[llvm] [SLP] SLP's copyable elements based upon Main/Alt operations. (PR #124242)

[Dinar Temirbulatov via llvm-commits]

================
@@ -1088,6 +1144,69 @@ static InstructionsState getSameOpcode(ArrayRef<Value *> VL,
   return InstructionsState(MainOp, AltOp);
 }
 
+/// \returns analysis of the Instructions in \p VL described in
+/// InstructionsState in propose to vectorize with copyable instructions.
+static InstructionsState getCopyableOpcode(ArrayRef<Value *> VL,
+                                           const TargetLibraryInfo &TLI) {
+  if (!all_of(VL, IsaPred<Instruction>) || !VectorizeCopyable)
+    return InstructionsState::invalid();
+  Instruction *MainOp = cast<Instruction>(VL[0]);
+  Instruction *AltOp = nullptr;
+  unsigned Opcode = MainOp->getOpcode();
+  unsigned AltOpcode = Opcode;
+  auto getAltOp = [Opcode, &AltOpcode, &AltOp](ArrayRef<Value *> VL) {
+    for (Value *V : VL) {
+      Instruction *I = cast<Instruction>(V);
+      if (I->isIntDivRem() || I->isFPDivRem())
+        return false;
+      if (isa<PHINode>(I)) {
+        return false;
+      }
----------------
dtemirbulatov wrote:

I think it could somehow regress the flow analisys later?

https://github.com/llvm/llvm-project/pull/124242