We can only move static allocas into the resume entry points

[aschwaighofer]

Summary

Dynamic allocas that still exist have been verified to be only used
'locally' not accross a suspend point.

rdar://73903220

[rjmccall]

The idea in this test case is that this gets lowered to a dynamic alloca before the code that adjusts for the entry point sees it? That might be worth spelling out in the test case. Otherwise LGTM.

[aschwaighofer]

Add comment to test case.

llvm/lib/Transforms/Coroutines/CoroSplit.cpp

case coro::ABI::RetconOnce: {
     auto *MappedCS = cast<AnyCoroSuspendInst>(VMap[ActiveSuspend]);
     auto Branch = cast<BranchInst>(MappedCS->getNextNode());
     assert(Branch->isUnconditional());
     Builder.CreateBr(Branch->getSuccessor(0));
     break;
   }
   }
 
-  // Any alloca that's still being used but not reachable from the new entry
-  // needs to be moved to the new entry.
+  // Any static alloca that's still being used but not reachable from the new
+  // entry needs to be moved to the new entry.
   Function *F = OldEntry->getParent();
   DominatorTree DT{*F};
   for (auto IT = inst_begin(F), End = inst_end(F); IT != End;) {
     Instruction &I = *IT++;
-    if (!isa<AllocaInst>(&I) || I.use_empty())
+    auto *Alloca = dyn_cast<AllocaInst>(&I);
+    if (!Alloca || I.use_empty())
       continue;
-    if (DT.isReachableFromEntry(I.getParent()))
+    if (DT.isReachableFromEntry(I.getParent()) ||
+        !isa<ConstantInt>(Alloca->getArraySize()))
       continue;
     I.moveBefore(*Entry, Entry->getFirstInsertionPt());
   }
 }
 
 /// Derive the value of the new frame pointer.
 Value *CoroCloner::deriveNewFramePointer() {
   // Builder should be inserting to the front of the new entry block.

llvm/test/Transforms/Coroutines/coro-retcon-alloca.ll

forward:
   %inc = add i32 %n.val, 1
   br label %back
 
 end:
   call i1 @llvm.coro.end(i8* %hdl, i1 0)
   unreachable
 }
 
+declare i32 @getSize()
+define {i8*, i32} @k(i8* %buffer, i32 %n, i1 %cond) {
+entry:
+  %id = call token @llvm.coro.id.retcon(i32 1024, i32 8, i8* %buffer, i8* bitcast ({i8*, i32} (i8*, i1)* @prototype_g to i8*), i8* bitcast (i8* (i32)* @allocate to i8*), i8* bitcast (void (i8*)* @deallocate to i8*))
+  %hdl = call i8* @llvm.coro.begin(token %id, i8* null)
+  br i1 %cond, label %alloca_block, label %non_alloca_block
+
+suspend:
+  %unwind = call i1 (...) @llvm.coro.suspend.retcon.i1(i32 %n)
+  br i1 %unwind, label %cleanup, label %resume
+
+resume:
+  br label %cleanup
+
+alloca_block:
+  %size = call i32 @getSize()
+  ; This will get lowered to a dynamic alloca.
+  ; Make sure code that runs after that lowering does not hoist the dynamic
+  ; alloca into the entry block of the resume function.
+  %alloca = call token @llvm.coro.alloca.alloc.i32(i32 %size, i32 8)
+  %ptr = call i8* @llvm.coro.alloca.get(token %alloca)
+  call void @use(i8* %ptr)
+  call void @llvm.coro.alloca.free(token %alloca)
+  br label %suspend
+
+non_alloca_block:
+  %ignore = call i32 @getSize()
+  br label %suspend
+
+cleanup:
+  call i1 @llvm.coro.end(i8* %hdl, i1 0)
+  unreachable
+}
+
 declare token @llvm.coro.id.retcon(i32, i32, i8*, i8*, i8*, i8*)
 declare i8* @llvm.coro.begin(token, i8*)
 declare i1 @llvm.coro.suspend.retcon.i1(...)
 declare void @llvm.coro.suspend.retcon.isVoid(...)
 declare i1 @llvm.coro.end(i8*, i1)
 declare i8* @llvm.coro.prepare.retcon(i8*)
 declare token @llvm.coro.alloca.alloc.i32(i32, i32)
 declare i8* @llvm.coro.alloca.get(token)
 declare void @llvm.coro.alloca.free(token)
 
 declare noalias i8* @allocate(i32 %size)
 declare void @deallocate(i8* %ptr)
 
 declare void @print(i32)
 declare void @use(i8*)