Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line | 
|---|---|---|---|
| 14 | pmbaty | 1 | //===- llvm/Support/TimeProfiler.h - Hierarchical Time Profiler -*- 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 provides lightweight and dependency-free machinery to trace execution | ||
| 10 | // time around arbitrary code. Two API flavors are available. | ||
| 11 | // | ||
| 12 | // The primary API uses a RAII object to trigger tracing: | ||
| 13 | // | ||
| 14 | // \code | ||
| 15 | //   { | ||
| 16 | //     TimeTraceScope scope("my_event_name"); | ||
| 17 | //     ...my code... | ||
| 18 | //   } | ||
| 19 | // \endcode | ||
| 20 | // | ||
| 21 | // If the code to be profiled does not have a natural lexical scope then | ||
| 22 | // it is also possible to start and end events with respect to an implicit | ||
| 23 | // per-thread stack of profiling entries: | ||
| 24 | // | ||
| 25 | // \code | ||
| 26 | //   timeTraceProfilerBegin("my_event_name"); | ||
| 27 | //   ...my code... | ||
| 28 | //   timeTraceProfilerEnd();  // must be called on all control flow paths | ||
| 29 | // \endcode | ||
| 30 | // | ||
| 31 | // Time profiling entries can be given an arbitrary name and, optionally, | ||
| 32 | // an arbitrary 'detail' string. The resulting trace will include 'Total' | ||
| 33 | // entries summing the time spent for each name. Thus, it's best to choose | ||
| 34 | // names to be fairly generic, and rely on the detail field to capture | ||
| 35 | // everything else of interest. | ||
| 36 | // | ||
| 37 | // To avoid lifetime issues name and detail strings are copied into the event | ||
| 38 | // entries at their time of creation. Care should be taken to make string | ||
| 39 | // construction cheap to prevent 'Heisenperf' effects. In particular, the | ||
| 40 | // 'detail' argument may be a string-returning closure: | ||
| 41 | // | ||
| 42 | // \code | ||
| 43 | //   int n; | ||
| 44 | //   { | ||
| 45 | //     TimeTraceScope scope("my_event_name", | ||
| 46 | //                          [n]() { return (Twine("x=") + Twine(n)).str(); }); | ||
| 47 | //     ...my code... | ||
| 48 | //   } | ||
| 49 | // \endcode | ||
| 50 | // The closure will not be called if tracing is disabled. Otherwise, the | ||
| 51 | // resulting string will be directly moved into the entry. | ||
| 52 | // | ||
| 53 | // The main process should begin with a timeTraceProfilerInitialize, and | ||
| 54 | // finish with timeTraceProfileWrite and timeTraceProfilerCleanup calls. | ||
| 55 | // Each new thread should begin with a timeTraceProfilerInitialize, and | ||
| 56 | // finish with a timeTraceProfilerFinishThread call. | ||
| 57 | // | ||
| 58 | // Timestamps come from std::chrono::stable_clock. Note that threads need | ||
| 59 | // not see the same time from that clock, and the resolution may not be | ||
| 60 | // the best available. | ||
| 61 | // | ||
| 62 | // Currently, there are a number of compatible viewers: | ||
| 63 | //  - chrome://tracing is the original chromium trace viewer. | ||
| 64 | //  - http://ui.perfetto.dev is the replacement for the above, under active | ||
| 65 | //    development by Google as part of the 'Perfetto' project. | ||
| 66 | //  - https://www.speedscope.app/ has also been reported as an option. | ||
| 67 | // | ||
| 68 | // Future work: | ||
| 69 | //  - Support akin to LLVM_DEBUG for runtime enable/disable of named tracing | ||
| 70 | //    families for non-debug builds which wish to support optional tracing. | ||
| 71 | //  - Evaluate the detail closures at profile write time to avoid | ||
| 72 | //    stringification costs interfering with tracing. | ||
| 73 | // | ||
| 74 | //===----------------------------------------------------------------------===// | ||
| 75 | |||
| 76 | #ifndef LLVM_SUPPORT_TIMEPROFILER_H | ||
| 77 | #define LLVM_SUPPORT_TIMEPROFILER_H | ||
| 78 | |||
| 79 | #include "llvm/ADT/STLFunctionalExtras.h" | ||
| 80 | #include "llvm/Support/Error.h" | ||
| 81 | |||
| 82 | namespace llvm { | ||
| 83 | |||
| 84 | class raw_pwrite_stream; | ||
| 85 | |||
| 86 | struct TimeTraceProfiler; | ||
| 87 | TimeTraceProfiler *getTimeTraceProfilerInstance(); | ||
| 88 | |||
| 89 | /// Initialize the time trace profiler. | ||
| 90 | /// This sets up the global \p TimeTraceProfilerInstance | ||
| 91 | /// variable to be the profiler instance. | ||
| 92 | void timeTraceProfilerInitialize(unsigned TimeTraceGranularity, | ||
| 93 | StringRef ProcName); | ||
| 94 | |||
| 95 | /// Cleanup the time trace profiler, if it was initialized. | ||
| 96 | void timeTraceProfilerCleanup(); | ||
| 97 | |||
| 98 | /// Finish a time trace profiler running on a worker thread. | ||
| 99 | void timeTraceProfilerFinishThread(); | ||
| 100 | |||
| 101 | /// Is the time trace profiler enabled, i.e. initialized? | ||
| 102 | inline bool timeTraceProfilerEnabled() { | ||
| 103 | return getTimeTraceProfilerInstance() != nullptr; | ||
| 104 | } | ||
| 105 | |||
| 106 | /// Write profiling data to output stream. | ||
| 107 | /// Data produced is JSON, in Chrome "Trace Event" format, see | ||
| 108 | /// https://docs.google.com/document/d/1CvAClvFfyA5R-PhYUmn5OOQtYMH4h6I0nSsKchNAySU/preview | ||
| 109 | void timeTraceProfilerWrite(raw_pwrite_stream &OS); | ||
| 110 | |||
| 111 | /// Write profiling data to a file. | ||
| 112 | /// The function will write to \p PreferredFileName if provided, if not | ||
| 113 | /// then will write to \p FallbackFileName appending .time-trace. | ||
| 114 | /// Returns a StringError indicating a failure if the function is | ||
| 115 | /// unable to open the file for writing. | ||
| 116 | Error timeTraceProfilerWrite(StringRef PreferredFileName, | ||
| 117 | StringRef FallbackFileName); | ||
| 118 | |||
| 119 | /// Manually begin a time section, with the given \p Name and \p Detail. | ||
| 120 | /// Profiler copies the string data, so the pointers can be given into | ||
| 121 | /// temporaries. Time sections can be hierarchical; every Begin must have a | ||
| 122 | /// matching End pair but they can nest. | ||
| 123 | void timeTraceProfilerBegin(StringRef Name, StringRef Detail); | ||
| 124 | void timeTraceProfilerBegin(StringRef Name, | ||
| 125 | llvm::function_ref<std::string()> Detail); | ||
| 126 | |||
| 127 | /// Manually end the last time section. | ||
| 128 | void timeTraceProfilerEnd(); | ||
| 129 | |||
| 130 | /// The TimeTraceScope is a helper class to call the begin and end functions | ||
| 131 | /// of the time trace profiler.  When the object is constructed, it begins | ||
| 132 | /// the section; and when it is destroyed, it stops it. If the time profiler | ||
| 133 | /// is not initialized, the overhead is a single branch. | ||
| 134 | struct TimeTraceScope { | ||
| 135 | |||
| 136 | TimeTraceScope() = delete; | ||
| 137 | TimeTraceScope(const TimeTraceScope &) = delete; | ||
| 138 | TimeTraceScope &operator=(const TimeTraceScope &) = delete; | ||
| 139 | TimeTraceScope(TimeTraceScope &&) = delete; | ||
| 140 | TimeTraceScope &operator=(TimeTraceScope &&) = delete; | ||
| 141 | |||
| 142 | TimeTraceScope(StringRef Name) { | ||
| 143 | if (getTimeTraceProfilerInstance() != nullptr) | ||
| 144 | timeTraceProfilerBegin(Name, StringRef("")); | ||
| 145 |   } | ||
| 146 | TimeTraceScope(StringRef Name, StringRef Detail) { | ||
| 147 | if (getTimeTraceProfilerInstance() != nullptr) | ||
| 148 | timeTraceProfilerBegin(Name, Detail); | ||
| 149 |   } | ||
| 150 | TimeTraceScope(StringRef Name, llvm::function_ref<std::string()> Detail) { | ||
| 151 | if (getTimeTraceProfilerInstance() != nullptr) | ||
| 152 | timeTraceProfilerBegin(Name, Detail); | ||
| 153 |   } | ||
| 154 | ~TimeTraceScope() { | ||
| 155 | if (getTimeTraceProfilerInstance() != nullptr) | ||
| 156 | timeTraceProfilerEnd(); | ||
| 157 |   } | ||
| 158 | }; | ||
| 159 | |||
| 160 | } // end namespace llvm | ||
| 161 | |||
| 162 | #endif |