|
| 1 | +#ifndef MEMPROF_DATA_INC |
| 2 | +#define MEMPROF_DATA_INC |
| 3 | +/*===-- MemProfData.inc - MemProf profiling runtime structures -*- C++ -*-=== *\ |
| 4 | +|* |
| 5 | +|* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 6 | +|* See https://llvm.org/LICENSE.txt for license information. |
| 7 | +|* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 8 | +|* |
| 9 | +\*===----------------------------------------------------------------------===*/ |
| 10 | +/* |
| 11 | + * This is the main file that defines all the data structure, signature, |
| 12 | + * constant literals that are shared across profiling runtime library, |
| 13 | + * and host tools (reader/writer). |
| 14 | + * |
| 15 | + * This file has two identical copies. The primary copy lives in LLVM and |
| 16 | + * the other one sits in compiler-rt/include/profile directory. To make changes |
| 17 | + * in this file, first modify the primary copy and copy it over to compiler-rt. |
| 18 | + * Testing of any change in this file can start only after the two copies are |
| 19 | + * synced up. |
| 20 | + * |
| 21 | +\*===----------------------------------------------------------------------===*/ |
| 22 | +#include <string.h> |
| 23 | + |
| 24 | +#ifdef _MSC_VER |
| 25 | +#define PACKED(...) __pragma(pack(push,1)) __VA_ARGS__ __pragma(pack(pop)) |
| 26 | +#else |
| 27 | +#define PACKED(...) __VA_ARGS__ __attribute__((__packed__)) |
| 28 | +#endif |
| 29 | + |
| 30 | +// A 64-bit magic number to uniquely identify the raw binary memprof profile file. |
| 31 | +#define MEMPROF_RAW_MAGIC_64 \ |
| 32 | + ((uint64_t)255 << 56 | (uint64_t)'m' << 48 | (uint64_t)'p' << 40 | (uint64_t)'r' << 32 | \ |
| 33 | + (uint64_t)'o' << 24 | (uint64_t)'f' << 16 | (uint64_t)'r' << 8 | (uint64_t)129) |
| 34 | + |
| 35 | +// The version number of the raw binary format. |
| 36 | +#define MEMPROF_RAW_VERSION 4ULL |
| 37 | + |
| 38 | +// Currently supported versions. |
| 39 | +#define MEMPROF_RAW_SUPPORTED_VERSIONS \ |
| 40 | + { 3ULL, 4ULL } |
| 41 | + |
| 42 | +#define MEMPROF_V3_MIB_SIZE 132ULL; |
| 43 | + |
| 44 | +#define MEMPROF_BUILDID_MAX_SIZE 32ULL |
| 45 | + |
| 46 | +namespace llvm { |
| 47 | +namespace memprof { |
| 48 | +// A struct describing the header used for the raw binary memprof profile format. |
| 49 | +PACKED(struct Header { |
| 50 | + uint64_t Magic; |
| 51 | + uint64_t Version; |
| 52 | + uint64_t TotalSize; |
| 53 | + uint64_t SegmentOffset; |
| 54 | + uint64_t MIBOffset; |
| 55 | + uint64_t StackOffset; |
| 56 | +}); |
| 57 | + |
| 58 | +// A struct describing the information necessary to describe a /proc/maps |
| 59 | +// segment entry for a particular binary/library identified by its build id. |
| 60 | +PACKED(struct SegmentEntry { |
| 61 | + uint64_t Start; |
| 62 | + uint64_t End; |
| 63 | + uint64_t Offset; |
| 64 | + uint64_t BuildIdSize; |
| 65 | + uint8_t BuildId[MEMPROF_BUILDID_MAX_SIZE] = {0}; |
| 66 | + |
| 67 | + // This constructor is only used in tests so don't set the BuildId. |
| 68 | + SegmentEntry(uint64_t S, uint64_t E, uint64_t O) |
| 69 | + : Start(S), End(E), Offset(O), BuildIdSize(0) {} |
| 70 | + |
| 71 | + SegmentEntry(const SegmentEntry& S) { |
| 72 | + Start = S.Start; |
| 73 | + End = S.End; |
| 74 | + Offset = S.Offset; |
| 75 | + BuildIdSize = S.BuildIdSize; |
| 76 | + memcpy(BuildId, S.BuildId, S.BuildIdSize); |
| 77 | + } |
| 78 | + |
| 79 | + SegmentEntry& operator=(const SegmentEntry& S) { |
| 80 | + Start = S.Start; |
| 81 | + End = S.End; |
| 82 | + Offset = S.Offset; |
| 83 | + BuildIdSize = S.BuildIdSize; |
| 84 | + memcpy(BuildId, S.BuildId, S.BuildIdSize); |
| 85 | + return *this; |
| 86 | + } |
| 87 | + |
| 88 | + bool operator==(const SegmentEntry& S) const { |
| 89 | + return Start == S.Start && End == S.End && Offset == S.Offset && |
| 90 | + BuildIdSize == S.BuildIdSize && |
| 91 | + memcmp(BuildId, S.BuildId, S.BuildIdSize) == 0; |
| 92 | + } |
| 93 | +}); |
| 94 | + |
| 95 | +// Packed struct definition for MSVC. We can't use the PACKED macro defined in |
| 96 | +// MemProfData.inc since it would mean we are embedding a directive (the |
| 97 | +// #include for MIBEntryDef) into the macros which is undefined behaviour. |
| 98 | +#ifdef _MSC_VER |
| 99 | +__pragma(pack(push,1)) |
| 100 | +#endif |
| 101 | + |
| 102 | +// A struct representing the heap allocation characteristics of a particular |
| 103 | +// runtime context. This struct is shared between the compiler-rt runtime and |
| 104 | +// the raw profile reader. The indexed format uses a separate, self-describing |
| 105 | +// backwards compatible format. |
| 106 | +struct MemInfoBlock{ |
| 107 | + |
| 108 | +#define MIBEntryDef(NameTag, Name, Type) Type Name; |
| 109 | +#include "MIBEntryDef.inc" |
| 110 | +#undef MIBEntryDef |
| 111 | + |
| 112 | +bool operator==(const MemInfoBlock& Other) const { |
| 113 | + bool IsEqual = true; |
| 114 | +#define MIBEntryDef(NameTag, Name, Type) \ |
| 115 | + IsEqual = (IsEqual && Name == Other.Name); |
| 116 | +#include "MIBEntryDef.inc" |
| 117 | +#undef MIBEntryDef |
| 118 | + return IsEqual; |
| 119 | +} |
| 120 | + |
| 121 | +MemInfoBlock() { |
| 122 | +#define MIBEntryDef(NameTag, Name, Type) Name = Type(); |
| 123 | +#include "MIBEntryDef.inc" |
| 124 | +#undef MIBEntryDef |
| 125 | +} |
| 126 | + |
| 127 | +MemInfoBlock(uint32_t Size, uint64_t AccessCount, uint32_t AllocTs, |
| 128 | + uint32_t DeallocTs, uint32_t AllocCpu, uint32_t DeallocCpu, |
| 129 | + uintptr_t Histogram, uint32_t HistogramSize) |
| 130 | + : MemInfoBlock() { |
| 131 | + AllocCount = 1U; |
| 132 | + TotalAccessCount = AccessCount; |
| 133 | + MinAccessCount = AccessCount; |
| 134 | + MaxAccessCount = AccessCount; |
| 135 | + TotalSize = Size; |
| 136 | + MinSize = Size; |
| 137 | + MaxSize = Size; |
| 138 | + AllocTimestamp = AllocTs; |
| 139 | + DeallocTimestamp = DeallocTs; |
| 140 | + TotalLifetime = DeallocTimestamp - AllocTimestamp; |
| 141 | + MinLifetime = TotalLifetime; |
| 142 | + MaxLifetime = TotalLifetime; |
| 143 | + // Access density is accesses per byte. Multiply by 100 to include the |
| 144 | + // fractional part. |
| 145 | + TotalAccessDensity = AccessCount * 100 / Size; |
| 146 | + MinAccessDensity = TotalAccessDensity; |
| 147 | + MaxAccessDensity = TotalAccessDensity; |
| 148 | + // Lifetime access density is the access density per second of lifetime. |
| 149 | + // Multiply by 1000 to convert denominator lifetime to seconds (using a |
| 150 | + // minimum lifetime of 1ms to avoid divide by 0. Do the multiplication first |
| 151 | + // to reduce truncations to 0. |
| 152 | + TotalLifetimeAccessDensity = |
| 153 | + TotalAccessDensity * 1000 / (TotalLifetime ? TotalLifetime : 1); |
| 154 | + MinLifetimeAccessDensity = TotalLifetimeAccessDensity; |
| 155 | + MaxLifetimeAccessDensity = TotalLifetimeAccessDensity; |
| 156 | + AllocCpuId = AllocCpu; |
| 157 | + DeallocCpuId = DeallocCpu; |
| 158 | + NumMigratedCpu = AllocCpuId != DeallocCpuId; |
| 159 | + AccessHistogramSize = HistogramSize; |
| 160 | + AccessHistogram = Histogram; |
| 161 | +} |
| 162 | + |
| 163 | +void Merge(const MemInfoBlock &newMIB) { |
| 164 | + AllocCount += newMIB.AllocCount; |
| 165 | + |
| 166 | + TotalAccessCount += newMIB.TotalAccessCount; |
| 167 | + MinAccessCount = newMIB.MinAccessCount < MinAccessCount ? newMIB.MinAccessCount : MinAccessCount; |
| 168 | + MaxAccessCount = newMIB.MaxAccessCount > MaxAccessCount ? newMIB.MaxAccessCount : MaxAccessCount; |
| 169 | + |
| 170 | + TotalSize += newMIB.TotalSize; |
| 171 | + MinSize = newMIB.MinSize < MinSize ? newMIB.MinSize : MinSize; |
| 172 | + MaxSize = newMIB.MaxSize > MaxSize ? newMIB.MaxSize : MaxSize; |
| 173 | + |
| 174 | + TotalLifetime += newMIB.TotalLifetime; |
| 175 | + MinLifetime = newMIB.MinLifetime < MinLifetime ? newMIB.MinLifetime : MinLifetime; |
| 176 | + MaxLifetime = newMIB.MaxLifetime > MaxLifetime ? newMIB.MaxLifetime : MaxLifetime; |
| 177 | + |
| 178 | + TotalAccessDensity += newMIB.TotalAccessDensity; |
| 179 | + MinAccessDensity = newMIB.MinAccessDensity < MinAccessDensity |
| 180 | + ? newMIB.MinAccessDensity |
| 181 | + : MinAccessDensity; |
| 182 | + MaxAccessDensity = newMIB.MaxAccessDensity > MaxAccessDensity |
| 183 | + ? newMIB.MaxAccessDensity |
| 184 | + : MaxAccessDensity; |
| 185 | + |
| 186 | + TotalLifetimeAccessDensity += newMIB.TotalLifetimeAccessDensity; |
| 187 | + MinLifetimeAccessDensity = |
| 188 | + newMIB.MinLifetimeAccessDensity < MinLifetimeAccessDensity |
| 189 | + ? newMIB.MinLifetimeAccessDensity |
| 190 | + : MinLifetimeAccessDensity; |
| 191 | + MaxLifetimeAccessDensity = |
| 192 | + newMIB.MaxLifetimeAccessDensity > MaxLifetimeAccessDensity |
| 193 | + ? newMIB.MaxLifetimeAccessDensity |
| 194 | + : MaxLifetimeAccessDensity; |
| 195 | + |
| 196 | + // We know newMIB was deallocated later, so just need to check if it was |
| 197 | + // allocated before last one deallocated. |
| 198 | + NumLifetimeOverlaps += newMIB.AllocTimestamp < DeallocTimestamp; |
| 199 | + AllocTimestamp = newMIB.AllocTimestamp; |
| 200 | + DeallocTimestamp = newMIB.DeallocTimestamp; |
| 201 | + |
| 202 | + NumSameAllocCpu += AllocCpuId == newMIB.AllocCpuId; |
| 203 | + NumSameDeallocCpu += DeallocCpuId == newMIB.DeallocCpuId; |
| 204 | + AllocCpuId = newMIB.AllocCpuId; |
| 205 | + DeallocCpuId = newMIB.DeallocCpuId; |
| 206 | + |
| 207 | + // For merging histograms, we always keep the longer histogram, and add |
| 208 | + // values of shorter histogram to larger one. |
| 209 | + uintptr_t ShorterHistogram; |
| 210 | + uint32_t ShorterHistogramSize; |
| 211 | + if (newMIB.AccessHistogramSize > AccessHistogramSize) { |
| 212 | + ShorterHistogram = AccessHistogram; |
| 213 | + ShorterHistogramSize = AccessHistogramSize; |
| 214 | + // Swap histogram of current to larger histogram |
| 215 | + AccessHistogram = newMIB.AccessHistogram; |
| 216 | + AccessHistogramSize = newMIB.AccessHistogramSize; |
| 217 | + } else { |
| 218 | + ShorterHistogram = newMIB.AccessHistogram; |
| 219 | + ShorterHistogramSize = newMIB.AccessHistogramSize; |
| 220 | + } |
| 221 | + for (size_t i = 0; i < ShorterHistogramSize; ++i) { |
| 222 | + ((uint64_t *)AccessHistogram)[i] += ((uint64_t *)ShorterHistogram)[i]; |
| 223 | + } |
| 224 | +} |
| 225 | + |
| 226 | +#ifdef _MSC_VER |
| 227 | +} __pragma(pack(pop)); |
| 228 | +#else |
| 229 | +} __attribute__((__packed__)); |
| 230 | +#endif |
| 231 | + |
| 232 | +} // namespace memprof |
| 233 | +} // namespace llvm |
| 234 | + |
| 235 | +#endif |
0 commit comments