[ObjCARC] Account for catchswitch in bitcast insertion

[smeenai]

Summary

A catchswitch is both a pad and a terminator, meaning it must be the
only non-phi instruction in its basic block. When we're inserting a
bitcast in the incoming basic block for a phi, if that incoming block is
a catchswitch, we should go up the dominator tree to find a valid
insertion point rather than attempting to insert before the catchswitch
(which would result in invalid IR).

[rjmccall]

Okay, LGTM.

llvm/trunk/lib/Transforms/ObjCARC/ObjCARCContract.cpp

auto ReplaceArgUses = [Inst, this](Value *Arg) {
           continue;
 
         Changed = true;
         Instruction *Replacement = Inst;
         Type *UseTy = U.get()->getType();
         if (PHINode *PHI = dyn_cast<PHINode>(U.getUser())) {
           // For PHI nodes, insert the bitcast in the predecessor block.
           unsigned ValNo = PHINode::getIncomingValueNumForOperand(OperandNo);
-          BasicBlock *BB = PHI->getIncomingBlock(ValNo);
-          if (Replacement->getType() != UseTy)
-            Replacement = new BitCastInst(Replacement, UseTy, "", &BB->back());
+          BasicBlock *IncomingBB = PHI->getIncomingBlock(ValNo);
+          if (Replacement->getType() != UseTy) {
+            // A catchswitch is both a pad and a terminator, meaning a basic
+            // block with a catchswitch has no insertion point. Keep going up
+            // the dominator tree until we find a non-catchswitch.
+            BasicBlock *InsertBB = IncomingBB;
+            while (isa<CatchSwitchInst>(InsertBB->getFirstNonPHI())) {
+              InsertBB = DT->getNode(InsertBB)->getIDom()->getBlock();
+            }
+
+            assert(DT->dominates(Inst, &InsertBB->back()) &&
+                   "Invalid insertion point for bitcast");
+            Replacement =
+                new BitCastInst(Replacement, UseTy, "", &InsertBB->back());
+          }
+
           // While we're here, rewrite all edges for this PHI, rather
           // than just one use at a time, to minimize the number of
           // bitcasts we emit.
           for (unsigned i = 0, e = PHI->getNumIncomingValues(); i != e; ++i)
-            if (PHI->getIncomingBlock(i) == BB) {
+            if (PHI->getIncomingBlock(i) == IncomingBB) {
               // Keep the UI iterator valid.
               if (UI != UE &&
                   &PHI->getOperandUse(
                       PHINode::getOperandNumForIncomingValue(i)) == &*UI)
                 ++UI;
               PHI->setIncomingValue(i, Replacement);
             }
         } else {

llvm/trunk/test/Transforms/ObjCARC/contract-catchswitch.ll

+; RUN: opt -S -objc-arc-contract < %s | FileCheck %s
+
+target datalayout = "e-m:x-p:32:32-i64:64-f80:32-n8:16:32-a:0:32-S32"
+target triple = "i686--windows-msvc19.11.0"
+
+%0 = type opaque
+
+declare i32 @__CxxFrameHandler3(...)
+declare dllimport void @objc_release(i8*) local_unnamed_addr
+declare dllimport i8* @objc_retain(i8* returned) local_unnamed_addr
+
+@p = global i8* null, align 4
+
+declare void @f() local_unnamed_addr
+
+define void @g() local_unnamed_addr personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
+entry:
+  %tmp = load i8*, i8** @p, align 4
+  %cast = bitcast i8* %tmp to %0*
+  %tmp1 = tail call i8* @objc_retain(i8* %tmp) #0
+  ; Split the basic block to ensure bitcast ends up in entry.split.
+  br label %entry.split
+
+entry.split:
+  invoke void @f()
+          to label %invoke.cont unwind label %catch.dispatch
+
+; Dummy nested catchswitch to test looping through the dominator tree.
+catch.dispatch:
+  %tmp2 = catchswitch within none [label %catch] unwind label %catch.dispatch1
+
+catch:
+  %tmp3 = catchpad within %tmp2 [i8* null, i32 64, i8* null]
+  catchret from %tmp3 to label %invoke.cont
+
+catch.dispatch1:
+  %tmp4 = catchswitch within none [label %catch1] unwind label %ehcleanup
+
+catch1:
+  %tmp5 = catchpad within %tmp4 [i8 *null, i32 64, i8* null]
+  catchret from %tmp5 to label %invoke.cont
+
+invoke.cont:
+  %tmp6 = load i8*, i8** @p, align 4
+  %cast1 = bitcast i8* %tmp6 to %0*
+  %tmp7 = tail call i8* @objc_retain(i8* %tmp6) #0
+  call void @objc_release(i8* %tmp) #0, !clang.imprecise_release !0
+  ; Split the basic block to ensure bitcast ends up in invoke.cont.split.
+  br label %invoke.cont.split
+
+invoke.cont.split:
+  invoke void @f()
+          to label %invoke.cont1 unwind label %ehcleanup
+
+invoke.cont1:
+  ret void
+
+ehcleanup:
+  %tmp8 = phi %0* [ %cast, %catch.dispatch1 ], [ %cast1, %invoke.cont.split ]
+  %tmp9 = cleanuppad within none []
+  %tmp10 = bitcast %0* %tmp8 to i8*
+  call void @objc_release(i8* %tmp10) #0 [ "funclet"(token %tmp9) ]
+  cleanupret from %tmp9 unwind to caller
+}
+
+; CHECK-LABEL: entry.split:
+; CHECK-NEXT:    %0 = bitcast i8* %tmp1 to %0*
+; CHECK-NEXT:    invoke void @f()
+; CHECK-NEXT:            to label %invoke.cont unwind label %catch.dispatch
+
+; CHECK-LABEL: invoke.cont.split:
+; CHECK-NEXT:    %1 = bitcast i8* %tmp7 to %0*
+; CHECK-NEXT:    invoke void @f()
+; CHECK-NEXT:            to label %invoke.cont1 unwind label %ehcleanup
+
+; CHECK-LABEL: ehcleanup:
+; CHECK-NEXT:    %tmp8 = phi %0* [ %0, %catch.dispatch1 ], [ %1, %invoke.cont.split ]
+
+attributes #0 = { nounwind }
+
+!0 = !{}