[MemorySSA] Don't hoist stores if interfering uses (as calls) exist.

llvm-svn: 373674

lib/Transforms/Scalar/LICM.cpp

@@ -1248,12 +1248,22 @@ bool llvm::canSinkOrHoistInst(Instruction &I, AAResults *AA, DominatorTree *DT,
               // FIXME: More precise: no Uses that alias SI.
               if (!Flags->IsSink && !MSSA->dominates(SIMD, MU))
                 return false;
-            } else if (const auto *MD = dyn_cast<MemoryDef>(&MA))
+            } else if (const auto *MD = dyn_cast<MemoryDef>(&MA)) {
               if (auto *LI = dyn_cast<LoadInst>(MD->getMemoryInst())) {
                 (void)LI; // Silence warning.
                 assert(!LI->isUnordered() && "Expected unordered load");
                 return false;
               }
+              // Any call, while it may not be clobbering SI, it may be a use.
+              if (auto *CI = dyn_cast<CallInst>(MD->getMemoryInst())) {
+                // Check if the call may read from the memory locattion written
+                // to by SI. Check CI's attributes and arguments; the number of
+                // such checks performed is limited above by NoOfMemAccTooLarge.
+                ModRefInfo MRI = AA->getModRefInfo(CI, MemoryLocation::get(SI));
+                if (isModOrRefSet(MRI))
+                  return false;
+              }
+            }
         }
 
       auto *Source = MSSA->getSkipSelfWalker()->getClobberingMemoryAccess(SI);

test/Analysis/MemorySSA/pr43540.ll

@@ -0,0 +1,34 @@
+; RUN: opt -S -licm -enable-mssa-loop-dependency=true %s | FileCheck %s
+@v_1 = global i8 0, align 1
+@v_2 =  global i8 0, align 1
+
+; CHECK-LABEL: @foo()
+; CHECK: for.cond:
+; CHECK-NOT: store
+; CHECK: for.body:
+; CHECK: call void @llvm.memcpy.p0i8.p0i8.i64
+; CHECK: store
+define void @foo() {
+entry:
+  br label %for.cond
+
+for.cond:                                         ; preds = %for.body, %entry
+  %0 = phi i16 [ %inc, %for.body ], [ 0, %entry ]
+  %cmp = icmp slt i16 %0, 1
+  br i1 %cmp, label %for.body, label %for.end
+
+for.body:                                         ; preds = %for.cond
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* @v_1, i8 * @v_2, i64 1, i1 false)
+  store i8 1, i8 * @v_2, align 1
+  %inc = add nsw i16 %0, 1
+  br label %for.cond
+
+for.end:                                          ; preds = %for.cond
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* noalias nocapture writeonly, i8 * noalias nocapture readonly, i64, i1 immarg) #2
+
+attributes #2 = { argmemonly nounwind willreturn }
+

test/Transforms/LICM/guards.ll

@@ -84,7 +84,7 @@ loop:
 }
 
 ; But can hoist if the side effect is hoisted with MSSA
-define void @test2b_prime(i1 %cond, i32* %ptr) {
+define void @test2b_prime(i1 %cond, i32* noalias %ptr) {
 ; MSSA-LABEL: @test2b_prime(
 ; MSSA-NEXT:  entry:
 ; MSSA-NEXT:    [[P2:%.*]] = getelementptr i32, i32* [[PTR:%.*]], i32 1