| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2013 x265 project |
| 3 | * |
| 4 | * Authors: Steve Borho <steve@borho.org> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. |
| 19 | * |
| 20 | * This program is also available under a commercial proprietary license. |
| 21 | * For more information, contact us at license @ x265.com |
| 22 | *****************************************************************************/ |
| 23 | |
| 24 | #include "common.h" |
| 25 | #include "threadpool.h" |
| 26 | #include "wavefront.h" |
| 27 | #include "threading.h" |
| 28 | #include "md5.h" |
| 29 | #include "PPA/ppa.h" |
| 30 | |
| 31 | #include <sstream> |
| 32 | #include <iostream> |
| 33 | |
| 34 | using namespace x265; |
| 35 | |
| 36 | struct CUData |
| 37 | { |
| 38 | CUData() |
| 39 | { |
| 40 | memset(digest, 0, sizeof(digest)); |
| 41 | } |
| 42 | |
| 43 | unsigned char digest[16]; |
| 44 | }; |
| 45 | |
| 46 | struct RowData |
| 47 | { |
| 48 | RowData() : active(false), curCol(0) {} |
| 49 | |
| 50 | Lock lock; |
| 51 | volatile bool active; |
| 52 | volatile int curCol; |
| 53 | }; |
| 54 | |
| 55 | // Create a fake frame class with manufactured data in each CU block. We |
| 56 | // need to create an MD5 hash such that each CU's hash includes the hashes |
| 57 | // of the blocks that would have HEVC data dependencies (left, top-left, |
| 58 | // top, top-right). This will give us one deterministic output hash. We |
| 59 | // then generate the same hash using the thread pool and wave-front parallelism |
| 60 | // to verify the thread-pool behavior and the wave-front schedule data |
| 61 | // structures. |
| 62 | class MD5Frame : public WaveFront |
| 63 | { |
| 64 | private: |
| 65 | |
| 66 | CUData *cu; |
| 67 | RowData *row; |
| 68 | int numrows; |
| 69 | int numcols; |
| 70 | Event complete; |
| 71 | |
| 72 | public: |
| 73 | |
| 74 | MD5Frame(ThreadPool *pool) : WaveFront(pool), cu(0), row(0) {} |
| 75 | |
| 76 | virtual ~MD5Frame() |
| 77 | { |
| 78 | // ensure no threads are lingering on FindJob() before allowing |
| 79 | // this object's vtable to be destroyed |
| 80 | JobProvider::flush(); |
| 81 | |
| 82 | delete[] this->cu; |
| 83 | delete[] this->row; |
| 84 | } |
| 85 | |
| 86 | void initialize(int cols, int rows); |
| 87 | |
| 88 | void encode(); |
| 89 | |
| 90 | void processRow(int row, int threadid); |
| 91 | }; |
| 92 | |
| 93 | void MD5Frame::initialize(int cols, int rows) |
| 94 | { |
| 95 | this->cu = new CUData[rows * cols]; |
| 96 | this->row = new RowData[rows]; |
| 97 | this->numrows = rows; |
| 98 | this->numcols = cols; |
| 99 | |
| 100 | if (!this->WaveFront::init(rows)) |
| 101 | { |
| 102 | assert(!"Unable to initialize job queue"); |
| 103 | } |
| 104 | } |
| 105 | |
| 106 | void MD5Frame::encode() |
| 107 | { |
| 108 | this->JobProvider::enqueue(); |
| 109 | |
| 110 | this->WaveFront::enqueueRow(0); |
| 111 | |
| 112 | // NOTE: When EnableRow after enqueueRow at first row, we'd better call pokeIdleThread, it will release a thread to do job |
| 113 | this->WaveFront::enableRow(0); |
| 114 | this->m_pool->pokeIdleThread(); |
| 115 | |
| 116 | this->complete.wait(); |
| 117 | |
| 118 | this->JobProvider::dequeue(); |
| 119 | |
| 120 | unsigned int *outdigest = (unsigned int*)this->cu[this->numrows * this->numcols - 1].digest; |
| 121 | |
| 122 | std::stringstream ss; |
| 123 | |
| 124 | for (int i = 0; i < 4; i++) |
| 125 | { |
| 126 | ss << std::hex << outdigest[i]; |
| 127 | } |
| 128 | |
| 129 | if (ss.str().compare("da667b741a7a9d0ee862158da2dd1882")) |
| 130 | std::cout << "Bad hash: " << ss.str() << std::endl; |
| 131 | } |
| 132 | |
| 133 | void MD5Frame::processRow(int rownum, int) |
| 134 | { |
| 135 | // Called by worker thread |
| 136 | RowData &curRow = this->row[rownum]; |
| 137 | |
| 138 | assert(rownum < this->numrows && rownum >= 0); |
| 139 | assert(curRow.curCol < this->numcols); |
| 140 | |
| 141 | while (curRow.curCol < this->numcols) |
| 142 | { |
| 143 | int id = rownum * this->numcols + curRow.curCol; |
| 144 | CUData &curCTU = this->cu[id]; |
| 145 | MD5 hash; |
| 146 | |
| 147 | // * Fake CTU processing * |
| 148 | PPAStartCpuEventFunc(encode_block); |
| 149 | memset(curCTU.digest, id, sizeof(curCTU.digest)); |
| 150 | hash.update(curCTU.digest, sizeof(curCTU.digest)); |
| 151 | if (curRow.curCol > 0) |
| 152 | hash.update(this->cu[id - 1].digest, sizeof(curCTU.digest)); |
| 153 | |
| 154 | if (rownum > 0) |
| 155 | { |
| 156 | if (curRow.curCol > 0) |
| 157 | hash.update(this->cu[id - this->numcols - 1].digest, sizeof(curCTU.digest)); |
| 158 | |
| 159 | hash.update(this->cu[id - this->numcols].digest, sizeof(curCTU.digest)); |
| 160 | if (curRow.curCol < this->numcols - 1) |
| 161 | hash.update(this->cu[id - this->numcols + 1].digest, sizeof(curCTU.digest)); |
| 162 | } |
| 163 | |
| 164 | hash.finalize(curCTU.digest); |
| 165 | PPAStopCpuEventFunc(encode_block); |
| 166 | |
| 167 | curRow.curCol++; |
| 168 | |
| 169 | if (curRow.curCol >= 2 && rownum < this->numrows - 1) |
| 170 | { |
| 171 | ScopedLock below(this->row[rownum + 1].lock); |
| 172 | |
| 173 | if (this->row[rownum + 1].active == false && |
| 174 | this->row[rownum + 1].curCol + 2 <= curRow.curCol) |
| 175 | { |
| 176 | // set active indicator so row is only enqueued once |
| 177 | // row stays marked active until blocked or done |
| 178 | this->row[rownum + 1].active = true; |
| 179 | this->WaveFront::enqueueRow(rownum + 1); |
| 180 | this->WaveFront::enableRow(rownum + 1); |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | ScopedLock self(curRow.lock); |
| 185 | |
| 186 | if (rownum > 0 && |
| 187 | curRow.curCol < this->numcols - 1 && |
| 188 | this->row[rownum - 1].curCol < curRow.curCol + 2) |
| 189 | { |
| 190 | // row is blocked, quit job |
| 191 | curRow.active = false; |
| 192 | return; |
| 193 | } |
| 194 | } |
| 195 | |
| 196 | // * Row completed * |
| 197 | |
| 198 | if (rownum == this->numrows - 1) |
| 199 | this->complete.trigger(); |
| 200 | } |
| 201 | |
| 202 | int main(int, char **) |
| 203 | { |
| 204 | ThreadPool *pool; |
| 205 | |
| 206 | PPA_INIT(); |
| 207 | |
| 208 | pool = ThreadPool::allocThreadPool(1); |
| 209 | { |
| 210 | MD5Frame frame(pool); |
| 211 | frame.initialize(60, 40); |
| 212 | frame.encode(); |
| 213 | } |
| 214 | pool->release(); |
| 215 | pool = ThreadPool::allocThreadPool(2); |
| 216 | { |
| 217 | MD5Frame frame(pool); |
| 218 | frame.initialize(60, 40); |
| 219 | frame.encode(); |
| 220 | } |
| 221 | pool->release(); |
| 222 | pool = ThreadPool::allocThreadPool(4); |
| 223 | { |
| 224 | MD5Frame frame(pool); |
| 225 | frame.initialize(60, 40); |
| 226 | frame.encode(); |
| 227 | } |
| 228 | pool->release(); |
| 229 | pool = ThreadPool::allocThreadPool(8); |
| 230 | { |
| 231 | MD5Frame frame(pool); |
| 232 | frame.initialize(60, 40); |
| 233 | frame.encode(); |
| 234 | } |
| 235 | pool->release(); |
| 236 | |
| 237 | return 0; |
| 238 | } |