[llvm-commits] [patch] Enable simplifycfg in bb with only a lifetime intrinsic

[Nick Lewycky]

Rafael Ávila de Espíndola wrote:
> I noticed that simplify cfg would fail to simplify bbs like
>
> bb:
>    call void @llvm.lifetime.end(i64 -1, i8* %a) nounwind
>    br label %bb1
>
> The attached patch enables that. It can cause a llvm.lifetime.* to be lost,
> but if I understand the manual correctly, that is fine: the argument now
> has a larger life range.

-  while (PHINode *PN = dyn_cast<PHINode>(&BB->front())) {
-    if (Succ->getSinglePredecessor()) {
-      // BB is the only predecessor of Succ, so Succ will end up with 
exactly
-      // the same predecessors BB had.
+  if (Succ->getSinglePredecessor()) {
+    // BB is the only predecessor of Succ, so Succ will end up with exactly
+    // the same predecessors BB had.
+
+    // Copy over any phi, debug or lifetime instruction.
+    BB->getTerminator()->eraseFromParent();
+    while (!BB->empty())
        Succ->getInstList().splice(Succ->begin(),
                                   BB->getInstList(), BB->begin());
-    } else {
+  } else {
+    while (PHINode *PN = dyn_cast<PHINode>(&BB->front())) {

Wow, I'm impressed with how wrong this loop was. :) I mean, it had 
correct behaviour, but did N splices when it just needed to splice N 
items. Anyhow, thanks for fixing it!

Is there any reason not to copy the lifetime/debug intrinsics from this 
block to its successor when the successor has a single predecessor?

/// TryToSimplifyUncondBranchFromEmptyBlock - BB is known to contain an
/// unconditional branch, and contains no instructions other than PHI nodes,
/// potential debug intrinsics and the branch.  If possible, eliminate BB by
/// rewriting all the predecessors to branch to the successor block and 
return
/// true.  If we can't transform, return false.

Change "potential debug intrinsics" to "potential side-effect-free 
intrinsics"?

Nick

>
> Cheers,
> Rafael
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits