Add -pass-remarks flag to 'opt'.

[dnovillo]

Summary

This adds support in 'opt' to filter pass remarks emitted by
optimization passes. A new flag -pass-remarks specifies which
passes should emit a diagnostic when LLVMContext::emitOptimizationRemark
is invoked.

This will allow the front end to simply pass along the regular
expression from its own -Rpass flag when launching the backend.

Depends on D3227.

[dnovillo]

• Make -pass-remarks additive when given multiple times.
  
  This implements the suggestion we discussed on IRC. To keep track of the patterns, I've had to introduce a separate string. Every pattern is added using parentheses and '|' to make sure that the sub-expressions are properly separated.

[qcolombet]

Hi Diego,

The question I am about to ask may be beyond the scope of this patch but is the option setting thread safe?
The implication is do we have to do something special to access and modify:

• OptimizationRemarkPattern.
• OptimizationRemarkExpr.

Thanks,
-Quentin

lib/IR/LLVMContextImpl.cpp

53

Add = NULL.
I am a bit paranoid with default initializer.

[dnovillo]

• Initialize OptimizationRemarkPattern explicitly.

include/llvm/IR/DiagnosticInfo.h

 namespace llvm {
 
 // Forward declarations.
 class DiagnosticPrinter;
 class Function;
 class Instruction;
 class Twine;
 class Value;
+class DebugLoc;
 
 /// \brief Defines the different supported severity of a diagnostic.
 enum DiagnosticSeverity {
   DS_Error,
   DS_Warning,
   DS_Remark,
   // A note attaches additional information to one of the previous diagnostic
   // types.
   DS_Note
 };
 
 /// \brief Defines the different supported kind of a diagnostic.
 /// This enum should be extended with a new ID for each added concrete subclass.
 enum DiagnosticKind {
   DK_InlineAsm,
   DK_StackSize,
   DK_DebugMetadataVersion,
   DK_SampleProfile,
+  DK_OptimizationRemark,
   DK_FirstPluginKind
 };
 
 /// \brief Get the next available kind ID for a plugin diagnostic.
 /// Each time this function is called, it returns a different number.
 /// Therefore, a plugin that wants to "identify" its own classes
 /// with a dynamic identifier, just have to use this method to get a new ID
 /// and assign it to each of its classes.
private:
   /// Line number where the diagnostic occured. If 0, no line number will
   /// be emitted in the message.
   unsigned LineNum;
 
   /// Message to report.
   const Twine &Msg;
 };
 
+/// Diagnostic information for optimization remarks.
+class DiagnosticInfoOptimizationRemark : public DiagnosticInfo {
+public:
+  /// \p PassName is the name of the pass emitting this diagnostic. If
+  /// this name matches the regular expression given in -Rpass=, then the
+  /// diagnostic will be emitted. \p Fn is the function where the diagnostic
+  /// is being emitted. \p DLoc is the location information to use in the
+  /// diagnostic. If line table information is available, the diagnostic
+  /// will include the source code location. \p Msg is the message to show.
+  /// Note that this class does not copy this message, so this reference
+  /// must be valid for the whole life time of the diagnostic.
+  DiagnosticInfoOptimizationRemark(const char *PassName, const Function &Fn,
+                                   const DebugLoc &DLoc, const Twine &Msg)
+      : DiagnosticInfo(DK_OptimizationRemark, DS_Remark), PassName(PassName),
+        Fn(Fn), DLoc(DLoc), Msg(Msg) {}
+
+  /// \see DiagnosticInfo::print.
+  void print(DiagnosticPrinter &DP) const override;
+
+  /// Hand rolled RTTI.
+  static bool classof(const DiagnosticInfo *DI) {
+    return DI->getKind() == DK_OptimizationRemark;
+  }
+
+  /// Return true if location information is available for this diagnostic.
+  bool isLocationAvailable() const;
+
+  /// Return a string with the location information for this diagnostic
+  /// in the format "file:line:col". If location information is not available,
+  /// it returns "<unknown>:0:0".
+  const StringRef getLocationStr() const;
+
+  /// Return location information for this diagnostic in three parts:
+  /// the source file name, line number and column.
+  void getLocation(StringRef *Filename, unsigned *Line, unsigned *Column) const;
+
+  const char *getPassName() const { return PassName; }
+  const Function &getFunction() const { return Fn; }
+  const DebugLoc &getDebugLoc() const { return DLoc; }
+  const Twine &getMsg() const { return Msg; }
+
+private:
+  /// Name of the pass that triggers this report. If this matches the
+  /// regular expression given in -Rpass=regexp, then the remark will
+  /// be emitted.
+  const char *PassName;
+
+  /// Function where this diagnostic is triggered.
+  const Function &Fn;
+
+  /// Debug location where this diagnostic is triggered.
+  const DebugLoc &DLoc;
+
+  /// Message to report.
+  const Twine &Msg;
+};
+
 } // End namespace llvm
 
 #endif

include/llvm/IR/LLVMContext.h

 class LLVMContextImpl;
 class StringRef;
 class Twine;
 class Instruction;
 class Module;
 class SMDiagnostic;
 class DiagnosticInfo;
 template <typename T> class SmallVectorImpl;
+class Function;
+class DebugLoc;
 
 /// This is an important class for using LLVM in a threaded context.  It
 /// (opaquely) owns and manages the core "global" data of LLVM's core
 /// infrastructure, including the type and constant uniquing tables.
 /// LLVMContext itself provides no locking guarantees, so you should be careful
 /// to have one context per thread.
 class LLVMContext {
 public:
public:
   /// with optional location information.  This function returns, so code should
   /// be prepared to drop the erroneous construct on the floor and "not crash".
   /// The generated code need not be correct.  The error message will be
   /// implicitly prefixed with "error: " and should not end with a ".".
   void emitError(unsigned LocCookie, const Twine &ErrorStr);
   void emitError(const Instruction *I, const Twine &ErrorStr);
   void emitError(const Twine &ErrorStr);
 
+  /// emitOptimizationRemark - Emit an optimization remark message. \p PassName
+  /// is the name of the pass emitting the message. If -Rpass= is given
+  /// and \p PassName matches the regular expression in -Rpass, then the
+  /// remark will be emitted. \p Fn is the function triggering the remark,
+  /// \p DLoc is the debug location where the diagnostic is generated.
+  /// \p Msg is the message string to use.
+  void emitOptimizationRemark(const char *PassName, const Function &Fn,
+                              const DebugLoc &DLoc, const Twine &Msg);
+
 private:
   LLVMContext(LLVMContext&) LLVM_DELETED_FUNCTION;
   void operator=(LLVMContext&) LLVM_DELETED_FUNCTION;
 
   /// addModule - Register a module as being instantiated in this context.  If
   /// the context is deleted, the module will be deleted as well.
   void addModule(Module*);
 

lib/IR/DiagnosticInfo.cpp

 //===- llvm/Support/DiagnosticInfo.cpp - Diagnostic Definitions -*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // This file defines the different classes involved in low level diagnostics.
 //
 // Diagnostics reporting is still done as part of the LLVMContext.
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ADT/Twine.h"
 #include "llvm/IR/Constants.h"
+#include "llvm/IR/DebugInfo.h"
 #include "llvm/IR/DiagnosticInfo.h"
 #include "llvm/IR/DiagnosticPrinter.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Instruction.h"
 #include "llvm/IR/Metadata.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Support/Atomic.h"
 #include <string>
 
 using namespace llvm;
 
 int llvm::getNextAvailablePluginDiagnosticKind() {
   static sys::cas_flag PluginKindID = DK_FirstPluginKind;
   return (int)sys::AtomicIncrement(&PluginKindID);

 
 void DiagnosticInfoSampleProfile::print(DiagnosticPrinter &DP) const {
   if (getFileName() && getLineNum() > 0)
     DP << getFileName() << ":" << getLineNum() << ": ";
   else if (getFileName())
     DP << getFileName() << ": ";
   DP << getMsg();
 }
+
+bool DiagnosticInfoOptimizationRemark::isLocationAvailable() const {
+  return getFunction().getParent()->getNamedMetadata("llvm.dbg.cu") != 0;
+}
+
+void DiagnosticInfoOptimizationRemark::getLocation(StringRef *Filename,
+                                                   unsigned *Line,
+                                                   unsigned *Column) const {
+  DILocation DIL(getDebugLoc().getAsMDNode(getFunction().getContext()));
+  *Filename = DIL.getFilename();
+  *Line = DIL.getLineNumber();
+  *Column = DIL.getColumnNumber();
+}
+
+const StringRef DiagnosticInfoOptimizationRemark::getLocationStr() const {
+  StringRef Filename("<unknown>");
+  unsigned Line = 0;
+  unsigned Column = 0;
+  if (isLocationAvailable())
+    getLocation(&Filename, &Line, &Column);
+  return Twine(Filename + ":" + Twine(Line) + ":" + Twine(Column)).str();
+}
+
+void DiagnosticInfoOptimizationRemark::print(DiagnosticPrinter &DP) const {
+  DP << getLocationStr() << ": " << getMsg();
+}

lib/IR/LLVMContext.cpp

 //  This file implements LLVMContext, as a wrapper around the opaque
 //  class LLVMContextImpl.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/IR/LLVMContext.h"
 #include "LLVMContextImpl.h"
 #include "llvm/IR/Constants.h"
+#include "llvm/IR/DebugLoc.h"
 #include "llvm/IR/DiagnosticInfo.h"
 #include "llvm/IR/DiagnosticPrinter.h"
 #include "llvm/IR/Instruction.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/Support/ManagedStatic.h"
 #include "llvm/Support/SourceMgr.h"
 #include <cctype>
 using namespace llvm;
case DS_Note:
     break;
   }
 }
 
 void LLVMContext::emitError(unsigned LocCookie, const Twine &ErrorStr) {
   diagnose(DiagnosticInfoInlineAsm(LocCookie, ErrorStr));
 }
 
+void LLVMContext::emitOptimizationRemark(const char *PassName,
+                                         const Function &Fn,
+                                         const DebugLoc &DLoc,
+                                         const Twine &Msg) {
+  if (pImpl->optimizationRemarksEnabledFor(PassName))
+    diagnose(DiagnosticInfoOptimizationRemark(PassName, Fn, DLoc, Msg));
+}
+
 //===----------------------------------------------------------------------===//
 // Metadata Kind Uniquing
 //===----------------------------------------------------------------------===//
 
 #ifndef NDEBUG
 /// isValidName - Return true if Name is a valid custom metadata handler name.
 static bool isValidName(StringRef MDName) {
   if (MDName.empty())

lib/IR/LLVMContextImpl.h

public:
   typedef DenseMap<const Function*, unsigned> IntrinsicIDCacheTy;
   IntrinsicIDCacheTy IntrinsicIDCache;
 
   /// \brief Mapping from a function to its prefix data, which is stored as the
   /// operand of an unparented ReturnInst so that the prefix data has a Use.
   typedef DenseMap<const Function *, ReturnInst *> PrefixDataMapTy;
   PrefixDataMapTy PrefixDataMap;
 
+  /// \brief Return true if the given pass name should emit optimization
+  /// remarks.
+  bool optimizationRemarksEnabledFor(const char *PassName) const;
+
   int getOrAddScopeRecordIdxEntry(MDNode *N, int ExistingIdx);
   int getOrAddScopeInlinedAtIdxEntry(MDNode *Scope, MDNode *IA,int ExistingIdx);
   
   LLVMContextImpl(LLVMContext &C);
   ~LLVMContextImpl();
 };
 
 }
 
 #endif

lib/IR/LLVMContextImpl.cpp

 //  This file implements the opaque LLVMContextImpl.
 //
 //===----------------------------------------------------------------------===//
 
 #include "LLVMContextImpl.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/IR/Attributes.h"
 #include "llvm/IR/Module.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Regex.h"
 #include <algorithm>
 using namespace llvm;
 
 LLVMContextImpl::LLVMContextImpl(LLVMContext &C)
   : TheTrueVal(0), TheFalseVal(0),
     VoidTy(C, Type::VoidTyID),
     LabelTy(C, Type::LabelTyID),
     HalfTy(C, Type::HalfTyID),
LLVMContextImpl::LLVMContextImpl(LLVMContext &C)
   InlineAsmDiagHandler = 0;
   InlineAsmDiagContext = 0;
   DiagnosticHandler = 0;
   DiagnosticContext = 0;
   NamedStructTypesUniqueID = 0;
 }
 
 namespace {
+
+/// \brief Regular expression corresponding to the value given in the
+/// command line flag -pass-remarks. Passes whose name matches this
+/// regexp will emit a diagnostic when calling
+/// LLVMContext::emitOptimizationRemark.
+static Regex *OptimizationRemarkPattern = 0;
+
+/// \brief String to hold all the values passed via -pass-remarks. Every
+/// instance of -pass-remarks on the command line will be concatenated
+/// to this string. Values are stored inside braces and concatenated with
+/// the '|' operator. This implements the expected semantics that multiple
+/// -pass-remarks are additive.
+static std::string OptimizationRemarkExpr;
+
+struct PassRemarksOpt {
+  void operator=(const std::string &Val) const {
+    // Create a regexp object to match pass names for emitOptimizationRemark.
+    if (!Val.empty()) {
+      if (!OptimizationRemarkExpr.empty())
+        OptimizationRemarkExpr += "|";
+      OptimizationRemarkExpr += "(" + Val + ")";
+      delete OptimizationRemarkPattern;
+      OptimizationRemarkPattern = new Regex(OptimizationRemarkExpr);
+      std::string RegexError;
+      if (!OptimizationRemarkPattern->isValid(RegexError))
+        report_fatal_error("Invalid regular expression '" + Val +
+                               "' in -pass-remarks: " + RegexError,
+                           false);
+    }
+  };
+};
+
+static PassRemarksOpt PassRemarksOptLoc;
+
+// -pass-remarks
+//    Command line flag to enable LLVMContext::emitOptimizationRemark()
+//    and LLVMContext::emitOptimizationNote() calls.
+static cl::opt<PassRemarksOpt, true, cl::parser<std::string>>
+PassRemarks("pass-remarks", cl::value_desc("pattern"),
+            cl::desc("Enable optimization remarks from passes whose name match "
+                     "the given regular expression"),
+            cl::Hidden, cl::location(PassRemarksOptLoc), cl::ValueRequired,
+            cl::ZeroOrMore);
+}
+
+bool
+LLVMContextImpl::optimizationRemarksEnabledFor(const char *PassName) const {
+  return OptimizationRemarkPattern &&
+         OptimizationRemarkPattern->match(PassName);
+}
+
+
+namespace {
 struct DropReferences {
   // Takes the value_type of a ConstantUniqueMap's internal map, whose 'second'
   // is a Constant*.
   template<typename PairT>
   void operator()(const PairT &P) {
     P.second->dropAllReferences();
   }
 };

lib/Transforms/IPO/Inliner.cpp

 #define DEBUG_TYPE "inline"
 #include "llvm/Transforms/IPO/InlinerPass.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/Analysis/InlineCost.h"
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DiagnosticInfo.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Cloning.h"
for (unsigned CSi = 0; CSi != CallSites.size(); ++CSi) {
         if (!shouldInline(CS))
           continue;
 
         // Attempt to inline the function.
         if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
                                   InlineHistoryID, InsertLifetime, DL))
           continue;
         ++NumInlined;
-        
+        // Report the inline decision.
+        Caller->getContext().emitOptimizationRemark(
+            DEBUG_TYPE, *Caller, CS.getInstruction()->getDebugLoc(),
+            Twine(Callee->getName() + " inlined into " + Caller->getName()));
+
         // If inlining this function gave us any new call sites, throw them
         // onto our worklist to process.  They are useful inline candidates.
         if (!InlineInfo.InlinedCalls.empty()) {
           // Create a new inline history entry for this, so that we remember
           // that these new callsites came about due to inlining Callee.
           int NewHistoryID = InlineHistory.size();
           InlineHistory.push_back(std::make_pair(Callee, InlineHistoryID));
 

test/Other/optimization-remarks-inline.ll

+; RUN: opt < %s -inline -pass-remarks='inline' -S 2>&1 | FileCheck %s
+; RUN: opt < %s -inline -pass-remarks='inl.*' -S 2>&1 | FileCheck %s
+; RUN: opt < %s -inline -pass-remarks='vector' -pass-remarks='inl' -S 2>&1 | FileCheck %s
+; RUN: opt < %s -inline -pass-remarks='vector' -S 2>&1 | FileCheck --check-prefix=REMARKS %s
+; RUN: opt < %s -inline -S 2>&1 | FileCheck --check-prefix=REMARKS %s
+; RUN: not opt < %s -pass-remarks='(' 2>&1 | FileCheck --check-prefix=BAD-REGEXP %s
+
+define i32 @foo(i32 %x, i32 %y) #0 {
+entry:
+  %x.addr = alloca i32, align 4
+  %y.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  store i32 %y, i32* %y.addr, align 4
+  %0 = load i32* %x.addr, align 4
+  %1 = load i32* %y.addr, align 4
+  %add = add nsw i32 %0, %1
+  ret i32 %add
+}
+
+define i32 @bar(i32 %j) #0 {
+entry:
+  %j.addr = alloca i32, align 4
+  store i32 %j, i32* %j.addr, align 4
+  %0 = load i32* %j.addr, align 4
+  %1 = load i32* %j.addr, align 4
+  %sub = sub nsw i32 %1, 2
+  %call = call i32 @foo(i32 %0, i32 %sub)
+; CHECK: foo inlined into bar
+; REMARKS-NOT: foo inlined into bar
+  ret i32 %call
+}
+
+; BAD-REGEXP: Invalid regular expression '(' in -pass-remarks: