[llvm] [VPlan] Introduce child regions as VPlan transform. (PR #129402)

[Florian Hahn via llvm-commits]

================
@@ -113,13 +88,6 @@ static bool isHeaderBB(BasicBlock *BB, Loop *L) {
   return L && BB == L->getHeader();
 }
 
-void PlainCFGBuilder::setRegionPredsFromBB(VPRegionBlock *Region,
-                                           BasicBlock *BB) {
-  // BB is a loop header block. Connect the region to the loop preheader.
-  Loop *LoopOfBB = LI->getLoopFor(BB);
-  Region->setPredecessors({getOrCreateVPBB(LoopOfBB->getLoopPredecessor())});
-}
-
 // Add operands to VPInstructions representing phi nodes from the input IR.
 void PlainCFGBuilder::fixHeaderPhis() {
   for (auto *Phi : PhisToFix) {
----------------
fhahn wrote:

> Let VPlanCFG0 denote the initial VPlan, constructed from IR. VPlanCFG0 may indeed be a consistent 1-1 translation of the CFG and IR, arguably with all its branches - including early-exits and latch-to-exit (and perhaps unconditional branches too), possibly also representing the BasicBlock operands of phi's and branches as VPValues wrapping these Values, as operands of the corresponding VPIRInstructions.


Yep, I have some follow-up patches to retain the exit branches and keep the exit blocks connected.

In a sense, the specific operand order (1st op = preheader, 2nd op = latch) for header phis is purely a consequence of introducing a region, which itself only has a single predecessor I think. So introducing the canonical order at this point seems like a good fit; the header phis are the only ones which need their order adjusted.

Using a separate recipe for the initial recipes sounds good, something like VPInstruction which always has an underlying instruction would be good, but already models the operands/defs in VPlan via operands. Unfortunately VPIRInstruction is already taken. We would need a version that defines result VPValues and uses VPvalue as operands.

https://github.com/llvm/llvm-project/pull/129402