[llvm] [SLPVectorizer] Use TargetFolder (PR #85800)

[Nikita Popov via llvm-commits]

https://github.com/nikic created https://github.com/llvm/llvm-project/pull/85800

Use IRBuilder with TargetFolder in SLPVectorizer to avoid the custom constant folding code.

This fixes the remaining part of https://github.com/llvm/llvm-project/issues/61240.

>From 71f72cf92d8d775665598d31bf41aabea4067a5e Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov at redhat.com>
Date: Tue, 19 Mar 2024 14:36:14 +0100
Subject: [PATCH] use target folder

---
 .../Transforms/Vectorize/SLPVectorizer.cpp    | 40 ++++++-------------
 1 file changed, 13 insertions(+), 27 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index f2ac060a2e3a9c..456aef2f13e601 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -1001,8 +1001,9 @@ class BoUpSLP {
           TargetLibraryInfo *TLi, AAResults *Aa, LoopInfo *Li,
           DominatorTree *Dt, AssumptionCache *AC, DemandedBits *DB,
           const DataLayout *DL, OptimizationRemarkEmitter *ORE)
-      : BatchAA(*Aa), F(Func), SE(Se), TTI(Tti), TLI(TLi), LI(Li),
-        DT(Dt), AC(AC), DB(DB), DL(DL), ORE(ORE), Builder(Se->getContext()) {
+      : BatchAA(*Aa), F(Func), SE(Se), TTI(Tti), TLI(TLi), LI(Li), DT(Dt),
+        AC(AC), DB(DB), DL(DL), ORE(ORE),
+        Builder(Se->getContext(), TargetFolder(*DL)) {
     CodeMetrics::collectEphemeralValues(F, AC, EphValues);
     // Use the vector register size specified by the target unless overridden
     // by a command-line option.
@@ -3621,7 +3622,7 @@ class BoUpSLP {
   unsigned MinVecRegSize; // Set by cl::opt (default: 128).
 
   /// Instruction builder to construct the vectorized tree.
-  IRBuilder<> Builder;
+  IRBuilder<TargetFolder> Builder;
 
   /// A map of scalar integer values to the smallest bit width with which they
   /// can legally be represented. The values map to (width, signed) pairs,
@@ -14692,7 +14693,6 @@ class HorizontalReduction {
   static Value *createOp(IRBuilderBase &Builder, RecurKind Kind, Value *LHS,
                          Value *RHS, const Twine &Name, bool UseSelect) {
     unsigned RdxOpcode = RecurrenceDescriptor::getOpcode(Kind);
-    bool IsConstant = isConstant(LHS) && isConstant(RHS);
     switch (Kind) {
     case RecurKind::Or:
       if (UseSelect &&
@@ -14714,49 +14714,33 @@ class HorizontalReduction {
       return Builder.CreateBinOp((Instruction::BinaryOps)RdxOpcode, LHS, RHS,
                                  Name);
     case RecurKind::FMax:
-      if (IsConstant)
-        return ConstantFP::get(LHS->getType(),
-                               maxnum(cast<ConstantFP>(LHS)->getValueAPF(),
-                                      cast<ConstantFP>(RHS)->getValueAPF()));
       return Builder.CreateBinaryIntrinsic(Intrinsic::maxnum, LHS, RHS);
     case RecurKind::FMin:
-      if (IsConstant)
-        return ConstantFP::get(LHS->getType(),
-                               minnum(cast<ConstantFP>(LHS)->getValueAPF(),
-                                      cast<ConstantFP>(RHS)->getValueAPF()));
       return Builder.CreateBinaryIntrinsic(Intrinsic::minnum, LHS, RHS);
     case RecurKind::FMaximum:
-      if (IsConstant)
-        return ConstantFP::get(LHS->getType(),
-                               maximum(cast<ConstantFP>(LHS)->getValueAPF(),
-                                       cast<ConstantFP>(RHS)->getValueAPF()));
       return Builder.CreateBinaryIntrinsic(Intrinsic::maximum, LHS, RHS);
     case RecurKind::FMinimum:
-      if (IsConstant)
-        return ConstantFP::get(LHS->getType(),
-                               minimum(cast<ConstantFP>(LHS)->getValueAPF(),
-                                       cast<ConstantFP>(RHS)->getValueAPF()));
       return Builder.CreateBinaryIntrinsic(Intrinsic::minimum, LHS, RHS);
     case RecurKind::SMax:
-      if (IsConstant || UseSelect) {
+      if (UseSelect) {
         Value *Cmp = Builder.CreateICmpSGT(LHS, RHS, Name);
         return Builder.CreateSelect(Cmp, LHS, RHS, Name);
       }
       return Builder.CreateBinaryIntrinsic(Intrinsic::smax, LHS, RHS);
     case RecurKind::SMin:
-      if (IsConstant || UseSelect) {
+      if (UseSelect) {
         Value *Cmp = Builder.CreateICmpSLT(LHS, RHS, Name);
         return Builder.CreateSelect(Cmp, LHS, RHS, Name);
       }
       return Builder.CreateBinaryIntrinsic(Intrinsic::smin, LHS, RHS);
     case RecurKind::UMax:
-      if (IsConstant || UseSelect) {
+      if (UseSelect) {
         Value *Cmp = Builder.CreateICmpUGT(LHS, RHS, Name);
         return Builder.CreateSelect(Cmp, LHS, RHS, Name);
       }
       return Builder.CreateBinaryIntrinsic(Intrinsic::umax, LHS, RHS);
     case RecurKind::UMin:
-      if (IsConstant || UseSelect) {
+      if (UseSelect) {
         Value *Cmp = Builder.CreateICmpULT(LHS, RHS, Name);
         return Builder.CreateSelect(Cmp, LHS, RHS, Name);
       }
@@ -15150,7 +15134,7 @@ class HorizontalReduction {
   }
 
   /// Attempt to vectorize the tree found by matchAssociativeReduction.
-  Value *tryToReduce(BoUpSLP &V, TargetTransformInfo *TTI,
+  Value *tryToReduce(BoUpSLP &V, const DataLayout &DL, TargetTransformInfo *TTI,
                      const TargetLibraryInfo &TLI) {
     constexpr int ReductionLimit = 4;
     constexpr unsigned RegMaxNumber = 4;
@@ -15176,7 +15160,9 @@ class HorizontalReduction {
       return nullptr;
     }
 
-    IRBuilder<> Builder(cast<Instruction>(ReductionRoot));
+    IRBuilder<TargetFolder> Builder(ReductionRoot->getContext(),
+                                    TargetFolder(DL));
+    Builder.SetInsertPoint(cast<Instruction>(ReductionRoot));
 
     // Track the reduced values in case if they are replaced by extractelement
     // because of the vectorization.
@@ -16272,7 +16258,7 @@ bool SLPVectorizerPass::vectorizeHorReduction(
     HorizontalReduction HorRdx;
     if (!HorRdx.matchAssociativeReduction(R, Inst, *SE, *DL, *TLI))
       return nullptr;
-    return HorRdx.tryToReduce(R, TTI, *TLI);
+    return HorRdx.tryToReduce(R, *DL, TTI, *TLI);
   };
   auto TryAppendToPostponedInsts = [&](Instruction *FutureSeed) {
     if (TryOperandsAsNewSeeds && FutureSeed == Root) {