)) {
     auto MaxVectorElementCountMaxBW = ElementCount::get(
         llvm::bit_floor(WidestRegister.getKnownMinValue() / SmallestType),
         ComputeScalableMaxVF);
@@ -4341,15 +4350,24 @@ VectorizationFactor LoopVectorizationPlanner::selectVectorizationFactor() {
   for (auto &P : VPlans) {
     ArrayRef<ElementCount> VFs(P->vectorFactors().begin(),
                                P->vectorFactors().end());
-    auto RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
-    for (auto [VF, RU] : zip_equal(VFs, RUs)) {
+
+    SmallVector<VPRegisterUsage, 8> RUs;
+    if (CM.useMaxBandwidth(TargetTransformInfo::RGK_ScalableVector) ||
+        CM.useMaxBandwidth(TargetTransformInfo::RGK_FixedWidthVector))
+      RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
+
+    for (unsigned I = 0; I < VFs.size(); I++) {
+      ElementCount VF = VFs[I];
       // The cost for scalar VF=1 is already calculated, so ignore it.
       if (VF.isScalar())
         continue;
 
       /// Don't consider the VF if it exceeds the number of registers for the
       /// target.
-      if (RU.exceedsMaxNumRegs(TTI))
+      if (CM.useMaxBandwidth(VF.isScalable()
+                                 ? TargetTransformInfo::RGK_ScalableVector
+                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
+          RUs[I].exceedsMaxNumRegs(TTI))
         continue;
 
       InstructionCost C = CM.expectedCost(VF);
@@ -7096,8 +7114,14 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
   for (auto &P : VPlans) {
     ArrayRef<ElementCount> VFs(P->vectorFactors().begin(),
                                P->vectorFactors().end());
-    auto RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
-    for (auto [VF, RU] : zip_equal(VFs, RUs)) {
+
+    SmallVector<VPRegisterUsage, 8> RUs;
+    if (CM.useMaxBandwidth(TargetTransformInfo::RGK_ScalableVector) ||
+        CM.useMaxBandwidth(TargetTransformInfo::RGK_FixedWidthVector))
+      RUs = calculateRegisterUsageForPlan(*P, VFs, TTI, CM.ValuesToIgnore);
+
+    for (unsigned I = 0; I < VFs.size(); I++) {
+      ElementCount VF = VFs[I];
       if (VF.isScalar())
         continue;
       if (!ForceVectorization && !willGenerateVectors(*P, VF, TTI)) {
@@ -7119,7 +7143,10 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
       InstructionCost Cost = cost(*P, VF);
       VectorizationFactor CurrentFactor(VF, Cost, ScalarCost);
 
-      if (RU.exceedsMaxNumRegs(TTI)) {
+      if (CM.useMaxBandwidth(VF.isScalable()
+                                 ? TargetTransformInfo::RGK_ScalableVector
+                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
+          RUs[I].exceedsMaxNumRegs(TTI)) {
         LLVM_DEBUG(dbgs() << "LV(REG): Not considering vector loop of width "
                           << VF << " because it uses too many registers\n");
         continue;
diff --git a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
index 23ddd74f5d2be..76da5b0314a8e 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
@@ -547,7 +547,7 @@ SmallVector<VPRegisterUsage, 8> llvm::calculateRegisterUsageForPlan(
 
         if (VFs[J].isScalar() ||
             isa<VPCanonicalIVPHIRecipe, VPReplicateRecipe, VPDerivedIVRecipe,
-                VPScalarIVStepsRecipe, VPWidenPointerInductionRecipe>(R) ||
+                VPScalarIVStepsRecipe>(R) ||
             (isa<VPInstruction>(R) &&
              all_of(cast<VPSingleDefRecipe>(R)->users(),
                     [&](VPUser *U) {
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
index cb071f989dafa..5a67b54c7a3d5 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-bf16.ll
@@ -3,8 +3,7 @@
 
 define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture readonly %src2, i32 signext %size, ptr noalias nocapture writeonly %result) {
 ; CHECK-LABEL: add
-; CHECK:       LV(REG): VF = 8
-; CHECK-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK:  LV(REG): Found max usage: 2 item
 ; CHECK-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; CHECK-NEXT:  LV(REG): Found invariant usage: 1 item
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
index 15facfc48137b..d4909fa61b4f5 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage-f16.ll
@@ -4,14 +4,12 @@
 
 define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture readonly %src2, i32 signext %size, ptr noalias nocapture writeonly %result) {
 ; CHECK-LABEL: add
-; ZVFH:       LV(REG): VF = 8
-; ZVFH-NEXT:  LV(REG): Found max usage: 2 item
+; ZVFH:  LV(REG): Found max usage: 2 item
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; ZVFH-NEXT:  LV(REG): Found invariant usage: 1 item
 ; ZVFH-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; ZVFHMIN:       LV(REG): VF = 8
-; ZVFHMIN-NEXT:  LV(REG): Found max usage: 2 item
+; ZVFHMIN:  LV(REG): Found max usage: 2 item
 ; ZVFHMIN-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; ZVFHMIN-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; ZVFHMIN-NEXT:  LV(REG): Found invariant usage: 1 item
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
index 870f52876c5a9..318ecd5934e13 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
@@ -28,28 +28,24 @@ define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture rea
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::FPRRC, 2 registers
 ; CHECK-SCALAR-NEXT: LV(REG): Found invariant usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL1:       LV(REG): VF = 2
-; CHECK-LMUL1-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL2:       LV(REG): VF = 4
-; CHECK-LMUL2-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
+; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL4:       LV(REG): VF = 8
-; CHECK-LMUL4-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
+; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL8:       LV(REG): VF = 16
-; CHECK-LMUL8-NEXT:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
+; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 16 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
 
@@ -80,21 +76,17 @@ define void @goo(ptr nocapture noundef %a, i32 noundef signext %n) {
 ; CHECK-SCALAR:      LV(REG): VF = 1
 ; CHECK-SCALAR-NEXT: LV(REG): Found max usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL1:       LV(REG): VF = 2
-; CHECK-LMUL1-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
-; CHECK-LMUL2:       LV(REG): VF = 4
-; CHECK-LMUL2-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
+; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
+; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 1 registers
+; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
-; CHECK-LMUL4:       LV(REG): VF = 8
-; CHECK-LMUL4-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
+; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
-; CHECK-LMUL8:       LV(REG): VF = 16
-; CHECK-LMUL8-NEXT:  LV(REG): Found max usage: 2 item
-; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
+; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 entry:
   %cmp3 = icmp sgt i32 %n, 0

>From ee9aa1571cc83a795fc2457b23cd1c52f489004d Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Tue, 10 Jun 2025 15:46:38 +0100
Subject: [PATCH 6/8] Add useMaxBandwidth for VF

---
 .../Transforms/Vectorize/LoopVectorize.cpp    | 22 +++++++++++--------
 1 file changed, 13 insertions(+), 9 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 83e7ed57f7bc4..d40bdd7a5accd 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -957,9 +957,13 @@ class LoopVectorizationCostModel {
   }
 
   /// \return True if maximizing vector bandwidth is enabled by the target or
-  /// user options.
+  /// user options, for the given register kind.
   bool useMaxBandwidth(TargetTransformInfo::RegisterKind RegKind);
 
+  /// \return True if maximizing vector bandwidth is enabled by the target or
+  /// user options, for the given vector factor.
+  bool useMaxBandwidth(ElementCount VF);
+
   /// \return The size (in bits) of the smallest and widest types in the code
   /// that needs to be vectorized. We ignore values that remain scalar such as
   /// 64 bit loop indices.
@@ -3922,6 +3926,12 @@ LoopVectorizationCostModel::computeMaxVF(ElementCount UserVF, unsigned UserIC) {
   return FixedScalableVFPair::getNone();
 }
 
+bool LoopVectorizationCostModel::useMaxBandwidth(ElementCount VF) {
+  return useMaxBandwidth(VF.isScalable()
+                             ? TargetTransformInfo::RGK_ScalableVector
+                             : TargetTransformInfo::RGK_FixedWidthVector);
+}
+
 bool LoopVectorizationCostModel::useMaxBandwidth(
     TargetTransformInfo::RegisterKind RegKind) {
   return MaximizeBandwidth || (MaximizeBandwidth.getNumOccurrences() == 0 &&
@@ -4364,10 +4374,7 @@ VectorizationFactor LoopVectorizationPlanner::selectVectorizationFactor() {
 
       /// Don't consider the VF if it exceeds the number of registers for the
       /// target.
-      if (CM.useMaxBandwidth(VF.isScalable()
-                                 ? TargetTransformInfo::RGK_ScalableVector
-                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
-          RUs[I].exceedsMaxNumRegs(TTI))
+      if (CM.useMaxBandwidth(VF) && RUs[I].exceedsMaxNumRegs(TTI))
         continue;
 
       InstructionCost C = CM.expectedCost(VF);
@@ -7143,10 +7150,7 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {
       InstructionCost Cost = cost(*P, VF);
       VectorizationFactor CurrentFactor(VF, Cost, ScalarCost);
 
-      if (CM.useMaxBandwidth(VF.isScalable()
-                                 ? TargetTransformInfo::RGK_ScalableVector
-                                 : TargetTransformInfo::RGK_FixedWidthVector) &&
-          RUs[I].exceedsMaxNumRegs(TTI)) {
+      if (CM.useMaxBandwidth(VF) && RUs[I].exceedsMaxNumRegs(TTI)) {
         LLVM_DEBUG(dbgs() << "LV(REG): Not considering vector loop of width "
                           << VF << " because it uses too many registers\n");
         continue;

>From 97973907721e71ba96b93438ef4226854f94c93b Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Tue, 10 Jun 2025 16:15:41 +0100
Subject: [PATCH 7/8] Clean up tests

---
 .../ARM/mve-reg-pressure-vmla.ll              | 53 ++++++++-----------
 .../LoopVectorize/RISCV/reg-usage.ll          | 16 +++---
 2 files changed, 29 insertions(+), 40 deletions(-)

diff --git a/llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll b/llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll
index 667022e0470e4..4c29a3a0d1d01 100644
--- a/llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll
+++ b/llvm/test/Transforms/LoopVectorize/ARM/mve-reg-pressure-vmla.ll
@@ -1,7 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "^scalar.ph:" --version 5
-; RUN: opt -passes=loop-vectorize < %s -S -o - | FileCheck %s
+; RUN: opt -mattr=+mve -passes=loop-vectorize < %s -S -o - | FileCheck %s
 
-source_filename = "<source>"
 target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
 target triple = "thumbv8.1m.main-unknown-none-eabihf"
 
@@ -12,7 +11,7 @@ define void @fn(i32 noundef %n, ptr %in, ptr %out) #0 {
 ; CHECK-SAME: i32 noundef [[N:%.*]], ptr [[IN:%.*]], ptr [[OUT:%.*]]) #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    [[CMP46_NOT:%.*]] = icmp eq i32 [[N]], 0
-; CHECK-NEXT:    br i1 [[CMP46_NOT]], [[FOR_COND_CLEANUP:label %.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK-NEXT:    br i1 [[CMP46_NOT]], [[EXIT:label %.*]], label %[[FOR_BODY_PREHEADER:.*]]
 ; CHECK:       [[FOR_BODY_PREHEADER]]:
 ; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
 ; CHECK:       [[VECTOR_MEMCHECK]]:
@@ -78,31 +77,22 @@ define void @fn(i32 noundef %n, ptr %in, ptr %out) #0 {
 ; CHECK-NEXT:    [[TMP32:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
 ; CHECK-NEXT:    br i1 [[TMP32]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
-; CHECK-NEXT:    br [[FOR_COND_CLEANUP_LOOPEXIT:label %.*]]
+; CHECK-NEXT:    br [[EXIT_LOOPEXIT:label %.*]]
 ; CHECK:       [[SCALAR_PH]]:
 ;
 entry:
   %cmp46.not = icmp eq i32 %n, 0
-  br i1 %cmp46.not, label %for.cond.cleanup, label %for.body.preheader
-
-for.body.preheader:                               ; preds = %entry
-  br label %for.body
-
-for.cond.cleanup.loopexit:                        ; preds = %for.body
-  br label %for.cond.cleanup
-
-for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
-  ret void
+  br i1 %cmp46.not, label %exit, label %for.body
 
 for.body:                                         ; preds = %for.body.preheader, %for.body
-  %in.addr.049 = phi ptr [ %incdec.ptr2, %for.body ], [ %in, %for.body.preheader ]
-  %out.addr.048 = phi ptr [ %incdec.ptr34, %for.body ], [ %out, %for.body.preheader ]
-  %i.047 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
-  %incdec.ptr = getelementptr inbounds nuw i8, ptr %in.addr.049, i32 1
-  %0 = load i8, ptr %in.addr.049, align 1
-  %incdec.ptr1 = getelementptr inbounds nuw i8, ptr %in.addr.049, i32 2
+  %ptr.iv.1 = phi ptr [ %in, %entry ], [ %ptr.iv.1.next, %for.body ]
+  %ptr.iv.2 = phi ptr [ %out, %entry ], [ %ptr.iv.2.next, %for.body ]
+  %iv = phi i32 [ %iv.next, %for.body ], [ 0, %entry ]
+  %incdec.ptr = getelementptr inbounds nuw i8, ptr %ptr.iv.1, i32 1
+  %0 = load i8, ptr %ptr.iv.1, align 1
+  %incdec.ptr1 = getelementptr inbounds nuw i8, ptr %ptr.iv.1, i32 2
   %1 = load i8, ptr %incdec.ptr, align 1
-  %incdec.ptr2 = getelementptr inbounds nuw i8, ptr %in.addr.049, i32 3
+  %ptr.iv.1.next = getelementptr inbounds nuw i8, ptr %ptr.iv.1, i32 3
   %2 = load i8, ptr %incdec.ptr1, align 1
   %conv = zext i8 %0 to i32
   %mul = mul nuw nsw i32 %conv, 19595
@@ -131,17 +121,16 @@ for.body:                                         ; preds = %for.body.preheader,
   %add29 = add nuw i32 %add28, %mul27
   %shr30 = lshr i32 %add29, 16
   %conv31 = trunc i32 %shr30 to i8
-  %incdec.ptr32 = getelementptr inbounds nuw i8, ptr %out.addr.048, i32 1
-  store i8 %conv9, ptr %out.addr.048, align 1
-  %incdec.ptr33 = getelementptr inbounds nuw i8, ptr %out.addr.048, i32 2
+  %incdec.ptr32 = getelementptr inbounds nuw i8, ptr %ptr.iv.2, i32 1
+  store i8 %conv9, ptr %ptr.iv.2, align 1
+  %incdec.ptr33 = getelementptr inbounds nuw i8, ptr %ptr.iv.2, i32 2
   store i8 %conv20, ptr %incdec.ptr32, align 1
-  %incdec.ptr34 = getelementptr inbounds nuw i8, ptr %out.addr.048, i32 3
+  %ptr.iv.2.next = getelementptr inbounds nuw i8, ptr %ptr.iv.2, i32 3
   store i8 %conv31, ptr %incdec.ptr33, align 1
-  %inc = add nuw i32 %i.047, 1
-  %exitcond.not = icmp eq i32 %inc, %n
-  br i1 %exitcond.not, label %for.cond.cleanup.loopexit, label %for.body, !llvm.loop !7
-}
-
-attributes #0 = { nofree norecurse nosync nounwind memory(argmem: readwrite) "target-features"="+mve" }
+  %iv.next = add nuw i32 %iv, 1
+  %exitcond.not = icmp eq i32 %iv.next, %n
+  br i1 %exitcond.not, label %exit, label %for.body
 
-!7 = !{!"llvm.loop.mustprogress"}
+exit:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret void
+}
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
index 318ecd5934e13..cee0b1222b6be 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/reg-usage.ll
@@ -28,22 +28,22 @@ define void @add(ptr noalias nocapture readonly %src1, ptr noalias nocapture rea
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::FPRRC, 2 registers
 ; CHECK-SCALAR-NEXT: LV(REG): Found invariant usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): Found invariant usage: 1 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 1 registers
-; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 16 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): Found invariant usage: 1 item
@@ -76,16 +76,16 @@ define void @goo(ptr nocapture noundef %a, i32 noundef signext %n) {
 ; CHECK-SCALAR:      LV(REG): VF = 1
 ; CHECK-SCALAR-NEXT: LV(REG): Found max usage: 1 item
 ; CHECK-SCALAR-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 3 registers
-; CHECK-LMUL1:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL1:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL1-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 1 registers
-; CHECK-LMUL2:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL2:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL2-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 2 registers
-; CHECK-LMUL4:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL4:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL4-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 4 registers
-; CHECK-LMUL8:  LV(REG): Found max usage: 2 item
+; CHECK-LMUL8:       LV(REG): Found max usage: 2 item
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::GPRRC, 2 registers
 ; CHECK-LMUL8-NEXT:  LV(REG): RegisterClass: RISCV::VRRC, 8 registers
 entry:

>From ea866e835dadff4465d8b02b4a7cf36b659d5add Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Wed, 11 Jun 2025 14:57:06 +0100
Subject: [PATCH 8/8] Update RISCV test

---
 .../RISCV/interleaved-masked-access.ll        | 284 +++++++++---------
 1 file changed, 142 insertions(+), 142 deletions(-)

diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/interleaved-masked-access.ll b/llvm/test/Transforms/LoopVectorize/RISCV/interleaved-masked-access.ll
index 1b0feef3e6664..b7c9612e57aec 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/interleaved-masked-access.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/interleaved-masked-access.ll
@@ -11,44 +11,44 @@ define void @masked_strided_factor2(ptr noalias nocapture readonly %p, ptr noali
 ; SCALAR_EPILOGUE-NEXT:  entry:
 ; SCALAR_EPILOGUE-NEXT:    [[CONV:%.*]] = zext i8 [[GUARD]] to i32
 ; SCALAR_EPILOGUE-NEXT:    [[TMP0:%.*]] = call i32 @llvm.vscale.i32()
-; SCALAR_EPILOGUE-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 3
+; SCALAR_EPILOGUE-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 4
 ; SCALAR_EPILOGUE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ugt i32 [[TMP1]], 1024
 ; SCALAR_EPILOGUE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
 ; SCALAR_EPILOGUE:       vector.ph:
 ; SCALAR_EPILOGUE-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vscale.i32()
-; SCALAR_EPILOGUE-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 3
+; SCALAR_EPILOGUE-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 4
 ; SCALAR_EPILOGUE-NEXT:    [[N_MOD_VF:%.*]] = urem i32 1024, [[TMP3]]
 ; SCALAR_EPILOGUE-NEXT:    [[N_VEC:%.*]] = sub nuw nsw i32 1024, [[N_MOD_VF]]
 ; SCALAR_EPILOGUE-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vscale.i32()
-; SCALAR_EPILOGUE-NEXT:    [[TMP5:%.*]] = shl i32 [[TMP4]], 3
-; SCALAR_EPILOGUE-NEXT:    [[TMP6:%.*]] = call <vscale x 8 x i32> @llvm.stepvector.nxv8i32()
-; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[CONV]], i64 0
-; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
-; SCALAR_EPILOGUE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[TMP5]], i64 0
-; SCALAR_EPILOGUE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[DOTSPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
+; SCALAR_EPILOGUE-NEXT:    [[TMP5:%.*]] = shl i32 [[TMP4]], 4
+; SCALAR_EPILOGUE-NEXT:    [[TMP6:%.*]] = call <vscale x 16 x i32> @llvm.stepvector.nxv16i32()
+; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[CONV]], i64 0
+; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
+; SCALAR_EPILOGUE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[TMP5]], i64 0
+; SCALAR_EPILOGUE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[DOTSPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
 ; SCALAR_EPILOGUE-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; SCALAR_EPILOGUE:       vector.body:
 ; SCALAR_EPILOGUE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; SCALAR_EPILOGUE-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 8 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
-; SCALAR_EPILOGUE-NEXT:    [[TMP7:%.*]] = icmp ugt <vscale x 8 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; SCALAR_EPILOGUE-NEXT:    [[TMP8:%.*]] = shl nuw nsw <vscale x 8 x i32> [[VEC_IND]], splat (i32 1)
-; SCALAR_EPILOGUE-NEXT:    [[TMP9:%.*]] = zext nneg <vscale x 8 x i32> [[TMP8]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP9]]
-; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP10]], i32 1, <vscale x 8 x i1> [[TMP7]], <vscale x 8 x i8> poison)
-; SCALAR_EPILOGUE-NEXT:    [[TMP11:%.*]] = or disjoint <vscale x 8 x i32> [[TMP8]], splat (i32 1)
-; SCALAR_EPILOGUE-NEXT:    [[TMP12:%.*]] = zext nneg <vscale x 8 x i32> [[TMP11]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP12]]
-; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP13]], i32 1, <vscale x 8 x i1> [[TMP7]], <vscale x 8 x i8> poison)
-; SCALAR_EPILOGUE-NEXT:    [[TMP14:%.*]] = call <vscale x 8 x i8> @llvm.smax.nxv8i8(<vscale x 8 x i8> [[WIDE_MASKED_GATHER]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER1]])
-; SCALAR_EPILOGUE-NEXT:    [[TMP15:%.*]] = zext nneg <vscale x 8 x i32> [[TMP8]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP15]]
-; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP14]], <vscale x 8 x ptr> [[TMP16]], i32 1, <vscale x 8 x i1> [[TMP7]])
-; SCALAR_EPILOGUE-NEXT:    [[TMP17:%.*]] = sub <vscale x 8 x i8> zeroinitializer, [[TMP14]]
-; SCALAR_EPILOGUE-NEXT:    [[TMP18:%.*]] = zext nneg <vscale x 8 x i32> [[TMP11]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP18]]
-; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP17]], <vscale x 8 x ptr> [[TMP19]], i32 1, <vscale x 8 x i1> [[TMP7]])
+; SCALAR_EPILOGUE-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 16 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; SCALAR_EPILOGUE-NEXT:    [[TMP7:%.*]] = icmp ugt <vscale x 16 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; SCALAR_EPILOGUE-NEXT:    [[TMP8:%.*]] = shl nuw nsw <vscale x 16 x i32> [[VEC_IND]], splat (i32 1)
+; SCALAR_EPILOGUE-NEXT:    [[TMP9:%.*]] = zext nneg <vscale x 16 x i32> [[TMP8]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP9]]
+; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP10]], i32 1, <vscale x 16 x i1> [[TMP7]], <vscale x 16 x i8> poison)
+; SCALAR_EPILOGUE-NEXT:    [[TMP11:%.*]] = or disjoint <vscale x 16 x i32> [[TMP8]], splat (i32 1)
+; SCALAR_EPILOGUE-NEXT:    [[TMP12:%.*]] = zext nneg <vscale x 16 x i32> [[TMP11]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP12]]
+; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP13]], i32 1, <vscale x 16 x i1> [[TMP7]], <vscale x 16 x i8> poison)
+; SCALAR_EPILOGUE-NEXT:    [[TMP14:%.*]] = call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i8> [[WIDE_MASKED_GATHER1]])
+; SCALAR_EPILOGUE-NEXT:    [[TMP15:%.*]] = zext nneg <vscale x 16 x i32> [[TMP8]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP15]]
+; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP14]], <vscale x 16 x ptr> [[TMP16]], i32 1, <vscale x 16 x i1> [[TMP7]])
+; SCALAR_EPILOGUE-NEXT:    [[TMP17:%.*]] = sub <vscale x 16 x i8> zeroinitializer, [[TMP14]]
+; SCALAR_EPILOGUE-NEXT:    [[TMP18:%.*]] = zext nneg <vscale x 16 x i32> [[TMP11]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP18]]
+; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP17]], <vscale x 16 x ptr> [[TMP19]], i32 1, <vscale x 16 x i1> [[TMP7]])
 ; SCALAR_EPILOGUE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP5]]
-; SCALAR_EPILOGUE-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 8 x i32> [[VEC_IND]], [[DOTSPLAT]]
+; SCALAR_EPILOGUE-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 16 x i32> [[VEC_IND]], [[DOTSPLAT]]
 ; SCALAR_EPILOGUE-NEXT:    [[TMP20:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
 ; SCALAR_EPILOGUE-NEXT:    br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; SCALAR_EPILOGUE:       middle.block:
@@ -63,42 +63,42 @@ define void @masked_strided_factor2(ptr noalias nocapture readonly %p, ptr noali
 ; PREDICATED_TAIL_FOLDING:       vector.ph:
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[CONV:%.*]] = zext i8 [[GUARD]] to i32
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP0:%.*]] = call i32 @llvm.vscale.i32()
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 3
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 4
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[N_RND_UP:%.*]] = add i32 [[TMP1]], 1023
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[N_RND_UP]], [[TMP1]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[N_MOD_VF]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vscale.i32()
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 3
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i32> @llvm.stepvector.nxv8i32()
-; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[CONV]], i64 0
-; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
-; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[TMP3]], i64 0
-; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[DOTSPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 4
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP4:%.*]] = call <vscale x 16 x i32> @llvm.stepvector.nxv16i32()
+; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[CONV]], i64 0
+; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
+; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[TMP3]], i64 0
+; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[DOTSPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
 ; PREDICATED_TAIL_FOLDING-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; PREDICATED_TAIL_FOLDING:       vector.body:
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 8 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i32(i32 [[INDEX]], i32 1024)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP5:%.*]] = icmp ugt <vscale x 8 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP6:%.*]] = select <vscale x 8 x i1> [[ACTIVE_LANE_MASK]], <vscale x 8 x i1> [[TMP5]], <vscale x 8 x i1> zeroinitializer
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP7:%.*]] = shl nuw nsw <vscale x 8 x i32> [[VEC_IND]], splat (i32 1)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP8:%.*]] = zext nneg <vscale x 8 x i32> [[TMP7]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP8]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP9]], i32 1, <vscale x 8 x i1> [[TMP6]], <vscale x 8 x i8> poison)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP10:%.*]] = or disjoint <vscale x 8 x i32> [[TMP7]], splat (i32 1)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP11:%.*]] = zext nneg <vscale x 8 x i32> [[TMP10]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP11]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP12]], i32 1, <vscale x 8 x i1> [[TMP6]], <vscale x 8 x i8> poison)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP13:%.*]] = call <vscale x 8 x i8> @llvm.smax.nxv8i8(<vscale x 8 x i8> [[WIDE_MASKED_GATHER]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER1]])
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP14:%.*]] = zext nneg <vscale x 8 x i32> [[TMP7]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP14]]
-; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP13]], <vscale x 8 x ptr> [[TMP15]], i32 1, <vscale x 8 x i1> [[TMP6]])
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP16:%.*]] = sub <vscale x 8 x i8> zeroinitializer, [[TMP13]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP17:%.*]] = zext nneg <vscale x 8 x i32> [[TMP10]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP17]]
-; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP16]], <vscale x 8 x ptr> [[TMP18]], i32 1, <vscale x 8 x i1> [[TMP6]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 16 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i32(i32 [[INDEX]], i32 1024)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP5:%.*]] = icmp ugt <vscale x 16 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP6:%.*]] = select <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i1> [[TMP5]], <vscale x 16 x i1> zeroinitializer
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP7:%.*]] = shl nuw nsw <vscale x 16 x i32> [[VEC_IND]], splat (i32 1)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP8:%.*]] = zext nneg <vscale x 16 x i32> [[TMP7]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP8]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP9]], i32 1, <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i8> poison)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP10:%.*]] = or disjoint <vscale x 16 x i32> [[TMP7]], splat (i32 1)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP11:%.*]] = zext nneg <vscale x 16 x i32> [[TMP10]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP11]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP12]], i32 1, <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i8> poison)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP13:%.*]] = call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i8> [[WIDE_MASKED_GATHER1]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP14:%.*]] = zext nneg <vscale x 16 x i32> [[TMP7]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP14]]
+; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP13]], <vscale x 16 x ptr> [[TMP15]], i32 1, <vscale x 16 x i1> [[TMP6]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP16:%.*]] = sub <vscale x 16 x i8> zeroinitializer, [[TMP13]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP17:%.*]] = zext nneg <vscale x 16 x i32> [[TMP10]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP17]]
+; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP16]], <vscale x 16 x ptr> [[TMP18]], i32 1, <vscale x 16 x i1> [[TMP6]])
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP3]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 8 x i32> [[VEC_IND]], [[DOTSPLAT]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 16 x i32> [[VEC_IND]], [[DOTSPLAT]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP19:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; PREDICATED_TAIL_FOLDING:       middle.block:
@@ -180,60 +180,60 @@ define void @masked_strided_factor4(ptr noalias nocapture readonly %p, ptr noali
 ; SCALAR_EPILOGUE-NEXT:  entry:
 ; SCALAR_EPILOGUE-NEXT:    [[CONV:%.*]] = zext i8 [[GUARD]] to i32
 ; SCALAR_EPILOGUE-NEXT:    [[TMP0:%.*]] = call i32 @llvm.vscale.i32()
-; SCALAR_EPILOGUE-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 3
+; SCALAR_EPILOGUE-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 4
 ; SCALAR_EPILOGUE-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ugt i32 [[TMP1]], 1024
 ; SCALAR_EPILOGUE-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
 ; SCALAR_EPILOGUE:       vector.ph:
 ; SCALAR_EPILOGUE-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vscale.i32()
-; SCALAR_EPILOGUE-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 3
+; SCALAR_EPILOGUE-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 4
 ; SCALAR_EPILOGUE-NEXT:    [[N_MOD_VF:%.*]] = urem i32 1024, [[TMP3]]
 ; SCALAR_EPILOGUE-NEXT:    [[N_VEC:%.*]] = sub nuw nsw i32 1024, [[N_MOD_VF]]
 ; SCALAR_EPILOGUE-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vscale.i32()
-; SCALAR_EPILOGUE-NEXT:    [[TMP5:%.*]] = shl i32 [[TMP4]], 3
-; SCALAR_EPILOGUE-NEXT:    [[TMP6:%.*]] = call <vscale x 8 x i32> @llvm.stepvector.nxv8i32()
-; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[CONV]], i64 0
-; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
-; SCALAR_EPILOGUE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[TMP5]], i64 0
-; SCALAR_EPILOGUE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[DOTSPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
+; SCALAR_EPILOGUE-NEXT:    [[TMP5:%.*]] = shl i32 [[TMP4]], 4
+; SCALAR_EPILOGUE-NEXT:    [[TMP6:%.*]] = call <vscale x 16 x i32> @llvm.stepvector.nxv16i32()
+; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[CONV]], i64 0
+; SCALAR_EPILOGUE-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
+; SCALAR_EPILOGUE-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[TMP5]], i64 0
+; SCALAR_EPILOGUE-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[DOTSPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
 ; SCALAR_EPILOGUE-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; SCALAR_EPILOGUE:       vector.body:
 ; SCALAR_EPILOGUE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; SCALAR_EPILOGUE-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 8 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
-; SCALAR_EPILOGUE-NEXT:    [[TMP7:%.*]] = icmp ugt <vscale x 8 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; SCALAR_EPILOGUE-NEXT:    [[TMP8:%.*]] = shl nuw nsw <vscale x 8 x i32> [[VEC_IND]], splat (i32 2)
-; SCALAR_EPILOGUE-NEXT:    [[TMP9:%.*]] = or disjoint <vscale x 8 x i32> [[TMP8]], splat (i32 1)
-; SCALAR_EPILOGUE-NEXT:    [[TMP10:%.*]] = or disjoint <vscale x 8 x i32> [[TMP8]], splat (i32 2)
-; SCALAR_EPILOGUE-NEXT:    [[TMP11:%.*]] = or disjoint <vscale x 8 x i32> [[TMP8]], splat (i32 3)
-; SCALAR_EPILOGUE-NEXT:    [[TMP12:%.*]] = zext nneg <vscale x 8 x i32> [[TMP8]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP12]]
-; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP13]], i32 1, <vscale x 8 x i1> [[TMP7]], <vscale x 8 x i8> poison)
-; SCALAR_EPILOGUE-NEXT:    [[TMP14:%.*]] = zext nneg <vscale x 8 x i32> [[TMP9]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP14]]
-; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP15]], i32 1, <vscale x 8 x i1> [[TMP7]], <vscale x 8 x i8> poison)
-; SCALAR_EPILOGUE-NEXT:    [[TMP16:%.*]] = zext nneg <vscale x 8 x i32> [[TMP10]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP16]]
-; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER2:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP17]], i32 1, <vscale x 8 x i1> [[TMP7]], <vscale x 8 x i8> poison)
-; SCALAR_EPILOGUE-NEXT:    [[TMP18:%.*]] = zext nneg <vscale x 8 x i32> [[TMP11]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP18]]
-; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER3:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP19]], i32 1, <vscale x 8 x i1> [[TMP7]], <vscale x 8 x i8> poison)
-; SCALAR_EPILOGUE-NEXT:    [[TMP20:%.*]] = call <vscale x 8 x i8> @llvm.smax.nxv8i8(<vscale x 8 x i8> [[WIDE_MASKED_GATHER]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER1]])
-; SCALAR_EPILOGUE-NEXT:    [[TMP21:%.*]] = sub <vscale x 8 x i8> zeroinitializer, [[TMP20]]
-; SCALAR_EPILOGUE-NEXT:    [[TMP22:%.*]] = call <vscale x 8 x i8> @llvm.smax.nxv8i8(<vscale x 8 x i8> [[WIDE_MASKED_GATHER2]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER3]])
-; SCALAR_EPILOGUE-NEXT:    [[TMP23:%.*]] = sub <vscale x 8 x i8> zeroinitializer, [[TMP22]]
-; SCALAR_EPILOGUE-NEXT:    [[TMP24:%.*]] = zext nneg <vscale x 8 x i32> [[TMP8]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP25:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP24]]
-; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP20]], <vscale x 8 x ptr> [[TMP25]], i32 1, <vscale x 8 x i1> [[TMP7]])
-; SCALAR_EPILOGUE-NEXT:    [[TMP26:%.*]] = zext nneg <vscale x 8 x i32> [[TMP9]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP26]]
-; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP21]], <vscale x 8 x ptr> [[TMP27]], i32 1, <vscale x 8 x i1> [[TMP7]])
-; SCALAR_EPILOGUE-NEXT:    [[TMP28:%.*]] = zext nneg <vscale x 8 x i32> [[TMP10]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP29:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP28]]
-; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP22]], <vscale x 8 x ptr> [[TMP29]], i32 1, <vscale x 8 x i1> [[TMP7]])
-; SCALAR_EPILOGUE-NEXT:    [[TMP30:%.*]] = zext nneg <vscale x 8 x i32> [[TMP11]] to <vscale x 8 x i64>
-; SCALAR_EPILOGUE-NEXT:    [[TMP31:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP30]]
-; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP23]], <vscale x 8 x ptr> [[TMP31]], i32 1, <vscale x 8 x i1> [[TMP7]])
+; SCALAR_EPILOGUE-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 16 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; SCALAR_EPILOGUE-NEXT:    [[TMP7:%.*]] = icmp ugt <vscale x 16 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; SCALAR_EPILOGUE-NEXT:    [[TMP8:%.*]] = shl nuw nsw <vscale x 16 x i32> [[VEC_IND]], splat (i32 2)
+; SCALAR_EPILOGUE-NEXT:    [[TMP9:%.*]] = or disjoint <vscale x 16 x i32> [[TMP8]], splat (i32 1)
+; SCALAR_EPILOGUE-NEXT:    [[TMP10:%.*]] = or disjoint <vscale x 16 x i32> [[TMP8]], splat (i32 2)
+; SCALAR_EPILOGUE-NEXT:    [[TMP11:%.*]] = or disjoint <vscale x 16 x i32> [[TMP8]], splat (i32 3)
+; SCALAR_EPILOGUE-NEXT:    [[TMP12:%.*]] = zext nneg <vscale x 16 x i32> [[TMP8]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP12]]
+; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP13]], i32 1, <vscale x 16 x i1> [[TMP7]], <vscale x 16 x i8> poison)
+; SCALAR_EPILOGUE-NEXT:    [[TMP14:%.*]] = zext nneg <vscale x 16 x i32> [[TMP9]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP14]]
+; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP15]], i32 1, <vscale x 16 x i1> [[TMP7]], <vscale x 16 x i8> poison)
+; SCALAR_EPILOGUE-NEXT:    [[TMP16:%.*]] = zext nneg <vscale x 16 x i32> [[TMP10]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP16]]
+; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER2:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP17]], i32 1, <vscale x 16 x i1> [[TMP7]], <vscale x 16 x i8> poison)
+; SCALAR_EPILOGUE-NEXT:    [[TMP18:%.*]] = zext nneg <vscale x 16 x i32> [[TMP11]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP18]]
+; SCALAR_EPILOGUE-NEXT:    [[WIDE_MASKED_GATHER3:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP19]], i32 1, <vscale x 16 x i1> [[TMP7]], <vscale x 16 x i8> poison)
+; SCALAR_EPILOGUE-NEXT:    [[TMP20:%.*]] = call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i8> [[WIDE_MASKED_GATHER1]])
+; SCALAR_EPILOGUE-NEXT:    [[TMP21:%.*]] = sub <vscale x 16 x i8> zeroinitializer, [[TMP20]]
+; SCALAR_EPILOGUE-NEXT:    [[TMP22:%.*]] = call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER2]], <vscale x 16 x i8> [[WIDE_MASKED_GATHER3]])
+; SCALAR_EPILOGUE-NEXT:    [[TMP23:%.*]] = sub <vscale x 16 x i8> zeroinitializer, [[TMP22]]
+; SCALAR_EPILOGUE-NEXT:    [[TMP24:%.*]] = zext nneg <vscale x 16 x i32> [[TMP8]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP25:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP24]]
+; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP20]], <vscale x 16 x ptr> [[TMP25]], i32 1, <vscale x 16 x i1> [[TMP7]])
+; SCALAR_EPILOGUE-NEXT:    [[TMP26:%.*]] = zext nneg <vscale x 16 x i32> [[TMP9]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP26]]
+; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP21]], <vscale x 16 x ptr> [[TMP27]], i32 1, <vscale x 16 x i1> [[TMP7]])
+; SCALAR_EPILOGUE-NEXT:    [[TMP28:%.*]] = zext nneg <vscale x 16 x i32> [[TMP10]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP29:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP28]]
+; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP22]], <vscale x 16 x ptr> [[TMP29]], i32 1, <vscale x 16 x i1> [[TMP7]])
+; SCALAR_EPILOGUE-NEXT:    [[TMP30:%.*]] = zext nneg <vscale x 16 x i32> [[TMP11]] to <vscale x 16 x i64>
+; SCALAR_EPILOGUE-NEXT:    [[TMP31:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP30]]
+; SCALAR_EPILOGUE-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP23]], <vscale x 16 x ptr> [[TMP31]], i32 1, <vscale x 16 x i1> [[TMP7]])
 ; SCALAR_EPILOGUE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP5]]
-; SCALAR_EPILOGUE-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 8 x i32> [[VEC_IND]], [[DOTSPLAT]]
+; SCALAR_EPILOGUE-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 16 x i32> [[VEC_IND]], [[DOTSPLAT]]
 ; SCALAR_EPILOGUE-NEXT:    [[TMP32:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
 ; SCALAR_EPILOGUE-NEXT:    br i1 [[TMP32]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
 ; SCALAR_EPILOGUE:       middle.block:
@@ -248,58 +248,58 @@ define void @masked_strided_factor4(ptr noalias nocapture readonly %p, ptr noali
 ; PREDICATED_TAIL_FOLDING:       vector.ph:
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[CONV:%.*]] = zext i8 [[GUARD]] to i32
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP0:%.*]] = call i32 @llvm.vscale.i32()
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 3
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP1:%.*]] = shl i32 [[TMP0]], 4
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[N_RND_UP:%.*]] = add i32 [[TMP1]], 1023
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[N_RND_UP]], [[TMP1]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[N_MOD_VF]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vscale.i32()
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 3
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP4:%.*]] = call <vscale x 8 x i32> @llvm.stepvector.nxv8i32()
-; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[CONV]], i64 0
-; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
-; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i32> poison, i32 [[TMP3]], i64 0
-; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i32> [[DOTSPLATINSERT]], <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP3:%.*]] = shl i32 [[TMP2]], 4
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP4:%.*]] = call <vscale x 16 x i32> @llvm.stepvector.nxv16i32()
+; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[CONV]], i64 0
+; PREDICATED_TAIL_FOLDING-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
+; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[TMP3]], i64 0
+; PREDICATED_TAIL_FOLDING-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[DOTSPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
 ; PREDICATED_TAIL_FOLDING-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; PREDICATED_TAIL_FOLDING:       vector.body:
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 8 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i32(i32 [[INDEX]], i32 1024)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP5:%.*]] = icmp ugt <vscale x 8 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP6:%.*]] = select <vscale x 8 x i1> [[ACTIVE_LANE_MASK]], <vscale x 8 x i1> [[TMP5]], <vscale x 8 x i1> zeroinitializer
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP7:%.*]] = shl nuw nsw <vscale x 8 x i32> [[VEC_IND]], splat (i32 2)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP8:%.*]] = or disjoint <vscale x 8 x i32> [[TMP7]], splat (i32 1)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP9:%.*]] = or disjoint <vscale x 8 x i32> [[TMP7]], splat (i32 2)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP10:%.*]] = or disjoint <vscale x 8 x i32> [[TMP7]], splat (i32 3)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP11:%.*]] = zext nneg <vscale x 8 x i32> [[TMP7]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP11]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP12]], i32 1, <vscale x 8 x i1> [[TMP6]], <vscale x 8 x i8> poison)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP13:%.*]] = zext nneg <vscale x 8 x i32> [[TMP8]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP13]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP14]], i32 1, <vscale x 8 x i1> [[TMP6]], <vscale x 8 x i8> poison)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP15:%.*]] = zext nneg <vscale x 8 x i32> [[TMP9]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP15]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER2:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP16]], i32 1, <vscale x 8 x i1> [[TMP6]], <vscale x 8 x i8> poison)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP17:%.*]] = zext nneg <vscale x 8 x i32> [[TMP10]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 8 x i64> [[TMP17]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER3:%.*]] = call <vscale x 8 x i8> @llvm.masked.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> [[TMP18]], i32 1, <vscale x 8 x i1> [[TMP6]], <vscale x 8 x i8> poison)
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP19:%.*]] = call <vscale x 8 x i8> @llvm.smax.nxv8i8(<vscale x 8 x i8> [[WIDE_MASKED_GATHER]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER1]])
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP20:%.*]] = sub <vscale x 8 x i8> zeroinitializer, [[TMP19]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP21:%.*]] = call <vscale x 8 x i8> @llvm.smax.nxv8i8(<vscale x 8 x i8> [[WIDE_MASKED_GATHER2]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER3]])
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP22:%.*]] = sub <vscale x 8 x i8> zeroinitializer, [[TMP21]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP23:%.*]] = zext nneg <vscale x 8 x i32> [[TMP7]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP24:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP23]]
-; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP19]], <vscale x 8 x ptr> [[TMP24]], i32 1, <vscale x 8 x i1> [[TMP6]])
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP25:%.*]] = zext nneg <vscale x 8 x i32> [[TMP8]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP25]]
-; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP20]], <vscale x 8 x ptr> [[TMP26]], i32 1, <vscale x 8 x i1> [[TMP6]])
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP27:%.*]] = zext nneg <vscale x 8 x i32> [[TMP9]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP27]]
-; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP21]], <vscale x 8 x ptr> [[TMP28]], i32 1, <vscale x 8 x i1> [[TMP6]])
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP29:%.*]] = zext nneg <vscale x 8 x i32> [[TMP10]] to <vscale x 8 x i64>
-; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP30:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 8 x i64> [[TMP29]]
-; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP22]], <vscale x 8 x ptr> [[TMP30]], i32 1, <vscale x 8 x i1> [[TMP6]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND:%.*]] = phi <vscale x 16 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i32(i32 [[INDEX]], i32 1024)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP5:%.*]] = icmp ugt <vscale x 16 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP6:%.*]] = select <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], <vscale x 16 x i1> [[TMP5]], <vscale x 16 x i1> zeroinitializer
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP7:%.*]] = shl nuw nsw <vscale x 16 x i32> [[VEC_IND]], splat (i32 2)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP8:%.*]] = or disjoint <vscale x 16 x i32> [[TMP7]], splat (i32 1)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP9:%.*]] = or disjoint <vscale x 16 x i32> [[TMP7]], splat (i32 2)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP10:%.*]] = or disjoint <vscale x 16 x i32> [[TMP7]], splat (i32 3)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP11:%.*]] = zext nneg <vscale x 16 x i32> [[TMP7]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP11]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP12]], i32 1, <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i8> poison)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP13:%.*]] = zext nneg <vscale x 16 x i32> [[TMP8]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP13]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER1:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP14]], i32 1, <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i8> poison)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP15:%.*]] = zext nneg <vscale x 16 x i32> [[TMP9]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP15]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER2:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP16]], i32 1, <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i8> poison)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP17:%.*]] = zext nneg <vscale x 16 x i32> [[TMP10]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[P]], <vscale x 16 x i64> [[TMP17]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[WIDE_MASKED_GATHER3:%.*]] = call <vscale x 16 x i8> @llvm.masked.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> [[TMP18]], i32 1, <vscale x 16 x i1> [[TMP6]], <vscale x 16 x i8> poison)
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP19:%.*]] = call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i8> [[WIDE_MASKED_GATHER1]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP20:%.*]] = sub <vscale x 16 x i8> zeroinitializer, [[TMP19]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP21:%.*]] = call <vscale x 16 x i8> @llvm.smax.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER2]], <vscale x 16 x i8> [[WIDE_MASKED_GATHER3]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP22:%.*]] = sub <vscale x 16 x i8> zeroinitializer, [[TMP21]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP23:%.*]] = zext nneg <vscale x 16 x i32> [[TMP7]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP24:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP23]]
+; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP19]], <vscale x 16 x ptr> [[TMP24]], i32 1, <vscale x 16 x i1> [[TMP6]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP25:%.*]] = zext nneg <vscale x 16 x i32> [[TMP8]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP25]]
+; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP20]], <vscale x 16 x ptr> [[TMP26]], i32 1, <vscale x 16 x i1> [[TMP6]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP27:%.*]] = zext nneg <vscale x 16 x i32> [[TMP9]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP27]]
+; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP21]], <vscale x 16 x ptr> [[TMP28]], i32 1, <vscale x 16 x i1> [[TMP6]])
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP29:%.*]] = zext nneg <vscale x 16 x i32> [[TMP10]] to <vscale x 16 x i64>
+; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP30:%.*]] = getelementptr inbounds i8, ptr [[Q]], <vscale x 16 x i64> [[TMP29]]
+; PREDICATED_TAIL_FOLDING-NEXT:    call void @llvm.masked.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP22]], <vscale x 16 x ptr> [[TMP30]], i32 1, <vscale x 16 x i1> [[TMP6]])
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], [[TMP3]]
-; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 8 x i32> [[VEC_IND]], [[DOTSPLAT]]
+; PREDICATED_TAIL_FOLDING-NEXT:    [[VEC_IND_NEXT]] = add <vscale x 16 x i32> [[VEC_IND]], [[DOTSPLAT]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    [[TMP31:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
 ; PREDICATED_TAIL_FOLDING-NEXT:    br i1 [[TMP31]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
 ; PREDICATED_TAIL_FOLDING:       middle.block:



More information about the llvm-commits mailing list