[InstCombine] fold sub of low-bit masked value from offset of same value

There might be some demanded/known bits way to generalize this,
but I'm not seeing it right now.

This came up as a regression when I was looking at a different
demanded bits improvement.

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

  Name: general
  Pre: ((-1 << countTrailingZeros(C1)) & C2) == 0
  %a1 = add i8 %x, C1
  %a2 = and i8 %x, C2
  %r = sub i8 %a1, %a2
  =>
  %r = and i8 %a1, ~C2

  Name: test 1
  %a1 = add i8 %x, 192
  %a2 = and i8 %x, 10
  %r = sub i8 %a1, %a2
  =>
  %r = and i8 %a1, -11

  Name: test 2
  %a1 = add i8 %x, -108
  %a2 = and i8 %x, 3
  %r = sub i8 %a1, %a2
  =>
  %r = and i8 %a1, -4

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index e78df30..abeddf4 100644 (file)
--- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -2051,6 +2051,20 @@ Instruction *InstCombinerImpl::visitSub(BinaryOperator &I) {
     return SelectInst::Create(Cmp, Neg, A);
   }
 
+  // If we are subtracting a low-bit masked subset of some value from an add
+  // of that same value with no low bits changed, that is clearing some low bits
+  // of the sum:
+  // sub (X + AddC), (X & AndC) --> and (X + AddC), ~AndC
+  const APInt *AddC, *AndC;
+  if (match(Op0, m_Add(m_Value(X), m_APInt(AddC))) &&
+      match(Op1, m_And(m_Specific(X), m_APInt(AndC)))) {
+    unsigned BitWidth = Ty->getScalarSizeInBits();
+    unsigned Cttz = AddC->countTrailingZeros();
+    APInt HighMask(APInt::getHighBitsSet(BitWidth, BitWidth - Cttz));
+    if ((HighMask & *AndC).isNullValue())
+      return BinaryOperator::CreateAnd(Op0, ConstantInt::get(Ty, ~(*AndC)));
+  }
+
   if (Instruction *V =
           canonicalizeCondSignextOfHighBitExtractToSignextHighBitExtract(I))
     return V;

diff --git a/llvm/test/Transforms/InstCombine/sub.ll b/llvm/test/Transforms/InstCombine/sub.ll
index d703004..3e9c066 100644 (file)
--- a/llvm/test/Transforms/InstCombine/sub.ll
+++ b/llvm/test/Transforms/InstCombine/sub.ll
@@ -1464,8 +1464,7 @@ define i8 @sub_add_sub_reassoc_use2(i8 %w, i8 %x, i8 %y, i8 %z) {
 define i8 @sub_mask_lowbits(i8 %x) {
 ; CHECK-LABEL: @sub_mask_lowbits(
 ; CHECK-NEXT:    [[A1:%.*]] = add i8 [[X:%.*]], -108
-; CHECK-NEXT:    [[A2:%.*]] = and i8 [[X]], 3
-; CHECK-NEXT:    [[R:%.*]] = sub i8 [[A1]], [[A2]]
+; CHECK-NEXT:    [[R:%.*]] = and i8 [[A1]], -4
 ; CHECK-NEXT:    ret i8 [[R]]
 ;
   %a1 = add i8 %x, 148 ; 0x94
@@ -1474,6 +1473,8 @@ define i8 @sub_mask_lowbits(i8 %x) {
   ret i8 %r
 }
 
+; Negative test - low-bit mask must not overlap with offset
+
 define i8 @sub_not_mask_lowbits(i8 %x) {
 ; CHECK-LABEL: @sub_not_mask_lowbits(
 ; CHECK-NEXT:    [[A1:%.*]] = add i8 [[X:%.*]], 4
@@ -1492,7 +1493,7 @@ define <2 x i8> @sub_mask_lowbits_splat_extra_use(<2 x i8> %x, <2 x i8>* %p) {
 ; CHECK-NEXT:    [[A1:%.*]] = add <2 x i8> [[X:%.*]], <i8 -64, i8 -64>
 ; CHECK-NEXT:    [[A2:%.*]] = and <2 x i8> [[X]], <i8 10, i8 10>
 ; CHECK-NEXT:    store <2 x i8> [[A2]], <2 x i8>* [[P:%.*]], align 2
-; CHECK-NEXT:    [[R:%.*]] = sub <2 x i8> [[A1]], [[A2]]
+; CHECK-NEXT:    [[R:%.*]] = and <2 x i8> [[A1]], <i8 -11, i8 -11>
 ; CHECK-NEXT:    ret <2 x i8> [[R]]
 ;
   %a1 = add <2 x i8> %x, <i8 192, i8 192> ; 0xc0