Remove add/sub handling

Author: goldsteinn

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -1975,10 +1975,10 @@ static bool foldFCmpToFPClassTest(CmpInst *Cmp, const TargetLowering &TLI,
   return true;
 }
 
-static bool isRemOfLoopIncrementWithLoopInvariant(
-    Instruction *Rem, const LoopInfo *LI, Value *&RemAmtOut,
-    std::optional<bool> &AddOrSubOut, Value *&AddOrSubInstOut,
-    Value *&AddOrSubOffsetOut, PHINode *&LoopIncrPNOut) {
+static bool isRemOfLoopIncrementWithLoopInvariant(Instruction *Rem,
+                                                  const LoopInfo *LI,
+                                                  Value *&RemAmtOut,
+                                                  PHINode *&LoopIncrPNOut) {
   Value *Incr, *RemAmt;
   // NB: If RemAmt is a power of 2 it *should* have been transformed by now.
   if (!match(Rem, m_URem(m_Value(Incr), m_Value(RemAmt))))
@@ -1989,33 +1989,8 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
   if (!L || !L->getLoopPreheader() || !L->getLoopLatch())
     return false;
 
-  std::optional<bool> AddOrSub;
-  Value *AddOrSubOffset;
   // Find out loop increment PHI.
   auto *PN = dyn_cast<PHINode>(Incr);
-  if (PN) {
-    AddOrSub = std::nullopt;
-    AddOrSubOffset = nullptr;
-  } else {
-    // Search through a NUW add/sub on top of the loop increment.
-    Value *V0, *V1;
-    if (match(Incr, m_NUWAddLike(m_Value(V0), m_Value(V1))))
-      AddOrSub = true;
-    else if (match(Incr, m_NUWSub(m_Value(V0), m_Value(V1))))
-      AddOrSub = false;
-    else
-      return false;
-
-    AddOrSubInstOut = Incr;
-
-    PN = dyn_cast<PHINode>(V0);
-    if (PN != nullptr) {
-      AddOrSubOffset = V1;
-    } else if (*AddOrSub) {
-      PN = dyn_cast<PHINode>(V1);
-      AddOrSubOffset = V0;
-    }
-  }
 
   if (!PN)
     return false;
@@ -2052,8 +2027,6 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
   // Set output variables.
   RemAmtOut = RemAmt;
   LoopIncrPNOut = PN;
-  AddOrSubOut = AddOrSub;
-  AddOrSubOffsetOut = AddOrSubOffset;
 
   return true;
 }
@@ -2069,16 +2042,14 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
 // for(i = Start; i < End; ++i, ++rem)
 //    Rem = rem == RemAmtLoopInvariant ? 0 : Rem;
 //
-// Currently only implemented for `Start` and `IncrLoopInvariant` being zero.
+// Currently only implemented for `IncrLoopInvariant` being zero.
 static bool foldURemOfLoopIncrement(Instruction *Rem, const DataLayout *DL,
                                     const LoopInfo *LI,
                                     SmallSet<BasicBlock *, 32> &FreshBBs,
                                     bool IsHuge) {
-  std::optional<bool> AddOrSub;
-  Value *AddOrSubOffset, *RemAmt, *AddOrSubInst;
+  Value *RemAmt;
   PHINode *LoopIncrPN;
-  if (!isRemOfLoopIncrementWithLoopInvariant(
-          Rem, LI, RemAmt, AddOrSub, AddOrSubInst, AddOrSubOffset, LoopIncrPN))
+  if (!isRemOfLoopIncrementWithLoopInvariant(Rem, LI, RemAmt, LoopIncrPN))
     return false;
 
   // Only non-constant remainder as the extra IV is probably not profitable
@@ -2102,26 +2073,6 @@ static bool foldURemOfLoopIncrement(Instruction *Rem, const DataLayout *DL,
   // Only proceed if the expression simplifies (otherwise we can't fully
   // optimize out the urem).
   Value *Start = LoopIncrPN->getIncomingValueForBlock(L->getLoopPreheader());
-  if (AddOrSub) {
-    assert(AddOrSubOffset && AddOrSubInst &&
-           "We found an add/sub but missing values");
-    // Without dom-condition/assumption cache we aren't likely to get much out
-    // of a context instruction.
-    const SimplifyQuery Q(*DL);
-    bool NSW = cast<OverflowingBinaryOperator>(AddOrSubInst)->hasNoSignedWrap();
-    if (*AddOrSub)
-      Start = simplifyAddInst(Start, AddOrSubOffset, /*IsNSW=*/NSW,
-                              /*IsNUW=*/true, Q);
-    else
-      Start = simplifySubInst(Start, AddOrSubOffset, /*IsNSW=*/NSW,
-                              /*IsNUW=*/true, Q);
-    if (!Start)
-      return false;
-
-    Start = simplifyURemInst(Start, RemAmt, Q);
-    if (!Start)
-      return false;
-  }
 
   // Create new remainder with induction variable.
   Type *Ty = Rem->getType();
@@ -2148,8 +2099,6 @@ static bool foldURemOfLoopIncrement(Instruction *Rem, const DataLayout *DL,
 
   replaceAllUsesWith(Rem, NewRem, FreshBBs, IsHuge);
   Rem->eraseFromParent();
-  if (AddOrSubInst && AddOrSubInst->use_empty())
-    cast<Instruction>(AddOrSubInst)->eraseFromParent();
   return true;
 }
 

llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll

@@ -706,12 +706,10 @@ define void @simple_urem_to_sel_non_zero_start_through_add(i32 %N, i32 %rem_amt_
 ; CHECK:       [[FOR_COND_CLEANUP]]:
 ; CHECK-NEXT:    ret void
 ; CHECK:       [[FOR_BODY]]:
-; CHECK-NEXT:    [[REM:%.*]] = phi i32 [ 7, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
 ; CHECK-NEXT:    [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 2, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[I_WITH_OFF:%.*]] = add nuw i32 [[I_04]], 5
+; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
 ; CHECK-NEXT:    tail call void @use.i32(i32 [[REM]])
-; CHECK-NEXT:    [[TMP1:%.*]] = add nuw i32 [[REM]], 1
-; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
-; CHECK-NEXT:    [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
 ; CHECK-NEXT:    [[INC]] = add nuw i32 [[I_04]], 1
 ; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
 ; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -819,12 +817,10 @@ define void @simple_urem_to_sel_non_zero_start_through_sub(i32 %N, i32 %rem_amt,
 ; CHECK:       [[FOR_COND_CLEANUP]]:
 ; CHECK-NEXT:    ret void
 ; CHECK:       [[FOR_BODY]]:
-; CHECK-NEXT:    [[REM:%.*]] = phi i32 [ 0, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
 ; CHECK-NEXT:    [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[START]], %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[I_WITH_OFF:%.*]] = sub nuw i32 [[I_04]], [[START]]
+; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
 ; CHECK-NEXT:    tail call void @use.i32(i32 [[REM]])
-; CHECK-NEXT:    [[TMP1:%.*]] = add nuw i32 [[REM]], 1
-; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
-; CHECK-NEXT:    [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
 ; CHECK-NEXT:    [[INC]] = add nuw i32 [[I_04]], 1
 ; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
 ; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]