[llvm] 275bee3 - [LoopUnroll] Forget block and loop dispositions during unrolling.

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2022-09-27T08:49:04+01:00
New Revision: 275bee32ad305eb5da121e8a60918256f3caf1e7

URL: https://github.com/llvm/llvm-project/commit/275bee32ad305eb5da121e8a60918256f3caf1e7
DIFF: https://github.com/llvm/llvm-project/commit/275bee32ad305eb5da121e8a60918256f3caf1e7.diff

LOG: [LoopUnroll] Forget block and loop dispositions during unrolling.

After unrolling a loop, the block and loop dispositions need to be
cleared. As we don't know which SCEVs in the loop/blocks may be
impacted, completely clear the cache. This should also fix some cases
where deleted loops remained in the LoopDispositions cache.

This fixes a verification failure surfaced by D134531.

I am planning on reviewing/updating the existing uses of
forgetLoopDispositions to check if they should be replaced by
forgetBlockAndLoopDispositions.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D134612

Added: 
    

Modified: 
    llvm/include/llvm/Analysis/ScalarEvolution.h
    llvm/lib/Analysis/ScalarEvolution.cpp
    llvm/lib/Transforms/Utils/LoopUnroll.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/Analysis/ScalarEvolution.h b/llvm/include/llvm/Analysis/ScalarEvolution.h
index 1de1296a63a9..83f8a9e94bfe 100644
--- a/llvm/include/llvm/Analysis/ScalarEvolution.h
+++ b/llvm/include/llvm/Analysis/ScalarEvolution.h
@@ -939,6 +939,13 @@ class ScalarEvolution {
   /// recompute is simpler.
   void forgetLoopDispositions();
 
+  /// Called when the client has changed the disposition of values in
+  /// a loop or block.
+  ///
+  /// We don't have a way to invalidate per-loop/per-block dispositions. Clear
+  /// and recompute is simpler.
+  void forgetBlockAndLoopDispositions();
+
   /// Determine the minimum number of zero bits that S is guaranteed to end in
   /// (at every loop iteration).  It is, at the same time, the minimum number
   /// of times S is divisible by 2.  For example, given {4,+,8} it returns 2.

diff  --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 966bb3594433..f56b9e7950c1 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -8384,6 +8384,11 @@ void ScalarEvolution::forgetValue(Value *V) {
 
 void ScalarEvolution::forgetLoopDispositions() { LoopDispositions.clear(); }
 
+void ScalarEvolution::forgetBlockAndLoopDispositions() {
+  BlockDispositions.clear();
+  LoopDispositions.clear();
+}
+
 /// Get the exact loop backedge taken count considering all loop exits. A
 /// computable result can only be returned for loops with all exiting blocks
 /// dominating the latch. howFarToZero assumes that the limit of each loop test

diff  --git a/llvm/lib/Transforms/Utils/LoopUnroll.cpp b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
index 75961a1d898c..6d6c3db034ca 100644
--- a/llvm/lib/Transforms/Utils/LoopUnroll.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
@@ -465,8 +465,10 @@ LoopUnrollResult llvm::UnrollLoop(Loop *L, UnrollLoopOptions ULO, LoopInfo *LI,
   if (SE) {
     if (ULO.ForgetAllSCEV)
       SE->forgetAllLoops();
-    else
+    else {
       SE->forgetTopmostLoop(L);
+      SE->forgetBlockAndLoopDispositions();
+    }
   }
 
   if (!LatchIsExiting)