[PATCH] D25963: [LoopUnroll] Implement profile-based loop peeling

[Michael Kuperstein via llvm-commits]

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


================
Comment at: lib/Transforms/Utils/LoopUnroll.cpp:357
+              << " peeling loop %" << Header->getName() << " by "
+              << NV("PeelCount", PeelCount) << ".\n");
   } else {
----------------
davidxl wrote:
> .\n --> iterations.\n
Sure.


================
Comment at: lib/Transforms/Utils/LoopUnrollPeel.cpp:63
+unsigned llvm::getPeelCount(Loop *L, unsigned LoopSize,
+                            TargetTransformInfo::UnrollingPreferences *UP) {
+  if (!canPeel(L))
----------------
davidxl wrote:
> Is UP parameter intended to be used?
Sorry, I don't quite understand what you mean here. It is used (in line 84 or below). Could you explain?


================
Comment at: lib/Transforms/Utils/LoopUnrollPeel.cpp:121
+                            BasicBlock *PreHeader, BasicBlock *Exit,
+                            uint64_t &TotalLoopWeight,
+                            SmallVectorImpl<BasicBlock *> &NewBlocks,
----------------
davidxl wrote:
> I think the following weights should be passed in:
> 
> 
> * TotalHeaderWeight of the currently peeled loop -- it is passed in for update:
>      TotalHeaderWeight -= PeeledIterationHeaderWeight;
> 
> * PeeledIterationHeaderWeight 
>    1) its initial value for the first peeled iteration is the PreheaderWeight or ExitWeight of the original loop
>    2) After peeling of one iteration, its value will be updated to the fallthrough weight of the peel
> 
> * TotalExitEdgeWeight -- passed in for update.
>     After peeling,  TotalExitEdgeWeight -= PeeledIterationExitWeight'
> 
> 
I don't think we need TotalExitEdgeWeight, but you're right, we need to accumulate the header weight to be able to update the final weight of the backedge properly.


================
Comment at: lib/Transforms/Utils/LoopUnrollPeel.cpp:168
+    uint64_t FallThruWeight =
+        TotalLoopWeight * ((float)(PeelCount - IterNumber) / PeelCount * 0.9);
+    uint64_t ExitWeight = TotalLoopWeight - FallThruWeight;
----------------
davidxl wrote:
> Should average trip count be used instead of PeelCount (which may be different)?
Right, I assumed they're the same (and currently they are), but it's not necessarily true in general, I'll add a parameter (and pass PeelCount).


https://reviews.llvm.org/D25963