LLVM 19.0.0git
SampleProfReader.h
Go to the documentation of this file.
1//===- SampleProfReader.h - Read LLVM sample profile data -------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file contains definitions needed for reading sample profiles.
10//
11// NOTE: If you are making changes to this file format, please remember
12// to document them in the Clang documentation at
13// tools/clang/docs/UsersManual.rst.
14//
15// Text format
16// -----------
17//
18// Sample profiles are written as ASCII text. The file is divided into
19// sections, which correspond to each of the functions executed at runtime.
20// Each section has the following format
21//
22// function1:total_samples:total_head_samples
23// offset1[.discriminator]: number_of_samples [fn1:num fn2:num ... ]
24// offset2[.discriminator]: number_of_samples [fn3:num fn4:num ... ]
25// ...
26// offsetN[.discriminator]: number_of_samples [fn5:num fn6:num ... ]
27// offsetA[.discriminator]: fnA:num_of_total_samples
28// offsetA1[.discriminator]: number_of_samples [fn7:num fn8:num ... ]
29// ...
30// !CFGChecksum: num
31// !Attribute: flags
32//
33// This is a nested tree in which the indentation represents the nesting level
34// of the inline stack. There are no blank lines in the file. And the spacing
35// within a single line is fixed. Additional spaces will result in an error
36// while reading the file.
37//
38// Any line starting with the '#' character is completely ignored.
39//
40// Inlined calls are represented with indentation. The Inline stack is a
41// stack of source locations in which the top of the stack represents the
42// leaf function, and the bottom of the stack represents the actual
43// symbol to which the instruction belongs.
44//
45// Function names must be mangled in order for the profile loader to
46// match them in the current translation unit. The two numbers in the
47// function header specify how many total samples were accumulated in the
48// function (first number), and the total number of samples accumulated
49// in the prologue of the function (second number). This head sample
50// count provides an indicator of how frequently the function is invoked.
51//
52// There are three types of lines in the function body.
53//
54// * Sampled line represents the profile information of a source location.
55// * Callsite line represents the profile information of a callsite.
56// * Metadata line represents extra metadata of the function.
57//
58// Each sampled line may contain several items. Some are optional (marked
59// below):
60//
61// a. Source line offset. This number represents the line number
62// in the function where the sample was collected. The line number is
63// always relative to the line where symbol of the function is
64// defined. So, if the function has its header at line 280, the offset
65// 13 is at line 293 in the file.
66//
67// Note that this offset should never be a negative number. This could
68// happen in cases like macros. The debug machinery will register the
69// line number at the point of macro expansion. So, if the macro was
70// expanded in a line before the start of the function, the profile
71// converter should emit a 0 as the offset (this means that the optimizers
72// will not be able to associate a meaningful weight to the instructions
73// in the macro).
74//
75// b. [OPTIONAL] Discriminator. This is used if the sampled program
76// was compiled with DWARF discriminator support
77// (http://wiki.dwarfstd.org/index.php?title=Path_Discriminators).
78// DWARF discriminators are unsigned integer values that allow the
79// compiler to distinguish between multiple execution paths on the
80// same source line location.
81//
82// For example, consider the line of code ``if (cond) foo(); else bar();``.
83// If the predicate ``cond`` is true 80% of the time, then the edge
84// into function ``foo`` should be considered to be taken most of the
85// time. But both calls to ``foo`` and ``bar`` are at the same source
86// line, so a sample count at that line is not sufficient. The
87// compiler needs to know which part of that line is taken more
88// frequently.
89//
90// This is what discriminators provide. In this case, the calls to
91// ``foo`` and ``bar`` will be at the same line, but will have
92// different discriminator values. This allows the compiler to correctly
93// set edge weights into ``foo`` and ``bar``.
94//
95// c. Number of samples. This is an integer quantity representing the
96// number of samples collected by the profiler at this source
97// location.
98//
99// d. [OPTIONAL] Potential call targets and samples. If present, this
100// line contains a call instruction. This models both direct and
101// number of samples. For example,
102//
103// 130: 7 foo:3 bar:2 baz:7
104//
105// The above means that at relative line offset 130 there is a call
106// instruction that calls one of ``foo()``, ``bar()`` and ``baz()``,
107// with ``baz()`` being the relatively more frequently called target.
108//
109// Each callsite line may contain several items. Some are optional.
110//
111// a. Source line offset. This number represents the line number of the
112// callsite that is inlined in the profiled binary.
113//
114// b. [OPTIONAL] Discriminator. Same as the discriminator for sampled line.
115//
116// c. Number of samples. This is an integer quantity representing the
117// total number of samples collected for the inlined instance at this
118// callsite
119//
120// Metadata line can occur in lines with one indent only, containing extra
121// information for the top-level function. Furthermore, metadata can only
122// occur after all the body samples and callsite samples.
123// Each metadata line may contain a particular type of metadata, marked by
124// the starting characters annotated with !. We process each metadata line
125// independently, hence each metadata line has to form an independent piece
126// of information that does not require cross-line reference.
127// We support the following types of metadata:
128//
129// a. CFG Checksum (a.k.a. function hash):
130// !CFGChecksum: 12345
131// b. CFG Checksum (see ContextAttributeMask):
132// !Atribute: 1
133//
134//
135// Binary format
136// -------------
137//
138// This is a more compact encoding. Numbers are encoded as ULEB128 values
139// and all strings are encoded in a name table. The file is organized in
140// the following sections:
141//
142// MAGIC (uint64_t)
143// File identifier computed by function SPMagic() (0x5350524f463432ff)
144//
145// VERSION (uint32_t)
146// File format version number computed by SPVersion()
147//
148// SUMMARY
149// TOTAL_COUNT (uint64_t)
150// Total number of samples in the profile.
151// MAX_COUNT (uint64_t)
152// Maximum value of samples on a line.
153// MAX_FUNCTION_COUNT (uint64_t)
154// Maximum number of samples at function entry (head samples).
155// NUM_COUNTS (uint64_t)
156// Number of lines with samples.
157// NUM_FUNCTIONS (uint64_t)
158// Number of functions with samples.
159// NUM_DETAILED_SUMMARY_ENTRIES (size_t)
160// Number of entries in detailed summary
161// DETAILED_SUMMARY
162// A list of detailed summary entry. Each entry consists of
163// CUTOFF (uint32_t)
164// Required percentile of total sample count expressed as a fraction
165// multiplied by 1000000.
166// MIN_COUNT (uint64_t)
167// The minimum number of samples required to reach the target
168// CUTOFF.
169// NUM_COUNTS (uint64_t)
170// Number of samples to get to the desrired percentile.
171//
172// NAME TABLE
173// SIZE (uint64_t)
174// Number of entries in the name table.
175// NAMES
176// A NUL-separated list of SIZE strings.
177//
178// FUNCTION BODY (one for each uninlined function body present in the profile)
179// HEAD_SAMPLES (uint64_t) [only for top-level functions]
180// Total number of samples collected at the head (prologue) of the
181// function.
182// NOTE: This field should only be present for top-level functions
183// (i.e., not inlined into any caller). Inlined function calls
184// have no prologue, so they don't need this.
185// NAME_IDX (uint64_t)
186// Index into the name table indicating the function name.
187// SAMPLES (uint64_t)
188// Total number of samples collected in this function.
189// NRECS (uint32_t)
190// Total number of sampling records this function's profile.
191// BODY RECORDS
192// A list of NRECS entries. Each entry contains:
193// OFFSET (uint32_t)
194// Line offset from the start of the function.
195// DISCRIMINATOR (uint32_t)
196// Discriminator value (see description of discriminators
197// in the text format documentation above).
198// SAMPLES (uint64_t)
199// Number of samples collected at this location.
200// NUM_CALLS (uint32_t)
201// Number of non-inlined function calls made at this location. In the
202// case of direct calls, this number will always be 1. For indirect
203// calls (virtual functions and function pointers) this will
204// represent all the actual functions called at runtime.
205// CALL_TARGETS
206// A list of NUM_CALLS entries for each called function:
207// NAME_IDX (uint64_t)
208// Index into the name table with the callee name.
209// SAMPLES (uint64_t)
210// Number of samples collected at the call site.
211// NUM_INLINED_FUNCTIONS (uint32_t)
212// Number of callees inlined into this function.
213// INLINED FUNCTION RECORDS
214// A list of NUM_INLINED_FUNCTIONS entries describing each of the inlined
215// callees.
216// OFFSET (uint32_t)
217// Line offset from the start of the function.
218// DISCRIMINATOR (uint32_t)
219// Discriminator value (see description of discriminators
220// in the text format documentation above).
221// FUNCTION BODY
222// A FUNCTION BODY entry describing the inlined function.
223//===----------------------------------------------------------------------===//
224
225#ifndef LLVM_PROFILEDATA_SAMPLEPROFREADER_H
226#define LLVM_PROFILEDATA_SAMPLEPROFREADER_H
227
228#include "llvm/ADT/SmallVector.h"
229#include "llvm/ADT/StringRef.h"
231#include "llvm/IR/LLVMContext.h"
236#include "llvm/Support/Debug.h"
238#include "llvm/Support/ErrorOr.h"
240#include <cstdint>
241#include <list>
242#include <memory>
243#include <optional>
244#include <string>
245#include <system_error>
246#include <unordered_set>
247#include <vector>
248
249namespace llvm {
250
251class raw_ostream;
252class Twine;
253
254namespace vfs {
255class FileSystem;
256} // namespace vfs
257
258namespace sampleprof {
259
260class SampleProfileReader;
261
262/// SampleProfileReaderItaniumRemapper remaps the profile data from a
263/// sample profile data reader, by applying a provided set of equivalences
264/// between components of the symbol names in the profile.
266public:
267 SampleProfileReaderItaniumRemapper(std::unique_ptr<MemoryBuffer> B,
268 std::unique_ptr<SymbolRemappingReader> SRR,
270 : Buffer(std::move(B)), Remappings(std::move(SRR)), Reader(R) {
271 assert(Remappings && "Remappings cannot be nullptr");
272 }
273
274 /// Create a remapper from the given remapping file. The remapper will
275 /// be used for profile read in by Reader.
277 create(const std::string Filename, vfs::FileSystem &FS,
279
280 /// Create a remapper from the given Buffer. The remapper will
281 /// be used for profile read in by Reader.
283 create(std::unique_ptr<MemoryBuffer> &B, SampleProfileReader &Reader,
284 LLVMContext &C);
285
286 /// Apply remappings to the profile read by Reader.
287 void applyRemapping(LLVMContext &Ctx);
288
289 bool hasApplied() { return RemappingApplied; }
290
291 /// Insert function name into remapper.
292 void insert(StringRef FunctionName) { Remappings->insert(FunctionName); }
293
294 /// Query whether there is equivalent in the remapper which has been
295 /// inserted.
296 bool exist(StringRef FunctionName) {
297 return Remappings->lookup(FunctionName);
298 }
299
300 /// Return the equivalent name in the profile for \p FunctionName if
301 /// it exists.
302 std::optional<StringRef> lookUpNameInProfile(StringRef FunctionName);
303
304private:
305 // The buffer holding the content read from remapping file.
306 std::unique_ptr<MemoryBuffer> Buffer;
307 std::unique_ptr<SymbolRemappingReader> Remappings;
308 // Map remapping key to the name in the profile. By looking up the
309 // key in the remapper, a given new name can be mapped to the
310 // cannonical name using the NameMap.
312 // The Reader the remapper is servicing.
313 SampleProfileReader &Reader;
314 // Indicate whether remapping has been applied to the profile read
315 // by Reader -- by calling applyRemapping.
316 bool RemappingApplied = false;
317};
318
319/// Sample-based profile reader.
320///
321/// Each profile contains sample counts for all the functions
322/// executed. Inside each function, statements are annotated with the
323/// collected samples on all the instructions associated with that
324/// statement.
325///
326/// For this to produce meaningful data, the program needs to be
327/// compiled with some debug information (at minimum, line numbers:
328/// -gline-tables-only). Otherwise, it will be impossible to match IR
329/// instructions to the line numbers collected by the profiler.
330///
331/// From the profile file, we are interested in collecting the
332/// following information:
333///
334/// * A list of functions included in the profile (mangled names).
335///
336/// * For each function F:
337/// 1. The total number of samples collected in F.
338///
339/// 2. The samples collected at each line in F. To provide some
340/// protection against source code shuffling, line numbers should
341/// be relative to the start of the function.
342///
343/// The reader supports two file formats: text and binary. The text format
344/// is useful for debugging and testing, while the binary format is more
345/// compact and I/O efficient. They can both be used interchangeably.
347public:
348 SampleProfileReader(std::unique_ptr<MemoryBuffer> B, LLVMContext &C,
350 : Profiles(), Ctx(C), Buffer(std::move(B)), Format(Format) {}
351
352 virtual ~SampleProfileReader() = default;
353
354 /// Read and validate the file header.
355 virtual std::error_code readHeader() = 0;
356
357 /// Set the bits for FS discriminators. Parameter Pass specify the sequence
358 /// number, Pass == i is for the i-th round of adding FS discriminators.
359 /// Pass == 0 is for using base discriminators.
362 }
363
364 /// Get the bitmask the discriminators: For FS profiles, return the bit
365 /// mask for this pass. For non FS profiles, return (unsigned) -1.
367 if (!ProfileIsFS)
368 return 0xFFFFFFFF;
369 assert((MaskedBitFrom != 0) && "MaskedBitFrom is not set properly");
370 return getN1Bits(MaskedBitFrom);
371 }
372
373 /// The interface to read sample profiles from the associated file.
374 std::error_code read() {
375 if (std::error_code EC = readImpl())
376 return EC;
377 if (Remapper)
378 Remapper->applyRemapping(Ctx);
381 }
382
383 /// The implementaion to read sample profiles from the associated file.
384 virtual std::error_code readImpl() = 0;
385
386 /// Print the profile for \p FunctionSamples on stream \p OS.
388
389 /// Collect functions with definitions in Module M. For reader which
390 /// support loading function profiles on demand, return true when the
391 /// reader has been given a module. Always return false for reader
392 /// which doesn't support loading function profiles on demand.
393 virtual bool collectFuncsFromModule() { return false; }
394
395 /// Print all the profiles on stream \p OS.
396 void dump(raw_ostream &OS = dbgs());
397
398 /// Print all the profiles on stream \p OS in the JSON format.
399 void dumpJson(raw_ostream &OS = dbgs());
400
401 /// Return the samples collected for function \p F.
403 // The function name may have been updated by adding suffix. Call
404 // a helper to (optionally) strip off suffixes so that we can
405 // match against the original function name in the profile.
407 return getSamplesFor(CanonName);
408 }
409
410 /// Return the samples collected for function \p F.
412 auto It = Profiles.find(FunctionId(Fname));
413 if (It != Profiles.end())
414 return &It->second;
415
416 if (Remapper) {
417 if (auto NameInProfile = Remapper->lookUpNameInProfile(Fname)) {
418 auto It = Profiles.find(FunctionId(*NameInProfile));
419 if (It != Profiles.end())
420 return &It->second;
421 }
422 }
423 return nullptr;
424 }
425
426 /// Return all the profiles.
428
429 /// Report a parse error message.
430 void reportError(int64_t LineNumber, const Twine &Msg) const {
431 Ctx.diagnose(DiagnosticInfoSampleProfile(Buffer->getBufferIdentifier(),
432 LineNumber, Msg));
433 }
434
435 /// Create a sample profile reader appropriate to the file format.
436 /// Create a remapper underlying if RemapFilename is not empty.
437 /// Parameter P specifies the FSDiscriminatorPass.
439 create(const std::string Filename, LLVMContext &C, vfs::FileSystem &FS,
441 const std::string RemapFilename = "");
442
443 /// Create a sample profile reader from the supplied memory buffer.
444 /// Create a remapper underlying if RemapFilename is not empty.
445 /// Parameter P specifies the FSDiscriminatorPass.
447 create(std::unique_ptr<MemoryBuffer> &B, LLVMContext &C, vfs::FileSystem &FS,
449 const std::string RemapFilename = "");
450
451 /// Return the profile summary.
452 ProfileSummary &getSummary() const { return *(Summary.get()); }
453
454 MemoryBuffer *getBuffer() const { return Buffer.get(); }
455
456 /// \brief Return the profile format.
458
459 /// Whether input profile is based on pseudo probes.
461
462 /// Whether input profile is fully context-sensitive.
463 bool profileIsCS() const { return ProfileIsCS; }
464
465 /// Whether input profile contains ShouldBeInlined contexts.
467
468 /// Whether input profile is flow-sensitive.
469 bool profileIsFS() const { return ProfileIsFS; }
470
471 virtual std::unique_ptr<ProfileSymbolList> getProfileSymbolList() {
472 return nullptr;
473 };
474
475 /// It includes all the names that have samples either in outline instance
476 /// or inline instance.
477 virtual std::vector<FunctionId> *getNameTable() { return nullptr; }
478 virtual bool dumpSectionInfo(raw_ostream &OS = dbgs()) { return false; };
479
480 /// Return whether names in the profile are all MD5 numbers.
481 bool useMD5() const { return ProfileIsMD5; }
482
483 /// Force the profile to use MD5 in Sample contexts, even if function names
484 /// are present.
485 virtual void setProfileUseMD5() { ProfileIsMD5 = true; }
486
487 /// Don't read profile without context if the flag is set. This is only meaningful
488 /// for ExtBinary format.
489 virtual void setSkipFlatProf(bool Skip) {}
490 /// Return whether any name in the profile contains ".__uniq." suffix.
491 virtual bool hasUniqSuffix() { return false; }
492
494
495 void setModule(const Module *Mod) { M = Mod; }
496
497protected:
498 /// Map every function to its associated profile.
499 ///
500 /// The profile of every function executed at runtime is collected
501 /// in the structure FunctionSamples. This maps function objects
502 /// to their corresponding profiles.
504
505 /// LLVM context used to emit diagnostics.
507
508 /// Memory buffer holding the profile file.
509 std::unique_ptr<MemoryBuffer> Buffer;
510
511 /// Profile summary information.
512 std::unique_ptr<ProfileSummary> Summary;
513
514 /// Take ownership of the summary of this reader.
515 static std::unique_ptr<ProfileSummary>
517 return std::move(Reader.Summary);
518 }
519
520 /// Compute summary for this profile.
521 void computeSummary();
522
523 std::unique_ptr<SampleProfileReaderItaniumRemapper> Remapper;
524
525 /// \brief Whether samples are collected based on pseudo probes.
527
528 /// Whether function profiles are context-sensitive flat profiles.
529 bool ProfileIsCS = false;
530
531 /// Whether function profile contains ShouldBeInlined contexts.
533
534 /// Number of context-sensitive profiles.
536
537 /// Whether the function profiles use FS discriminators.
538 bool ProfileIsFS = false;
539
540 /// \brief The format of sample.
542
543 /// \brief The current module being compiled if SampleProfileReader
544 /// is used by compiler. If SampleProfileReader is used by other
545 /// tools which are not compiler, M is usually nullptr.
546 const Module *M = nullptr;
547
548 /// Zero out the discriminator bits higher than bit MaskedBitFrom (0 based).
549 /// The default is to keep all the bits.
551
552 /// Whether the profile uses MD5 for Sample Contexts and function names. This
553 /// can be one-way overriden by the user to force use MD5.
554 bool ProfileIsMD5 = false;
555};
556
558public:
559 SampleProfileReaderText(std::unique_ptr<MemoryBuffer> B, LLVMContext &C)
561
562 /// Read and validate the file header.
563 std::error_code readHeader() override { return sampleprof_error::success; }
564
565 /// Read sample profiles from the associated file.
566 std::error_code readImpl() override;
567
568 /// Return true if \p Buffer is in the format supported by this class.
569 static bool hasFormat(const MemoryBuffer &Buffer);
570
571 /// Text format sample profile does not support MD5 for now.
572 void setProfileUseMD5() override {}
573
574private:
575 /// CSNameTable is used to save full context vectors. This serves as an
576 /// underlying immutable buffer for all clients.
577 std::list<SampleContextFrameVector> CSNameTable;
578};
579
581public:
582 SampleProfileReaderBinary(std::unique_ptr<MemoryBuffer> B, LLVMContext &C,
585
586 /// Read and validate the file header.
587 std::error_code readHeader() override;
588
589 /// Read sample profiles from the associated file.
590 std::error_code readImpl() override;
591
592 /// It includes all the names that have samples either in outline instance
593 /// or inline instance.
594 std::vector<FunctionId> *getNameTable() override {
595 return &NameTable;
596 }
597
598protected:
599 /// Read a numeric value of type T from the profile.
600 ///
601 /// If an error occurs during decoding, a diagnostic message is emitted and
602 /// EC is set.
603 ///
604 /// \returns the read value.
605 template <typename T> ErrorOr<T> readNumber();
606
607 /// Read a numeric value of type T from the profile. The value is saved
608 /// without encoded.
609 template <typename T> ErrorOr<T> readUnencodedNumber();
610
611 /// Read a string from the profile.
612 ///
613 /// If an error occurs during decoding, a diagnostic message is emitted and
614 /// EC is set.
615 ///
616 /// \returns the read value.
618
619 /// Read the string index and check whether it overflows the table.
620 template <typename T> inline ErrorOr<size_t> readStringIndex(T &Table);
621
622 /// Read the next function profile instance.
623 std::error_code readFuncProfile(const uint8_t *Start);
624
625 /// Read the contents of the given profile instance.
626 std::error_code readProfile(FunctionSamples &FProfile);
627
628 /// Read the contents of Magic number and Version number.
629 std::error_code readMagicIdent();
630
631 /// Read profile summary.
632 std::error_code readSummary();
633
634 /// Read the whole name table.
635 std::error_code readNameTable();
636
637 /// Read a string indirectly via the name table. Optionally return the index.
638 ErrorOr<FunctionId> readStringFromTable(size_t *RetIdx = nullptr);
639
640 /// Read a context indirectly via the CSNameTable. Optionally return the
641 /// index.
642 ErrorOr<SampleContextFrames> readContextFromTable(size_t *RetIdx = nullptr);
643
644 /// Read a context indirectly via the CSNameTable if the profile has context,
645 /// otherwise same as readStringFromTable, also return its hash value.
647
648 /// Points to the current location in the buffer.
649 const uint8_t *Data = nullptr;
650
651 /// Points to the end of the buffer.
652 const uint8_t *End = nullptr;
653
654 /// Function name table.
655 std::vector<FunctionId> NameTable;
656
657 /// CSNameTable is used to save full context vectors. It is the backing buffer
658 /// for SampleContextFrames.
659 std::vector<SampleContextFrameVector> CSNameTable;
660
661 /// Table to cache MD5 values of sample contexts corresponding to
662 /// readSampleContextFromTable(), used to index into Profiles or
663 /// FuncOffsetTable.
664 std::vector<uint64_t> MD5SampleContextTable;
665
666 /// The starting address of the table of MD5 values of sample contexts. For
667 /// fixed length MD5 non-CS profile it is same as MD5NameMemStart because
668 /// hashes of non-CS contexts are already in the profile. Otherwise it points
669 /// to the start of MD5SampleContextTable.
671
672private:
673 std::error_code readSummaryEntry(std::vector<ProfileSummaryEntry> &Entries);
674 virtual std::error_code verifySPMagic(uint64_t Magic) = 0;
675};
676
678private:
679 std::error_code verifySPMagic(uint64_t Magic) override;
680
681public:
682 SampleProfileReaderRawBinary(std::unique_ptr<MemoryBuffer> B, LLVMContext &C,
685
686 /// \brief Return true if \p Buffer is in the format supported by this class.
687 static bool hasFormat(const MemoryBuffer &Buffer);
688};
689
690/// SampleProfileReaderExtBinaryBase/SampleProfileWriterExtBinaryBase defines
691/// the basic structure of the extensible binary format.
692/// The format is organized in sections except the magic and version number
693/// at the beginning. There is a section table before all the sections, and
694/// each entry in the table describes the entry type, start, size and
695/// attributes. The format in each section is defined by the section itself.
696///
697/// It is easy to add a new section while maintaining the backward
698/// compatibility of the profile. Nothing extra needs to be done. If we want
699/// to extend an existing section, like add cache misses information in
700/// addition to the sample count in the profile body, we can add a new section
701/// with the extension and retire the existing section, and we could choose
702/// to keep the parser of the old section if we want the reader to be able
703/// to read both new and old format profile.
704///
705/// SampleProfileReaderExtBinary/SampleProfileWriterExtBinary define the
706/// commonly used sections of a profile in extensible binary format. It is
707/// possible to define other types of profile inherited from
708/// SampleProfileReaderExtBinaryBase/SampleProfileWriterExtBinaryBase.
710private:
711 std::error_code decompressSection(const uint8_t *SecStart,
712 const uint64_t SecSize,
713 const uint8_t *&DecompressBuf,
714 uint64_t &DecompressBufSize);
715
716 BumpPtrAllocator Allocator;
717
718protected:
719 std::vector<SecHdrTableEntry> SecHdrTable;
720 std::error_code readSecHdrTableEntry(uint64_t Idx);
721 std::error_code readSecHdrTable();
722
723 std::error_code readFuncMetadata(bool ProfileHasAttribute);
724 std::error_code readFuncMetadata(bool ProfileHasAttribute,
725 FunctionSamples *FProfile);
726 std::error_code readFuncOffsetTable();
727 std::error_code readFuncProfiles();
728 std::error_code readNameTableSec(bool IsMD5, bool FixedLengthMD5);
729 std::error_code readCSNameTableSec();
730 std::error_code readProfileSymbolList();
731
732 std::error_code readHeader() override;
733 std::error_code verifySPMagic(uint64_t Magic) override = 0;
734 virtual std::error_code readOneSection(const uint8_t *Start, uint64_t Size,
735 const SecHdrTableEntry &Entry);
736 // placeholder for subclasses to dispatch their own section readers.
737 virtual std::error_code readCustomSection(const SecHdrTableEntry &Entry) = 0;
738
739 /// Determine which container readFuncOffsetTable() should populate, the list
740 /// FuncOffsetList or the map FuncOffsetTable.
741 bool useFuncOffsetList() const;
742
743 std::unique_ptr<ProfileSymbolList> ProfSymList;
744
745 /// The table mapping from a function context's MD5 to the offset of its
746 /// FunctionSample towards file start.
747 /// At most one of FuncOffsetTable and FuncOffsetList is populated.
749
750 /// The list version of FuncOffsetTable. This is used if every entry is
751 /// being accessed.
752 std::vector<std::pair<SampleContext, uint64_t>> FuncOffsetList;
753
754 /// The set containing the functions to use when compiling a module.
756
757 /// If SkipFlatProf is true, skip the sections with
758 /// SecFlagFlat flag.
759 bool SkipFlatProf = false;
760
761public:
762 SampleProfileReaderExtBinaryBase(std::unique_ptr<MemoryBuffer> B,
765
766 /// Read sample profiles in extensible format from the associated file.
767 std::error_code readImpl() override;
768
769 /// Get the total size of all \p Type sections.
771 /// Get the total size of header and all sections.
773 bool dumpSectionInfo(raw_ostream &OS = dbgs()) override;
774
775 /// Collect functions with definitions in Module M. Return true if
776 /// the reader has been given a module.
777 bool collectFuncsFromModule() override;
778
779 std::unique_ptr<ProfileSymbolList> getProfileSymbolList() override {
780 return std::move(ProfSymList);
781 };
782
783 void setSkipFlatProf(bool Skip) override { SkipFlatProf = Skip; }
784};
785
787private:
788 std::error_code verifySPMagic(uint64_t Magic) override;
789 std::error_code readCustomSection(const SecHdrTableEntry &Entry) override {
790 // Update the data reader pointer to the end of the section.
791 Data = End;
793 };
794
795public:
796 SampleProfileReaderExtBinary(std::unique_ptr<MemoryBuffer> B, LLVMContext &C,
799
800 /// \brief Return true if \p Buffer is in the format supported by this class.
801 static bool hasFormat(const MemoryBuffer &Buffer);
802};
803
805
806// Supported histogram types in GCC. Currently, we only need support for
807// call target histograms.
818
820public:
821 SampleProfileReaderGCC(std::unique_ptr<MemoryBuffer> B, LLVMContext &C)
823 GcovBuffer(Buffer.get()) {}
824
825 /// Read and validate the file header.
826 std::error_code readHeader() override;
827
828 /// Read sample profiles from the associated file.
829 std::error_code readImpl() override;
830
831 /// Return true if \p Buffer is in the format supported by this class.
832 static bool hasFormat(const MemoryBuffer &Buffer);
833
834protected:
835 std::error_code readNameTable();
836 std::error_code readOneFunctionProfile(const InlineCallStack &InlineStack,
837 bool Update, uint32_t Offset);
838 std::error_code readFunctionProfiles();
839 std::error_code skipNextWord();
840 template <typename T> ErrorOr<T> readNumber();
842
843 /// Read the section tag and check that it's the same as \p Expected.
844 std::error_code readSectionTag(uint32_t Expected);
845
846 /// GCOV buffer containing the profile.
848
849 /// Function names in this profile.
850 std::vector<std::string> Names;
851
852 /// GCOV tags used to separate sections in the profile file.
853 static const uint32_t GCOVTagAFDOFileNames = 0xaa000000;
854 static const uint32_t GCOVTagAFDOFunction = 0xac000000;
855};
856
857} // end namespace sampleprof
858
859} // end namespace llvm
860
861#endif // LLVM_PROFILEDATA_SAMPLEPROFREADER_H
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
Returns the sub type a function will return at a given Idx Should correspond to the result type of an ExtractValue instruction executed with just that one unsigned Idx
uint64_t Size
Provides ErrorOr<T> smart pointer.
#define F(x, y, z)
Definition: MD5.cpp:55
#define P(N)
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
raw_pwrite_stream & OS
This file defines the SmallVector class.
Allocate memory in an ever growing pool, as if by bump-pointer.
Definition: Allocator.h:66
Implements a dense probed hash-table based set.
Definition: DenseSet.h:271
Diagnostic information for the sample profiler.
Represents either an error or a value T.
Definition: ErrorOr.h:56
Tagged union holding either a T or a Error.
Definition: Error.h:474
GCOVBuffer - A wrapper around MemoryBuffer to provide GCOV specific read operations.
Definition: GCOV.h:73
This is an important class for using LLVM in a threaded context.
Definition: LLVMContext.h:67
void diagnose(const DiagnosticInfo &DI)
Report a message to the currently installed diagnostic handler.
This interface provides simple read-only access to a block of memory, and provides simple methods for...
Definition: MemoryBuffer.h:51
A Module instance is used to store all the information related to an LLVM module.
Definition: Module.h:65
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
Definition: SmallVector.h:1209
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:50
Twine - A lightweight data structure for efficiently representing the concatenation of temporary valu...
Definition: Twine.h:81
The instances of the Type class are immutable: once they are created, they are never changed.
Definition: Type.h:45
This class implements an extremely fast bulk output stream that can only output to a stream.
Definition: raw_ostream.h:52
This class represents a function that is read from a sample profile.
Definition: FunctionId.h:36
Representation of the samples collected for a function.
Definition: SampleProf.h:744
static StringRef getCanonicalFnName(const Function &F)
Return the canonical name for a function, taking into account suffix elision policy attributes.
Definition: SampleProf.h:1085
static bool UseMD5
Whether the profile uses MD5 to represent string.
Definition: SampleProf.h:1190
This class provides operator overloads to the map container using MD5 as the key type,...
Definition: SampleProf.h:1306
iterator find(const SampleContext &Ctx)
Definition: SampleProf.h:1317
std::error_code readProfile(FunctionSamples &FProfile)
Read the contents of the given profile instance.
std::error_code readNameTable()
Read the whole name table.
const uint8_t * Data
Points to the current location in the buffer.
ErrorOr< StringRef > readString()
Read a string from the profile.
std::vector< FunctionId > NameTable
Function name table.
ErrorOr< T > readNumber()
Read a numeric value of type T from the profile.
ErrorOr< SampleContextFrames > readContextFromTable(size_t *RetIdx=nullptr)
Read a context indirectly via the CSNameTable.
ErrorOr< std::pair< SampleContext, uint64_t > > readSampleContextFromTable()
Read a context indirectly via the CSNameTable if the profile has context, otherwise same as readStrin...
std::error_code readHeader() override
Read and validate the file header.
const uint64_t * MD5SampleContextStart
The starting address of the table of MD5 values of sample contexts.
std::vector< SampleContextFrameVector > CSNameTable
CSNameTable is used to save full context vectors.
std::error_code readImpl() override
Read sample profiles from the associated file.
ErrorOr< FunctionId > readStringFromTable(size_t *RetIdx=nullptr)
Read a string indirectly via the name table. Optionally return the index.
SampleProfileReaderBinary(std::unique_ptr< MemoryBuffer > B, LLVMContext &C, SampleProfileFormat Format=SPF_None)
std::vector< uint64_t > MD5SampleContextTable
Table to cache MD5 values of sample contexts corresponding to readSampleContextFromTable(),...
ErrorOr< size_t > readStringIndex(T &Table)
Read the string index and check whether it overflows the table.
const uint8_t * End
Points to the end of the buffer.
ErrorOr< T > readUnencodedNumber()
Read a numeric value of type T from the profile.
std::error_code readFuncProfile(const uint8_t *Start)
Read the next function profile instance.
std::error_code readSummary()
Read profile summary.
std::error_code readMagicIdent()
Read the contents of Magic number and Version number.
std::vector< FunctionId > * getNameTable() override
It includes all the names that have samples either in outline instance or inline instance.
SampleProfileReaderExtBinaryBase/SampleProfileWriterExtBinaryBase defines the basic structure of the ...
std::error_code readFuncMetadata(bool ProfileHasAttribute)
bool collectFuncsFromModule() override
Collect functions with definitions in Module M.
uint64_t getSectionSize(SecType Type)
Get the total size of all Type sections.
void setSkipFlatProf(bool Skip) override
Don't read profile without context if the flag is set.
virtual std::error_code readCustomSection(const SecHdrTableEntry &Entry)=0
std::vector< std::pair< SampleContext, uint64_t > > FuncOffsetList
The list version of FuncOffsetTable.
DenseSet< StringRef > FuncsToUse
The set containing the functions to use when compiling a module.
std::unique_ptr< ProfileSymbolList > ProfSymList
bool useFuncOffsetList() const
Determine which container readFuncOffsetTable() should populate, the list FuncOffsetList or the map F...
std::error_code readNameTableSec(bool IsMD5, bool FixedLengthMD5)
std::unique_ptr< ProfileSymbolList > getProfileSymbolList() override
std::error_code readImpl() override
Read sample profiles in extensible format from the associated file.
virtual std::error_code readOneSection(const uint8_t *Start, uint64_t Size, const SecHdrTableEntry &Entry)
bool dumpSectionInfo(raw_ostream &OS=dbgs()) override
std::error_code verifySPMagic(uint64_t Magic) override=0
SampleProfileReaderExtBinaryBase(std::unique_ptr< MemoryBuffer > B, LLVMContext &C, SampleProfileFormat Format)
bool SkipFlatProf
If SkipFlatProf is true, skip the sections with SecFlagFlat flag.
DenseMap< hash_code, uint64_t > FuncOffsetTable
The table mapping from a function context's MD5 to the offset of its FunctionSample towards file star...
std::error_code readHeader() override
Read and validate the file header.
uint64_t getFileSize()
Get the total size of header and all sections.
SampleProfileReaderExtBinary(std::unique_ptr< MemoryBuffer > B, LLVMContext &C, SampleProfileFormat Format=SPF_Ext_Binary)
static bool hasFormat(const MemoryBuffer &Buffer)
Return true if Buffer is in the format supported by this class.
GCOVBuffer GcovBuffer
GCOV buffer containing the profile.
std::vector< std::string > Names
Function names in this profile.
std::error_code readImpl() override
Read sample profiles from the associated file.
SampleProfileReaderGCC(std::unique_ptr< MemoryBuffer > B, LLVMContext &C)
std::error_code readHeader() override
Read and validate the file header.
std::error_code readOneFunctionProfile(const InlineCallStack &InlineStack, bool Update, uint32_t Offset)
static const uint32_t GCOVTagAFDOFileNames
GCOV tags used to separate sections in the profile file.
static bool hasFormat(const MemoryBuffer &Buffer)
Return true if Buffer is in the format supported by this class.
std::error_code readSectionTag(uint32_t Expected)
Read the section tag and check that it's the same as Expected.
SampleProfileReaderItaniumRemapper remaps the profile data from a sample profile data reader,...
bool exist(StringRef FunctionName)
Query whether there is equivalent in the remapper which has been inserted.
static ErrorOr< std::unique_ptr< SampleProfileReaderItaniumRemapper > > create(const std::string Filename, vfs::FileSystem &FS, SampleProfileReader &Reader, LLVMContext &C)
Create a remapper from the given remapping file.
void applyRemapping(LLVMContext &Ctx)
Apply remappings to the profile read by Reader.
SampleProfileReaderItaniumRemapper(std::unique_ptr< MemoryBuffer > B, std::unique_ptr< SymbolRemappingReader > SRR, SampleProfileReader &R)
void insert(StringRef FunctionName)
Insert function name into remapper.
std::optional< StringRef > lookUpNameInProfile(StringRef FunctionName)
Return the equivalent name in the profile for FunctionName if it exists.
SampleProfileReaderRawBinary(std::unique_ptr< MemoryBuffer > B, LLVMContext &C, SampleProfileFormat Format=SPF_Binary)
static bool hasFormat(const MemoryBuffer &Buffer)
Return true if Buffer is in the format supported by this class.
SampleProfileReaderText(std::unique_ptr< MemoryBuffer > B, LLVMContext &C)
void setProfileUseMD5() override
Text format sample profile does not support MD5 for now.
std::error_code readHeader() override
Read and validate the file header.
std::error_code readImpl() override
Read sample profiles from the associated file.
static bool hasFormat(const MemoryBuffer &Buffer)
Return true if Buffer is in the format supported by this class.
Sample-based profile reader.
uint32_t MaskedBitFrom
Zero out the discriminator bits higher than bit MaskedBitFrom (0 based).
bool ProfileIsPreInlined
Whether function profile contains ShouldBeInlined contexts.
SampleProfileMap & getProfiles()
Return all the profiles.
uint32_t CSProfileCount
Number of context-sensitive profiles.
bool profileIsProbeBased() const
Whether input profile is based on pseudo probes.
FunctionSamples * getSamplesFor(const Function &F)
Return the samples collected for function F.
void dump(raw_ostream &OS=dbgs())
Print all the profiles on stream OS.
bool useMD5() const
Return whether names in the profile are all MD5 numbers.
const Module * M
The current module being compiled if SampleProfileReader is used by compiler.
std::unique_ptr< MemoryBuffer > Buffer
Memory buffer holding the profile file.
std::unique_ptr< SampleProfileReaderItaniumRemapper > Remapper
bool profileIsPreInlined() const
Whether input profile contains ShouldBeInlined contexts.
std::error_code read()
The interface to read sample profiles from the associated file.
bool profileIsFS() const
Whether input profile is flow-sensitive.
SampleProfileReaderItaniumRemapper * getRemapper()
bool ProfileIsCS
Whether function profiles are context-sensitive flat profiles.
bool ProfileIsMD5
Whether the profile uses MD5 for Sample Contexts and function names.
static std::unique_ptr< ProfileSummary > takeSummary(SampleProfileReader &Reader)
Take ownership of the summary of this reader.
ProfileSummary & getSummary() const
Return the profile summary.
SampleProfileFormat Format
The format of sample.
SampleProfileReader(std::unique_ptr< MemoryBuffer > B, LLVMContext &C, SampleProfileFormat Format=SPF_None)
std::unique_ptr< ProfileSummary > Summary
Profile summary information.
virtual bool hasUniqSuffix()
Return whether any name in the profile contains ".__uniq." suffix.
void computeSummary()
Compute summary for this profile.
uint32_t getDiscriminatorMask() const
Get the bitmask the discriminators: For FS profiles, return the bit mask for this pass.
virtual bool dumpSectionInfo(raw_ostream &OS=dbgs())
SampleProfileFormat getFormat() const
Return the profile format.
virtual void setProfileUseMD5()
Force the profile to use MD5 in Sample contexts, even if function names are present.
void setDiscriminatorMaskedBitFrom(FSDiscriminatorPass P)
Set the bits for FS discriminators.
virtual std::vector< FunctionId > * getNameTable()
It includes all the names that have samples either in outline instance or inline instance.
bool profileIsCS() const
Whether input profile is fully context-sensitive.
bool ProfileIsFS
Whether the function profiles use FS discriminators.
virtual bool collectFuncsFromModule()
Collect functions with definitions in Module M.
virtual void setSkipFlatProf(bool Skip)
Don't read profile without context if the flag is set.
FunctionSamples * getSamplesFor(StringRef Fname)
Return the samples collected for function F.
void dumpJson(raw_ostream &OS=dbgs())
Print all the profiles on stream OS in the JSON format.
static ErrorOr< std::unique_ptr< SampleProfileReader > > create(const std::string Filename, LLVMContext &C, vfs::FileSystem &FS, FSDiscriminatorPass P=FSDiscriminatorPass::Base, const std::string RemapFilename="")
Create a sample profile reader appropriate to the file format.
SampleProfileMap Profiles
Map every function to its associated profile.
virtual std::error_code readHeader()=0
Read and validate the file header.
void dumpFunctionProfile(const FunctionSamples &FS, raw_ostream &OS=dbgs())
Print the profile for FunctionSamples on stream OS.
bool ProfileIsProbeBased
Whether samples are collected based on pseudo probes.
void reportError(int64_t LineNumber, const Twine &Msg) const
Report a parse error message.
virtual std::unique_ptr< ProfileSymbolList > getProfileSymbolList()
LLVMContext & Ctx
LLVM context used to emit diagnostics.
virtual std::error_code readImpl()=0
The implementaion to read sample profiles from the associated file.
The virtual file system interface.
@ C
The default llvm calling convention, compatible with C.
Definition: CallingConv.h:34
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
@ Offset
Definition: DWP.cpp:456
static unsigned getFSPassBitEnd(sampleprof::FSDiscriminatorPass P)
Definition: Discriminator.h:87
decltype(auto) get(const PointerIntPair< PointerTy, IntBits, IntType, PtrTraits, Info > &Pair)
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
Definition: Debug.cpp:163
@ Mod
The access may modify the value stored in memory.
static unsigned getN1Bits(int N)
OutputIt move(R &&Range, OutputIt Out)
Provide wrappers to std::move which take ranges instead of having to pass begin/end explicitly.
Definition: STLExtras.h:1858
Implement std::hash so that hash_code can be used in STL containers.
Definition: BitVector.h:858