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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@352152 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp b/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
index 6170b792a68999f0b0b707dc57a25f4354605909..a6ffff418db959bdbfb48db0625abf7262127e28 100644 (file)
--- a/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
+++ b/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp
@@ -1038,6 +1038,21 @@ bool HWAddressSanitizer::runOnFunction(Function &F) {
     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/test/Instrumentation/HWAddressSanitizer/lazy-thread-init.ll b/test/Instrumentation/HWAddressSanitizer/lazy-thread-init.ll
index b02d0f88a0f8c6c943a94aa1d4a4f19869c6d8ab..d5a2541eb7763505cdb66c51f62339ad707b5354 100644 (file)
--- a/test/Instrumentation/HWAddressSanitizer/lazy-thread-init.ll
+++ b/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 @@ define void @alloca() sanitize_hwaddress "hwasan-abi"="interceptor" {
 
   ; 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
 }