[AMDGPU] Test codegen'ing True16 additions.
[llvm-project.git] / llvm / lib / ProfileData / MemProf.cpp
blob3255cba4dd0ca8634d980751eae5405bcf532df8
1 #include "llvm/ProfileData/MemProf.h"
2 #include "llvm/ADT/SmallVector.h"
3 #include "llvm/IR/Function.h"
4 #include "llvm/ProfileData/InstrProf.h"
5 #include "llvm/ProfileData/SampleProf.h"
6 #include "llvm/Support/Endian.h"
7 #include "llvm/Support/EndianStream.h"
9 namespace llvm {
10 namespace memprof {
12 void IndexedMemProfRecord::serialize(const MemProfSchema &Schema,
13 raw_ostream &OS) {
14 using namespace support;
16 endian::Writer LE(OS, little);
18 LE.write<uint64_t>(AllocSites.size());
19 for (const IndexedAllocationInfo &N : AllocSites) {
20 LE.write<uint64_t>(N.CallStack.size());
21 for (const FrameId &Id : N.CallStack)
22 LE.write<FrameId>(Id);
23 N.Info.serialize(Schema, OS);
26 // Related contexts.
27 LE.write<uint64_t>(CallSites.size());
28 for (const auto &Frames : CallSites) {
29 LE.write<uint64_t>(Frames.size());
30 for (const FrameId &Id : Frames)
31 LE.write<FrameId>(Id);
35 IndexedMemProfRecord
36 IndexedMemProfRecord::deserialize(const MemProfSchema &Schema,
37 const unsigned char *Ptr) {
38 using namespace support;
40 IndexedMemProfRecord Record;
42 // Read the meminfo nodes.
43 const uint64_t NumNodes = endian::readNext<uint64_t, little, unaligned>(Ptr);
44 for (uint64_t I = 0; I < NumNodes; I++) {
45 IndexedAllocationInfo Node;
46 const uint64_t NumFrames =
47 endian::readNext<uint64_t, little, unaligned>(Ptr);
48 for (uint64_t J = 0; J < NumFrames; J++) {
49 const FrameId Id = endian::readNext<FrameId, little, unaligned>(Ptr);
50 Node.CallStack.push_back(Id);
52 Node.Info.deserialize(Schema, Ptr);
53 Ptr += PortableMemInfoBlock::serializedSize();
54 Record.AllocSites.push_back(Node);
57 // Read the callsite information.
58 const uint64_t NumCtxs = endian::readNext<uint64_t, little, unaligned>(Ptr);
59 for (uint64_t J = 0; J < NumCtxs; J++) {
60 const uint64_t NumFrames =
61 endian::readNext<uint64_t, little, unaligned>(Ptr);
62 llvm::SmallVector<FrameId> Frames;
63 Frames.reserve(NumFrames);
64 for (uint64_t K = 0; K < NumFrames; K++) {
65 const FrameId Id = endian::readNext<FrameId, little, unaligned>(Ptr);
66 Frames.push_back(Id);
68 Record.CallSites.push_back(Frames);
71 return Record;
74 GlobalValue::GUID IndexedMemProfRecord::getGUID(const StringRef FunctionName) {
75 // Canonicalize the function name to drop suffixes such as ".llvm.", ".uniq."
76 // etc. We can then match functions in the profile use phase prior to the
77 // addition of these suffixes. Note that this applies to both instrumented and
78 // sampled function names.
79 StringRef CanonicalName =
80 sampleprof::FunctionSamples::getCanonicalFnName(FunctionName);
82 // We use the function guid which we expect to be a uint64_t. At
83 // this time, it is the lower 64 bits of the md5 of the canonical
84 // function name.
85 return Function::getGUID(CanonicalName);
88 Expected<MemProfSchema> readMemProfSchema(const unsigned char *&Buffer) {
89 using namespace support;
91 const unsigned char *Ptr = Buffer;
92 const uint64_t NumSchemaIds =
93 endian::readNext<uint64_t, little, unaligned>(Ptr);
94 if (NumSchemaIds > static_cast<uint64_t>(Meta::Size)) {
95 return make_error<InstrProfError>(instrprof_error::malformed,
96 "memprof schema invalid");
99 MemProfSchema Result;
100 for (size_t I = 0; I < NumSchemaIds; I++) {
101 const uint64_t Tag = endian::readNext<uint64_t, little, unaligned>(Ptr);
102 if (Tag >= static_cast<uint64_t>(Meta::Size)) {
103 return make_error<InstrProfError>(instrprof_error::malformed,
104 "memprof schema invalid");
106 Result.push_back(static_cast<Meta>(Tag));
108 // Advace the buffer to one past the schema if we succeeded.
109 Buffer = Ptr;
110 return Result;
113 } // namespace memprof
114 } // namespace llvm