[clang] Add optional pass to remove UBSAN traps using PGO (#84214)

With #83471 it reduces UBSAN overhead from 44% to 6%.
Measured as "Geomean difference" on "test-suite/MultiSource/Benchmarks"
with PGO build.

On real large server binary we see 95% of code is still instrumented,
with 10% -> 1.5% UBSAN overhead improvements. We can pass this test only
with subset of UBSAN, so base overhead is smaller.

We have followup patches to improve it even further.

1 files changed, 21 insertions, 0 deletions

diff --git a/clang/lib/CodeGen/BackendUtil.cpp b/clang/lib/CodeGen/BackendUtil.cpp
index 7310e38..82b30b8 100644
--- a/clang/lib/CodeGen/BackendUtil.cpp
+++ b/clang/lib/CodeGen/BackendUtil.cpp
@@ -76,6 +76,7 @@
 #include "llvm/Transforms/Instrumentation/MemProfiler.h"
 #include "llvm/Transforms/Instrumentation/MemorySanitizer.h"
 #include "llvm/Transforms/Instrumentation/PGOInstrumentation.h"
+#include "llvm/Transforms/Instrumentation/RemoveTrapsPass.h"
 #include "llvm/Transforms/Instrumentation/SanitizerBinaryMetadata.h"
 #include "llvm/Transforms/Instrumentation/SanitizerCoverage.h"
 #include "llvm/Transforms/Instrumentation/ThreadSanitizer.h"
@@ -83,6 +84,7 @@
 #include "llvm/Transforms/Scalar/EarlyCSE.h"
 #include "llvm/Transforms/Scalar/GVN.h"
 #include "llvm/Transforms/Scalar/JumpThreading.h"
+#include "llvm/Transforms/Scalar/SimplifyCFG.h"
 #include "llvm/Transforms/Utils/Debugify.h"
 #include "llvm/Transforms/Utils/EntryExitInstrumenter.h"
 #include "llvm/Transforms/Utils/ModuleUtils.h"
@@ -98,6 +100,10 @@ using namespace llvm;
 namespace llvm {
 extern cl::opt<bool> PrintPipelinePasses;
 
+cl::opt<bool> ClRemoveTraps("clang-remove-traps", cl::Optional,
+                            cl::desc("Insert remove-traps pass."),
+                            cl::init(false));
+
 // Experiment to move sanitizers earlier.
 static cl::opt<bool> ClSanitizeOnOptimizerEarlyEP(
     "sanitizer-early-opt-ep", cl::Optional,
@@ -744,6 +750,21 @@ static void addSanitizers(const Triple &TargetTriple,
     // LastEP does not need GlobalsAA.
     PB.registerOptimizerLastEPCallback(SanitizersCallback);
   }
+
+  if (ClRemoveTraps) {
+    // We can optimize after inliner, and PGO profile matching. The hook below
+    // is called at the end `buildFunctionSimplificationPipeline`, which called
+    // from `buildInlinerPipeline`, which called after profile matching.
+    PB.registerScalarOptimizerLateEPCallback(
+        [](FunctionPassManager &FPM, OptimizationLevel Level) {
+          // RemoveTrapsPass expects trap blocks preceded by conditional
+          // branches, which usually is not the case without SimplifyCFG.
+          // TODO: Remove `SimplifyCFGPass` after switching to dedicated
+          // intrinsic.
+          FPM.addPass(SimplifyCFGPass());
+          FPM.addPass(RemoveTrapsPass());
+        });
+  }
 }
 
 void EmitAssemblyHelper::RunOptimizationPipeline(