[X86] Use shift+add/sub for vXi8 splat multiplies

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -4841,7 +4841,8 @@ template <class MatchContextClass> SDValue DAGCombiner::visitMUL(SDNode *N) {
     //       callback to determine that sequence (similar to sqrt expansion).
     unsigned MathOp = ISD::DELETED_NODE;
     APInt MulC = ConstValue1.abs();
-    // The constant `2` should be treated as (2^0 + 1).
+    // Decompose 2^N ± 2^M as 2^(A+B) ± 2^B. The constant `2` should be
+    // treated as (2^0 + 1).
     unsigned TZeros = MulC == 2 ? 0 : MulC.countr_zero();
     MulC.lshrInPlace(TZeros);
     if ((MulC - 1).isPowerOf2())

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -3389,6 +3389,15 @@ bool X86TargetLowering::decomposeMulByConstant(LLVMContext &Context, EVT VT,
   if (!ISD::isConstantSplatVector(C.getNode(), MulC))
     return false;
 
+  if (VT.isVector() && VT.getScalarSizeInBits() == 8) {
+    // Decompose 2^m ± 2^n as 2^(a+b) ± 2^b
+    APInt ShiftedMulC = MulC.abs();
+    unsigned TZeros = ShiftedMulC == 2 ? 0 : ShiftedMulC.countr_zero();
+    ShiftedMulC.lshrInPlace(TZeros);
+    if ((ShiftedMulC - 1).isPowerOf2() || (ShiftedMulC + 1).isPowerOf2())
+      return true;
+  }
+
   // Find the type this will be legalized too. Otherwise we might prematurely
   // convert this to shl+add/sub and then still have to type legalize those ops.
   // Another choice would be to defer the decision for illegal types until