hwasan: If we split the entry block, move static allocas back into the entry block.

Otherwise they are treated as dynamic allocas, which ends up increasing
code size significantly. This reduces size of Chromium base_unittests
by 2MB (6.7%).

Differential Revision: https://reviews.llvm.org/D57205

llvm-svn: 352152
diff --git a/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
index 6170b79..a6ffff4 100644
--- a/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
@@ -1038,6 +1038,21 @@
     Changed |= instrumentStack(AllocasToInstrument, RetVec, StackTag);
   }
 
+  // If we split the entry block, move any allocas that were originally in the
+  // entry block back into the entry block so that they aren't treated as
+  // dynamic allocas.
+  if (EntryIRB.GetInsertBlock() != &F.getEntryBlock()) {
+    InsertPt = &*F.getEntryBlock().begin();
+    for (auto II = EntryIRB.GetInsertBlock()->begin(),
+              IE = EntryIRB.GetInsertBlock()->end();
+         II != IE;) {
+      Instruction *I = &*II++;
+      if (auto *AI = dyn_cast<AllocaInst>(I))
+        if (isa<ConstantInt>(AI->getArraySize()))
+          I->moveBefore(InsertPt);
+    }
+  }
+
   for (auto Inst : ToInstrument)
     Changed |= instrumentMemAccess(Inst);
 
diff --git a/llvm/test/Instrumentation/HWAddressSanitizer/lazy-thread-init.ll b/llvm/test/Instrumentation/HWAddressSanitizer/lazy-thread-init.ll
index b02d0f8..d5a2541 100644
--- a/llvm/test/Instrumentation/HWAddressSanitizer/lazy-thread-init.ll
+++ b/llvm/test/Instrumentation/HWAddressSanitizer/lazy-thread-init.ll
@@ -2,9 +2,10 @@
 
 target triple = "aarch64--linux-android"
 
-declare void @bar([16 x i32]* %p)
+declare i32 @bar([16 x i32]* %p)
 
 define void @alloca() sanitize_hwaddress "hwasan-abi"="interceptor" {
+  ; CHECK: alloca [16 x i32]
   ; CHECK: [[A:%[^ ]*]] = call i8* @llvm.thread.pointer()
   ; CHECK: [[B:%[^ ]*]] = getelementptr i8, i8* [[A]], i32 48
   ; CHECK: [[C:%[^ ]*]] = bitcast i8* [[B]] to i64*
@@ -19,9 +20,11 @@
 
   ; CHECK: [[CONT]]:
   ; CHECK: phi i64 [ [[LOAD]], %0 ], [ [[RELOAD]], %[[INIT]] ]
+  ; CHECK: alloca i8
 
   %p = alloca [16 x i32]
-  call void @bar([16 x i32]* %p)
+  %size = call i32 @bar([16 x i32]* %p)
+  %q = alloca i8, i32 %size
   ret void
 }