[VPlan] Connect Entry to scalar preheader during initial construction. #140132

fhahn · 2025-05-15T20:14:20Z

Update initial construction to connect the Plan's entry to the scalar preheader during initial construction. This moves a small part of the
skeleton creation out of ILV and will also enable replacing
VPInstruction::ResumePhi with regular VPPhi recipes.

Resume phis need 2 incoming values to start with, the second being the bypass value from the scalar ph (and used to replicate the incoming value for other bypass blocks). Adding the extra edge ensures we incoming values for resume phis match the incoming blocks.

llvmbot · 2025-05-15T20:30:22Z

@llvm/pr-subscribers-backend-risc-v
@llvm/pr-subscribers-backend-powerpc
@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-vectorizers

Author: Florian Hahn (fhahn)

Changes

Update initial construction to connect the Plan's entry to the scalar preheader during initial construction. This moves a small part of the
skeleton creation out of ILV and will also enable replacing
VPInstruction::ResumePhi with regular VPPhi recipes.

Resume phis need 2 incoming values to start with, the second being the bypass value from the scalar ph (and used to replicate the incoming value for other bypass blocks). Adding the extra edge ensures we incoming values for resume phis match the incoming blocks.

Still needs various VPlan printing tests to be updated. vplan-printing.ll shows the changes.

Patch is 22.17 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/140132.diff

5 Files Affected:

(modified) llvm/lib/Transforms/Vectorize/LoopVectorize.cpp (+10-16)
(modified) llvm/lib/Transforms/Vectorize/VPlan.h (+2-1)
(modified) llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp (+6)
(modified) llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp (+1-1)
(modified) llvm/test/Transforms/LoopVectorize/vplan-printing.ll (+53-53)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 7ad02956a5b69..a43c72451eaff 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -490,7 +490,7 @@ class InnerLoopVectorizer {
         MinProfitableTripCount(MinProfitableTripCount), UF(UnrollFactor),
         Builder(PSE.getSE()->getContext()), Cost(CM), BFI(BFI), PSI(PSI),
         RTChecks(RTChecks), Plan(Plan),
-        VectorPHVPB(Plan.getEntry()->getSingleSuccessor()) {}
+        VectorPHVPB(Plan.getEntry()->getSuccessors()[1]) {}
 
   virtual ~InnerLoopVectorizer() = default;
 
@@ -2365,14 +2365,11 @@ InnerLoopVectorizer::getOrCreateVectorTripCount(BasicBlock *InsertBlock) {
 void InnerLoopVectorizer::introduceCheckBlockInVPlan(BasicBlock *CheckIRBB) {
   VPBlockBase *ScalarPH = Plan.getScalarPreheader();
   VPBlockBase *PreVectorPH = VectorPHVPB->getSinglePredecessor();
-  if (PreVectorPH->getNumSuccessors() != 1) {
-    assert(PreVectorPH->getNumSuccessors() == 2 && "Expected 2 successors");
-    assert(PreVectorPH->getSuccessors()[0] == ScalarPH &&
-           "Unexpected successor");
-    VPIRBasicBlock *CheckVPIRBB = Plan.createVPIRBasicBlock(CheckIRBB);
-    VPBlockUtils::insertOnEdge(PreVectorPH, VectorPHVPB, CheckVPIRBB);
-    PreVectorPH = CheckVPIRBB;
-  }
+  assert(PreVectorPH->getNumSuccessors() == 2 && "Expected 2 successors");
+  assert(PreVectorPH->getSuccessors()[0] == ScalarPH && "Unexpected successor");
+  VPIRBasicBlock *CheckVPIRBB = Plan.createVPIRBasicBlock(CheckIRBB);
+  VPBlockUtils::insertOnEdge(PreVectorPH, VectorPHVPB, CheckVPIRBB);
+  PreVectorPH = CheckVPIRBB;
   VPBlockUtils::connectBlocks(PreVectorPH, ScalarPH);
   PreVectorPH->swapSuccessors();
 
@@ -2463,9 +2460,6 @@ void InnerLoopVectorizer::emitIterationCountCheck(BasicBlock *Bypass) {
     setBranchWeights(BI, MinItersBypassWeights, /*IsExpected=*/false);
   ReplaceInstWithInst(TCCheckBlock->getTerminator(), &BI);
   LoopBypassBlocks.push_back(TCCheckBlock);
-
-  // TODO: Wrap LoopVectorPreHeader in VPIRBasicBlock here.
-  introduceCheckBlockInVPlan(TCCheckBlock);
 }
 
 BasicBlock *InnerLoopVectorizer::emitSCEVChecks(BasicBlock *Bypass) {
@@ -7838,7 +7832,7 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan(
 
   // 1. Set up the skeleton for vectorization, including vector pre-header and
   // middle block. The vector loop is created during VPlan execution.
-  VPBasicBlock *VectorPH = cast<VPBasicBlock>(Entry->getSingleSuccessor());
+  VPBasicBlock *VectorPH = cast<VPBasicBlock>(Entry->getSuccessors()[1]);
   State.CFG.PrevBB = ILV.createVectorizedLoopSkeleton();
   if (VectorizingEpilogue)
     VPlanTransforms::removeDeadRecipes(BestVPlan);
@@ -8068,7 +8062,8 @@ EpilogueVectorizerMainLoop::emitIterationCountCheck(BasicBlock *Bypass,
     setBranchWeights(BI, MinItersBypassWeights, /*IsExpected=*/false);
   ReplaceInstWithInst(TCCheckBlock->getTerminator(), &BI);
 
-  introduceCheckBlockInVPlan(TCCheckBlock);
+  if (!ForEpilogue)
+    introduceCheckBlockInVPlan(TCCheckBlock);
   return TCCheckBlock;
 }
 
@@ -8198,7 +8193,6 @@ EpilogueVectorizerEpilogueLoop::emitMinimumVectorEpilogueIterCountCheck(
   Plan.setEntry(NewEntry);
   // OldEntry is now dead and will be cleaned up when the plan gets destroyed.
 
-  introduceCheckBlockInVPlan(Insert);
   return Insert;
 }
 
@@ -9158,7 +9152,7 @@ static void addScalarResumePhis(VPRecipeBuilder &Builder, VPlan &Plan,
                                 DenseMap<VPValue *, VPValue *> &IVEndValues) {
   VPTypeAnalysis TypeInfo(Plan.getCanonicalIV()->getScalarType());
   auto *ScalarPH = Plan.getScalarPreheader();
-  auto *MiddleVPBB = cast<VPBasicBlock>(ScalarPH->getSinglePredecessor());
+  auto *MiddleVPBB = cast<VPBasicBlock>(ScalarPH->getPredecessors()[0]);
   VPRegionBlock *VectorRegion = Plan.getVectorLoopRegion();
   VPBuilder VectorPHBuilder(
       cast<VPBasicBlock>(VectorRegion->getSinglePredecessor()));
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 5fd7a369bf735..df4ba857d4e3a 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -3995,7 +3995,8 @@ class VPlan {
   /// that this relies on unneeded branches to the scalar tail loop being
   /// removed.
   bool hasScalarTail() const {
-    return getScalarPreheader()->getNumPredecessors() != 0;
+    return getScalarPreheader()->getNumPredecessors() != 0 &&
+           getScalarPreheader()->getSinglePredecessor() != getEntry();
   }
 };
 
diff --git a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
index 287bc93ce496a..78c6e2f98387f 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp
@@ -546,6 +546,9 @@ void VPlanTransforms::prepareForVectorization(
     if (auto *LatchExitVPB = MiddleVPBB->getSingleSuccessor())
       VPBlockUtils::disconnectBlocks(MiddleVPBB, LatchExitVPB);
     VPBlockUtils::connectBlocks(MiddleVPBB, ScalarPH);
+    VPBlockUtils::connectBlocks(Plan.getEntry(), ScalarPH);
+    Plan.getEntry()->swapSuccessors();
+
     // The exit blocks are unreachable, remove their recipes to make sure no
     // users remain that may pessimize transforms.
     for (auto *EB : Plan.getExitBlocks()) {
@@ -558,6 +561,9 @@ void VPlanTransforms::prepareForVectorization(
   // The connection order corresponds to the operands of the conditional branch,
   // with the middle block already connected to the exit block.
   VPBlockUtils::connectBlocks(MiddleVPBB, ScalarPH);
+  // Also connect the entry block to the scalar preheader.
+  VPBlockUtils::connectBlocks(Plan.getEntry(), ScalarPH);
+  Plan.getEntry()->swapSuccessors();
 
   auto *ScalarLatchTerm = TheLoop->getLoopLatch()->getTerminator();
   // Here we use the same DebugLoc as the scalar loop latch terminator instead
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 806c20ef8cf73..0dcb6536e9985 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -1733,7 +1733,7 @@ static void removeBranchOnCondTrue(VPlan &Plan) {
   using namespace llvm::VPlanPatternMatch;
   for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
            vp_depth_first_shallow(Plan.getEntry()))) {
-    if (VPBB->getNumSuccessors() != 2 ||
+    if (VPBB->getNumSuccessors() != 2 || isa<VPIRBasicBlock>(VPBB) ||
         !match(&VPBB->back(), m_BranchOnCond(m_True())))
       continue;
 
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-printing.ll b/llvm/test/Transforms/LoopVectorize/vplan-printing.ll
index 727784de2cb6d..3504558d1b545 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-printing.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-printing.ll
@@ -15,7 +15,7 @@ define void @print_call_and_memory(i64 %n, ptr noalias %y, ptr noalias %x) nounw
 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<for.body.preheader>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -42,6 +42,9 @@ define void @print_call_and_memory(i64 %n, ptr noalias %y, ptr noalias %x) nounw
 ; CHECK-NEXT:    EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT:  Successor(s): ir-bb<for.end.loopexit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT:  ir-bb<for.end.loopexit>
+; CHECK-NEXT:  No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT:  scalar.ph
 ; CHECK-NEXT:    EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<for.body>
@@ -50,9 +53,6 @@ define void @print_call_and_memory(i64 %n, ptr noalias %y, ptr noalias %x) nounw
 ; CHECK-NEXT:    IR   %iv = phi i64 [ %iv.next, %for.body ], [ 0, %for.body.preheader ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %exitcond = icmp eq i64 %iv.next, %n
 ; CHECK-NEXT:  No successors
-; CHECK-EMPTY:
-; CHECK-NEXT:  ir-bb<for.end.loopexit>
-; CHECK-NEXT:  No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -83,7 +83,7 @@ define void @print_widen_gep_and_select(i64 %n, ptr noalias %y, ptr noalias %x,
 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<for.body.preheader>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -113,6 +113,9 @@ define void @print_widen_gep_and_select(i64 %n, ptr noalias %y, ptr noalias %x,
 ; CHECK-NEXT:    EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT:  Successor(s): ir-bb<for.end.loopexit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT:  ir-bb<for.end.loopexit>
+; CHECK-NEXT:  No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT:  scalar.ph
 ; CHECK-NEXT:    EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<for.body>
@@ -121,9 +124,6 @@ define void @print_widen_gep_and_select(i64 %n, ptr noalias %y, ptr noalias %x,
 ; CHECK-NEXT:    IR   %iv = phi i64 [ %iv.next, %for.body ], [ 0, %for.body.preheader ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %exitcond = icmp eq i64 %iv.next, %n
 ; CHECK-NEXT:  No successors
-; CHECK-EMPTY:
-; CHECK-NEXT:  ir-bb<for.end.loopexit>
-; CHECK-NEXT:  No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -157,7 +157,7 @@ define void @print_replicate_predicated_phi(i64 %n, ptr %x) {
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
 ; CHECK-NEXT:  EMIT vp<[[TC]]> = EXPAND SCEV (1 smax %n)
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -202,6 +202,9 @@ define void @print_replicate_predicated_phi(i64 %n, ptr %x) {
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<for.end>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<for.end>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<for.body>
@@ -210,9 +213,6 @@ define void @print_replicate_predicated_phi(i64 %n, ptr %x) {
 ; CHECK-NEXT:    IR   %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK-NEXT:    IR   %cmp = icmp ult i64 %i, 5
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<for.end>
-; CHECK-NEXT: No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -251,7 +251,7 @@ define void @print_interleave_groups(i32 %C, i32 %D) {
 ; CHECK-NEXT: Live-in ir<256> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT:   vp<[[IV_END:%.+]]> = DERIVED-IV ir<0> + vp<[[VTC]]> * ir<4>
@@ -285,6 +285,9 @@ define void @print_interleave_groups(i32 %C, i32 %D) {
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<for.end>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<for.end>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[IV_END]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<for.body>
@@ -293,9 +296,6 @@ define void @print_interleave_groups(i32 %C, i32 %D) {
 ; CHECK-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %cmp = icmp slt i64 %iv.next, 1024
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<for.end>
-; CHECK-NEXT: No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -338,7 +338,7 @@ define void @debug_loc_vpinstruction(ptr nocapture %asd, ptr nocapture %bsd) !db
 ; CHECK-NEXT: Live-in ir<128> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -390,6 +390,9 @@ define void @debug_loc_vpinstruction(ptr nocapture %asd, ptr nocapture %bsd) !db
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<exit>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph:
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<loop>
@@ -398,9 +401,6 @@ define void @debug_loc_vpinstruction(ptr nocapture %asd, ptr nocapture %bsd) !db
 ; CHECK-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %if.end ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %cmp1 = icmp slt i32 %lsd, 100
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<exit>
-; CHECK-NEXT: No successors
 ; CHECK-NEXT:}
 ;
 entry:
@@ -449,7 +449,7 @@ define void @print_expand_scev(i64 %y, ptr %ptr) {
 ; CHECK-NEXT:   IR %inc = add i64 %div, 1
 ; CHECK-NEXT:   EMIT vp<[[TC]]> = EXPAND SCEV (1 + ((15 + (%y /u 492802768830814060))<nuw><nsw> /u (1 + (%y /u 492802768830814060))<nuw><nsw>))<nuw><nsw>
 ; CHECK-NEXT:   EMIT vp<[[EXP_SCEV:%.+]]> = EXPAND SCEV (1 + (%y /u 492802768830814060))<nuw><nsw>
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT:   vp<[[IV_END:%.+]]> = DERIVED-IV ir<0> + vp<[[VTC]]> * vp<[[EXP_SCEV]]>
@@ -475,6 +475,9 @@ define void @print_expand_scev(i64 %y, ptr %ptr) {
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<loop.exit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<loop.exit>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[IV_END]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<loop>
@@ -483,9 +486,6 @@ define void @print_expand_scev(i64 %y, ptr %ptr) {
 ; CHECK-NEXT:    IR   %iv = phi i64 [ %iv.next, %loop ], [ 0, %entry ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %iv.next = add i64 %iv, %inc
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<loop.exit>
-; CHECK-NEXT: No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -517,7 +517,7 @@ define i32 @print_exit_value(ptr %ptr, i32 %off) {
 ; CHECK-NEXT: Live-in ir<1000> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -543,6 +543,10 @@ define i32 @print_exit_value(ptr %ptr, i32 %off) {
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<exit>
+; CHECK-NEXT:   IR %lcssa = phi i32 [ %add, %loop ] (extra operand: vp<[[EXIT]]> from middle.block)
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<loop>
@@ -551,10 +555,6 @@ define i32 @print_exit_value(ptr %ptr, i32 %off) {
 ; CHECK-NEXT:    IR   %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %ec = icmp eq i32 %iv.next, 1000
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<exit>
-; CHECK-NEXT:   IR %lcssa = phi i32 [ %add, %loop ] (extra operand: vp<[[EXIT]]> from middle.block)
-; CHECK-NEXT: No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -583,7 +583,7 @@ define void @print_fast_math_flags(i64 %n, ptr noalias %y, ptr noalias %x, ptr %
 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -612,6 +612,9 @@ define void @print_fast_math_flags(i64 %n, ptr noalias %y, ptr noalias %x, ptr %
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<exit>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<loop>
@@ -620,9 +623,6 @@ define void @print_fast_math_flags(i64 %n, ptr noalias %y, ptr noalias %x, ptr %
 ; CHECK-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
 ; CHECK:         IR   %exitcond = icmp eq i64 %iv.next, %n
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<exit>
-; CHECK-NEXT: No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -654,7 +654,7 @@ define void @print_exact_flags(i64 %n, ptr noalias %x) {
 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -682,6 +682,9 @@ define void @print_exact_flags(i64 %n, ptr noalias %x) {
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<exit>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<loop>
@@ -690,10 +693,7 @@ define void @print_exact_flags(i64 %n, ptr noalias %x) {
 ; CHECK-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %exitcond = icmp eq i64 %iv.next, %n
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<exit>
-; CHECK-NEXT: No successors
-; CHECK-NEXT: }
+; ; CHECK-NEXT: }
 ;
 entry:
   br label %loop
@@ -723,7 +723,7 @@ define void @print_call_flags(ptr readonly %src, ptr noalias %dest, i64 %n) {
 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -772,6 +772,9 @@ define void @print_call_flags(ptr readonly %src, ptr noalias %dest, i64 %n) {
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<end>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<end>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<for.body>
@@ -780,9 +783,6 @@ define void @print_call_flags(ptr readonly %src, ptr noalias %dest, i64 %n) {
 ; CHECK-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.loop ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %ifcond = fcmp oeq float %ld.value, 5.0
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<end>
-; CHECK-NEXT: No successors
 ; CHECK-NEXT: }
 ;
 entry:
@@ -823,7 +823,7 @@ define void @print_disjoint_flags(i64 %n, ptr noalias %x) {
 ; CHECK-NEXT: Live-in ir<%n> = original trip-count
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<entry>:
-; CHECK-NEXT: Successor(s): vector.ph
+; CHECK-NEXT: Successor(s): scalar.ph, vector.ph
 ; CHECK-EMPTY:
 ; CHECK-NEXT: vector.ph:
 ; CHECK-NEXT: Successor(s): vector loop
@@ -851,6 +851,9 @@ define void @print_disjoint_flags(i64 %n, ptr noalias %x) {
 ; CHECK-NEXT:   EMIT branch-on-cond vp<[[CMP]]>
 ; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph
 ; CHECK-EMPTY:
+; CHECK-NEXT: ir-bb<exit>
+; CHECK-NEXT: No successors
+; CHECK-EMPTY:
 ; CHECK-NEXT: scalar.ph
 ; CHECK-NEXT:   EMIT vp<[[RESUME_IV:%.+]]> = resume-phi vp<[[VTC]]>, ir<0>
 ; CHECK-NEXT:  Successor(s): ir-bb<loop>
@@ -859,9 +862,6 @@ define void @print_disjoint_flags(i64 %n, ptr noalias %x) {
 ; CHECK-NEXT:    IR   %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] (extra operand: vp<[[RESUME_IV]]> from scalar.ph)
 ; CHECK:         IR   %exitcond = icmp eq i64 %iv.next, %n
 ; CHECK-NEXT: No successors
-; CHECK-EMPTY:
-; CHECK-NEXT: ir-bb<exit...
[truncated]

Update initial construction to connect the Plan's entry to the scalar preheader during initial construction. This moves a small part of the skeleton creation out of ILV and will also enable replacing VPInstruction::ResumePhi with regular VPPhi recipes. Resume phis need 2 incoming values to start with, the second being the bypass value from the scalar ph (and used to replicate the incoming value for other bypass blocks). Adding the extra edge ensures we incoming values for resume phis match the incoming blocks. Still needs various VPlan printing tests to be updated. vplan-printing.ll shows the changes.

Use regular VPPhi instead of a separate opcode for resume phis. This removes an unneeded specialized opcode and unifies the code (verification, printing, updating when CFG is changed). Depends on llvm#140132.

ayalz · 2025-05-16T22:13:25Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -490,7 +490,7 @@ class InnerLoopVectorizer {
        MinProfitableTripCount(MinProfitableTripCount), UF(UnrollFactor),
        Builder(PSE.getSE()->getContext()), Cost(CM), BFI(BFI), PSI(PSI),
        RTChecks(RTChecks), Plan(Plan),
-        VectorPHVPB(Plan.getEntry()->getSingleSuccessor()) {}
+        VectorPHVPB(Plan.getEntry()->getSuccessors()[1]) {}


So now Plan->entry has ScalarPH as its first successor and VectorPH as its second, rather than the latter as its only successor.

Good to assert there are two successors. Perhaps some API to retrieve "1of2" and "2of2", analogous to existing "1of1"?

Better look instead for first predecessor (among two) of first header block? Possibly as follow-up.

Updated to get the single predecessor of the vector loop region, which should be clearer, thanks

ayalz · 2025-05-20T17:52:44Z

llvm/lib/Transforms/Vectorize/VPlan.h

+    return getScalarPreheader()->getNumPredecessors() != 0 &&
+           getScalarPreheader()->getSinglePredecessor() != getEntry();


Perhaps clearer as:

Suggested change

return getScalarPreheader()->getNumPredecessors() != 0 &&

getScalarPreheader()->getSinglePredecessor() != getEntry();

return !(getScalarPreheader()->getNumPredecessors() == 0 ||

getScalarPreheader()->getSinglePredecessor() == getEntry());

I.e., scalar loop does not take care of leftover iterations if (a) it doesn't take care of anything, or (b) it only takes care of all iterations as an alternative to the vector loop?
The latter may reach scalar loop from various bypassing blocks?

Updated thanks

ayalz · 2025-05-20T19:46:29Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

-
-  // TODO: Wrap LoopVectorPreHeader in VPIRBasicBlock here.
-  introduceCheckBlockInVPlan(TCCheckBlock);


Connecting entry VPBB to scalar preheader VPBB is done here as part of introducing the conditional branch at end of entry IRBB?

TCCheckBlock is the same block as the VPlan entry block. Before the change, we needed special logic to not create a new VPIRBB for this call.

TCCheckBlock is the same block as the VPlan entry block.

Worth an assert to clarify?

Added thanks

ayalz · 2025-05-20T19:47:27Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -8070,7 +8064,8 @@ EpilogueVectorizerMainLoop::emitIterationCountCheck(BasicBlock *Bypass,
    setBranchWeights(BI, MinItersBypassWeights, /*IsExpected=*/false);
  ReplaceInstWithInst(TCCheckBlock->getTerminator(), &BI);

-  introduceCheckBlockInVPlan(TCCheckBlock);
+  if (!ForEpilogue)


Suggested change

if (!ForEpilogue)

// Worth explaining?

if (!ForEpilogue)

Added thanks

ayalz · 2025-05-20T19:51:39Z

llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp

+  // Also connect the entry block to the scalar preheader.
+  VPBlockUtils::connectBlocks(Plan.getEntry(), ScalarPH);


The CFG is less consistent with VPlan's recipes, connecting entry VPBB to scalarPH VPBB now, when it is already connected to vectorPH, w/o introducing a conditional branch recipe at the end of entry - the branch instruction is generated outside VPlan's execute.

Yep, we still need to allow terminator-less VPIRBBs for various parts of the skeleton. One of the next steps after adding it early here is adding a branch recipe.

Temporary inconsistency is worth clarifying in a TODO.

Added thanks

…ph-early

ayalz

LGTM, thanks! Adding several comments.

ayalz · 2025-05-26T16:53:17Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -490,7 +490,7 @@ class InnerLoopVectorizer {
        MinProfitableTripCount(MinProfitableTripCount), UF(UnrollFactor),
        Builder(PSE.getSE()->getContext()), Cost(CM), BFI(BFI), PSI(PSI),
        RTChecks(RTChecks), Plan(Plan),
-        VectorPHVPB(Plan.getEntry()->getSingleSuccessor()) {}
+        VectorPHVPB(Plan.getVectorLoopRegion()->getSinglePredecessor()) {}


Note that ILV now depends on HCFG here, and elsewhere. Maybe better to retrieve the first hierarchical predecessor of first header block, to relax this dependence, possibly as follow up.

Sounds good!

ayalz · 2025-05-26T19:37:46Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -2368,14 +2368,11 @@ InnerLoopVectorizer::getOrCreateVectorTripCount(BasicBlock *InsertBlock) {
 void InnerLoopVectorizer::introduceCheckBlockInVPlan(BasicBlock *CheckIRBB) {


Clarify that CheckBlock now excludes the initial trip-count check, which is expected to be already introduced before calling introduceCheckBlockInVPlan()?

Added a note, thanks

ayalz · 2025-05-26T22:31:41Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

-
-  // TODO: Wrap LoopVectorPreHeader in VPIRBasicBlock here.
-  introduceCheckBlockInVPlan(TCCheckBlock);


TCCheckBlock is the same block as the VPlan entry block.

Worth an assert to clarify?

ayalz · 2025-05-27T08:51:37Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

+  // Only generate add a new block for the trip-count check for the main loop.
+  // The epilogue will use the already existing VPlan entry block.


Suggested change

// Only generate add a new block for the trip-count check for the main loop.

// The epilogue will use the already existing VPlan entry block.

// When vectorizing the main loop, its trip-count check is placed in a new block, whereas the overall trip-count check is placed in the VPlan entry block.

// When vectorizing the epilogue loop, its trip-count check is placed in the VPlan entry block.

sounds clearer? Referring to https://llvm.org/docs/Vectorizers.html#epilogue-vectorization

updated thanks

ayalz · 2025-05-27T08:54:21Z

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

  auto *ScalarPH = Plan.getScalarPreheader();
-  auto *MiddleVPBB = cast<VPBasicBlock>(ScalarPH->getSinglePredecessor());
+  auto *MiddleVPBB = cast<VPBasicBlock>(ScalarPH->getPredecessors()[0]);


Suggesting a getMiddleBlock(Plan) utility?

Sounds good, will check where this could be useful separately.

ayalz · 2025-05-27T09:01:04Z

llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp

@@ -540,6 +540,9 @@ void VPlanTransforms::prepareForVectorization(
    if (auto *LatchExitVPB = MiddleVPBB->getSingleSuccessor())


Independent: worth clarifying that not requiring a scalar epilog check means the scalar epilog is (always) required, i.e., case 1.

ayalz · 2025-05-27T09:03:16Z

llvm/lib/Transforms/Vectorize/VPlanConstruction.cpp

+  // Also connect the entry block to the scalar preheader.
+  VPBlockUtils::connectBlocks(Plan.getEntry(), ScalarPH);


Temporary inconsistency is worth clarifying in a TODO.

ayalz · 2025-05-27T09:05:54Z

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

@@ -1846,7 +1846,7 @@ static void removeBranchOnCondTrue(VPlan &Plan) {
  using namespace llvm::VPlanPatternMatch;
  for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
           vp_depth_first_shallow(Plan.getEntry()))) {
-    if (VPBB->getNumSuccessors() != 2 ||
+    if (VPBB->getNumSuccessors() != 2 || isa<VPIRBasicBlock>(VPBB) ||


Excluding VPIRBB's only because of the temporary terminator-less entry block with two successors? If so, suffice to exclude VPBB == Plan.getEntry()? But from an earlier comment: "we still need to allow terminator-less VPIRBBs for various parts of the skeleton."

Updated, thanks! We don't run removeBranchOnCond after introducing the other, terminator-less VPIRBBs.

ayalz · 2025-05-27T09:12:23Z

llvm/test/Transforms/LoopVectorize/first-order-recurrence-sink-replicate-region.ll

@@ -484,7 +479,6 @@ define void @need_new_block_after_sinking_pr56146(i32 %x, ptr %src, ptr noalias
 ; CHECK-EMPTY:
 ; CHECK-NEXT: ir-bb<exit>
 ; CHECK-NEXT: No successors
-; CHECK-NEXT: }


Need to drop the check for }? Here and above.

Yep, otherwise we would have to match scalar.ph and scalar header here as well, which isn't really relevant for the test

…ph-early

llvm-ci · 2025-05-27T15:14:37Z

LLVM Buildbot has detected a new failure on builder lldb-x86_64-debian running on lldb-x86_64-debian while building llvm at step 6 "test".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/162/builds/23251

Here is the relevant piece of the build log for the reference

Step 6 (test) failure: build (failure)
...
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Lua/persistent_state.test (2950 of 2961)
UNSUPPORTED: lldb-shell :: Process/Windows/exception_access_violation.cpp (2951 of 2961)
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Lua/lua-python.test (2952 of 2961)
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Lua/io.test (2953 of 2961)
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Lua/watchpoint_callback.test (2954 of 2961)
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Lua/lua.test (2955 of 2961)
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Python/Crashlog/parser_json.test (2956 of 2961)
PASS: lldb-api :: api/multithreaded/TestMultithreaded.py (2957 of 2961)
PASS: lldb-api :: terminal/TestEditlineCompletions.py (2958 of 2961)
UNRESOLVED: lldb-api :: tools/lldb-dap/launch/TestDAP_launch.py (2959 of 2961)
******************** TEST 'lldb-api :: tools/lldb-dap/launch/TestDAP_launch.py' FAILED ********************
Script:
--
/usr/bin/python3 /home/worker/2.0.1/lldb-x86_64-debian/llvm-project/lldb/test/API/dotest.py -u CXXFLAGS -u CFLAGS --env LLVM_LIBS_DIR=/home/worker/2.0.1/lldb-x86_64-debian/build/./lib --env LLVM_INCLUDE_DIR=/home/worker/2.0.1/lldb-x86_64-debian/build/include --env LLVM_TOOLS_DIR=/home/worker/2.0.1/lldb-x86_64-debian/build/./bin --arch x86_64 --build-dir /home/worker/2.0.1/lldb-x86_64-debian/build/lldb-test-build.noindex --lldb-module-cache-dir /home/worker/2.0.1/lldb-x86_64-debian/build/lldb-test-build.noindex/module-cache-lldb/lldb-api --clang-module-cache-dir /home/worker/2.0.1/lldb-x86_64-debian/build/lldb-test-build.noindex/module-cache-clang/lldb-api --executable /home/worker/2.0.1/lldb-x86_64-debian/build/./bin/lldb --compiler /home/worker/2.0.1/lldb-x86_64-debian/build/./bin/clang --dsymutil /home/worker/2.0.1/lldb-x86_64-debian/build/./bin/dsymutil --make /usr/bin/gmake --llvm-tools-dir /home/worker/2.0.1/lldb-x86_64-debian/build/./bin --lldb-obj-root /home/worker/2.0.1/lldb-x86_64-debian/build/tools/lldb --lldb-libs-dir /home/worker/2.0.1/lldb-x86_64-debian/build/./lib --cmake-build-type Release -t /home/worker/2.0.1/lldb-x86_64-debian/llvm-project/lldb/test/API/tools/lldb-dap/launch -p TestDAP_launch.py
--
Exit Code: 1

Command Output (stdout):
--
lldb version 21.0.0git (https://github.com/llvm/llvm-project.git revision d56deea1e4fb16bdf2789fe1b9a18084f42fd933)
  clang revision d56deea1e4fb16bdf2789fe1b9a18084f42fd933
  llvm revision d56deea1e4fb16bdf2789fe1b9a18084f42fd933
Skipping the following test categories: ['libc++', 'dsym', 'gmodules', 'debugserver', 'objc']

--
Command Output (stderr):
--
Change dir to: /home/worker/2.0.1/lldb-x86_64-debian/llvm-project/lldb/test/API/tools/lldb-dap/launch
runCmd: settings clear --all

output: 

runCmd: settings set symbols.enable-external-lookup false

output: 

runCmd: settings set target.inherit-tcc true

output: 

runCmd: settings set target.disable-aslr false

output: 

runCmd: settings set target.detach-on-error false

output: 

runCmd: settings set target.auto-apply-fixits false

…onstruction. (#140132) Update initial construction to connect the Plan's entry to the scalar preheader during initial construction. This moves a small part of the skeleton creation out of ILV and will also enable replacing VPInstruction::ResumePhi with regular VPPhi recipes. Resume phis need 2 incoming values to start with, the second being the bypass value from the scalar ph (and used to replicate the incoming value for other bypass blocks). Adding the extra edge ensures we incoming values for resume phis match the incoming blocks. PR: llvm/llvm-project#140132

Use regular VPPhi instead of a separate opcode for resume phis. This removes an unneeded specialized opcode and unifies the code (verification, printing, updating when CFG is changed). Depends on llvm#140132.

) Use regular VPPhi instead of a separate opcode for resume phis. This removes an unneeded specialized opcode and unifies the code (verification, printing, updating when CFG is changed). Depends on #140132. PR: #140405

…NFC). (#140405) Use regular VPPhi instead of a separate opcode for resume phis. This removes an unneeded specialized opcode and unifies the code (verification, printing, updating when CFG is changed). Depends on llvm/llvm-project#140132. PR: llvm/llvm-project#140405

llvm#140132) Update initial construction to connect the Plan's entry to the scalar preheader during initial construction. This moves a small part of the skeleton creation out of ILV and will also enable replacing VPInstruction::ResumePhi with regular VPPhi recipes. Resume phis need 2 incoming values to start with, the second being the bypass value from the scalar ph (and used to replicate the incoming value for other bypass blocks). Adding the extra edge ensures we incoming values for resume phis match the incoming blocks. PR: llvm#140132

…#140405) Use regular VPPhi instead of a separate opcode for resume phis. This removes an unneeded specialized opcode and unifies the code (verification, printing, updating when CFG is changed). Depends on llvm#140132. PR: llvm#140405

fhahn requested review from rengolin, ayalz and aniragil May 15, 2025 20:14

fhahn marked this pull request as ready for review May 15, 2025 20:29

llvmbot added vectorizers llvm:transforms labels May 15, 2025

fhahn force-pushed the vplan-connect-scalar-ph-early branch from 49578c3 to 4acf4d4 Compare May 17, 2025 16:56

fhahn force-pushed the vplan-connect-scalar-ph-early branch from 4acf4d4 to d481278 Compare May 17, 2025 21:12

llvmbot added backend:RISC-V backend:PowerPC labels May 17, 2025

fhahn mentioned this pull request May 17, 2025

[VPlan] Remove ResumePhi opcode, use regular PHI instead (NFC). #140405

Merged

ayalz reviewed May 20, 2025

View reviewed changes

fhahn added 2 commits May 23, 2025 15:02

Merge remote-tracking branch 'origin/main' into vplan-connect-scalar-…

76212e6

…ph-early

!fixup address latest comments, thanks!

a324d27

ayalz approved these changes May 27, 2025

View reviewed changes

fhahn added 2 commits May 27, 2025 13:17

Merge remote-tracking branch 'origin/main' into vplan-connect-scalar-…

2ad3e76

…ph-early

!fixup address latest comments, thanks

f4a28d4

fhahn merged commit d56deea into llvm:main May 27, 2025
9 of 10 checks passed

fhahn deleted the vplan-connect-scalar-ph-early branch May 27, 2025 15:08

		return getScalarPreheader()->getNumPredecessors() != 0 &&
		getScalarPreheader()->getSinglePredecessor() != getEntry();


		// TODO: Wrap LoopVectorPreHeader in VPIRBasicBlock here.
		introduceCheckBlockInVPlan(TCCheckBlock);

		// Also connect the entry block to the scalar preheader.
		VPBlockUtils::connectBlocks(Plan.getEntry(), ScalarPH);

		@@ -2368,14 +2368,11 @@ InnerLoopVectorizer::getOrCreateVectorTripCount(BasicBlock *InsertBlock) {
		void InnerLoopVectorizer::introduceCheckBlockInVPlan(BasicBlock *CheckIRBB) {

		// Only generate add a new block for the trip-count check for the main loop.
		// The epilogue will use the already existing VPlan entry block.

		@@ -540,6 +540,9 @@ void VPlanTransforms::prepareForVectorization(
		if (auto *LatchExitVPB = MiddleVPBB->getSingleSuccessor())

[VPlan] Connect Entry to scalar preheader during initial construction. #140132

[VPlan] Connect Entry to scalar preheader during initial construction. #140132

Uh oh!

Conversation

fhahn commented May 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented May 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ayalz left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

llvm-ci commented May 27, 2025

Uh oh!

Uh oh!

fhahn commented May 15, 2025 •

edited

Loading

llvmbot commented May 15, 2025 •

edited

Loading