[InstCombine] move add after smin/smax

Follow-up to rL355221.
This isn't specifically called for within PR14613,
but we'll get there eventually if it's not already
requested in some other bug report.

https://rise4fun.com/Alive/5b0

  Name: smax
  Pre: WillNotOverflowSignedSub(C1,C0)
  %a = add nsw i8 %x, C0
  %cond = icmp sgt i8 %a, C1
  %r = select i1 %cond, i8 %a, i8 C1
  =>
  %c2 = icmp sgt i8 %x, C1-C0
  %u2 = select i1 %c2, i8 %x, i8 C1-C0
  %r = add nsw i8 %u2, C0

  Name: smin
  Pre: WillNotOverflowSignedSub(C1,C0)
  %a = add nsw i32 %x, C0
  %cond = icmp slt i32 %a, C1
  %r = select i1 %cond, i32 %a, i32 C1
  =>
  %c2 = icmp slt i32 %x, C1-C0
  %u2 = select i1 %c2, i32 %x, i32 C1-C0
  %r = add nsw i32 %u2, C0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355272 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/InstCombine/InstCombineSelect.cpp b/lib/Transforms/InstCombine/InstCombineSelect.cpp
index b39a4648579a24f599257182c9f4f9bce31ff46d..582353eba1c762a4270b9f2c599df77e0f7a993f 100644 (file)
--- a/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -1566,7 +1566,7 @@ static Instruction *foldSelectCmpXchg(SelectInst &SI) {
 static Instruction *moveAddAfterMinMax(SelectPatternFlavor SPF, Value *X,
                                        Value *Y,
                                        InstCombiner::BuilderTy &Builder) {
-  assert (SelectPatternResult::isMinOrMax(SPF) && "Expected min/max pattern");
+  assert(SelectPatternResult::isMinOrMax(SPF) && "Expected min/max pattern");
   bool IsUnsigned = SPF == SelectPatternFlavor::SPF_UMIN ||
                     SPF == SelectPatternFlavor::SPF_UMAX;
   // TODO: If InstSimplify could fold all cases where C2 <= C1, we could change
@@ -1582,8 +1582,21 @@ static Instruction *moveAddAfterMinMax(SelectPatternFlavor SPF, Value *X,
     return BinaryOperator::CreateNUW(BinaryOperator::Add, NewMinMax,
                                      ConstantInt::get(X->getType(), *C1));
   }
-  // TODO: Handle SMIN/SMAX (similar to unsigned, but the signed subtraction of
-  // the constants must not overflow).
+
+  if (!IsUnsigned && match(X, m_NSWAdd(m_Value(A), m_APInt(C1))) &&
+      match(Y, m_APInt(C2)) && X->hasNUses(2)) {
+    bool Overflow;
+    APInt Diff = C2->ssub_ov(*C1, Overflow);
+    if (!Overflow) {
+      // smin (add nsw A, C1), C2 --> add nsw (smin A, C2 - C1), C1
+      // smax (add nsw A, C1), C2 --> add nsw (smax A, C2 - C1), C1
+      Value *NewMinMax = createMinMax(Builder, SPF, A,
+                                      ConstantInt::get(X->getType(), Diff));
+      return BinaryOperator::CreateNSW(BinaryOperator::Add, NewMinMax,
+                                       ConstantInt::get(X->getType(), *C1));
+    }
+  }
+
   return nullptr;
 }
 

diff --git a/test/Transforms/InstCombine/minmax-fold.ll b/test/Transforms/InstCombine/minmax-fold.ll
index 0e232df9582c691b0cda7b2dd93c6d183b2dc659..d312155c523e18f1cfc262ebdafb35c1614435a9 100644 (file)
--- a/test/Transforms/InstCombine/minmax-fold.ll
+++ b/test/Transforms/InstCombine/minmax-fold.ll
@@ -1164,9 +1164,9 @@ define i8 @PR14613_umax(i8 %x) {
 
 define i32 @add_smin(i32 %x) {
 ; CHECK-LABEL: @add_smin(
-; CHECK-NEXT:    [[A:%.*]] = add nsw i32 [[X:%.*]], 15
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i32 [[A]], 42
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[C]], i32 [[A]], i32 42
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp slt i32 [[X:%.*]], 27
+; CHECK-NEXT:    [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[X]], i32 27
+; CHECK-NEXT:    [[R:%.*]] = add nsw i32 [[TMP2]], 15
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %a = add nsw i32 %x, 15
@@ -1177,9 +1177,9 @@ define i32 @add_smin(i32 %x) {
 
 define i32 @add_smin_constant_limit(i32 %x) {
 ; CHECK-LABEL: @add_smin_constant_limit(
-; CHECK-NEXT:    [[A:%.*]] = add nsw i32 [[X:%.*]], -3
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i32 [[A]], 2147483643
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[C]], i32 [[A]], i32 2147483643
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp slt i32 [[X:%.*]], 2147483646
+; CHECK-NEXT:    [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[X]], i32 2147483646
+; CHECK-NEXT:    [[R:%.*]] = add nsw i32 [[TMP2]], -3
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %a = add nsw i32 %x, -3
@@ -1193,9 +1193,7 @@ define i32 @add_smin_constant_limit(i32 %x) {
 
 define i32 @add_smin_simplify(i32 %x) {
 ; CHECK-LABEL: @add_smin_simplify(
-; CHECK-NEXT:    [[A:%.*]] = add nsw i32 [[X:%.*]], -3
-; CHECK-NEXT:    [[C:%.*]] = icmp slt i32 [[A]], 2147483644
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[C]], i32 [[A]], i32 2147483644
+; CHECK-NEXT:    [[R:%.*]] = add nsw i32 [[X:%.*]], -3
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %a = add nsw i32 %x, -3
@@ -1267,9 +1265,9 @@ define i32 @add_smin_extra_use(i32 %x, i32* %p) {
 
 define <2 x i16> @add_smin_vec(<2 x i16> %x) {
 ; CHECK-LABEL: @add_smin_vec(
-; CHECK-NEXT:    [[A:%.*]] = add nsw <2 x i16> [[X:%.*]], <i16 15, i16 15>
-; CHECK-NEXT:    [[C:%.*]] = icmp slt <2 x i16> [[A]], <i16 240, i16 240>
-; CHECK-NEXT:    [[R:%.*]] = select <2 x i1> [[C]], <2 x i16> [[A]], <2 x i16> <i16 240, i16 240>
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp slt <2 x i16> [[X:%.*]], <i16 225, i16 225>
+; CHECK-NEXT:    [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x i16> [[X]], <2 x i16> <i16 225, i16 225>
+; CHECK-NEXT:    [[R:%.*]] = add nsw <2 x i16> [[TMP2]], <i16 15, i16 15>
 ; CHECK-NEXT:    ret <2 x i16> [[R]]
 ;
   %a = add nsw <2 x i16> %x, <i16 15, i16 15>
@@ -1280,9 +1278,9 @@ define <2 x i16> @add_smin_vec(<2 x i16> %x) {
 
 define i37 @add_smax(i37 %x) {
 ; CHECK-LABEL: @add_smax(
-; CHECK-NEXT:    [[A:%.*]] = add nsw i37 [[X:%.*]], 5
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt i37 [[A]], 42
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[C]], i37 [[A]], i37 42
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i37 [[X:%.*]], 37
+; CHECK-NEXT:    [[TMP2:%.*]] = select i1 [[TMP1]], i37 [[X]], i37 37
+; CHECK-NEXT:    [[R:%.*]] = add nuw nsw i37 [[TMP2]], 5
 ; CHECK-NEXT:    ret i37 [[R]]
 ;
   %a = add nsw i37 %x, 5
@@ -1293,9 +1291,9 @@ define i37 @add_smax(i37 %x) {
 
 define i8 @add_smax_constant_limit(i8 %x) {
 ; CHECK-LABEL: @add_smax_constant_limit(
-; CHECK-NEXT:    [[A:%.*]] = add nsw i8 [[X:%.*]], 125
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt i8 [[A]], -2
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[C]], i8 [[A]], i8 -2
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i8 [[X:%.*]], -127
+; CHECK-NEXT:    [[TMP2:%.*]] = select i1 [[TMP1]], i8 [[X]], i8 -127
+; CHECK-NEXT:    [[R:%.*]] = add nsw i8 [[TMP2]], 125
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %a = add nsw i8 %x, 125
@@ -1309,9 +1307,7 @@ define i8 @add_smax_constant_limit(i8 %x) {
 
 define i8 @add_smax_simplify(i8 %x) {
 ; CHECK-LABEL: @add_smax_simplify(
-; CHECK-NEXT:    [[A:%.*]] = add nsw i8 [[X:%.*]], 126
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt i8 [[A]], -2
-; CHECK-NEXT:    [[R:%.*]] = select i1 [[C]], i8 [[A]], i8 -2
+; CHECK-NEXT:    [[R:%.*]] = add nsw i8 [[X:%.*]], 126
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %a = add nsw i8 %x, 126
@@ -1383,9 +1379,9 @@ define i32 @add_smax_extra_use(i32 %x, i32* %p) {
 
 define <2 x i33> @add_smax_vec(<2 x i33> %x) {
 ; CHECK-LABEL: @add_smax_vec(
-; CHECK-NEXT:    [[A:%.*]] = add nsw <2 x i33> [[X:%.*]], <i33 5, i33 5>
-; CHECK-NEXT:    [[C:%.*]] = icmp sgt <2 x i33> [[A]], <i33 240, i33 240>
-; CHECK-NEXT:    [[R:%.*]] = select <2 x i1> [[C]], <2 x i33> [[A]], <2 x i33> <i33 240, i33 240>
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <2 x i33> [[X:%.*]], <i33 235, i33 235>
+; CHECK-NEXT:    [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x i33> [[X]], <2 x i33> <i33 235, i33 235>
+; CHECK-NEXT:    [[R:%.*]] = add nuw nsw <2 x i33> [[TMP2]], <i33 5, i33 5>
 ; CHECK-NEXT:    ret <2 x i33> [[R]]
 ;
   %a = add nsw <2 x i33> %x, <i33 5, i33 5>
@@ -1396,12 +1392,10 @@ define <2 x i33> @add_smax_vec(<2 x i33> %x) {
 
 define i8 @PR14613_smin(i8 %x) {
 ; CHECK-LABEL: @PR14613_smin(
-; CHECK-NEXT:    [[U4:%.*]] = sext i8 [[X:%.*]] to i32
-; CHECK-NEXT:    [[U5:%.*]] = add nuw nsw i32 [[U4]], 15
-; CHECK-NEXT:    [[U6:%.*]] = icmp slt i32 [[U5]], 55
-; CHECK-NEXT:    [[U7:%.*]] = select i1 [[U6]], i32 [[U5]], i32 55
-; CHECK-NEXT:    [[R:%.*]] = trunc i32 [[U7]] to i8
-; CHECK-NEXT:    ret i8 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp slt i8 [[X:%.*]], 40
+; CHECK-NEXT:    [[TMP2:%.*]] = select i1 [[TMP1]], i8 [[X]], i8 40
+; CHECK-NEXT:    [[U7:%.*]] = add i8 [[TMP2]], 15
+; CHECK-NEXT:    ret i8 [[U7]]
 ;
   %u4 = sext i8 %x to i32
   %u5 = add nuw nsw i32 %u4, 15
@@ -1413,12 +1407,10 @@ define i8 @PR14613_smin(i8 %x) {
 
 define i8 @PR14613_smax(i8 %x) {
 ; CHECK-LABEL: @PR14613_smax(
-; CHECK-NEXT:    [[U4:%.*]] = sext i8 [[X:%.*]] to i32
-; CHECK-NEXT:    [[U5:%.*]] = add nuw nsw i32 [[U4]], 15
-; CHECK-NEXT:    [[U6:%.*]] = icmp sgt i32 [[U5]], 55
-; CHECK-NEXT:    [[U7:%.*]] = select i1 [[U6]], i32 [[U5]], i32 55
-; CHECK-NEXT:    [[R:%.*]] = trunc i32 [[U7]] to i8
-; CHECK-NEXT:    ret i8 [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i8 [[X:%.*]], 40
+; CHECK-NEXT:    [[TMP2:%.*]] = select i1 [[TMP1]], i8 [[X]], i8 40
+; CHECK-NEXT:    [[U7:%.*]] = add nuw i8 [[TMP2]], 15
+; CHECK-NEXT:    ret i8 [[U7]]
 ;
   %u4 = sext i8 %x to i32
   %u5 = add nuw nsw i32 %u4, 15