[llvm] [CGP]: Optimize mul.overflow. (PR #148343)

[David Green via llvm-commits]

================
@@ -6389,6 +6395,235 @@ bool CodeGenPrepare::optimizeGatherScatterInst(Instruction *MemoryInst,
   return true;
 }
 
+// Rewrite the mul_with_overflow intrinsic by checking if both of the
+// operands' value range is within the legal type. If so, we can optimize the
+// multiplication algorithm. This code is supposed to be written during the step
+// of type legalization, but given that we need to reconstruct the IR which is
+// not doable there, we do it here.
+bool CodeGenPrepare::optimizeMulWithOverflow(Instruction *I, bool IsSigned,
+                                             ModifyDT &ModifiedDT) {
+  if (!TLI->shouldOptimizeMulOverflowIntrinsic())
+    return false;
+
+  if (TLI->getTypeAction(
+          I->getContext(),
+          TLI->getValueType(*DL, I->getType()->getContainedType(0))) !=
+      TargetLowering::TypeExpandInteger)
+    return false;
+
+  Value *LHS = I->getOperand(0);
+  Value *RHS = I->getOperand(1);
+  Type *Ty = LHS->getType();
+  unsigned VTBitWidth = Ty->getScalarSizeInBits();
+  unsigned VTHalfBitWidth = VTBitWidth / 2;
+  IntegerType *LegalTy =
+      IntegerType::getIntNTy(I->getContext(), VTHalfBitWidth);
+
+  // Skip the optimization if the type with HalfBitWidth is not legal for the
+  // target.
+  if (TLI->getTypeAction(I->getContext(), TLI->getValueType(*DL, LegalTy)) !=
+      TargetLowering::TypeLegal)
+    return false;
+
+  // Make sure that the I->getType() is a struct type with two elements.
+  if (!I->getType()->isStructTy() || I->getType()->getStructNumElements() != 2)
+    return false;
+
+  // Keep track of the instruction to stop reoptimizing it again.
+  InsertedInsts.insert(I);
+  // ----------------------------
+
+  // For the simple case where IR just checks the overflow flag, new blocks
+  // should be:
+  //  entry:
+  //    if signed:
+  //      (lhs_lo ^ lhs_hi) || (rhs_lo ^ rhs_hi) ? overflow, overflow_no
----------------
davemgreen wrote:

Does there needs to be a `>> BW-1` on the lhs_lo/rhs_lo?

https://github.com/llvm/llvm-project/pull/148343