[llvm] d83a96f - [DAG] Make it clear mul(x,x) knownbits bit[1] == 0 check should be for x is undef only

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2022-02-01T11:32:14Z
New Revision: d83a96f59f93290aa7771388498bb620974de6b7

URL: https://github.com/llvm/llvm-project/commit/d83a96f59f93290aa7771388498bb620974de6b7
DIFF: https://github.com/llvm/llvm-project/commit/d83a96f59f93290aa7771388498bb620974de6b7.diff

LOG: [DAG] Make it clear mul(x,x) knownbits bit[1] == 0 check should be for x is undef only

As raised on rGffd0e464b4b9, if x is poison, this fold is still ok.

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/test/CodeGen/X86/combine-mul.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index cf42a2200564b..3eff266bc89d4 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -3082,6 +3082,7 @@ KnownBits SelectionDAG::computeKnownBits(SDValue Op, const APInt &DemandedElts,
     Known = computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
     Known2 = computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
     bool SelfMultiply = Op.getOperand(0) == Op.getOperand(1);
+    // TODO: SelfMultiply can be poison, but not undef.
     SelfMultiply &= isGuaranteedNotToBeUndefOrPoison(
         Op.getOperand(0), DemandedElts, false, Depth + 1);
     Known = KnownBits::mul(Known, Known2, SelfMultiply);

diff  --git a/llvm/test/CodeGen/X86/combine-mul.ll b/llvm/test/CodeGen/X86/combine-mul.ll
index 403443da60ee3..0d0aff21550a5 100644
--- a/llvm/test/CodeGen/X86/combine-mul.ll
+++ b/llvm/test/CodeGen/X86/combine-mul.ll
@@ -393,7 +393,7 @@ define <4 x i32> @combine_mul_self_knownbits_vector(<4 x i32> %x) {
   ret <4 x i32> %2
 }
 
-; mul(x,x) - bit[1] is 0, but if demanding the other bits the source must not be undef/poison
+; mul(x,x) - bit[1] is 0, but if demanding the other bits the source must not be undef
 
 define i64 @combine_mul_self_demandedbits(i64 %x) {
 ; SSE-LABEL: combine_mul_self_demandedbits: