[1966] | 1 | /* |
---|
| 2 | Bullet Continuous Collision Detection and Physics Library |
---|
| 3 | Copyright (c) 2003-2007 Erwin Coumans http://bulletphysics.com |
---|
| 4 | |
---|
| 5 | This software is provided 'as-is', without any express or implied warranty. |
---|
| 6 | In no event will the authors be held liable for any damages arising from the use of this software. |
---|
| 7 | Permission is granted to anyone to use this software for any purpose, |
---|
| 8 | including commercial applications, and to alter it and redistribute it freely, |
---|
| 9 | subject to the following restrictions: |
---|
| 10 | |
---|
| 11 | 1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required. |
---|
| 12 | 2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software. |
---|
| 13 | 3. This notice may not be removed or altered from any source distribution. |
---|
| 14 | */ |
---|
| 15 | |
---|
| 16 | |
---|
| 17 | //#define DEBUG_SPU_TASK_SCHEDULING 1 |
---|
| 18 | |
---|
| 19 | |
---|
| 20 | //class OptimizedBvhNode; |
---|
| 21 | |
---|
| 22 | #include "SpuCollisionTaskProcess.h" |
---|
| 23 | |
---|
| 24 | |
---|
| 25 | |
---|
| 26 | |
---|
| 27 | |
---|
| 28 | |
---|
| 29 | SpuCollisionTaskProcess::SpuCollisionTaskProcess(class btThreadSupportInterface* threadInterface, unsigned int maxNumOutstandingTasks) |
---|
| 30 | :m_threadInterface(threadInterface), |
---|
| 31 | m_maxNumOutstandingTasks(maxNumOutstandingTasks) |
---|
| 32 | { |
---|
| 33 | m_workUnitTaskBuffers = (unsigned char *)0; |
---|
| 34 | m_taskBusy.resize(m_maxNumOutstandingTasks); |
---|
| 35 | m_spuGatherTaskDesc.resize(m_maxNumOutstandingTasks); |
---|
| 36 | |
---|
| 37 | for (int i = 0; i < m_maxNumOutstandingTasks; i++) |
---|
| 38 | { |
---|
| 39 | m_taskBusy[i] = false; |
---|
| 40 | } |
---|
| 41 | m_numBusyTasks = 0; |
---|
| 42 | m_currentTask = 0; |
---|
| 43 | m_currentPage = 0; |
---|
| 44 | m_currentPageEntry = 0; |
---|
| 45 | |
---|
| 46 | #ifdef DEBUG_SpuCollisionTaskProcess |
---|
| 47 | m_initialized = false; |
---|
| 48 | #endif |
---|
| 49 | |
---|
| 50 | m_threadInterface->startSPU(); |
---|
| 51 | |
---|
| 52 | //printf("sizeof vec_float4: %d\n", sizeof(vec_float4)); |
---|
| 53 | printf("sizeof SpuGatherAndProcessWorkUnitInput: %d\n", sizeof(SpuGatherAndProcessWorkUnitInput)); |
---|
| 54 | |
---|
| 55 | } |
---|
| 56 | |
---|
| 57 | SpuCollisionTaskProcess::~SpuCollisionTaskProcess() |
---|
| 58 | { |
---|
| 59 | |
---|
| 60 | if (m_workUnitTaskBuffers != 0) |
---|
| 61 | { |
---|
| 62 | btAlignedFree(m_workUnitTaskBuffers); |
---|
| 63 | m_workUnitTaskBuffers = 0; |
---|
| 64 | } |
---|
| 65 | |
---|
| 66 | |
---|
| 67 | |
---|
| 68 | m_threadInterface->stopSPU(); |
---|
| 69 | |
---|
| 70 | } |
---|
| 71 | |
---|
| 72 | |
---|
| 73 | |
---|
| 74 | void SpuCollisionTaskProcess::initialize2(bool useEpa) |
---|
| 75 | { |
---|
| 76 | |
---|
| 77 | #ifdef DEBUG_SPU_TASK_SCHEDULING |
---|
| 78 | printf("SpuCollisionTaskProcess::initialize()\n"); |
---|
| 79 | #endif //DEBUG_SPU_TASK_SCHEDULING |
---|
| 80 | if (!m_workUnitTaskBuffers) |
---|
| 81 | { |
---|
| 82 | m_workUnitTaskBuffers = (unsigned char *)btAlignedAlloc(MIDPHASE_WORKUNIT_TASK_SIZE*m_maxNumOutstandingTasks, 128); |
---|
| 83 | } |
---|
| 84 | |
---|
| 85 | |
---|
| 86 | for (int i = 0; i < m_maxNumOutstandingTasks; i++) |
---|
| 87 | { |
---|
| 88 | m_taskBusy[i] = false; |
---|
| 89 | } |
---|
| 90 | m_numBusyTasks = 0; |
---|
| 91 | m_currentTask = 0; |
---|
| 92 | m_currentPage = 0; |
---|
| 93 | m_currentPageEntry = 0; |
---|
| 94 | m_useEpa = useEpa; |
---|
| 95 | |
---|
| 96 | #ifdef DEBUG_SpuCollisionTaskProcess |
---|
| 97 | m_initialized = true; |
---|
| 98 | assert(MIDPHASE_NUM_WORKUNITS_PER_TASK*sizeof(SpuGatherAndProcessWorkUnitInput) <= MIDPHASE_WORKUNIT_TASK_SIZE); |
---|
| 99 | #endif |
---|
| 100 | } |
---|
| 101 | |
---|
| 102 | |
---|
| 103 | void SpuCollisionTaskProcess::issueTask2() |
---|
| 104 | { |
---|
| 105 | |
---|
| 106 | #ifdef DEBUG_SPU_TASK_SCHEDULING |
---|
| 107 | printf("SpuCollisionTaskProcess::issueTask (m_currentTask= %d\n)", m_currentTask); |
---|
| 108 | #endif //DEBUG_SPU_TASK_SCHEDULING |
---|
| 109 | |
---|
| 110 | m_taskBusy[m_currentTask] = true; |
---|
| 111 | m_numBusyTasks++; |
---|
| 112 | |
---|
| 113 | |
---|
| 114 | SpuGatherAndProcessPairsTaskDesc& taskDesc = m_spuGatherTaskDesc[m_currentTask]; |
---|
| 115 | taskDesc.m_useEpa = m_useEpa; |
---|
| 116 | |
---|
| 117 | { |
---|
| 118 | // send task description in event message |
---|
| 119 | // no error checking here... |
---|
| 120 | // but, currently, event queue can be no larger than NUM_WORKUNIT_TASKS. |
---|
| 121 | |
---|
| 122 | taskDesc.inPtr = reinterpret_cast<uint64_t>(MIDPHASE_TASK_PTR(m_currentTask)); |
---|
| 123 | |
---|
| 124 | taskDesc.taskId = m_currentTask; |
---|
| 125 | taskDesc.numPages = m_currentPage+1; |
---|
| 126 | taskDesc.numOnLastPage = m_currentPageEntry; |
---|
| 127 | } |
---|
| 128 | |
---|
| 129 | |
---|
| 130 | |
---|
| 131 | m_threadInterface->sendRequest(CMD_GATHER_AND_PROCESS_PAIRLIST, (uint32_t) &taskDesc,m_currentTask); |
---|
| 132 | |
---|
| 133 | // if all tasks busy, wait for spu event to clear the task. |
---|
| 134 | |
---|
| 135 | |
---|
| 136 | if (m_numBusyTasks >= m_maxNumOutstandingTasks) |
---|
| 137 | { |
---|
| 138 | unsigned int taskId; |
---|
| 139 | unsigned int outputSize; |
---|
| 140 | |
---|
| 141 | |
---|
| 142 | for (int i=0;i<m_maxNumOutstandingTasks;i++) |
---|
| 143 | { |
---|
| 144 | if (m_taskBusy[i]) |
---|
| 145 | { |
---|
| 146 | taskId = i; |
---|
| 147 | break; |
---|
| 148 | } |
---|
| 149 | } |
---|
| 150 | |
---|
| 151 | btAssert(taskId>=0); |
---|
| 152 | |
---|
| 153 | |
---|
| 154 | m_threadInterface->waitForResponse(&taskId, &outputSize); |
---|
| 155 | |
---|
| 156 | // printf("issueTask taskId %d completed, numBusy=%d\n",taskId,m_numBusyTasks); |
---|
| 157 | |
---|
| 158 | //printf("PPU: after issue, received event: %u %d\n", taskId, outputSize); |
---|
| 159 | |
---|
| 160 | //postProcess(taskId, outputSize); |
---|
| 161 | |
---|
| 162 | m_taskBusy[taskId] = false; |
---|
| 163 | |
---|
| 164 | m_numBusyTasks--; |
---|
| 165 | } |
---|
| 166 | |
---|
| 167 | } |
---|
| 168 | |
---|
| 169 | void SpuCollisionTaskProcess::addWorkToTask(void* pairArrayPtr,int startIndex,int endIndex) |
---|
| 170 | { |
---|
| 171 | #ifdef DEBUG_SPU_TASK_SCHEDULING |
---|
| 172 | printf("#"); |
---|
| 173 | #endif //DEBUG_SPU_TASK_SCHEDULING |
---|
| 174 | |
---|
| 175 | #ifdef DEBUG_SpuCollisionTaskProcess |
---|
| 176 | assert(m_initialized); |
---|
| 177 | assert(m_workUnitTaskBuffers); |
---|
| 178 | |
---|
| 179 | #endif |
---|
| 180 | |
---|
| 181 | bool batch = true; |
---|
| 182 | |
---|
| 183 | if (batch) |
---|
| 184 | { |
---|
| 185 | if (m_currentPageEntry == MIDPHASE_NUM_WORKUNITS_PER_PAGE) |
---|
| 186 | { |
---|
| 187 | if (m_currentPage == MIDPHASE_NUM_WORKUNIT_PAGES-1) |
---|
| 188 | { |
---|
| 189 | // task buffer is full, issue current task. |
---|
| 190 | // if all task buffers busy, this waits until SPU is done. |
---|
| 191 | issueTask2(); |
---|
| 192 | |
---|
| 193 | // find new task buffer |
---|
| 194 | for (unsigned int i = 0; i < m_maxNumOutstandingTasks; i++) |
---|
| 195 | { |
---|
| 196 | if (!m_taskBusy[i]) |
---|
| 197 | { |
---|
| 198 | m_currentTask = i; |
---|
| 199 | //init the task data |
---|
| 200 | |
---|
| 201 | break; |
---|
| 202 | } |
---|
| 203 | } |
---|
| 204 | |
---|
| 205 | m_currentPage = 0; |
---|
| 206 | } |
---|
| 207 | else |
---|
| 208 | { |
---|
| 209 | m_currentPage++; |
---|
| 210 | } |
---|
| 211 | |
---|
| 212 | m_currentPageEntry = 0; |
---|
| 213 | } |
---|
| 214 | } |
---|
| 215 | |
---|
| 216 | { |
---|
| 217 | |
---|
| 218 | |
---|
| 219 | |
---|
| 220 | SpuGatherAndProcessWorkUnitInput &wuInput = |
---|
| 221 | *(reinterpret_cast<SpuGatherAndProcessWorkUnitInput*> |
---|
| 222 | (MIDPHASE_ENTRY_PTR(m_currentTask, m_currentPage, m_currentPageEntry))); |
---|
| 223 | |
---|
| 224 | wuInput.m_pairArrayPtr = reinterpret_cast<uint64_t>(pairArrayPtr); |
---|
| 225 | wuInput.m_startIndex = startIndex; |
---|
| 226 | wuInput.m_endIndex = endIndex; |
---|
| 227 | |
---|
| 228 | |
---|
| 229 | |
---|
| 230 | m_currentPageEntry++; |
---|
| 231 | |
---|
| 232 | if (!batch) |
---|
| 233 | { |
---|
| 234 | issueTask2(); |
---|
| 235 | |
---|
| 236 | // find new task buffer |
---|
| 237 | for (unsigned int i = 0; i < m_maxNumOutstandingTasks; i++) |
---|
| 238 | { |
---|
| 239 | if (!m_taskBusy[i]) |
---|
| 240 | { |
---|
| 241 | m_currentTask = i; |
---|
| 242 | //init the task data |
---|
| 243 | |
---|
| 244 | break; |
---|
| 245 | } |
---|
| 246 | } |
---|
| 247 | |
---|
| 248 | m_currentPage = 0; |
---|
| 249 | m_currentPageEntry =0; |
---|
| 250 | } |
---|
| 251 | } |
---|
| 252 | } |
---|
| 253 | |
---|
| 254 | |
---|
| 255 | void |
---|
| 256 | SpuCollisionTaskProcess::flush2() |
---|
| 257 | { |
---|
| 258 | #ifdef DEBUG_SPU_TASK_SCHEDULING |
---|
| 259 | printf("\nSpuCollisionTaskProcess::flush()\n"); |
---|
| 260 | #endif //DEBUG_SPU_TASK_SCHEDULING |
---|
| 261 | |
---|
| 262 | // if there's a partially filled task buffer, submit that task |
---|
| 263 | if (m_currentPage > 0 || m_currentPageEntry > 0) |
---|
| 264 | { |
---|
| 265 | issueTask2(); |
---|
| 266 | } |
---|
| 267 | |
---|
| 268 | |
---|
| 269 | // all tasks are issued, wait for all tasks to be complete |
---|
| 270 | while(m_numBusyTasks > 0) |
---|
| 271 | { |
---|
| 272 | // Consolidating SPU code |
---|
| 273 | unsigned int taskId=-1; |
---|
| 274 | unsigned int outputSize; |
---|
| 275 | |
---|
| 276 | for (int i=0;i<m_maxNumOutstandingTasks;i++) |
---|
| 277 | { |
---|
| 278 | if (m_taskBusy[i]) |
---|
| 279 | { |
---|
| 280 | taskId = i; |
---|
| 281 | break; |
---|
| 282 | } |
---|
| 283 | } |
---|
| 284 | |
---|
| 285 | btAssert(taskId>=0); |
---|
| 286 | |
---|
| 287 | |
---|
| 288 | { |
---|
| 289 | |
---|
| 290 | // SPURS support. |
---|
| 291 | m_threadInterface->waitForResponse(&taskId, &outputSize); |
---|
| 292 | } |
---|
| 293 | // printf("flush2 taskId %d completed, numBusy =%d \n",taskId,m_numBusyTasks); |
---|
| 294 | //printf("PPU: flushing, received event: %u %d\n", taskId, outputSize); |
---|
| 295 | |
---|
| 296 | //postProcess(taskId, outputSize); |
---|
| 297 | |
---|
| 298 | m_taskBusy[taskId] = false; |
---|
| 299 | |
---|
| 300 | m_numBusyTasks--; |
---|
| 301 | } |
---|
| 302 | |
---|
| 303 | |
---|
| 304 | } |
---|