[llvm] [ctx_profile] Profile reader and writer (PR #91859)

[Snehasish Kumar via llvm-commits]

================
@@ -0,0 +1,167 @@
+//===- PGOCtxProfReader.cpp - Contextual Instrumentation profile reader ---===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Read a contextual profile into a datastructure suitable for maintenance
+// throughout IPO
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ProfileData/PGOCtxProfReader.h"
+#include "llvm/Bitstream/BitCodeEnums.h"
+#include "llvm/Bitstream/BitstreamReader.h"
+#include "llvm/ProfileData/InstrProf.h"
+#include "llvm/ProfileData/PGOCtxProfWriter.h"
+#include "llvm/Support/Errc.h"
+#include "llvm/Support/Error.h"
+
+using namespace llvm;
+
+// FIXME(#92054) - these Error handling macros are (re-)invented in a few
+// places.
+#define EXPECT_OR_RET(LHS, RHS)                                                \
+  auto LHS = RHS;                                                              \
+  if (!LHS)                                                                    \
+    return LHS.takeError();
+
+#define RET_ON_ERR(EXPR)                                                       \
+  if (auto Err = (EXPR))                                                       \
+    return Err;
+
+Expected<PGOContextualProfile &>
+PGOContextualProfile::getOrEmplace(uint32_t Index, GlobalValue::GUID G,
+                                   SmallVectorImpl<uint64_t> &&Counters) {
+  auto I = Callsites[Index].insert(
+      {G, PGOContextualProfile(G, std::move(Counters))});
+  if (!I.second)
+    return make_error<StringError>(llvm::errc::invalid_argument,
+                                   "Duplicate GUID for same callsite.");
+  return I.first->second;
+}
+
+void PGOContextualProfile::getContainedGuids(
+    DenseSet<GlobalValue::GUID> &Guids) const {
+  Guids.insert(GUID);
+  for (const auto &[_, Callsite] : Callsites)
+    for (const auto &[_, Callee] : Callsite)
+      Callee.getContainedGuids(Guids);
+}
+
+Expected<BitstreamEntry> PGOCtxProfileReader::advance() {
+  return Cursor.advance(BitstreamCursor::AF_DontAutoprocessAbbrevs);
+}
+
+Error PGOCtxProfileReader::wrongValue(const Twine &Msg) {
+  return make_error<InstrProfError>(instrprof_error::invalid_prof, Msg);
+}
+
+Error PGOCtxProfileReader::unsupported(const Twine &Msg) {
+  return make_error<InstrProfError>(instrprof_error::unsupported_version, Msg);
+}
+
+bool PGOCtxProfileReader::canReadContext() {
+  auto Blk = advance();
+  if (!Blk) {
+    consumeError(Blk.takeError());
+    return false;
+  }
+  return Blk->Kind == BitstreamEntry::SubBlock &&
+         Blk->ID == PGOCtxProfileBlockIDs::ContextNodeBlockID;
+}
+
+Expected<std::pair<std::optional<uint32_t>, PGOContextualProfile>>
+PGOCtxProfileReader::readContext(bool ExpectIndex) {
+  RET_ON_ERR(Cursor.EnterSubBlock(PGOCtxProfileBlockIDs::ContextNodeBlockID));
+
+  std::optional<ctx_profile::GUID> Guid;
+  std::optional<SmallVector<uint64_t, 16>> Counters;
+  std::optional<uint32_t> CallsiteIndex;
+
+  SmallVector<uint64_t, 1> RecordValues;
+
+  // We don't prescribe the order in which the records come in, and we are ok
+  // if other unsupported records appear. We seek in the current subblock until
+  // we get all we know.
+  while (!Guid || !Counters || (ExpectIndex && !CallsiteIndex)) {
----------------
snehasish wrote:

nit: The negative booleans are a little hard to read. Consider this alternative loop structure:

```
while(true) {

... existing logic ...

// Exit the loop once we have gathered all the values we expect.
if (Guid.has_value() && Counters.has_value && (ExpectIndex && CallsiteIndex.has_value) 
  break;
}
```

https://github.com/llvm/llvm-project/pull/91859