[PM] Introduce options to enable the (still experimental) new pass

manager, and a code path to use it.

The option is actually a top-level option but does contain
'experimental' in the name. This is the compromise suggested by Richard
in discussions. We expect this option will be around long enough and
have enough users towards the end that it merits not being relegated to
CC1, but it still needs to be clear that this option will go away at
some point.

The backend code is a fresh codepath dedicated to handling the flow with
the new pass manager. This was also Richard's suggested code structuring
to essentially leave a clean path for development rather than carrying
complexity or idiosyncracies of how we do things just to share code with
the parts of this in common with the legacy pass manager. And it turns
out, not much is really in common even though we use the legacy pass
manager for codegen at this point.

I've switched a couple of tests to run with the new pass manager, and
they appear to work. There are still plenty of bugs that need squashing
(just with basic experiments I've found two already!) but they aren't in
this code, and the whole point is to expose the necessary hooks to start
experimenting with the pass manager in more realistic scenarios.

That said, I want to *strongly caution* anyone itching to play with
this: it is still *very shaky*. Several large components have not yet
been shaken down. For example I have bugs in both the always inliner and
inliner that I have already spotted and will be fixing independently.

Still, this is a fun milestone. =D

One thing not in this patch (but that might be very reasonable to add)
is some level of support for raw textual pass pipelines such as what
Sean had a patch for some time ago. I'm mostly interested in the more
traditional flow of getting the IR out of Clang and then running it
through opt, but I can see other use cases so someone may want to add
it.

And of course, *many* features are not yet supported!
- O1 is currently more like O2
- None of the sanitizers are wired up
- ObjC ARC optimizer isn't wired up
- ...

So plenty of stuff still lef to do!

Differential Revision: https://reviews.llvm.org/D28077

llvm-svn: 290450

clang/include/clang/Driver/Options.td

@@ -823,6 +823,9 @@ def finline_functions : Flag<["-"], "finline-functions">, Group<f_clang_Group>,
 def finline_hint_functions: Flag<["-"], "finline-hint-functions">, Group<f_clang_Group>, Flags<[CC1Option]>,
   HelpText<"Inline functions wich are (explicitly or implicitly) marked inline">;
 def finline : Flag<["-"], "finline">, Group<clang_ignored_f_Group>;
+def fexperimental_new_pass_manager : Flag<["-"], "fexperimental-new-pass-manager">,
+  Group<f_clang_Group>, Flags<[CC1Option]>,
+  HelpText<"Enables an experimental new pass manager in LLVM.">;
 def finput_charset_EQ : Joined<["-"], "finput-charset=">, Group<f_Group>;
 def fexec_charset_EQ : Joined<["-"], "fexec-charset=">, Group<f_Group>;
 def finstrument_functions : Flag<["-"], "finstrument-functions">, Group<f_Group>, Flags<[CC1Option]>,
@@ -1000,6 +1003,9 @@ def fno_exceptions : Flag<["-"], "fno-exceptions">, Group<f_Group>;
 def fno_gnu_keywords : Flag<["-"], "fno-gnu-keywords">, Group<f_Group>, Flags<[CC1Option]>;
 def fno_inline_functions : Flag<["-"], "fno-inline-functions">, Group<f_clang_Group>, Flags<[CC1Option]>;
 def fno_inline : Flag<["-"], "fno-inline">, Group<f_clang_Group>, Flags<[CC1Option]>;
+def fno_experimental_new_pass_manager : Flag<["-"], "fno-experimental-new-pass-manager">,
+  Group<f_clang_Group>, Flags<[CC1Option]>,
+  HelpText<"Disables an experimental new pass manager in LLVM.">;
 def fveclib : Joined<["-"], "fveclib=">, Group<f_Group>, Flags<[CC1Option]>,
     HelpText<"Use the given vector functions library">;
 def fno_lax_vector_conversions : Flag<["-"], "fno-lax-vector-conversions">, Group<f_Group>,

clang/include/clang/Frontend/CodeGenOptions.def

@@ -52,6 +52,8 @@ CODEGENOPT(DisableGCov       , 1, 0) ///< Don't run the GCov pass, for testing.
 CODEGENOPT(DisableLLVMPasses , 1, 0) ///< Don't run any LLVM IR passes to get
                                      ///< the pristine IR generated by the
                                      ///< frontend.
+CODEGENOPT(ExperimentalNewPassManager, 1, 0) ///< Enables the new, experimental
+                                             ///< pass manager.
 CODEGENOPT(DisableRedZone    , 1, 0) ///< Set when -mno-red-zone is enabled.
 CODEGENOPT(DisableTailCalls  , 1, 0) ///< Do not emit tail calls.
 CODEGENOPT(EmitDeclMetadata  , 1, 0) ///< Emit special metadata indicating what

clang/lib/CodeGen/BackendUtil.cpp

@@ -20,20 +20,21 @@
 #include "llvm/ADT/Triple.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
-#include "llvm/Bitcode/BitcodeWriterPass.h"
-#include "llvm/Bitcode/BitcodeWriter.h"
 #include "llvm/Bitcode/BitcodeReader.h"
+#include "llvm/Bitcode/BitcodeWriter.h"
+#include "llvm/Bitcode/BitcodeWriterPass.h"
 #include "llvm/CodeGen/RegAllocRegistry.h"
 #include "llvm/CodeGen/SchedulerRegistry.h"
 #include "llvm/IR/DataLayout.h"
-#include "llvm/IR/ModuleSummaryIndex.h"
 #include "llvm/IR/IRPrintingPasses.h"
 #include "llvm/IR/LegacyPassManager.h"
 #include "llvm/IR/Module.h"
+#include "llvm/IR/ModuleSummaryIndex.h"
 #include "llvm/IR/Verifier.h"
 #include "llvm/LTO/LTOBackend.h"
 #include "llvm/MC/SubtargetFeature.h"
 #include "llvm/Object/ModuleSummaryIndexObjectFile.h"
+#include "llvm/Passes/PassBuilder.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/PrettyStackTrace.h"
@@ -114,6 +115,9 @@ class EmitAssemblyHelper {
 
   void EmitAssembly(BackendAction Action,
                     std::unique_ptr<raw_pwrite_stream> OS);
+
+  void EmitAssemblyWithNewPassManager(BackendAction Action,
+                                      std::unique_ptr<raw_pwrite_stream> OS);
 };
 
 // We need this wrapper to access LangOpts and CGOpts from extension functions
@@ -709,6 +713,134 @@ void EmitAssemblyHelper::EmitAssembly(BackendAction Action,
   }
 }
 
+static PassBuilder::OptimizationLevel mapToLevel(const CodeGenOptions &Opts) {
+  switch (Opts.OptimizationLevel) {
+  default:
+    llvm_unreachable("Invalid optimization level!");
+
+  case 1:
+    return PassBuilder::O1;
+
+  case 2:
+    switch (Opts.OptimizeSize) {
+    default:
+      llvm_unreachable("Invalide optimization level for size!");
+
+    case 0:
+      return PassBuilder::O2;
+
+    case 1:
+      return PassBuilder::Os;
+
+    case 2:
+      return PassBuilder::Oz;
+    }
+
+  case 3:
+    return PassBuilder::O3;
+  }
+}
+
+/// A clean version of `EmitAssembly` that uses the new pass manager.
+///
+/// Not all features are currently supported in this system, but where
+/// necessary it falls back to the legacy pass manager to at least provide
+/// basic functionality.
+///
+/// This API is planned to have its functionality finished and then to replace
+/// `EmitAssembly` at some point in the future when the default switches.
+void EmitAssemblyHelper::EmitAssemblyWithNewPassManager(
+    BackendAction Action, std::unique_ptr<raw_pwrite_stream> OS) {
+  TimeRegion Region(llvm::TimePassesIsEnabled ? &CodeGenerationTime : nullptr);
+  setCommandLineOpts();
+
+  // The new pass manager always makes a target machine available to passes
+  // during construction.
+  CreateTargetMachine(/*MustCreateTM*/ true);
+  if (!TM)
+    // This will already be diagnosed, just bail.
+    return;
+  TheModule->setDataLayout(TM->createDataLayout());
+
+  PassBuilder PB(TM.get());
+
+  LoopAnalysisManager LAM;
+  FunctionAnalysisManager FAM;
+  CGSCCAnalysisManager CGAM;
+  ModuleAnalysisManager MAM;
+
+  // Register the AA manager first so that our version is the one used.
+  FAM.registerPass([&] { return PB.buildDefaultAAPipeline(); });
+
+  // Register all the basic analyses with the managers.
+  PB.registerModuleAnalyses(MAM);
+  PB.registerCGSCCAnalyses(CGAM);
+  PB.registerFunctionAnalyses(FAM);
+  PB.registerLoopAnalyses(LAM);
+  PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
+
+  ModulePassManager MPM;
+  if (CodeGenOpts.OptimizationLevel == 0) {
+    // Build a minimal pipeline based on the semantics required by Clang, which
+    // is just that always inlining occurs.
+    MPM.addPass(AlwaysInlinerPass());
+  } else {
+    // Otherwise, use the default pass pipeline. We also have to map our
+    // optimization levels into one of the distinct levels used to configure
+    // the pipeline.
+    PassBuilder::OptimizationLevel Level = mapToLevel(CodeGenOpts);
+
+    MPM = PB.buildPerModuleDefaultPipeline(Level);
+  }
+
+  // FIXME: We still use the legacy pass manager to do code generation. We
+  // create that pass manager here and use it as needed below.
+  legacy::PassManager CodeGenPasses;
+  bool NeedCodeGen = false;
+
+  // Append any output we need to the pass manager.
+  switch (Action) {
+  case Backend_EmitNothing:
+    break;
+
+  case Backend_EmitBC:
+    MPM.addPass(BitcodeWriterPass(*OS, CodeGenOpts.EmitLLVMUseLists,
+                                  CodeGenOpts.EmitSummaryIndex,
+                                  CodeGenOpts.EmitSummaryIndex));
+    break;
+
+  case Backend_EmitLL:
+    MPM.addPass(PrintModulePass(*OS, "", CodeGenOpts.EmitLLVMUseLists));
+    break;
+
+  case Backend_EmitAssembly:
+  case Backend_EmitMCNull:
+  case Backend_EmitObj:
+    NeedCodeGen = true;
+    CodeGenPasses.add(
+        createTargetTransformInfoWrapperPass(getTargetIRAnalysis()));
+    if (!AddEmitPasses(CodeGenPasses, Action, *OS))
+      // FIXME: Should we handle this error differently?
+      return;
+    break;
+  }
+
+  // Before executing passes, print the final values of the LLVM options.
+  cl::PrintOptionValues();
+
+  // Now that we have all of the passes ready, run them.
+  {
+    PrettyStackTraceString CrashInfo("Optimizer");
+    MPM.run(*TheModule, MAM);
+  }
+
+  // Now if needed, run the legacy PM for codegen.
+  if (NeedCodeGen) {
+    PrettyStackTraceString CrashInfo("Code generation");
+    CodeGenPasses.run(*TheModule);
+  }
+}
+
 static void runThinLTOBackend(const CodeGenOptions &CGOpts, Module *M,
                               std::unique_ptr<raw_pwrite_stream> OS) {
   // If we are performing a ThinLTO importing compile, load the function index
@@ -801,7 +933,10 @@ void clang::EmitBackendOutput(DiagnosticsEngine &Diags,
 
   EmitAssemblyHelper AsmHelper(Diags, CGOpts, TOpts, LOpts, M);
 
-  AsmHelper.EmitAssembly(Action, std::move(OS));
+  if (CGOpts.ExperimentalNewPassManager)
+    AsmHelper.EmitAssemblyWithNewPassManager(Action, std::move(OS));
+  else
+    AsmHelper.EmitAssembly(Action, std::move(OS));
 
   // Verify clang's TargetInfo DataLayout against the LLVM TargetMachine's
   // DataLayout.

clang/lib/Driver/Tools.cpp

@@ -5909,6 +5909,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                                        options::OPT_fno_inline_functions))
     InlineArg->render(Args, CmdArgs);
 
+  Args.AddLastArg(CmdArgs, options::OPT_fexperimental_new_pass_manager,
+                  options::OPT_fno_experimental_new_pass_manager);
+
   ObjCRuntime objcRuntime = AddObjCRuntimeArgs(Args, CmdArgs, rewriteKind);
 
   // -fobjc-dispatch-method is only relevant with the nonfragile-abi, and

clang/lib/Frontend/CompilerInvocation.cpp

@@ -462,6 +462,10 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
     }
   }
 
+  Opts.ExperimentalNewPassManager = Args.hasFlag(
+      OPT_fexperimental_new_pass_manager, OPT_fno_experimental_new_pass_manager,
+      /* Default */ false);
+
   if (Arg *A = Args.getLastArg(OPT_fveclib)) {
     StringRef Name = A->getValue();
     if (Name == "Accelerate")

clang/test/CodeGen/arm64_crypto.c

@@ -1,4 +1,5 @@
 // RUN: %clang_cc1 -triple arm64-apple-ios7.0 -target-feature +neon -target-feature +crypto -ffreestanding -Os -S -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple arm64-apple-ios7.0 -target-feature +neon -target-feature +crypto -ffreestanding -fexperimental-new-pass-manager -Os -S -o - %s | FileCheck %s
 // REQUIRES: aarch64-registered-target
 
 #include <arm_neon.h>

clang/test/CodeGen/inline-optim.c

@@ -1,9 +1,13 @@
 // Make sure -finline-functions family flags are behaving correctly.
 
 // RUN: %clang_cc1 -triple i686-pc-win32 -emit-llvm %s -o - | FileCheck -check-prefix=NOINLINE %s
+// RUN: %clang_cc1 -triple i686-pc-win32 -fexperimental-new-pass-manager -emit-llvm %s -o - | FileCheck -check-prefix=NOINLINE %s
 // RUN: %clang_cc1 -triple i686-pc-win32 -O3 -fno-inline-functions -emit-llvm %s -o - | FileCheck -check-prefix=NOINLINE %s
+// RUN: %clang_cc1 -triple i686-pc-win32 -fexperimental-new-pass-manager -O3 -fno-inline-functions -emit-llvm %s -o - | FileCheck -check-prefix=NOINLINE %s
 // RUN: %clang_cc1 -triple i686-pc-win32 -O3 -finline-hint-functions -emit-llvm %s -o - | FileCheck -check-prefix=HINT %s
+// RUN: %clang_cc1 -triple i686-pc-win32 -fexperimental-new-pass-manager -O3 -finline-hint-functions -emit-llvm %s -o - | FileCheck -check-prefix=HINT %s
 // RUN: %clang_cc1 -triple i686-pc-win32 -O3 -finline-functions -emit-llvm %s -o - | FileCheck -check-prefix=INLINE %s
+// RUN: %clang_cc1 -triple i686-pc-win32 -fexperimental-new-pass-manager -O3 -finline-functions -emit-llvm %s -o - | FileCheck -check-prefix=INLINE %s
 
 inline int inline_hint(int a, int b) { return(a+b); }
 

clang/test/Driver/clang_f_opts.c

@@ -469,3 +469,11 @@
 // CHECK-WCHAR2: -fshort-wchar
 // CHECK-WCHAR2-NOT: -fno-short-wchar
 // DELIMITERS: {{^ *"}}
+
+// RUN: %clang -### -fno-experimental-new-pass-manager -fexperimental-new-pass-manager %s 2>&1 | FileCheck --check-prefix=CHECK-PM --check-prefix=CHECK-NEW-PM %s
+// RUN: %clang -### -fexperimental-new-pass-manager -fno-experimental-new-pass-manager %s 2>&1 | FileCheck --check-prefix=CHECK-PM --check-prefix=CHECK-NO-NEW-PM %s
+// CHECK-PM-NOT: argument unused
+// CHECK-NEW-PM: -fexperimental-new-pass-manager
+// CHECK-NEW-PM-NOT: -fno-experimental-new-pass-manager
+// CHECK-NO-NEW-PM: -fno-experimental-new-pass-manager
+// CHECK-NO-NEW-PM-NOT: -fexperimental-new-pass-manager