1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
|
/*
* Copyright (C) 2017-2020 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/helpers/aligned_memory.h"
#include "shared/source/helpers/debug_helpers.h"
#include <algorithm>
#include <cstdint>
#include <unordered_map>
#include <vector>
namespace NEO {
struct HeapChunk {
HeapChunk(uint64_t ptr, size_t size) : ptr(ptr), size(size) {}
uint64_t ptr;
size_t size;
};
bool operator<(const HeapChunk &hc1, const HeapChunk &hc2);
class HeapAllocator {
public:
HeapAllocator(uint64_t address, uint64_t size) : HeapAllocator(address, size, 4 * MemoryConstants::megaByte) {
}
HeapAllocator(uint64_t address, uint64_t size, size_t threshold) : size(size), availableSize(size), sizeThreshold(threshold) {
pLeftBound = address;
pRightBound = address + size;
freedChunksBig.reserve(10);
freedChunksSmall.reserve(50);
}
uint64_t allocate(size_t &sizeToAllocate) {
sizeToAllocate = alignUp(sizeToAllocate, allocationAlignment);
std::lock_guard<std::mutex> lock(mtx);
DBG_LOG(PrintDebugMessages, __FUNCTION__, "Allocator usage == ", this->getUsage());
if (availableSize < sizeToAllocate) {
return 0llu;
}
std::vector<HeapChunk> &freedChunks = (sizeToAllocate > sizeThreshold) ? freedChunksBig : freedChunksSmall;
uint32_t defragmentCount = 0;
for (;;) {
size_t sizeOfFreedChunk = 0;
uint64_t ptrReturn = getFromFreedChunks(sizeToAllocate, freedChunks, sizeOfFreedChunk);
if (ptrReturn == 0llu) {
if (sizeToAllocate > sizeThreshold) {
if (pLeftBound + sizeToAllocate <= pRightBound) {
ptrReturn = pLeftBound;
pLeftBound += sizeToAllocate;
}
} else {
if (pRightBound - sizeToAllocate >= pLeftBound) {
pRightBound -= sizeToAllocate;
ptrReturn = pRightBound;
}
}
}
if (ptrReturn != 0llu) {
if (sizeOfFreedChunk > 0) {
availableSize -= sizeOfFreedChunk;
sizeToAllocate = sizeOfFreedChunk;
} else {
availableSize -= sizeToAllocate;
}
return ptrReturn;
}
if (defragmentCount == 1)
return 0llu;
defragment();
defragmentCount++;
}
}
void free(uint64_t ptr, size_t size) {
if (ptr == 0llu)
return;
std::lock_guard<std::mutex> lock(mtx);
DBG_LOG(PrintDebugMessages, __FUNCTION__, "Allocator usage == ", this->getUsage());
if (ptr == pRightBound) {
pRightBound = ptr + size;
mergeLastFreedSmall();
} else if (ptr == pLeftBound - size) {
pLeftBound = ptr;
mergeLastFreedBig();
} else if (ptr < pLeftBound) {
DEBUG_BREAK_IF(size <= sizeThreshold);
storeInFreedChunks(ptr, size, freedChunksBig);
} else {
storeInFreedChunks(ptr, size, freedChunksSmall);
}
availableSize += size;
}
uint64_t getLeftSize() const {
return availableSize;
}
uint64_t getUsedSize() const {
return size - availableSize;
}
NO_SANITIZE
double getUsage() const {
return static_cast<double>(size - availableSize) / size;
}
protected:
const uint64_t size;
uint64_t availableSize;
uint64_t pLeftBound;
uint64_t pRightBound;
const size_t sizeThreshold;
size_t allocationAlignment = MemoryConstants::pageSize;
std::vector<HeapChunk> freedChunksSmall;
std::vector<HeapChunk> freedChunksBig;
std::mutex mtx;
uint64_t getFromFreedChunks(size_t size, std::vector<HeapChunk> &freedChunks, size_t &sizeOfFreedChunk) {
size_t elements = freedChunks.size();
size_t bestFitIndex = -1;
size_t bestFitSize = 0;
sizeOfFreedChunk = 0;
for (size_t i = 0; i < elements; i++) {
if (freedChunks[i].size == size) {
auto ptr = freedChunks[i].ptr;
freedChunks.erase(freedChunks.begin() + i);
return ptr;
}
if (freedChunks[i].size > size) {
if (freedChunks[i].size < bestFitSize || bestFitSize == 0) {
bestFitIndex = i;
bestFitSize = freedChunks[i].size;
}
}
}
if (bestFitSize != 0) {
if (bestFitSize < (size << 1)) {
auto ptr = freedChunks[bestFitIndex].ptr;
sizeOfFreedChunk = freedChunks[bestFitIndex].size;
freedChunks.erase(freedChunks.begin() + bestFitIndex);
return ptr;
} else {
size_t sizeDelta = freedChunks[bestFitIndex].size - size;
DEBUG_BREAK_IF(!(size <= sizeThreshold || (size > sizeThreshold && sizeDelta > sizeThreshold)));
auto ptr = freedChunks[bestFitIndex].ptr + sizeDelta;
freedChunks[bestFitIndex].size = sizeDelta;
return ptr;
}
}
return 0llu;
}
void storeInFreedChunks(uint64_t ptr, size_t size, std::vector<HeapChunk> &freedChunks) {
for (auto &freedChunk : freedChunks) {
if (freedChunk.ptr == ptr + size) {
freedChunk.ptr = ptr;
freedChunk.size += size;
return;
}
if (freedChunk.ptr + freedChunk.size == ptr) {
freedChunk.size += size;
return;
}
}
freedChunks.emplace_back(ptr, size);
}
void mergeLastFreedSmall() {
size_t maxSizeOfSmallChunks = freedChunksSmall.size();
if (maxSizeOfSmallChunks > 0) {
auto ptr = freedChunksSmall[maxSizeOfSmallChunks - 1].ptr;
size_t chunkSize = freedChunksSmall[maxSizeOfSmallChunks - 1].size;
if (ptr == pRightBound) {
pRightBound = ptr + chunkSize;
freedChunksSmall.pop_back();
}
}
}
void mergeLastFreedBig() {
size_t maxSizeOfBigChunks = freedChunksBig.size();
if (maxSizeOfBigChunks > 0) {
auto ptr = freedChunksBig[maxSizeOfBigChunks - 1].ptr;
size_t chunkSize = freedChunksBig[maxSizeOfBigChunks - 1].size;
if (ptr == pLeftBound - chunkSize) {
pLeftBound = ptr;
freedChunksBig.pop_back();
}
}
}
void defragment() {
if (freedChunksSmall.size() > 1) {
std::sort(freedChunksSmall.rbegin(), freedChunksSmall.rend());
size_t maxSize = freedChunksSmall.size();
for (size_t i = maxSize - 1; i > 0; --i) {
auto ptr = freedChunksSmall[i].ptr;
size_t chunkSize = freedChunksSmall[i].size;
if (freedChunksSmall[i - 1].ptr == ptr + chunkSize) {
freedChunksSmall[i - 1].ptr = ptr;
freedChunksSmall[i - 1].size += chunkSize;
freedChunksSmall.erase(freedChunksSmall.begin() + i);
}
}
}
mergeLastFreedSmall();
if (freedChunksBig.size() > 1) {
std::sort(freedChunksBig.begin(), freedChunksBig.end());
size_t maxSize = freedChunksBig.size();
for (size_t i = maxSize - 1; i > 0; --i) {
auto ptr = freedChunksBig[i].ptr;
size_t chunkSize = freedChunksBig[i].size;
if ((freedChunksBig[i - 1].ptr + freedChunksBig[i - 1].size) == ptr) {
freedChunksBig[i - 1].size += chunkSize;
freedChunksBig.erase(freedChunksBig.begin() + i);
}
}
}
mergeLastFreedBig();
DBG_LOG(PrintDebugMessages, __FUNCTION__, "Allocator usage == ", this->getUsage());
}
};
} // namespace NEO
|