[MemorySSA] Add previous def to cache when found, even if trivial.

Summary:
When inserting a new Def, MemorySSA may be have non-minimal number of Phis.
While inserting, the walk to find the previous definition may cleanup minimal Phis.
When the last definition is trivial to obtain, we do not cache it.

It is possible while getting the previous definition for a Def to get two different answers:
- one that was straight-forward to find when walking the first path (a trivial phi in this case), and
- another that follows a cleanup of the trivial phi, it determines it may need additional Phi nodes, it inserts them and returns a new phi in the same position as the former trivial one.
While the Phis added for the second path are all redundant, they are not complete (the walk is only done upwards), and they are not properly cleaned up afterwards.

A way to fix this problem is to cache the straight-forward answer we got on the first walk.
The caching is only kept for the duration of a getPreviousDef call, and for Phis we use TrackingVH, so removing the trivial phi will lead to replacing it with the next dominating phi in the cache.
Resolves PR40749.

Reviewers: george.burgess.iv

Subscribers: jlebar, Prazek, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D60634

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@358313 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Analysis/MemorySSAUpdater.cpp b/lib/Analysis/MemorySSAUpdater.cpp
index fe84ca1207ed2f89d4ef250809077286c8bee157..3f3fd998d5cc4bd9aa8f65079d98bfe243a21eb0 100644 (file)
--- a/lib/Analysis/MemorySSAUpdater.cpp
+++ b/lib/Analysis/MemorySSAUpdater.cpp
@@ -156,8 +156,10 @@ MemoryAccess *MemorySSAUpdater::getPreviousDefFromEnd(
     DenseMap<BasicBlock *, TrackingVH<MemoryAccess>> &CachedPreviousDef) {
   auto *Defs = MSSA->getWritableBlockDefs(BB);
 
-  if (Defs)
+  if (Defs) {
+    CachedPreviousDef.insert({BB, &*Defs->rbegin()});
     return &*Defs->rbegin();
+  }
 
   return getPreviousDefRecursive(BB, CachedPreviousDef);
 }

diff --git a/test/Analysis/MemorySSA/pr40749_2.ll b/test/Analysis/MemorySSA/pr40749_2.ll
new file mode 100644 (file)
index 0000000..1f36291
--- /dev/null
+++ b/test/Analysis/MemorySSA/pr40749_2.ll
@@ -0,0 +1,62 @@
+; RUN: opt -S -licm -loop-unswitch -enable-mssa-loop-dependency -verify-memoryssa %s | FileCheck %s
+; REQUIRES: asserts
+target datalayout = "E-m:e-i1:8:16-i8:8:16-i64:64-f128:64-v128:64-a:8:16-n32:64"
+target triple = "s390x-ibm-linux"
+
+@g_92 = external dso_local local_unnamed_addr global i16, align 2
+@g_993 = external dso_local local_unnamed_addr global i32, align 4
+
+; CHECK-LABEL: @ff6
+define dso_local fastcc void @ff6(i16 %arg1) unnamed_addr #0 {
+bb:
+  %tmp6.i = icmp sgt i16 %arg1, 0
+  br label %bb10
+
+bb10:                                             ; preds = %bb81.loopexit, %bb
+  %tmp17 = load i16, i16* @g_92, align 2
+  %tmp18 = add i16 %tmp17, 1
+  store i16 %tmp18, i16* @g_92, align 2
+  br label %bb19
+
+bb19:                                             ; preds = %bb42, %bb10
+  br label %bb24.preheader
+
+bb24.preheader:                                   ; preds = %bb75, %bb19
+  store i32 0, i32* @g_993, align 4
+  br i1 %tmp6.i, label %bb24.preheader.split.us, label %bb24.preheader.split
+
+bb24.preheader.split.us:                          ; preds = %bb24.preheader
+  br label %bb61.us
+
+bb67.us.loopexit:                                 ; preds = %bb65.us
+  br label %bb75
+
+bb61.us:                                          ; preds = %bb65.us, %bb24.preheader.split.us
+  br i1 false, label %bb65.us, label %bb81.loopexit
+
+bb65.us:                                          ; preds = %bb61.us
+  br i1 false, label %bb61.us, label %bb67.us.loopexit
+
+bb24.preheader.split:                             ; preds = %bb24.preheader
+  br label %bb27
+
+bb27:                                             ; preds = %bb24.preheader.split
+  br i1 false, label %bb42, label %bb67
+
+bb42:                                             ; preds = %bb27
+  br label %bb19
+
+bb67:                                             ; preds = %bb27
+  br label %bb75
+
+bb75:                                             ; preds = %bb67, %bb67.us.loopexit
+  br i1 undef, label %bb24.preheader, label %bb84.loopexit
+
+bb81.loopexit:                                    ; preds = %bb61.us
+  br label %bb10
+
+bb84.loopexit:                                    ; preds = %bb75
+  ret void
+}
+
+attributes #0 = { "target-features"="+transactional-execution,+vector" }