[PATCH] D64224: Keep the order of the basic blocks in the cloned loop as the original loop

[Whitney via Phabricator via llvm-commits]

Whitney marked an inline comment as done.
Whitney added inline comments.


================
Comment at: llvm/lib/Transforms/Utils/CloneFunction.cpp:783
+  for (BasicBlock *BB : OrigLoop->getBlocks()) {
+    Loop *CurLoop = LI->getLoopFor(BB);
+    Loop *&NewLoop = LMap[CurLoop];
----------------
hfinkel wrote:
> Is this logic equivalent? The original code is going a pre-order traversal, and for each loop, it excludes blocks in an inner loop (when `CurLoop != LI->getLoopFor(BB)`). Here we're doing something for all blocks in the loop including those also in inner loops?
The resulting cloned basic blocks will be the same, which are all the blocks in OrigLoop. The different is the order of the basic block. Example:
Given OrigLoop:
```
OuterHeader:
  br InnerHeader
InnerHeader:
  br InnerHeader or OuterLatch
OuterLatch:
  br OuterHeader or OuterExit
```
Output before this change:
```
ClonedOuterHeader:
  br ClonedInnerHeader
ClonedOuterLatch:
  br ClonedOuterHeader or ClonedOuterExit
ClonedInnerHeader:
  br ClonedInnerHeader or ClonedOuterLatch
```
Output after this change:
```
ClonedOuterHeader:
  br ClonedInnerHeader
ClonedInnerHeader:
  br ClonedInnerHeader or ClonedOuterLatch
ClonedOuterLatch:
  br ClonedOuterHeader or ClonedOuterExit
```

FYI - this function is extended by me in https://reviews.llvm.org/rG7c1deeff4a67296654823a871fea5c1a2aef3b8a to support cloning a loop nest instead of only allow the innermost loop. But I didn't think properly about the ordering of the basic block at that time.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D64224/new/

https://reviews.llvm.org/D64224