[llvm] [VPlan] Simplify the computation of the block entry mask. (PR #173265)

Mel Chen via llvm-commits llvm-commits at lists.llvm.org
Thu Feb 5 07:17:38 PST 2026


https://github.com/Mel-Chen updated https://github.com/llvm/llvm-project/pull/173265

>From 5f993f272a23e7d4c197121d778aca9b04db1221 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Thu, 18 Dec 2025 01:32:12 -0800
Subject: [PATCH 1/2] use post-dom tree to prove bb is control-flow-equivalent
 to header

---
 .../Transforms/Vectorize/VPlanDominatorTree.h |  9 ++++++
 .../Transforms/Vectorize/VPlanPredicator.cpp  | 29 ++++++++++++++++--
 .../RISCV/blocks-with-dead-instructions.ll    | 12 +-------
 .../RISCV/tail-folding-complex-mask.ll        | 30 +++++++------------
 .../LoopVectorize/cse-replicate-regions.ll    |  5 ++--
 ...predicated-loads-with-predicated-stores.ll | 18 +++++------
 6 files changed, 56 insertions(+), 47 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h b/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h
index 44506f5ac3e81..2864670f44913 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h
@@ -46,6 +46,15 @@ class VPDominatorTree : public DominatorTreeBase<VPBlockBase, false> {
   bool properlyDominates(const VPRecipeBase *A, const VPRecipeBase *B);
 };
 
+/// Template specialization of the standard LLVM post-dominator tree utility for
+/// VPBlockBases.
+class VPPostDominatorTree : public PostDomTreeBase<VPBlockBase> {
+  using Base = PostDomTreeBase<VPBlockBase>;
+
+public:
+  explicit VPPostDominatorTree(VPlan &Plan) { recalculate(Plan); }
+};
+
 using VPDomTreeNode = DomTreeNodeBase<VPBlockBase>;
 
 /// Template specializations of GraphTraits for VPDomTreeNode.
diff --git a/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp b/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp
index f7e7fc29bc203..b35d073c43192 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp
@@ -14,6 +14,7 @@
 #include "VPRecipeBuilder.h"
 #include "VPlan.h"
 #include "VPlanCFG.h"
+#include "VPlanDominatorTree.h"
 #include "VPlanPatternMatch.h"
 #include "VPlanTransforms.h"
 #include "VPlanUtils.h"
@@ -73,9 +74,18 @@ class VPPredicator {
     return EdgeMaskCache.lookup({Src, Dst});
   }
 
+  /// Copy the entry mask of block \p From to block \p To.
+  void copyBlockInMask(VPBasicBlock *To, VPBasicBlock *From) {
+    assert(BlockMaskCache.count(From) && "Source block mask not set");
+    setBlockInMask(To, getBlockInMask(From));
+  }
+
   /// Compute and return the mask for the vector loop header block.
   void createHeaderMask(VPBasicBlock *HeaderVPBB, bool FoldTail);
 
+  /// Compute the edge masks for all incoming edges to \p VPBB.
+  void createIncomingEdgeMasks(VPBasicBlock *VPBB);
+
   /// Compute and return the predicate of \p VPBB, assuming that the header
   /// block of the loop is set to True, or to the loop mask when tail folding.
   VPValue *createBlockInMask(VPBasicBlock *VPBB);
@@ -128,16 +138,22 @@ VPValue *VPPredicator::createEdgeMask(VPBasicBlock *Src, VPBasicBlock *Dst) {
   return setEdgeMask(Src, Dst, EdgeMask);
 }
 
-VPValue *VPPredicator::createBlockInMask(VPBasicBlock *VPBB) {
+void VPPredicator::createIncomingEdgeMasks(VPBasicBlock *VPBB) {
   // Start inserting after the block's phis, which be replaced by blends later.
   Builder.setInsertPoint(VPBB, VPBB->getFirstNonPhi());
+  for (auto *Predecessor : SetVector<VPBlockBase *>(
+           VPBB->getPredecessors().begin(), VPBB->getPredecessors().end()))
+    createEdgeMask(cast<VPBasicBlock>(Predecessor), VPBB);
+}
+
+VPValue *VPPredicator::createBlockInMask(VPBasicBlock *VPBB) {
   // All-one mask is modelled as no-mask following the convention for masked
   // load/store/gather/scatter. Initialize BlockMask to no-mask.
   VPValue *BlockMask = nullptr;
   // This is the block mask. We OR all unique incoming edges.
   for (auto *Predecessor : SetVector<VPBlockBase *>(
            VPBB->getPredecessors().begin(), VPBB->getPredecessors().end())) {
-    VPValue *EdgeMask = createEdgeMask(cast<VPBasicBlock>(Predecessor), VPBB);
+    VPValue *EdgeMask = getEdgeMask(cast<VPBasicBlock>(Predecessor), VPBB);
     if (!EdgeMask) { // Mask of predecessor is all-one so mask of block is
                      // too.
       setBlockInMask(VPBB, EdgeMask);
@@ -268,6 +284,7 @@ VPlanTransforms::introduceMasksAndLinearize(VPlan &Plan, bool FoldTail) {
   VPBasicBlock *Header = LoopRegion->getEntryBasicBlock();
   ReversePostOrderTraversal<VPBlockShallowTraversalWrapper<VPBlockBase *>> RPOT(
       Header);
+  VPPostDominatorTree VPPDT(Plan);
   VPPredicator Predicator;
   for (VPBlockBase *VPB : RPOT) {
     // Non-outer regions with VPBBs only are supported at the moment.
@@ -280,7 +297,13 @@ VPlanTransforms::introduceMasksAndLinearize(VPlan &Plan, bool FoldTail) {
       continue;
     }
 
-    Predicator.createBlockInMask(VPBB);
+    Predicator.createIncomingEdgeMasks(VPBB);
+    // Reuse the mask of header block if VPBB is control-flow equivalant to
+    // header.
+    if (VPPDT.properlyDominates(VPBB, Header))
+      Predicator.copyBlockInMask(VPBB, Header);
+    else
+      Predicator.createBlockInMask(VPBB);
     Predicator.convertPhisToBlends(VPBB);
   }
 
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll b/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
index 263c200c28801..180bc045f4eab 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
@@ -310,9 +310,6 @@ define void @multiple_blocks_with_dead_inst_multiple_successors_6(ptr %src, i1 %
 ; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
 ; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
 ; CHECK:       [[VECTOR_PH]]:
-; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i1> poison, i1 [[IC]], i64 0
-; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i1> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP8:%.*]] = xor <vscale x 8 x i1> [[BROADCAST_SPLAT]], splat (i1 true)
 ; CHECK-NEXT:    [[TMP11:%.*]] = call <vscale x 8 x i64> @llvm.stepvector.nxv8i64()
 ; CHECK-NEXT:    [[TMP5:%.*]] = mul nsw <vscale x 8 x i64> [[TMP11]], splat (i64 3)
 ; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
@@ -325,14 +322,7 @@ define void @multiple_blocks_with_dead_inst_multiple_successors_6(ptr %src, i1 %
 ; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[TMP16]], i64 0
 ; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i64> [[DOTSPLATINSERT]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = getelementptr i16, ptr [[SRC]], <vscale x 8 x i64> [[VEC_IND]]
-; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 8 x i16> @llvm.vp.gather.nxv8i16.nxv8p0(<vscale x 8 x ptr> align 2 [[TMP20]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP27]])
-; CHECK-NEXT:    [[TMP17:%.*]] = icmp eq <vscale x 8 x i16> [[WIDE_MASKED_GATHER]], zeroinitializer
-; CHECK-NEXT:    [[TMP14:%.*]] = select <vscale x 8 x i1> [[TMP17]], <vscale x 8 x i1> [[TMP8]], <vscale x 8 x i1> zeroinitializer
-; CHECK-NEXT:    [[TMP28:%.*]] = xor <vscale x 8 x i1> [[TMP17]], splat (i1 true)
-; CHECK-NEXT:    [[TMP22:%.*]] = or <vscale x 8 x i1> [[TMP14]], [[TMP28]]
-; CHECK-NEXT:    [[TMP23:%.*]] = select <vscale x 8 x i1> [[TMP17]], <vscale x 8 x i1> [[BROADCAST_SPLAT]], <vscale x 8 x i1> zeroinitializer
-; CHECK-NEXT:    [[TMP24:%.*]] = or <vscale x 8 x i1> [[TMP22]], [[TMP23]]
-; CHECK-NEXT:    call void @llvm.vp.scatter.nxv8i16.nxv8p0(<vscale x 8 x i16> zeroinitializer, <vscale x 8 x ptr> align 2 [[TMP20]], <vscale x 8 x i1> [[TMP24]], i32 [[TMP27]])
+; CHECK-NEXT:    call void @llvm.vp.scatter.nxv8i16.nxv8p0(<vscale x 8 x i16> zeroinitializer, <vscale x 8 x ptr> align 2 [[TMP20]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP27]])
 ; CHECK-NEXT:    [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP12]]
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nsw <vscale x 8 x i64> [[VEC_IND]], [[DOTSPLAT]]
 ; CHECK-NEXT:    [[TMP26:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-complex-mask.ll b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-complex-mask.ll
index 2ef5f55126c95..baa6da593716a 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-complex-mask.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-complex-mask.ll
@@ -21,8 +21,6 @@ define void @test(i64 %n, ptr noalias %src0, ptr noalias %src1, ptr noalias %src
 ; IF-EVL-NEXT:    [[TMP1:%.*]] = or <vscale x 4 x i1> [[BROADCAST_SPLAT]], [[BROADCAST_SPLAT2]]
 ; IF-EVL-NEXT:    [[TMP3:%.*]] = select <vscale x 4 x i1> [[TMP2]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i1> zeroinitializer
 ; IF-EVL-NEXT:    [[TMP4:%.*]] = or <vscale x 4 x i1> [[BROADCAST_SPLAT]], [[TMP3]]
-; IF-EVL-NEXT:    [[TMP5:%.*]] = xor <vscale x 4 x i1> [[TMP1]], splat (i1 true)
-; IF-EVL-NEXT:    [[TMP6:%.*]] = select <vscale x 4 x i1> [[TMP2]], <vscale x 4 x i1> [[TMP5]], <vscale x 4 x i1> zeroinitializer
 ; IF-EVL-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 4 x i1> poison, i1 [[C3]], i64 0
 ; IF-EVL-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 4 x i1> [[BROADCAST_SPLATINSERT3]], <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer
 ; IF-EVL-NEXT:    br label %[[VECTOR_BODY:.*]]
@@ -36,22 +34,18 @@ define void @test(i64 %n, ptr noalias %src0, ptr noalias %src1, ptr noalias %src
 ; IF-EVL-NEXT:    [[TMP9:%.*]] = icmp ult <vscale x 4 x i32> [[TMP8]], [[BROADCAST_SPLAT6]]
 ; IF-EVL-NEXT:    [[TMP10:%.*]] = getelementptr i32, ptr [[SRC0]], i64 [[EVL_BASED_IV]]
 ; IF-EVL-NEXT:    [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP10]], <vscale x 4 x i1> [[BROADCAST_SPLAT]], i32 [[TMP7]])
+; IF-EVL-NEXT:    [[TMP13:%.*]] = select <vscale x 4 x i1> [[TMP9]], <vscale x 4 x i1> [[TMP4]], <vscale x 4 x i1> zeroinitializer
 ; IF-EVL-NEXT:    [[PREDPHI:%.*]] = select <vscale x 4 x i1> [[TMP3]], <vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> [[VP_OP_LOAD]]
 ; IF-EVL-NEXT:    [[TMP11:%.*]] = getelementptr i32, ptr [[SRC1]], i64 [[EVL_BASED_IV]]
 ; IF-EVL-NEXT:    [[VP_OP_LOAD7:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP11]], <vscale x 4 x i1> [[TMP4]], i32 [[TMP7]])
 ; IF-EVL-NEXT:    [[TMP12:%.*]] = add <vscale x 4 x i32> [[VP_OP_LOAD7]], [[PREDPHI]]
-; IF-EVL-NEXT:    [[TMP13:%.*]] = select <vscale x 4 x i1> [[TMP9]], <vscale x 4 x i1> [[TMP6]], <vscale x 4 x i1> zeroinitializer
-; IF-EVL-NEXT:    [[TMP14:%.*]] = or <vscale x 4 x i1> [[TMP4]], [[TMP6]]
-; IF-EVL-NEXT:    [[TMP15:%.*]] = select <vscale x 4 x i1> [[TMP9]], <vscale x 4 x i1> [[TMP14]], <vscale x 4 x i1> zeroinitializer
-; IF-EVL-NEXT:    [[PREDPHI8:%.*]] = select <vscale x 4 x i1> [[TMP13]], <vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> [[TMP12]]
-; IF-EVL-NEXT:    [[TMP16:%.*]] = select <vscale x 4 x i1> [[TMP14]], <vscale x 4 x i1> [[BROADCAST_SPLAT4]], <vscale x 4 x i1> zeroinitializer
-; IF-EVL-NEXT:    [[TMP17:%.*]] = select <vscale x 4 x i1> [[TMP9]], <vscale x 4 x i1> [[TMP16]], <vscale x 4 x i1> zeroinitializer
+; IF-EVL-NEXT:    [[PREDPHI8:%.*]] = select <vscale x 4 x i1> [[TMP13]], <vscale x 4 x i32> [[TMP12]], <vscale x 4 x i32> zeroinitializer
 ; IF-EVL-NEXT:    [[TMP18:%.*]] = getelementptr i32, ptr [[SRC2]], i64 [[EVL_BASED_IV]]
-; IF-EVL-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr align 4 [[TMP18]], <vscale x 4 x i1> [[TMP17]], <vscale x 4 x i32> poison)
+; IF-EVL-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP18]], <vscale x 4 x i1> [[BROADCAST_SPLAT4]], i32 [[TMP7]])
 ; IF-EVL-NEXT:    [[TMP19:%.*]] = add <vscale x 4 x i32> [[WIDE_MASKED_LOAD]], [[PREDPHI8]]
 ; IF-EVL-NEXT:    [[PREDPHI9:%.*]] = select i1 [[C3]], <vscale x 4 x i32> [[TMP19]], <vscale x 4 x i32> [[PREDPHI8]]
 ; IF-EVL-NEXT:    [[TMP20:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 [[EVL_BASED_IV]]
-; IF-EVL-NEXT:    call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> [[PREDPHI9]], ptr align 4 [[TMP20]], <vscale x 4 x i1> [[TMP15]])
+; IF-EVL-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[PREDPHI9]], ptr align 4 [[TMP20]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP7]])
 ; IF-EVL-NEXT:    [[TMP21:%.*]] = zext i32 [[TMP7]] to i64
 ; IF-EVL-NEXT:    [[INDEX_EVL_NEXT]] = add i64 [[TMP21]], [[EVL_BASED_IV]]
 ; IF-EVL-NEXT:    [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP21]]
@@ -74,31 +68,27 @@ define void @test(i64 %n, ptr noalias %src0, ptr noalias %src1, ptr noalias %src
 ; NO-VP-NEXT:    [[TMP3:%.*]] = shl nuw i64 [[TMP2]], 2
 ; NO-VP-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
 ; NO-VP-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
-; NO-VP-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i1> poison, i1 [[C3]], i64 0
-; NO-VP-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i1> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer
 ; NO-VP-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 4 x i1> poison, i1 [[C1]], i64 0
 ; NO-VP-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 4 x i1> [[BROADCAST_SPLATINSERT1]], <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer
 ; NO-VP-NEXT:    [[TMP6:%.*]] = xor <vscale x 4 x i1> [[BROADCAST_SPLAT2]], splat (i1 true)
 ; NO-VP-NEXT:    [[TMP4:%.*]] = xor i1 [[C2]], true
 ; NO-VP-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 4 x i1> poison, i1 [[TMP4]], i64 0
 ; NO-VP-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 4 x i1> [[BROADCAST_SPLATINSERT3]], <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer
-; NO-VP-NEXT:    [[TMP5:%.*]] = or <vscale x 4 x i1> [[BROADCAST_SPLAT2]], [[BROADCAST_SPLAT4]]
-; NO-VP-NEXT:    [[TMP7:%.*]] = select <vscale x 4 x i1> [[TMP6]], <vscale x 4 x i1> [[TMP5]], <vscale x 4 x i1> zeroinitializer
-; NO-VP-NEXT:    [[TMP8:%.*]] = or <vscale x 4 x i1> [[BROADCAST_SPLAT2]], [[TMP7]]
-; NO-VP-NEXT:    [[TMP9:%.*]] = xor <vscale x 4 x i1> [[TMP5]], splat (i1 true)
+; NO-VP-NEXT:    [[TMP9:%.*]] = or <vscale x 4 x i1> [[BROADCAST_SPLAT2]], [[BROADCAST_SPLAT4]]
 ; NO-VP-NEXT:    [[TMP10:%.*]] = select <vscale x 4 x i1> [[TMP6]], <vscale x 4 x i1> [[TMP9]], <vscale x 4 x i1> zeroinitializer
-; NO-VP-NEXT:    [[TMP11:%.*]] = or <vscale x 4 x i1> [[TMP8]], [[TMP10]]
-; NO-VP-NEXT:    [[TMP12:%.*]] = select <vscale x 4 x i1> [[TMP11]], <vscale x 4 x i1> [[BROADCAST_SPLAT]], <vscale x 4 x i1> zeroinitializer
+; NO-VP-NEXT:    [[TMP8:%.*]] = or <vscale x 4 x i1> [[BROADCAST_SPLAT2]], [[TMP10]]
+; NO-VP-NEXT:    [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <vscale x 4 x i1> poison, i1 [[C3]], i64 0
+; NO-VP-NEXT:    [[TMP12:%.*]] = shufflevector <vscale x 4 x i1> [[BROADCAST_SPLATINSERT4]], <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer
 ; NO-VP-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; NO-VP:       [[VECTOR_BODY]]:
 ; NO-VP-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
 ; NO-VP-NEXT:    [[TMP13:%.*]] = getelementptr i32, ptr [[SRC0]], i64 [[INDEX]]
 ; NO-VP-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr align 4 [[TMP13]], <vscale x 4 x i1> [[BROADCAST_SPLAT2]], <vscale x 4 x i32> poison)
-; NO-VP-NEXT:    [[PREDPHI:%.*]] = select <vscale x 4 x i1> [[TMP7]], <vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> [[WIDE_MASKED_LOAD]]
+; NO-VP-NEXT:    [[PREDPHI:%.*]] = select <vscale x 4 x i1> [[TMP10]], <vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> [[WIDE_MASKED_LOAD]]
 ; NO-VP-NEXT:    [[TMP14:%.*]] = getelementptr i32, ptr [[SRC1]], i64 [[INDEX]]
 ; NO-VP-NEXT:    [[WIDE_MASKED_LOAD5:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr align 4 [[TMP14]], <vscale x 4 x i1> [[TMP8]], <vscale x 4 x i32> poison)
 ; NO-VP-NEXT:    [[TMP15:%.*]] = add <vscale x 4 x i32> [[WIDE_MASKED_LOAD5]], [[PREDPHI]]
-; NO-VP-NEXT:    [[PREDPHI6:%.*]] = select <vscale x 4 x i1> [[TMP10]], <vscale x 4 x i32> zeroinitializer, <vscale x 4 x i32> [[TMP15]]
+; NO-VP-NEXT:    [[PREDPHI6:%.*]] = select <vscale x 4 x i1> [[TMP8]], <vscale x 4 x i32> [[TMP15]], <vscale x 4 x i32> zeroinitializer
 ; NO-VP-NEXT:    [[TMP16:%.*]] = getelementptr i32, ptr [[SRC2]], i64 [[INDEX]]
 ; NO-VP-NEXT:    [[WIDE_MASKED_LOAD7:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr align 4 [[TMP16]], <vscale x 4 x i1> [[TMP12]], <vscale x 4 x i32> poison)
 ; NO-VP-NEXT:    [[TMP17:%.*]] = add <vscale x 4 x i32> [[WIDE_MASKED_LOAD7]], [[PREDPHI6]]
diff --git a/llvm/test/Transforms/LoopVectorize/cse-replicate-regions.ll b/llvm/test/Transforms/LoopVectorize/cse-replicate-regions.ll
index c0d603c22a5c8..7952745c1c703 100644
--- a/llvm/test/Transforms/LoopVectorize/cse-replicate-regions.ll
+++ b/llvm/test/Transforms/LoopVectorize/cse-replicate-regions.ll
@@ -97,8 +97,7 @@ define void @multiple_vppredinstphi_with_different_predicate(ptr %A, i32 %d) {
 ; CHECK-NEXT:    [[TMP9:%.*]] = phi <2 x i32> [ [[TMP5]], %[[PRED_SDIV_CONTINUE]] ], [ [[TMP8]], %[[PRED_SDIV_IF1]] ]
 ; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP1]], <2 x i32> [[TMP9]], <2 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP12:%.*]] = icmp sgt <2 x i32> [[WIDE_LOAD]], splat (i32 20)
-; CHECK-NEXT:    [[TMP13:%.*]] = select <2 x i1> splat (i1 true), <2 x i1> [[TMP12]], <2 x i1> zeroinitializer
-; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP13]], i32 0
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP12]], i32 0
 ; CHECK-NEXT:    br i1 [[TMP14]], label %[[PRED_SDIV_IF3:.*]], label %[[PRED_SDIV_CONTINUE4:.*]]
 ; CHECK:       [[PRED_SDIV_IF3]]:
 ; CHECK-NEXT:    [[TMP15:%.*]] = sdiv i32 -10, [[D]]
@@ -106,7 +105,7 @@ define void @multiple_vppredinstphi_with_different_predicate(ptr %A, i32 %d) {
 ; CHECK-NEXT:    br label %[[PRED_SDIV_CONTINUE4]]
 ; CHECK:       [[PRED_SDIV_CONTINUE4]]:
 ; CHECK-NEXT:    [[TMP17:%.*]] = phi <2 x i32> [ poison, %[[PRED_SDIV_CONTINUE2]] ], [ [[TMP16]], %[[PRED_SDIV_IF3]] ]
-; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <2 x i1> [[TMP13]], i32 1
+; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <2 x i1> [[TMP12]], i32 1
 ; CHECK-NEXT:    br i1 [[TMP18]], label %[[PRED_SDIV_IF5:.*]], label %[[PRED_SDIV_CONTINUE6]]
 ; CHECK:       [[PRED_SDIV_IF5]]:
 ; CHECK-NEXT:    [[TMP19:%.*]] = sdiv i32 -10, [[D]]
diff --git a/llvm/test/Transforms/LoopVectorize/hoist-predicated-loads-with-predicated-stores.ll b/llvm/test/Transforms/LoopVectorize/hoist-predicated-loads-with-predicated-stores.ll
index dca1c3fd08382..52a6693db84df 100644
--- a/llvm/test/Transforms/LoopVectorize/hoist-predicated-loads-with-predicated-stores.ll
+++ b/llvm/test/Transforms/LoopVectorize/hoist-predicated-loads-with-predicated-stores.ll
@@ -26,7 +26,7 @@ define void @test_stores_noalias_via_rt_checks_after_loads(ptr %dst, ptr %src, p
 ; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[INDEX]], 1
 ; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[COND]], i32 [[TMP4]]
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP6]], align 4, !alias.scope [[META0:![0-9]+]]
-; CHECK-NEXT:    [[TMP7:%.*]] = icmp ule <2 x i32> [[WIDE_LOAD]], splat (i32 11)
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ugt <2 x i32> [[WIDE_LOAD]], splat (i32 11)
 ; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 [[TMP4]]
 ; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[SRC]], i32 [[TMP5]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP10]], align 4, !alias.scope [[META3:![0-9]+]]
@@ -37,7 +37,7 @@ define void @test_stores_noalias_via_rt_checks_after_loads(ptr %dst, ptr %src, p
 ; CHECK-NEXT:    [[TMP36:%.*]] = add <2 x i32> [[TMP17]], splat (i32 10)
 ; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[TMP4]]
 ; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[TMP5]]
-; CHECK-NEXT:    [[TMP14:%.*]] = select <2 x i1> [[TMP7]], <2 x i32> [[TMP36]], <2 x i32> [[TMP19]]
+; CHECK-NEXT:    [[TMP14:%.*]] = select <2 x i1> [[TMP3]], <2 x i32> [[TMP19]], <2 x i32> [[TMP36]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <2 x i32> [[TMP14]], i32 0
 ; CHECK-NEXT:    store i32 [[TMP18]], ptr [[TMP21]], align 4, !alias.scope [[META5:![0-9]+]], !noalias [[META7:![0-9]+]]
 ; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <2 x i32> [[TMP14]], i32 1
@@ -1070,32 +1070,30 @@ define void @test_three_stores_with_different_predicates(ptr %dst, ptr %src, ptr
 ; CHECK-NEXT:    br label %[[PRED_STORE_CONTINUE3]]
 ; CHECK:       [[PRED_STORE_CONTINUE3]]:
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ule <2 x i32> [[WIDE_LOAD]], splat (i32 10)
-; CHECK-NEXT:    [[TMP11:%.*]] = select <2 x i1> splat (i1 true), <2 x i1> [[TMP10]], <2 x i1> zeroinitializer
-; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x i1> [[TMP11]], i32 0
+; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x i1> [[TMP10]], i32 0
 ; CHECK-NEXT:    br i1 [[TMP12]], label %[[PRED_STORE_IF4:.*]], label %[[PRED_STORE_CONTINUE5:.*]]
 ; CHECK:       [[PRED_STORE_IF4]]:
 ; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[TMP0]]
 ; CHECK-NEXT:    store i32 2, ptr [[TMP13]], align 4, !alias.scope [[META95]], !noalias [[META92]]
 ; CHECK-NEXT:    br label %[[PRED_STORE_CONTINUE5]]
 ; CHECK:       [[PRED_STORE_CONTINUE5]]:
-; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP11]], i32 1
-; CHECK-NEXT:    br i1 [[TMP14]], label %[[PRED_STORE_IF6:.*]], label %[[PRED_STORE_CONTINUE7:.*]]
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i1> [[TMP10]], i32 1
+; CHECK-NEXT:    br i1 [[TMP11]], label %[[PRED_STORE_IF6:.*]], label %[[PRED_STORE_CONTINUE7:.*]]
 ; CHECK:       [[PRED_STORE_IF6]]:
 ; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[TMP1]]
 ; CHECK-NEXT:    store i32 2, ptr [[TMP15]], align 4, !alias.scope [[META95]], !noalias [[META92]]
 ; CHECK-NEXT:    br label %[[PRED_STORE_CONTINUE7]]
 ; CHECK:       [[PRED_STORE_CONTINUE7]]:
 ; CHECK-NEXT:    [[TMP16:%.*]] = icmp ule <2 x i32> [[WIDE_LOAD]], splat (i32 9)
-; CHECK-NEXT:    [[TMP17:%.*]] = select <2 x i1> splat (i1 true), <2 x i1> [[TMP16]], <2 x i1> zeroinitializer
-; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <2 x i1> [[TMP17]], i32 0
+; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <2 x i1> [[TMP16]], i32 0
 ; CHECK-NEXT:    br i1 [[TMP18]], label %[[PRED_STORE_IF8:.*]], label %[[PRED_STORE_CONTINUE9:.*]]
 ; CHECK:       [[PRED_STORE_IF8]]:
 ; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[TMP0]]
 ; CHECK-NEXT:    store i32 3, ptr [[TMP19]], align 4, !alias.scope [[META95]], !noalias [[META92]]
 ; CHECK-NEXT:    br label %[[PRED_STORE_CONTINUE9]]
 ; CHECK:       [[PRED_STORE_CONTINUE9]]:
-; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <2 x i1> [[TMP17]], i32 1
-; CHECK-NEXT:    br i1 [[TMP20]], label %[[PRED_STORE_IF10:.*]], label %[[PRED_STORE_CONTINUE11]]
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <2 x i1> [[TMP16]], i32 1
+; CHECK-NEXT:    br i1 [[TMP17]], label %[[PRED_STORE_IF10:.*]], label %[[PRED_STORE_CONTINUE11]]
 ; CHECK:       [[PRED_STORE_IF10]]:
 ; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[TMP1]]
 ; CHECK-NEXT:    store i32 3, ptr [[TMP21]], align 4, !alias.scope [[META95]], !noalias [[META92]]

>From 813c2084b97cb62dd28d63a6dde3deeb8d1a813a Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Thu, 5 Feb 2026 07:09:17 -0800
Subject: [PATCH 2/2] fixup

---
 .../Transforms/Vectorize/VPlanPredicator.cpp  | 44 +++++++++++--------
 1 file changed, 26 insertions(+), 18 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp b/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp
index b35d073c43192..8a820e2f1f1b4 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp
@@ -28,6 +28,9 @@ class VPPredicator {
   /// Builder to construct recipes to compute masks.
   VPBuilder Builder;
 
+  /// Post-dominator tree for the VPlan.
+  VPPostDominatorTree VPPDT;
+
   /// When we if-convert we need to create edge masks. We have to cache values
   /// so that we don't end up with exponential recursion/IR.
   using EdgeMaskCacheTy =
@@ -45,6 +48,9 @@ class VPPredicator {
   /// possibly inserting new recipes at \p Dst (using Builder's insertion point)
   VPValue *createEdgeMask(VPBasicBlock *Src, VPBasicBlock *Dst);
 
+  /// Compute the edge masks for all incoming edges to \p VPBB.
+  void createIncomingEdgeMasks(VPBasicBlock *VPBB);
+
   /// Record \p Mask as the *entry* mask of \p VPBB, which is expected to not
   /// already have a mask.
   void setBlockInMask(VPBasicBlock *VPBB, VPValue *Mask) {
@@ -64,6 +70,8 @@ class VPPredicator {
   }
 
 public:
+  VPPredicator(VPlan &Plan) : VPPDT(Plan) {}
+
   /// Returns the *entry* mask for \p VPBB.
   VPValue *getBlockInMask(VPBasicBlock *VPBB) const {
     return BlockMaskCache.lookup(VPBB);
@@ -74,18 +82,9 @@ class VPPredicator {
     return EdgeMaskCache.lookup({Src, Dst});
   }
 
-  /// Copy the entry mask of block \p From to block \p To.
-  void copyBlockInMask(VPBasicBlock *To, VPBasicBlock *From) {
-    assert(BlockMaskCache.count(From) && "Source block mask not set");
-    setBlockInMask(To, getBlockInMask(From));
-  }
-
   /// Compute and return the mask for the vector loop header block.
   void createHeaderMask(VPBasicBlock *HeaderVPBB, bool FoldTail);
 
-  /// Compute the edge masks for all incoming edges to \p VPBB.
-  void createIncomingEdgeMasks(VPBasicBlock *VPBB);
-
   /// Compute and return the predicate of \p VPBB, assuming that the header
   /// block of the loop is set to True, or to the loop mask when tail folding.
   VPValue *createBlockInMask(VPBasicBlock *VPBB);
@@ -150,6 +149,22 @@ VPValue *VPPredicator::createBlockInMask(VPBasicBlock *VPBB) {
   // All-one mask is modelled as no-mask following the convention for masked
   // load/store/gather/scatter. Initialize BlockMask to no-mask.
   VPValue *BlockMask = nullptr;
+
+  // TODO: Skip creating edge masks for blocks that are control-flow equivalent
+  // to header and have no phis.
+  createIncomingEdgeMasks(VPBB);
+
+  // Reuse the mask of header block if VPBB is control-flow equivalent to
+  // header.
+  // TODO: Generalize to reuse mask of immediate dominator.
+  VPBasicBlock *Header =
+      VPBB->getPlan()->getVectorLoopRegion()->getEntryBasicBlock();
+  if (VPPDT.properlyDominates(VPBB, Header)) {
+    BlockMask = getBlockInMask(Header);
+    setBlockInMask(VPBB, BlockMask);
+    return BlockMask;
+  }
+
   // This is the block mask. We OR all unique incoming edges.
   for (auto *Predecessor : SetVector<VPBlockBase *>(
            VPBB->getPredecessors().begin(), VPBB->getPredecessors().end())) {
@@ -284,8 +299,7 @@ VPlanTransforms::introduceMasksAndLinearize(VPlan &Plan, bool FoldTail) {
   VPBasicBlock *Header = LoopRegion->getEntryBasicBlock();
   ReversePostOrderTraversal<VPBlockShallowTraversalWrapper<VPBlockBase *>> RPOT(
       Header);
-  VPPostDominatorTree VPPDT(Plan);
-  VPPredicator Predicator;
+  VPPredicator Predicator(Plan);
   for (VPBlockBase *VPB : RPOT) {
     // Non-outer regions with VPBBs only are supported at the moment.
     auto *VPBB = cast<VPBasicBlock>(VPB);
@@ -297,13 +311,7 @@ VPlanTransforms::introduceMasksAndLinearize(VPlan &Plan, bool FoldTail) {
       continue;
     }
 
-    Predicator.createIncomingEdgeMasks(VPBB);
-    // Reuse the mask of header block if VPBB is control-flow equivalant to
-    // header.
-    if (VPPDT.properlyDominates(VPBB, Header))
-      Predicator.copyBlockInMask(VPBB, Header);
-    else
-      Predicator.createBlockInMask(VPBB);
+    Predicator.createBlockInMask(VPBB);
     Predicator.convertPhisToBlends(VPBB);
   }
 



More information about the llvm-commits mailing list