[MemorySSA] Don't optimize incomplete phis.

Summary:
MemoryPhis cannot be optimized out until they are complete.
Resolves PR41254.

Reviewers: george.burgess.iv

Subscribers: sanjoy, jlebar, Prazek, jdoerfert, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D59966

llvm-svn: 357315

llvm/lib/Analysis/MemorySSAUpdater.cpp

@@ -309,8 +309,15 @@ void MemorySSAUpdater::insertDef(MemoryDef *MD, bool RenameUses) {
       IDFs.calculate(IDFBlocks);
       SmallVector<AssertingVH<MemoryPhi>, 4> NewInsertedPHIs;
       for (auto *BBIDF : IDFBlocks)
-        if (!MSSA->getMemoryAccess(BBIDF))
-          NewInsertedPHIs.push_back(MSSA->createMemoryPhi(BBIDF));
+        if (!MSSA->getMemoryAccess(BBIDF)) {
+          auto *MPhi = MSSA->createMemoryPhi(BBIDF);
+          NewInsertedPHIs.push_back(MPhi);
+          // Add the phis created into the IDF blocks to NonOptPhis, so they are
+          // not optimized out as trivial by the call to getPreviousDefFromEnd
+          // below. Once they are complete, all these Phis are added to the
+          // FixupList, and removed from NonOptPhis inside fixupDefs().
+          NonOptPhis.insert(MPhi);
+        }
 
       for (auto &MPhi : NewInsertedPHIs) {
         auto *BBIDF = MPhi->getBlock();

llvm/test/Analysis/MemorySSA/pr41254.ll

@@ -0,0 +1,62 @@
+; RUN: opt -licm -enable-mssa-loop-dependency -verify-memoryssa -S < %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_328 = external dso_local local_unnamed_addr global { i32, i16, i32, i8, i8, i32, i32 }, align 4
+
+define dso_local void @func_45() local_unnamed_addr {
+; CHECK-LABEL: @func_45()
+bb:
+  br label %bb7
+
+bb7:                                              ; preds = %bb
+  br label %bb8
+
+bb8:                                              ; preds = %bb80, %bb7
+  %tmp10 = load i32, i32* getelementptr inbounds ({ i32, i16, i32, i8, i8, i32, i32 }, { i32, i16, i32, i8, i8, i32, i32 }* @g_328, i64 0, i32 5), align 4
+  %0 = or i32 %tmp10, 9
+  store i32 %0, i32* getelementptr inbounds ({ i32, i16, i32, i8, i8, i32, i32 }, { i32, i16, i32, i8, i8, i32, i32 }* @g_328, i64 0, i32 5), align 4
+  br label %bb41.preheader.preheader
+
+bb41.preheader.preheader:                         ; preds = %bb80.thread, %bb8
+  br label %bb68
+
+bb84.thread.split.loop.exit67:                    ; preds = %bb71.1
+  br label %bb84.thread
+
+bb84.thread.split.loop.exit71:                    ; preds = %bb71.2
+  br label %bb84.thread
+
+bb84.thread.split.loop.exit91:                    ; preds = %bb71.1.2
+  br label %bb84.thread
+
+bb84.thread:                                      ; preds = %bb84.thread.split.loop.exit91, %bb84.thread.split.loop.exit71, %bb84.thread.split.loop.exit67
+  unreachable
+
+bb68:                                             ; preds = %bb41.preheader.preheader
+  br i1 false, label %bb71, label %bb80
+
+bb71:                                             ; preds = %bb68
+  br label %bb71.1
+
+bb80.thread:                                      ; preds = %bb71.1.2
+  br label %bb41.preheader.preheader
+
+bb80:                                             ; preds = %bb68
+  br label %bb8
+
+bb71.1:                                           ; preds = %bb71
+  br i1 true, label %bb84.thread.split.loop.exit67, label %bb71.2
+
+bb71.2:                                           ; preds = %bb71.1
+  br i1 true, label %bb84.thread.split.loop.exit71, label %bb71.145
+
+bb71.145:                                         ; preds = %bb71.2
+  br label %bb71.1.2
+
+bb71.1.2:                                         ; preds = %bb71.145
+  br i1 true, label %bb84.thread.split.loop.exit91, label %bb80.thread
+}
+