[HWASAN] Inline fast pass of instrumentMemAccessOutline

Usually pointer tag will match tag in the shadow, so we can keep
inlining this check keeping the rest in outlined part.

It imroves performance by about 25%, but increases code size by 30%.
Existing outlining reduces performance by 30%, but saves code size by 80%.
So we still significantly reduce code size with minimal performance loss.

Reviewed By: fmayer

Differential Revision: https://reviews.llvm.org/D159172

Author: vitalybuka

llvm/lib/Transforms/Instrumentation/HWAddressSanitizer.cpp

@@ -894,9 +894,8 @@ void HWAddressSanitizer::instrumentMemAccessOutline(Value *Ptr, bool IsWrite,
   assert(!UsePageAliases);
   const int64_t AccessInfo = getAccessInfo(IsWrite, AccessSizeIndex);
 
-  if (InlineFastPath) {
-    // TODO.
-  }
+  if (InlineFastPath)
+    InsertBefore = insertShadowTagCheck(Ptr, InsertBefore).TagMismatchTerm;
 
   IRBuilder<> IRB(InsertBefore);
   Module *M = IRB.GetInsertBlock()->getParent()->getParent();

llvm/test/Instrumentation/HWAddressSanitizer/RISCV/atomic.ll

@@ -1,7 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
 ; Test basic address sanitizer instrumentation.
 ;
-; RUN: opt < %s -passes=hwasan -S | FileCheck %s
+; RUN: opt < %s -passes=hwasan -hwasan-inline-fast-path-checks=0 -S | FileCheck %s
 
 target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
 target triple = "riscv64-unknown-linux"