diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 17b8db6b4145..aa4f0bdffe10 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -1829,15 +1829,11 @@ Instruction *InstCombinerImpl::visitSub(BinaryOperator &I) {
     return BinaryOperator::CreateNot(Op1);
 
   // (~X) - (~Y) --> Y - X
-  if (isFreeToInvert(Op0, Op0->hasOneUse()) &&
-      isFreeToInvert(Op1, Op1->hasOneUse())) {
-    Value *NotOp0 = Builder.CreateNot(Op0);
-    Value *NotOp1 = Builder.CreateNot(Op1);
-    return BinaryOperator::CreateSub(NotOp1, NotOp0);
-  }
+  Value *X, *Y;
+  if (match(Op0, m_Not(m_Value(X))) && match(Op1, m_Not(m_Value(Y))))
+    return BinaryOperator::CreateSub(Y, X);
 
   // (X + -1) - Y --> ~Y + X
-  Value *X, *Y;
   if (match(Op0, m_OneUse(m_Add(m_Value(X), m_AllOnes()))))
     return BinaryOperator::CreateAdd(Builder.CreateNot(Op1), X);
 
diff --git a/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll b/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
index ce7cc81b60e6..2e4224b11ccf 100644
--- a/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/minmax-intrinsics.ll
@@ -1157,8 +1157,8 @@ define i8 @freeToInvertSub(i8 %x, i8 %y, i8 %z) {
 ; CHECK-NEXT:    call void @use(i8 [[NX]])
 ; CHECK-NEXT:    call void @use(i8 [[NY]])
 ; CHECK-NEXT:    call void @use(i8 [[NZ]])
-; CHECK-NEXT:    [[TMP1:%.*]] = call i8 @llvm.umin.i8(i8 [[X]], i8 [[Y]])
-; CHECK-NEXT:    [[SUB:%.*]] = sub i8 [[TMP1]], [[Z]]
+; CHECK-NEXT:    [[M:%.*]] = call i8 @llvm.umax.i8(i8 [[NX]], i8 [[NY]])
+; CHECK-NEXT:    [[SUB:%.*]] = sub i8 [[NZ]], [[M]]
 ; CHECK-NEXT:    ret i8 [[SUB]]
 ;
   %nx = xor i8 %x, -1
diff --git a/llvm/test/Transforms/InstCombine/reassociate-nuw.ll b/llvm/test/Transforms/InstCombine/reassociate-nuw.ll
index 6045f1ec8806..a40d427fd29f 100644
--- a/llvm/test/Transforms/InstCombine/reassociate-nuw.ll
+++ b/llvm/test/Transforms/InstCombine/reassociate-nuw.ll
@@ -79,8 +79,9 @@ define i32 @reassoc_x2_mul_nuw(i32 %x, i32 %y) {
 
 define i32 @reassoc_x2_sub_nuw(i32 %x, i32 %y) {
 ; CHECK-LABEL: @reassoc_x2_sub_nuw(
-; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 [[X:%.*]], [[Y:%.*]]
-; CHECK-NEXT:    [[SUB2:%.*]] = add i32 [[TMP1]], 4
+; CHECK-NEXT:    [[SUB0:%.*]] = add i32 [[X:%.*]], -4
+; CHECK-NEXT:    [[SUB1:%.*]] = add i32 [[Y:%.*]], -8
+; CHECK-NEXT:    [[SUB2:%.*]] = sub nuw i32 [[SUB0]], [[SUB1]]
 ; CHECK-NEXT:    ret i32 [[SUB2]]
 ;
   %sub0 = sub nuw i32 %x, 4
diff --git a/llvm/test/Transforms/InstCombine/sub.ll b/llvm/test/Transforms/InstCombine/sub.ll
index 4cbad77b8c33..e228ac84c47c 100644
--- a/llvm/test/Transforms/InstCombine/sub.ll
+++ b/llvm/test/Transforms/InstCombine/sub.ll
@@ -1109,9 +1109,14 @@ define i32 @test57(i32 %A, i32 %B) {
 @dummy_global2 = external global i8*
 
 define i64 @test58([100 x [100 x i8]]* %foo, i64 %i, i64 %j) {
+; Note the reassociate pass and another instcombine pass will further optimize this to
+; "%sub = i64 %i, %j, ret i64 %sub"
+; gep1 and gep2 have only one use
 ; CHECK-LABEL: @test58(
-; CHECK-NEXT:    [[TMP1:%.*]] = sub i64 [[I:%.*]], [[J:%.*]]
-; CHECK-NEXT:    ret i64 [[TMP1]]
+; CHECK-NEXT:    [[GEP1_OFFS:%.*]] = add nsw i64 [[I:%.*]], 4200
+; CHECK-NEXT:    [[GEP2_OFFS:%.*]] = add nsw i64 [[J:%.*]], 4200
+; CHECK-NEXT:    [[GEPDIFF:%.*]] = sub nsw i64 [[GEP1_OFFS]], [[GEP2_OFFS]]
+; CHECK-NEXT:    ret i64 [[GEPDIFF]]
 ;
   %gep1 = getelementptr inbounds [100 x [100 x i8]], [100 x [100 x i8]]* %foo, i64 0, i64 42, i64 %i
   %gep2 = getelementptr inbounds [100 x [100 x i8]], [100 x [100 x i8]]* %foo, i64 0, i64 42, i64 %j