Re-land "[PDB] Merge types in parallel when using ghashing"

Stored Error objects have to be checked, even if they are success
values.

This reverts commit 8d250ac.
Relands commit 49b3459..

Original commit message:
-----------------------------------------

This makes type merging much faster (-24% on chrome.dll) when multiple
threads are available, but it slightly increases the time to link (+10%)
when /threads:1 is passed. With only one more thread, the new type
merging is faster (-11%). The output PDB should be identical to what it
was before this change.

To give an idea, here is the /time output placed side by side:
                              BEFORE    | AFTER
  Input File Reading:           956 ms  |  968 ms
  Code Layout:                  258 ms  |  190 ms
  Commit Output File:             6 ms  |    7 ms
  PDB Emission (Cumulative):   6691 ms  | 4253 ms
    Add Objects:               4341 ms  | 2927 ms
      Type Merging:            2814 ms  | 1269 ms  -55%!
      Symbol Merging:          1509 ms  | 1645 ms
    Publics Stream Layout:      111 ms  |  112 ms
    TPI Stream Layout:          764 ms  |   26 ms  trivial
    Commit to Disk:            1322 ms  | 1036 ms  -300ms
----------------------------------------- --------
Total Link Time:               8416 ms    5882 ms  -30% overall

The main source of the additional overhead in the single-threaded case
is the need to iterate all .debug$T sections up front to check which
type records should go in the IPI stream. See fillIsItemIndexFromDebugT.
With changes to the .debug$H section, we could pre-calculate this info
and eliminate the need to do this walk up front. That should restore
single-threaded performance back to what it was before this change.

This change will cause LLD to be much more parallel than it used to, and
for users who do multiple links in parallel, it could regress
performance. However, when the user is only doing one link, it's a huge
improvement. In the future, we can use NT worker threads to avoid
oversaturating the machine with work, but for now, this is such an
improvement for the single-link use case that I think we should land
this as is.

Algorithm
----------

Before this change, we essentially used a
DenseMap<GloballyHashedType, TypeIndex> to check if a type has already
been seen, and if it hasn't been seen, insert it now and use the next
available type index for it in the destination type stream. DenseMap
does not support concurrent insertion, and even if it did, the linker
must be deterministic: it cannot produce different PDBs by using
different numbers of threads. The output type stream must be in the same
order regardless of the order of hash table insertions.

In order to create a hash table that supports concurrent insertion, the
table cells must be small enough that they can be updated atomically.
The algorithm I used for updating the table using linear probing is
described in this paper, "Concurrent Hash Tables: Fast and General(?)!":
https://dl.acm.org/doi/10.1145/3309206

The GHashCell in this change is essentially a pair of 32-bit integer
indices: <sourceIndex, typeIndex>. The sourceIndex is the index of the
TpiSource object, and it represents an input type stream. The typeIndex
is the index of the type in the stream. Together, we have something like
a ragged 2D array of ghashes, which can be looked up as:
  tpiSources[tpiSrcIndex]->ghashes[typeIndex]

By using these side tables, we can omit the key data from the hash
table, and keep the table cell small. There is a cost to this: resolving
hash table collisions requires many more loads than simply looking at
the key in the same cache line as the insertion position. However, most
supported platforms should have a 64-bit CAS operation to update the
cell atomically.

To make the result of concurrent insertion deterministic, the cell
payloads must have a priority function. Defining one is pretty
straightforward: compare the two 32-bit numbers as a combined 64-bit
number. This means that types coming from inputs earlier on the command
line have a higher priority and are more likely to appear earlier in the
final PDB type stream than types from an input appearing later on the
link line.

After table insertion, the non-empty cells in the table can be copied
out of the main table and sorted by priority to determine the ordering
of the final type index stream. At this point, item and type records
must be separated, either by sorting or by splitting into two arrays,
and I chose sorting. This is why the GHashCell must contain the isItem
bit.

Once the final PDB TPI stream ordering is known, we need to compute a
mapping from source type index to PDB type index. To avoid starting over
from scratch and looking up every type again by its ghash, we save the
insertion position of every hash table insertion during the first
insertion phase. Because the table does not support rehashing, the
insertion position is stable. Using the array of insertion positions
indexed by source type index, we can replace the source type indices in
the ghash table cells with the PDB type indices.

Once the table cells have been updated to contain PDB type indices, the
mapping for each type source can be computed in parallel. Simply iterate
the list of cell positions and replace them with the PDB type index,
since the insertion positions are no longer needed.

Once we have a source to destination type index mapping for every type
source, there are no more data dependencies. We know which type records
are "unique" (not duplicates), and what their final type indices will
be. We can do the remapping in parallel, and accumulate type sizes and
type hashes in parallel by type source.

Lastly, TPI stream layout must be done serially. Accumulate all the type
records, sizes, and hashes, and add them to the PDB.

Differential Revision: https://reviews.llvm.org/D87805

Author: rnk

lld/COFF/DebugTypes.cpp

@@ -10,32 +10,37 @@
 #define LLD_COFF_DEBUGTYPES_H
 
 #include "lld/Common/LLVM.h"
-#include "llvm/DebugInfo/CodeView/TypeIndex.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/DebugInfo/CodeView/TypeIndexDiscovery.h"
+#include "llvm/DebugInfo/CodeView/TypeRecord.h"
 #include "llvm/Support/Error.h"
 #include "llvm/Support/MemoryBuffer.h"
 
 namespace llvm {
 namespace codeview {
-class PrecompRecord;
-class TypeServer2Record;
+struct GloballyHashedType;
 } // namespace codeview
 namespace pdb {
 class NativeSession;
+class TpiStream;
 }
 } // namespace llvm
 
 namespace lld {
 namespace coff {
 
+using llvm::codeview::GloballyHashedType;
 using llvm::codeview::TypeIndex;
 
 class ObjFile;
 class PDBInputFile;
 class TypeMerger;
+struct GHashState;
 
 class TpiSource {
 public:
-  enum TpiKind { Regular, PCH, UsingPCH, PDB, PDBIpi, UsingPDB };
+  enum TpiKind : uint8_t { Regular, PCH, UsingPCH, PDB, PDBIpi, UsingPDB };
 
   TpiSource(TpiKind k, ObjFile *f);
   virtual ~TpiSource();
@@ -53,21 +58,97 @@ class TpiSource {
   /// caller-provided ObjectIndexMap.
   virtual Error mergeDebugT(TypeMerger *m);
 
+  /// Load global hashes, either by hashing types directly, or by loading them
+  /// from LLVM's .debug$H section.
+  virtual void loadGHashes();
+
+  /// Use global hashes to merge type information.
+  virtual void remapTpiWithGHashes(GHashState *g);
+
+  // Remap a type index in place.
+  bool remapTypeIndex(TypeIndex &ti, llvm::codeview::TiRefKind refKind) const;
+
+protected:
+  void remapRecord(MutableArrayRef<uint8_t> rec,
+                   ArrayRef<llvm::codeview::TiReference> typeRefs);
+
+  void mergeTypeRecord(llvm::codeview::CVType ty);
+
+  // Merge the type records listed in uniqueTypes. beginIndex is the TypeIndex
+  // of the first record in this source, typically 0x1000. When PCHs are
+  // involved, it may start higher.
+  void mergeUniqueTypeRecords(
+      ArrayRef<uint8_t> debugTypes,
+      TypeIndex beginIndex = TypeIndex(TypeIndex::FirstNonSimpleIndex));
+
+  // Use the ghash table to construct a map from source type index to
+  // destination PDB type index. Usable for either TPI or IPI.
+  void fillMapFromGHashes(GHashState *m,
+                          llvm::SmallVectorImpl<TypeIndex> &indexMap);
+
+  // Copies ghashes from a vector into an array. These are long lived, so it's
+  // worth the time to copy these into an appropriately sized vector to reduce
+  // memory usage.
+  void assignGHashesFromVector(std::vector<GloballyHashedType> &&hashVec);
+
+  // Walk over file->debugTypes and fill in the isItemIndex bit vector.
+  void fillIsItemIndexFromDebugT();
+
+public:
+  bool remapTypesInSymbolRecord(MutableArrayRef<uint8_t> rec);
+
+  void remapTypesInTypeRecord(MutableArrayRef<uint8_t> rec);
+
   /// Is this a dependent file that needs to be processed first, before other
   /// OBJs?
   virtual bool isDependency() const { return false; }
 
-  static void forEachSource(llvm::function_ref<void(TpiSource *)> fn);
+  /// Returns true if this type record should be omitted from the PDB, even if
+  /// it is unique. This prevents a record from being added to the input ghash
+  /// table.
+  bool shouldOmitFromPdb(uint32_t ghashIdx) {
+    return ghashIdx == endPrecompGHashIdx;
+  }
+
+  /// All sources of type information in the program.
+  static std::vector<TpiSource *> instances;
+
+  /// Dependency type sources, such as type servers or PCH object files. These
+  /// must be processed before objects that rely on them. Set by
+  /// TpiSources::sortDependencies.
+  static ArrayRef<TpiSource *> dependencySources;
+
+  /// Object file sources. These must be processed after dependencySources.
+  static ArrayRef<TpiSource *> objectSources;
+
+  /// Sorts the dependencies and reassigns TpiSource indices.
+  static void sortDependencies();
 
   static uint32_t countTypeServerPDBs();
   static uint32_t countPrecompObjs();
 
+  /// Free heap allocated ghashes.
+  static void clearGHashes();
+
   /// Clear global data structures for TpiSources.
   static void clear();
 
   const TpiKind kind;
+  bool ownedGHashes = true;
+  uint32_t tpiSrcIdx = 0;
+
+protected:
+  /// The ghash index (zero based, not 0x1000-based) of the LF_ENDPRECOMP record
+  /// in this object, if one exists. This is the all ones value otherwise. It is
+  /// recorded here so that it can be omitted from the final ghash table.
+  uint32_t endPrecompGHashIdx = ~0U;
+
+public:
   ObjFile *file;
 
+  /// An error encountered during type merging, if any.
+  Error typeMergingError = Error::success();
+
   // Storage for tpiMap or ipiMap, depending on the kind of source.
   llvm::SmallVector<TypeIndex, 0> indexMapStorage;
 
@@ -76,6 +157,31 @@ class TpiSource {
   // objects.
   llvm::ArrayRef<TypeIndex> tpiMap;
   llvm::ArrayRef<TypeIndex> ipiMap;
+
+  /// Array of global type hashes, indexed by TypeIndex. May be calculated on
+  /// demand, or present in input object files.
+  llvm::ArrayRef<llvm::codeview::GloballyHashedType> ghashes;
+
+  /// When ghashing is used, record the mapping from LF_[M]FUNC_ID to function
+  /// type index here. Both indices are PDB indices, not object type indexes.
+  llvm::DenseMap<TypeIndex, TypeIndex> funcIdToType;
+
+  /// Indicates if a type record is an item index or a type index.
+  llvm::BitVector isItemIndex;
+
+  /// A list of all "unique" type indices which must be merged into the final
+  /// PDB. GHash type deduplication produces this list, and it should be
+  /// considerably smaller than the input.
+  std::vector<uint32_t> uniqueTypes;
+
+  struct MergedInfo {
+    std::vector<uint8_t> recs;
+    std::vector<uint16_t> recSizes;
+    std::vector<uint32_t> recHashes;
+  };
+
+  MergedInfo mergedTpi;
+  MergedInfo mergedIpi;
 };
 
 TpiSource *makeTpiSource(ObjFile *file);

lld/COFF/DebugTypes.h

@@ -69,13 +69,13 @@ bool link(ArrayRef<const char *> args, bool canExitEarly, raw_ostream &stdoutOS,
   lld::stderrOS = &stderrOS;
 
   errorHandler().cleanupCallback = []() {
+    TpiSource::clear();
     freeArena();
     ObjFile::instances.clear();
     PDBInputFile::instances.clear();
     ImportFile::instances.clear();
     BitcodeFile::instances.clear();
     memset(MergeChunk::instances, 0, sizeof(MergeChunk::instances));
-    TpiSource::clear();
     OutputSection::clear();
   };