[InstCombine] Add vector support to icmp(add(X,C1),add(Y,C2)) -> icmp(add(X,C1-C2...

As discussed on Issue #32161 this fold can be generalized a lot more than it currently is, but this patch at least adds vector support.

Differential Revision: https://reviews.llvm.org/D121358

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 7d4e9b5..e405dc7 100644 (file)
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -4110,29 +4110,31 @@ Instruction *InstCombinerImpl::foldICmpBinOp(ICmpInst &I,
   //  icmp (A + C1), (C + C2) -> icmp A, (C + C3)
   //  s.t. C3 = C2 - C1
   if (A && C && NoOp0WrapProblem && NoOp1WrapProblem &&
-      (BO0->hasOneUse() || BO1->hasOneUse()) && !I.isUnsigned())
-    if (ConstantInt *C1 = dyn_cast<ConstantInt>(B))
-      if (ConstantInt *C2 = dyn_cast<ConstantInt>(D)) {
-        const APInt &AP1 = C1->getValue();
-        const APInt &AP2 = C2->getValue();
-        if (AP1.isNegative() == AP2.isNegative()) {
-          APInt AP1Abs = C1->getValue().abs();
-          APInt AP2Abs = C2->getValue().abs();
-          if (AP1Abs.uge(AP2Abs)) {
-            ConstantInt *C3 = Builder.getInt(AP1 - AP2);
-            bool HasNUW = BO0->hasNoUnsignedWrap() && C3->getValue().ule(AP1);
-            bool HasNSW = BO0->hasNoSignedWrap();
-            Value *NewAdd = Builder.CreateAdd(A, C3, "", HasNUW, HasNSW);
-            return new ICmpInst(Pred, NewAdd, C);
-          } else {
-            ConstantInt *C3 = Builder.getInt(AP2 - AP1);
-            bool HasNUW = BO1->hasNoUnsignedWrap() && C3->getValue().ule(AP2);
-            bool HasNSW = BO1->hasNoSignedWrap();
-            Value *NewAdd = Builder.CreateAdd(C, C3, "", HasNUW, HasNSW);
-            return new ICmpInst(Pred, A, NewAdd);
-          }
-        }
+      (BO0->hasOneUse() || BO1->hasOneUse()) && !I.isUnsigned()) {
+    const APInt *AP1, *AP2;
+    // TODO: Support non-uniform vectors.
+    // TODO: Allow undef passthrough if B AND D's element is undef.
+    if (match(B, m_APIntAllowUndef(AP1)) && match(D, m_APIntAllowUndef(AP2)) &&
+        AP1->isNegative() == AP2->isNegative()) {
+      APInt AP1Abs = AP1->abs();
+      APInt AP2Abs = AP2->abs();
+      if (AP1Abs.uge(AP2Abs)) {
+        APInt Diff = *AP1 - *AP2;
+        bool HasNUW = BO0->hasNoUnsignedWrap() && Diff.ule(*AP1);
+        bool HasNSW = BO0->hasNoSignedWrap();
+        Constant *C3 = Constant::getIntegerValue(BO0->getType(), Diff);
+        Value *NewAdd = Builder.CreateAdd(A, C3, "", HasNUW, HasNSW);
+        return new ICmpInst(Pred, NewAdd, C);
+      } else {
+        APInt Diff = *AP2 - *AP1;
+        bool HasNUW = BO1->hasNoUnsignedWrap() && Diff.ule(*AP2);
+        bool HasNSW = BO1->hasNoSignedWrap();
+        Constant *C3 = Constant::getIntegerValue(BO0->getType(), Diff);
+        Value *NewAdd = Builder.CreateAdd(C, C3, "", HasNUW, HasNSW);
+        return new ICmpInst(Pred, A, NewAdd);
       }
+    }
+  }
 
   // Analyze the case when either Op0 or Op1 is a sub instruction.
   // Op0 = A - B (or A and B are null); Op1 = C - D (or C and D are null).

diff --git a/llvm/test/Transforms/InstCombine/icmp.ll b/llvm/test/Transforms/InstCombine/icmp.ll
index 8da2ed3..3c8f86c 100644 (file)
--- a/llvm/test/Transforms/InstCombine/icmp.ll
+++ b/llvm/test/Transforms/InstCombine/icmp.ll
@@ -1581,9 +1581,8 @@ define i1 @icmp_add20_sge_add57(i32 %x, i32 %y) {
 
 define <2 x i1> @icmp_add20_sge_add57_splat(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: @icmp_add20_sge_add57_splat(
-; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[X:%.*]], <i32 20, i32 20>
-; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <2 x i32> [[Y:%.*]], <i32 57, i32 57>
-; CHECK-NEXT:    [[CMP:%.*]] = icmp sge <2 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[Y:%.*]], <i32 37, i32 37>
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sle <2 x i32> [[TMP1]], [[X:%.*]]
 ; CHECK-NEXT:    ret <2 x i1> [[CMP]]
 ;
   %1 = add nsw <2 x i32> %x, <i32 20, i32 20>
@@ -1594,9 +1593,8 @@ define <2 x i1> @icmp_add20_sge_add57_splat(<2 x i32> %x, <2 x i32> %y) {
 
 define <2 x i1> @icmp_add20_sge_add57_undef(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: @icmp_add20_sge_add57_undef(
-; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[X:%.*]], <i32 20, i32 20>
-; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <2 x i32> [[Y:%.*]], <i32 57, i32 undef>
-; CHECK-NEXT:    [[CMP:%.*]] = icmp sge <2 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[Y:%.*]], <i32 37, i32 37>
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sle <2 x i32> [[TMP1]], [[X:%.*]]
 ; CHECK-NEXT:    ret <2 x i1> [[CMP]]
 ;
   %1 = add nsw <2 x i32> %x, <i32 20, i32 20>
@@ -1632,9 +1630,8 @@ define i1 @icmp_sub57_sge_sub20(i32 %x, i32 %y) {
 
 define <2 x i1> @icmp_sub57_sge_sub20_splat(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: @icmp_sub57_sge_sub20_splat(
-; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[X:%.*]], <i32 -57, i32 -57>
-; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <2 x i32> [[Y:%.*]], <i32 -20, i32 -20>
-; CHECK-NEXT:    [[CMP:%.*]] = icmp sge <2 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[X:%.*]], <i32 -37, i32 -37>
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sge <2 x i32> [[TMP1]], [[Y:%.*]]
 ; CHECK-NEXT:    ret <2 x i1> [[CMP]]
 ;
   %1 = add nsw <2 x i32> %x, <i32 -57, i32 -57>
@@ -1645,9 +1642,8 @@ define <2 x i1> @icmp_sub57_sge_sub20_splat(<2 x i32> %x, <2 x i32> %y) {
 
 define <2 x i1> @icmp_sub57_sge_sub20_vec_undef(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: @icmp_sub57_sge_sub20_vec_undef(
-; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[X:%.*]], <i32 -57, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <2 x i32> [[Y:%.*]], <i32 -20, i32 undef>
-; CHECK-NEXT:    [[CMP:%.*]] = icmp sge <2 x i32> [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <2 x i32> [[X:%.*]], <i32 -37, i32 -37>
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sge <2 x i32> [[TMP1]], [[Y:%.*]]
 ; CHECK-NEXT:    ret <2 x i1> [[CMP]]
 ;
   %1 = add nsw <2 x i32> %x, <i32 -57, i32 undef>