[llvm] [InstCombine] Tighten use constraint in factorization transforms (PR #102943)

[via llvm-commits]

================
@@ -1424,6 +1424,12 @@ static Instruction *factorizeMathWithShlOps(BinaryOperator &I,
       !match(Op1, m_Shl(m_Value(Y), m_Specific(ShAmt))))
     return nullptr;
 
+  // This transform is only profitiable if both operations or one operation and
+  // the resulting add/sub can be eliminated/folded.
+  if (!(Op0->hasOneUse() && Op1->hasOneUse()) &&
+      !(isa<Constant>(X) && isa<Constant>(Y)))
----------------
goldsteinn wrote:

Probably more precise is `!simplifyBinOp(I.getOpcode(), X, Y)`

https://github.com/llvm/llvm-project/pull/102943