[InstCombine] Simplify (X / C0) * C1 + (X % C0) * C2 to (X / C0) * (C1 - C2 * C0) + X * C2

Author: dtcxzyw

llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp

@@ -1132,6 +1132,8 @@ static bool MulWillOverflow(APInt &C0, APInt &C1, bool IsSigned) {
 
 // Simplifies X % C0 + (( X / C0 ) % C1) * C0 to X % (C0 * C1), where (C0 * C1)
 // does not overflow.
+// Simplifies (X / C0) * C1 + (X % C0) * C2 to
+// (X / C0) * (C1 - C2 * C0) + X * C2
 Value *InstCombinerImpl::SimplifyAddWithRemainder(BinaryOperator &I) {
   Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
   Value *X, *MulOpV;
@@ -1159,6 +1161,33 @@ Value *InstCombinerImpl::SimplifyAddWithRemainder(BinaryOperator &I) {
     }
   }
 
+  // Match I = (X / C0) * C1 + (X % C0) * C2
+  Value *Div, *Rem;
+  APInt C1, C2;
+  if (!LHS->hasOneUse() || !MatchMul(LHS, Div, C1))
+    Div = LHS, C1 = APInt(I.getType()->getScalarSizeInBits(), 1);
+  if (!RHS->hasOneUse() || !MatchMul(RHS, Rem, C2))
+    Rem = RHS, C2 = APInt(I.getType()->getScalarSizeInBits(), 1);
+  if (match(Div, m_IRem(m_Value(), m_Value()))) {
+    std::swap(Div, Rem);
+    std::swap(C1, C2);
+  }
+  Value *DivOpV;
+  APInt DivOpC;
+  if (MatchRem(Rem, X, C0, IsSigned) &&
+      MatchDiv(Div, DivOpV, DivOpC, IsSigned) && X == DivOpV && C0 == DivOpC) {
+    if (!isGuaranteedNotToBeUndef(X, &AC, &I, &DT))
+      return nullptr;
+    APInt NewC = C1 - C2 * C0;
+    if (!NewC.isZero() && !Rem->hasOneUse())
+      return nullptr;
+    Value *MulXC2 = Builder.CreateMul(X, ConstantInt::get(X->getType(), C2));
+    if (NewC.isZero())
+      return MulXC2;
+    return Builder.CreateAdd(
+        Builder.CreateMul(Div, ConstantInt::get(X->getType(), NewC)), MulXC2);
+  }
+
   return nullptr;
 }
 

llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp

@@ -3910,6 +3910,10 @@ Instruction *InstCombinerImpl::visitOr(BinaryOperator &I) {
 
   if (Instruction *Res = foldBitwiseLogicWithIntrinsics(I, Builder))
     return Res;
+  
+  if (cast<PossiblyDisjointInst>(I).isDisjoint())
+    if (Value *V = SimplifyAddWithRemainder(I))
+      return replaceInstUsesWith(I, V);
 
   return nullptr;
 }

llvm/test/Transforms/InstCombine/add4.ll

@@ -135,9 +135,8 @@ define i32 @fold_add_udiv_urem(i32 noundef %val) {
 ; CHECK-LABEL: @fold_add_udiv_urem(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[VAL:%.*]], 10
-; CHECK-NEXT:    [[SHL:%.*]] = shl i32 [[DIV]], 4
-; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[VAL]], 10
-; CHECK-NEXT:    [[ADD:%.*]] = or disjoint i32 [[SHL]], [[REM]]
+; CHECK-NEXT:    [[TMP0:%.*]] = mul nuw i32 [[DIV]], 6
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP0]], [[VAL]]
 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
 entry:
@@ -151,9 +150,8 @@ define i32 @fold_add_sdiv_srem(i32 noundef %val) {
 ; CHECK-LABEL: @fold_add_sdiv_srem(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[DIV:%.*]] = sdiv i32 [[VAL:%.*]], 10
-; CHECK-NEXT:    [[SHL:%.*]] = shl i32 [[DIV]], 4
-; CHECK-NEXT:    [[REM:%.*]] = srem i32 [[VAL]], 10
-; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[SHL]], [[REM]]
+; CHECK-NEXT:    [[TMP0:%.*]] = mul nsw i32 [[DIV]], 6
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP0]], [[VAL]]
 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
 entry:
@@ -166,11 +164,7 @@ entry:
 define i32 @fold_add_udiv_urem_to_mul(i32 noundef %val) {
 ; CHECK-LABEL: @fold_add_udiv_urem_to_mul(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[VAL:%.*]], 7
-; CHECK-NEXT:    [[MUL1:%.*]] = mul i32 [[DIV]], 21
-; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[VAL]], 7
-; CHECK-NEXT:    [[MUL2:%.*]] = mul nuw nsw i32 [[REM]], 3
-; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[MUL1]], [[MUL2]]
+; CHECK-NEXT:    [[ADD:%.*]] = mul i32 [[VAL:%.*]], 3
 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
 entry:
@@ -184,12 +178,9 @@ entry:
 define i32 @fold_add_udiv_urem_to_mul_multiuse(i32 noundef %val) {
 ; CHECK-LABEL: @fold_add_udiv_urem_to_mul_multiuse(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[VAL:%.*]], 7
-; CHECK-NEXT:    [[MUL1:%.*]] = mul i32 [[DIV]], 21
-; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[VAL]], 7
+; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[VAL:%.*]], 7
 ; CHECK-NEXT:    call void @use(i32 [[REM]])
-; CHECK-NEXT:    [[MUL2:%.*]] = mul nuw nsw i32 [[REM]], 3
-; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[MUL1]], [[MUL2]]
+; CHECK-NEXT:    [[ADD:%.*]] = mul i32 [[VAL]], 3
 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
 entry:
@@ -205,9 +196,8 @@ define i32 @fold_add_udiv_urem_commuted(i32 noundef %val) {
 ; CHECK-LABEL: @fold_add_udiv_urem_commuted(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[VAL:%.*]], 10
-; CHECK-NEXT:    [[SHL:%.*]] = shl i32 [[DIV]], 4
-; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[VAL]], 10
-; CHECK-NEXT:    [[ADD:%.*]] = or disjoint i32 [[REM]], [[SHL]]
+; CHECK-NEXT:    [[TMP0:%.*]] = mul nuw i32 [[DIV]], 6
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP0]], [[VAL]]
 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
 entry:
@@ -221,9 +211,8 @@ define i32 @fold_add_udiv_urem_or_disjoint(i32 noundef %val) {
 ; CHECK-LABEL: @fold_add_udiv_urem_or_disjoint(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[VAL:%.*]], 10
-; CHECK-NEXT:    [[SHL:%.*]] = shl i32 [[DIV]], 4
-; CHECK-NEXT:    [[REM:%.*]] = urem i32 [[VAL]], 10
-; CHECK-NEXT:    [[ADD:%.*]] = or disjoint i32 [[SHL]], [[REM]]
+; CHECK-NEXT:    [[TMP0:%.*]] = mul nuw i32 [[DIV]], 6
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP0]], [[VAL]]
 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
 entry: