[InstCombine] Fold ((X / C) cmp X) and ((X >> C) cmp X) into X ~cmp 0

Author: Poseydon42

llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp

@@ -7103,6 +7103,46 @@ Instruction *InstCombinerImpl::foldICmpCommutative(ICmpInst::Predicate Pred,
   if (Value *V = foldICmpWithLowBitMaskedVal(Pred, Op0, Op1, Q, *this))
     return replaceInstUsesWith(CxtI, V);
 
+  // Folding (X / Y) cmp X => X ~cmp 0 for some constant Y other than 0 or 1
+  {
+    Value *Dividend;
+    if (match(Op0,
+              m_UDiv(m_Value(Dividend),
+                     m_SpecificInt_ICMP(
+                         CmpInst::ICMP_UGT,
+                         APInt::getOneBitSet(
+                             Op0->getType()->getScalarSizeInBits(), 0)))) &&
+        Op1 == Dividend) {
+      return new ICmpInst(ICmpInst::getSwappedPredicate(Pred), Dividend,
+                          Constant::getNullValue(Dividend->getType()));
+    }
+
+    if (match(Op0,
+              m_SDiv(m_Value(Dividend),
+                     m_SpecificInt_ICMP(
+                         CmpInst::ICMP_UGT,
+                         APInt::getOneBitSet(
+                             Op0->getType()->getScalarSizeInBits(), 0)))) &&
+        Op1 == Dividend && !ICmpInst::isUnsigned(Pred)) {
+      return new ICmpInst(ICmpInst::getSwappedPredicate(Pred), Dividend,
+                          Constant::getNullValue(Dividend->getType()));
+    }
+  }
+
+  // Another case of this fold is (X >> Y) cmp X => X ~cmp 0 if Y != 0
+  {
+    Value *V;
+    if (match(Op0, m_LShr(m_Value(V),
+                          m_SpecificInt_ICMP(
+                              CmpInst::ICMP_NE,
+                              APInt::getZero(
+                                  Op0->getType()->getScalarSizeInBits())))) &&
+        Op1 == V) {
+      return new ICmpInst(ICmpInst::getInversePredicate(Pred), V,
+                          Constant::getNullValue(V->getType()));
+    }
+  }
+
   return nullptr;
 }
 

llvm/test/Transforms/InstCombine/icmp-div-constant.ll

@@ -381,8 +381,7 @@ define i1 @sdiv_eq_smin_use(i32 %x, i32 %y) {
 
 define i1 @sdiv_x_by_const_cmp_x(i32 %x) {
 ; CHECK-LABEL: @sdiv_x_by_const_cmp_x(
-; CHECK-NEXT:    [[V:%.*]] = udiv i32 [[X:%.*]], 13
-; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[V]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[X:%.*]], 0
 ; CHECK-NEXT:    ret i1 [[TMP1]]
 ;
   %v = udiv i32 %x, 13
@@ -392,8 +391,7 @@ define i1 @sdiv_x_by_const_cmp_x(i32 %x) {
 
 define i1 @udiv_x_by_const_cmp_x(i32 %x) {
 ; CHECK-LABEL: @udiv_x_by_const_cmp_x(
-; CHECK-NEXT:    [[TMP2:%.*]] = udiv i32 [[X:%.*]], 123
-; CHECK-NEXT:    [[TMP1:%.*]] = icmp slt i32 [[TMP2]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[X:%.*]], 0
 ; CHECK-NEXT:    ret i1 [[TMP1]]
 ;
   %1 = udiv i32 %x, 123
@@ -405,8 +403,7 @@ define i1 @udiv_x_by_const_cmp_x(i32 %x) {
 
 define i1 @lshr_x_by_const_cmp_x(i32 %x) {
 ; CHECK-LABEL: @lshr_x_by_const_cmp_x(
-; CHECK-NEXT:    [[V:%.*]] = lshr i32 [[X:%.*]], 1
-; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[V]], [[X]]
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne i32 [[X:%.*]], 0
 ; CHECK-NEXT:    ret i1 [[TMP1]]
 ;
   %v = lshr i32 %x, 1
@@ -416,8 +413,7 @@ define i1 @lshr_x_by_const_cmp_x(i32 %x) {
 
 define <4 x i1> @lshr_by_const_cmp_sle_value(<4 x i32> %x) {
 ; CHECK-LABEL: @lshr_by_const_cmp_sle_value(
-; CHECK-NEXT:    [[V:%.*]] = lshr <4 x i32> [[X:%.*]], <i32 2, i32 3, i32 2, i32 4>
-; CHECK-NEXT:    [[R:%.*]] = icmp sle <4 x i32> [[V]], [[X]]
+; CHECK-NEXT:    [[R:%.*]] = icmp sgt <4 x i32> [[X:%.*]], zeroinitializer
 ; CHECK-NEXT:    ret <4 x i1> [[R]]
 ;
   %v = lshr <4 x i32> %x, <i32 2, i32 3, i32 2, i32 4>
@@ -427,8 +423,7 @@ define <4 x i1> @lshr_by_const_cmp_sle_value(<4 x i32> %x) {
 
 define i1 @lshr_by_const_cmp_uge_value(i32 %x) {
 ; CHECK-LABEL: @lshr_by_const_cmp_uge_value(
-; CHECK-NEXT:    [[V:%.*]] = lshr i32 [[X:%.*]], 3
-; CHECK-NEXT:    [[R:%.*]] = icmp sle i32 [[V]], [[X]]
+; CHECK-NEXT:    [[R:%.*]] = icmp sgt i32 [[X:%.*]], 0
 ; CHECK-NEXT:    ret i1 [[R]]
 ;
   %v = lshr i32 %x, 3