[Clang] Enable -print-pipeline-passes in clang.

[arsenm]

I have no idea how you're supposed to work with clang without this

[jcranmer-intel]

Rebase to trunk.

[aaron.ballman]

Adding codegen code owners as reviewers just so they're aware (I think you're fine to land these changes if you don't hear back in the next day or two).

[aeubanks]

clang/lib/CodeGen/BackendUtil.cpp

1102

I wouldn't return here, doesn't seem right that we'll skip running the opt pipeline but continue with compilation. we should either bail out entirely of producing any output files (which would probably require code changes elsewhere), or run everything as normal, not do something weird where we don't run the optimization pipeline but still output files

clang/test/CodeGen/print-pipeline-passes.c

9

I wouldn't test BitcodeWriterPass, we may have a proper textual name for it at some point

[jcranmer-intel]

clang/lib/CodeGen/BackendUtil.cpp

1102

This is basically doing the same thing that opt -print-pipeline-passes is doing: https://github.com/llvm/llvm-project/blob/d1b418f55263ec48d14f220ad020d55f126cfcab/llvm/tools/opt/NewPMDriver.cpp#L500-L524.

In the case of emitting LLVM IR or bitcode, this logic is sufficient to not emit any output. In the case of .s or .o files, it looks like I have to also bail out of running RunCodegenPipeline

[jcranmer-intel]

Address review comments.

[aeubanks]

lg with one test comment

clang/lib/CodeGen/BackendUtil.cpp

1102

if skipping RunCodegenPipeline doesn't produce any output file, sgtm

clang/test/CodeGen/print-pipeline-passes.c

9

I believe this will fail in non-assert builds

9

since we don't run the verifier in non-assert builds

[efriedma]

I don't really like extern cl::opt<bool> PrintPipelinePasses; (as opposed to implementing a proper driver option that calls a proper API); secret handshakes like this make it harder for people to write non-clang frontends. But I won't block the patch just for that, I guess.

[jcranmer-intel]

Replace verify with annotation-remarks

[jcranmer-intel]

clang/test/CodeGen/print-pipeline-passes.c

9

Apparently verify is added by clang unless -disable-llvm-verifier is present on the command line, but I've switched it to annotation-remarks nevertheless.

[aeubanks]

clang/test/CodeGen/print-pipeline-passes.c

9

-disable-llvm-verifier is added by clang in non-assert builds

clang/lib/CodeGen/BackendUtil.cpp

 using namespace llvm;
 
 #define HANDLE_EXTENSION(Ext)                                                  \
   llvm::PassPluginLibraryInfo get##Ext##PluginInfo();
 #include "llvm/Support/Extension.def"
 
 namespace llvm {
 extern cl::opt<bool> DebugInfoCorrelate;
+extern cl::opt<bool> PrintPipelinePasses;
 
 // Experiment to move sanitizers earlier.
 static cl::opt<bool> ClSanitizeOnOptimizerEarlyEP(
     "sanitizer-early-opt-ep", cl::Optional,
     cl::desc("Insert sanitizers on OptimizerEarlyEP."), cl::init(false));
 }
 
 namespace {
if (!TheModule->getModuleFlag("ThinLTO"))
                                uint32_t(CodeGenOpts.PrepareForThinLTO));
     if (!TheModule->getModuleFlag("EnableSplitLTOUnit"))
       TheModule->addModuleFlag(Module::Error, "EnableSplitLTOUnit",
                                uint32_t(CodeGenOpts.EnableSplitLTOUnit));
     if (CodeGenOpts.UnifiedLTO && !TheModule->getModuleFlag("UnifiedLTO"))
       TheModule->addModuleFlag(Module::Error, "UnifiedLTO", uint32_t(1));
   }
 
+  // Print a textual, '-passes=' compatible, representation of pipeline if
+  // requested.
+  if (PrintPipelinePasses) {
+    MPM.printPipeline(outs(), [&PIC](StringRef ClassName) {
+      auto PassName = PIC.getPassNameForClassName(ClassName);
+      return PassName.empty() ? ClassName : PassName;
+    });
+    outs() << "\n";
+    return;
+  }
+
   // Now that we have all of the passes ready, run them.
   {
     PrettyStackTraceString CrashInfo("Optimizer");
     llvm::TimeTraceScope TimeScope("Optimizer");
     MPM.run(*TheModule, MAM);
   }
 }
 
if (!AddEmitPasses(CodeGenPasses, Action, *OS,
                        DwoOS ? &DwoOS->os() : nullptr))
       // FIXME: Should we handle this error differently?
       return;
     break;
   default:
     return;
   }
 
+  // If -print-pipeline-passes is requested, don't run the legacy pass manager.
+  // FIXME: when codegen is switched to use the new pass manager, it should also
+  // emit pass names here.
+  if (PrintPipelinePasses) {
+    return;
+  }
+
   {
     PrettyStackTraceString CrashInfo("Code generation");
     llvm::TimeTraceScope TimeScope("CodeGenPasses");
     CodeGenPasses.run(*TheModule);
   }
 }
 
 void EmitAssemblyHelper::EmitAssembly(BackendAction Action,

clang/test/CodeGen/print-pipeline-passes.c

+// Test that -print-pipeline-passes works in Clang
+
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm-bc -o /dev/null -mllvm -print-pipeline-passes -O0 %s 2>&1 | FileCheck %s
+
+// Don't try to check all passes, just a few to make sure that something is
+// actually printed.
+// CHECK: always-inline
+// CHECK-SAME: annotation-remarks
+void Foo(void) {}