rsdBcc.cpp revision cbb0b8aceedb9146ba901cba1fbd6d1e9ca88f51
1/* 2 * Copyright (C) 2011 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 18#include "rsdCore.h" 19#include "rsdBcc.h" 20#include "rsdRuntime.h" 21 22#include <bcinfo/MetadataExtractor.h> 23 24#include "rsContext.h" 25#include "rsScriptC.h" 26 27#include "utils/Timers.h" 28#include "utils/StopWatch.h" 29extern "C" { 30#include "libdex/ZipArchive.h" 31} 32 33 34using namespace android; 35using namespace android::renderscript; 36 37struct DrvScript { 38 int (*mRoot)(); 39 void (*mInit)(); 40 41 BCCScriptRef mBccScript; 42 43 bcinfo::MetadataExtractor *ME; 44 45 InvokeFunc_t *mInvokeFunctions; 46 void ** mFieldAddress; 47 bool * mFieldIsObject; 48 49 const uint8_t * mScriptText; 50 uint32_t mScriptTextLength; 51}; 52 53 54static Script * setTLS(Script *sc) { 55 ScriptTLSStruct * tls = (ScriptTLSStruct *)pthread_getspecific(rsdgThreadTLSKey); 56 rsAssert(tls); 57 Script *old = tls->mScript; 58 tls->mScript = sc; 59 return old; 60} 61 62 63bool rsdScriptInit(const Context *rsc, 64 ScriptC *script, 65 char const *resName, 66 char const *cacheDir, 67 uint8_t const *bitcode, 68 size_t bitcodeSize, 69 uint32_t flags) { 70 //LOGE("rsdScriptCreate %p %p %p %p %i %i %p", rsc, resName, cacheDir, bitcode, bitcodeSize, flags, lookupFunc); 71 72 pthread_mutex_lock(&rsdgInitMutex); 73 char *cachePath = NULL; 74 size_t exportFuncCount = 0; 75 size_t exportVarCount = 0; 76 size_t objectSlotCount = 0; 77 78 DrvScript *drv = (DrvScript *)calloc(1, sizeof(DrvScript)); 79 if (drv == NULL) { 80 goto error; 81 } 82 script->mHal.drv = drv; 83 84 drv->mBccScript = bccCreateScript(); 85 script->mHal.info.isThreadable = true; 86 drv->mScriptText = bitcode; 87 drv->mScriptTextLength = bitcodeSize; 88 89 90 drv->ME = new bcinfo::MetadataExtractor((const char*)drv->mScriptText, 91 drv->mScriptTextLength); 92 if (!drv->ME->extract()) { 93 LOGE("bcinfo: failed to read script metadata"); 94 goto error; 95 } 96 97 //LOGE("mBccScript %p", script->mBccScript); 98 99 if (bccRegisterSymbolCallback(drv->mBccScript, &rsdLookupRuntimeStub, script) != 0) { 100 LOGE("bcc: FAILS to register symbol callback"); 101 goto error; 102 } 103 104 if (bccReadBC(drv->mBccScript, 105 resName, 106 (char const *)drv->mScriptText, 107 drv->mScriptTextLength, 0) != 0) { 108 LOGE("bcc: FAILS to read bitcode"); 109 goto error; 110 } 111 112 if (bccLinkFile(drv->mBccScript, "/system/lib/libclcore.bc", 0) != 0) { 113 LOGE("bcc: FAILS to link bitcode"); 114 goto error; 115 } 116 117 if (bccPrepareExecutableEx(drv->mBccScript, cacheDir, resName, 0) != 0) { 118 LOGE("bcc: FAILS to prepare executable"); 119 goto error; 120 } 121 122 free(cachePath); 123 124 drv->mRoot = reinterpret_cast<int (*)()>(bccGetFuncAddr(drv->mBccScript, "root")); 125 drv->mInit = reinterpret_cast<void (*)()>(bccGetFuncAddr(drv->mBccScript, "init")); 126 127 exportFuncCount = drv->ME->getExportFuncCount(); 128 if (exportFuncCount > 0) { 129 drv->mInvokeFunctions = (InvokeFunc_t*) calloc(exportFuncCount, 130 sizeof(InvokeFunc_t)); 131 bccGetExportFuncList(drv->mBccScript, exportFuncCount, 132 (void **) drv->mInvokeFunctions); 133 } else { 134 drv->mInvokeFunctions = NULL; 135 } 136 137 exportVarCount = drv->ME->getExportVarCount(); 138 if (exportVarCount > 0) { 139 drv->mFieldAddress = (void **) calloc(exportVarCount, sizeof(void*)); 140 drv->mFieldIsObject = (bool *) calloc(exportVarCount, sizeof(bool)); 141 bccGetExportVarList(drv->mBccScript, exportVarCount, 142 (void **) drv->mFieldAddress); 143 } else { 144 drv->mFieldAddress = NULL; 145 drv->mFieldIsObject = NULL; 146 } 147 148 objectSlotCount = drv->ME->getObjectSlotCount(); 149 if (objectSlotCount > 0) { 150 const uint32_t *objectSlotList = drv->ME->getObjectSlotList(); 151 for (uint32_t ct=0; ct < objectSlotCount; ct++) { 152 drv->mFieldIsObject[objectSlotList[ct]] = true; 153 } 154 } 155 156 // Copy info over to runtime 157 script->mHal.info.exportedFunctionCount = drv->ME->getExportFuncCount(); 158 script->mHal.info.exportedVariableCount = drv->ME->getExportVarCount(); 159 script->mHal.info.exportedPragmaCount = drv->ME->getPragmaCount(); 160 script->mHal.info.exportedPragmaKeyList = drv->ME->getPragmaKeyList(); 161 script->mHal.info.exportedPragmaValueList = drv->ME->getPragmaValueList(); 162 script->mHal.info.root = drv->mRoot; 163 164 pthread_mutex_unlock(&rsdgInitMutex); 165 return true; 166 167error: 168 169 pthread_mutex_unlock(&rsdgInitMutex); 170 if (drv->ME) { 171 delete drv->ME; 172 drv->ME = NULL; 173 } 174 free(drv); 175 return false; 176 177} 178 179typedef struct { 180 Context *rsc; 181 Script *script; 182 const Allocation * ain; 183 Allocation * aout; 184 const void * usr; 185 186 uint32_t mSliceSize; 187 volatile int mSliceNum; 188 189 const uint8_t *ptrIn; 190 uint32_t eStrideIn; 191 uint8_t *ptrOut; 192 uint32_t eStrideOut; 193 194 uint32_t xStart; 195 uint32_t xEnd; 196 uint32_t yStart; 197 uint32_t yEnd; 198 uint32_t zStart; 199 uint32_t zEnd; 200 uint32_t arrayStart; 201 uint32_t arrayEnd; 202 203 uint32_t dimX; 204 uint32_t dimY; 205 uint32_t dimZ; 206 uint32_t dimArray; 207} MTLaunchStruct; 208typedef int (*rs_t)(const void *, void *, const void *, uint32_t, uint32_t, uint32_t, uint32_t); 209 210static void wc_xy(void *usr, uint32_t idx) { 211 MTLaunchStruct *mtls = (MTLaunchStruct *)usr; 212 213 while (1) { 214 uint32_t slice = (uint32_t)android_atomic_inc(&mtls->mSliceNum); 215 uint32_t yStart = mtls->yStart + slice * mtls->mSliceSize; 216 uint32_t yEnd = yStart + mtls->mSliceSize; 217 yEnd = rsMin(yEnd, mtls->yEnd); 218 if (yEnd <= yStart) { 219 return; 220 } 221 222 //LOGE("usr idx %i, x %i,%i y %i,%i", idx, mtls->xStart, mtls->xEnd, yStart, yEnd); 223 //LOGE("usr ptr in %p, out %p", mtls->ptrIn, mtls->ptrOut); 224 for (uint32_t y = yStart; y < yEnd; y++) { 225 uint32_t offset = mtls->dimX * y; 226 uint8_t *xPtrOut = mtls->ptrOut + (mtls->eStrideOut * offset); 227 const uint8_t *xPtrIn = mtls->ptrIn + (mtls->eStrideIn * offset); 228 229 for (uint32_t x = mtls->xStart; x < mtls->xEnd; x++) { 230 ((rs_t)mtls->script->mHal.info.root) (xPtrIn, xPtrOut, mtls->usr, x, y, 0, 0); 231 xPtrIn += mtls->eStrideIn; 232 xPtrOut += mtls->eStrideOut; 233 } 234 } 235 } 236} 237 238static void wc_x(void *usr, uint32_t idx) { 239 MTLaunchStruct *mtls = (MTLaunchStruct *)usr; 240 241 while (1) { 242 uint32_t slice = (uint32_t)android_atomic_inc(&mtls->mSliceNum); 243 uint32_t xStart = mtls->xStart + slice * mtls->mSliceSize; 244 uint32_t xEnd = xStart + mtls->mSliceSize; 245 xEnd = rsMin(xEnd, mtls->xEnd); 246 if (xEnd <= xStart) { 247 return; 248 } 249 250 //LOGE("usr idx %i, x %i,%i y %i,%i", idx, mtls->xStart, mtls->xEnd, yStart, yEnd); 251 //LOGE("usr ptr in %p, out %p", mtls->ptrIn, mtls->ptrOut); 252 uint8_t *xPtrOut = mtls->ptrOut + (mtls->eStrideOut * xStart); 253 const uint8_t *xPtrIn = mtls->ptrIn + (mtls->eStrideIn * xStart); 254 for (uint32_t x = xStart; x < xEnd; x++) { 255 ((rs_t)mtls->script->mHal.info.root) (xPtrIn, xPtrOut, mtls->usr, x, 0, 0, 0); 256 xPtrIn += mtls->eStrideIn; 257 xPtrOut += mtls->eStrideOut; 258 } 259 } 260} 261 262void rsdScriptInvokeForEach(const Context *rsc, 263 Script *s, 264 uint32_t slot, 265 const Allocation * ain, 266 Allocation * aout, 267 const void * usr, 268 uint32_t usrLen, 269 const RsScriptCall *sc) { 270 271 RsdHal * dc = (RsdHal *)rsc->mHal.drv; 272 273 MTLaunchStruct mtls; 274 memset(&mtls, 0, sizeof(mtls)); 275 276 if (ain) { 277 mtls.dimX = ain->getType()->getDimX(); 278 mtls.dimY = ain->getType()->getDimY(); 279 mtls.dimZ = ain->getType()->getDimZ(); 280 //mtls.dimArray = ain->getType()->getDimArray(); 281 } else if (aout) { 282 mtls.dimX = aout->getType()->getDimX(); 283 mtls.dimY = aout->getType()->getDimY(); 284 mtls.dimZ = aout->getType()->getDimZ(); 285 //mtls.dimArray = aout->getType()->getDimArray(); 286 } else { 287 rsc->setError(RS_ERROR_BAD_SCRIPT, "rsForEach called with null allocations"); 288 return; 289 } 290 291 if (!sc || (sc->xEnd == 0)) { 292 mtls.xEnd = mtls.dimX; 293 } else { 294 rsAssert(sc->xStart < mtls.dimX); 295 rsAssert(sc->xEnd <= mtls.dimX); 296 rsAssert(sc->xStart < sc->xEnd); 297 mtls.xStart = rsMin(mtls.dimX, sc->xStart); 298 mtls.xEnd = rsMin(mtls.dimX, sc->xEnd); 299 if (mtls.xStart >= mtls.xEnd) return; 300 } 301 302 if (!sc || (sc->yEnd == 0)) { 303 mtls.yEnd = mtls.dimY; 304 } else { 305 rsAssert(sc->yStart < mtls.dimY); 306 rsAssert(sc->yEnd <= mtls.dimY); 307 rsAssert(sc->yStart < sc->yEnd); 308 mtls.yStart = rsMin(mtls.dimY, sc->yStart); 309 mtls.yEnd = rsMin(mtls.dimY, sc->yEnd); 310 if (mtls.yStart >= mtls.yEnd) return; 311 } 312 313 mtls.xEnd = rsMax((uint32_t)1, mtls.xEnd); 314 mtls.yEnd = rsMax((uint32_t)1, mtls.yEnd); 315 mtls.zEnd = rsMax((uint32_t)1, mtls.zEnd); 316 mtls.arrayEnd = rsMax((uint32_t)1, mtls.arrayEnd); 317 318 rsAssert(!ain || (ain->getType()->getDimZ() == 0)); 319 320 Context *mrsc = (Context *)rsc; 321 Script * oldTLS = setTLS(s); 322 323 mtls.rsc = mrsc; 324 mtls.ain = ain; 325 mtls.aout = aout; 326 mtls.script = s; 327 mtls.usr = usr; 328 mtls.mSliceSize = 10; 329 mtls.mSliceNum = 0; 330 331 mtls.ptrIn = NULL; 332 mtls.eStrideIn = 0; 333 if (ain) { 334 mtls.ptrIn = (const uint8_t *)ain->getPtr(); 335 mtls.eStrideIn = ain->getType()->getElementSizeBytes(); 336 } 337 338 mtls.ptrOut = NULL; 339 mtls.eStrideOut = 0; 340 if (aout) { 341 mtls.ptrOut = (uint8_t *)aout->getPtr(); 342 mtls.eStrideOut = aout->getType()->getElementSizeBytes(); 343 } 344 345 if ((dc->mWorkers.mCount > 1) && s->mHal.info.isThreadable) { 346 if (mtls.dimY > 1) { 347 rsdLaunchThreads(mrsc, wc_xy, &mtls); 348 } else { 349 rsdLaunchThreads(mrsc, wc_x, &mtls); 350 } 351 352 //LOGE("launch 1"); 353 } else { 354 //LOGE("launch 3"); 355 for (uint32_t ar = mtls.arrayStart; ar < mtls.arrayEnd; ar++) { 356 for (uint32_t z = mtls.zStart; z < mtls.zEnd; z++) { 357 for (uint32_t y = mtls.yStart; y < mtls.yEnd; y++) { 358 uint32_t offset = mtls.dimX * mtls.dimY * mtls.dimZ * ar + 359 mtls.dimX * mtls.dimY * z + 360 mtls.dimX * y; 361 uint8_t *xPtrOut = mtls.ptrOut + (mtls.eStrideOut * offset); 362 const uint8_t *xPtrIn = mtls.ptrIn + (mtls.eStrideIn * offset); 363 364 for (uint32_t x = mtls.xStart; x < mtls.xEnd; x++) { 365 ((rs_t)s->mHal.info.root) (xPtrIn, xPtrOut, usr, x, y, z, ar); 366 xPtrIn += mtls.eStrideIn; 367 xPtrOut += mtls.eStrideOut; 368 } 369 } 370 } 371 } 372 } 373 374 setTLS(oldTLS); 375} 376 377 378int rsdScriptInvokeRoot(const Context *dc, Script *script) { 379 DrvScript *drv = (DrvScript *)script->mHal.drv; 380 381 Script * oldTLS = setTLS(script); 382 int ret = drv->mRoot(); 383 setTLS(oldTLS); 384 385 return ret; 386} 387 388void rsdScriptInvokeInit(const Context *dc, Script *script) { 389 DrvScript *drv = (DrvScript *)script->mHal.drv; 390 391 if (drv->mInit) { 392 drv->mInit(); 393 } 394} 395 396 397void rsdScriptInvokeFunction(const Context *dc, Script *script, 398 uint32_t slot, 399 const void *params, 400 size_t paramLength) { 401 DrvScript *drv = (DrvScript *)script->mHal.drv; 402 //LOGE("invoke %p %p %i %p %i", dc, script, slot, params, paramLength); 403 404 Script * oldTLS = setTLS(script); 405 ((void (*)(const void *, uint32_t)) 406 drv->mInvokeFunctions[slot])(params, paramLength); 407 setTLS(oldTLS); 408} 409 410void rsdScriptSetGlobalVar(const Context *dc, const Script *script, 411 uint32_t slot, void *data, size_t dataLength) { 412 DrvScript *drv = (DrvScript *)script->mHal.drv; 413 //rsAssert(!script->mFieldIsObject[slot]); 414 //LOGE("setGlobalVar %p %p %i %p %i", dc, script, slot, data, dataLength); 415 416 int32_t *destPtr = ((int32_t **)drv->mFieldAddress)[slot]; 417 if (!destPtr) { 418 //LOGV("Calling setVar on slot = %i which is null", slot); 419 return; 420 } 421 422 memcpy(destPtr, data, dataLength); 423} 424 425void rsdScriptSetGlobalBind(const Context *dc, const Script *script, uint32_t slot, void *data) { 426 DrvScript *drv = (DrvScript *)script->mHal.drv; 427 //rsAssert(!script->mFieldIsObject[slot]); 428 //LOGE("setGlobalBind %p %p %i %p", dc, script, slot, data); 429 430 int32_t *destPtr = ((int32_t **)drv->mFieldAddress)[slot]; 431 if (!destPtr) { 432 //LOGV("Calling setVar on slot = %i which is null", slot); 433 return; 434 } 435 436 memcpy(destPtr, &data, sizeof(void *)); 437} 438 439void rsdScriptSetGlobalObj(const Context *dc, const Script *script, uint32_t slot, ObjectBase *data) { 440 DrvScript *drv = (DrvScript *)script->mHal.drv; 441 //rsAssert(script->mFieldIsObject[slot]); 442 //LOGE("setGlobalObj %p %p %i %p", dc, script, slot, data); 443 444 int32_t *destPtr = ((int32_t **)drv->mFieldAddress)[slot]; 445 if (!destPtr) { 446 //LOGV("Calling setVar on slot = %i which is null", slot); 447 return; 448 } 449 450 rsrSetObject(dc, script, (ObjectBase **)destPtr, data); 451} 452 453void rsdScriptDestroy(const Context *dc, Script *script) { 454 DrvScript *drv = (DrvScript *)script->mHal.drv; 455 456 if (drv->mFieldAddress) { 457 size_t exportVarCount = drv->ME->getExportVarCount(); 458 for (size_t ct = 0; ct < exportVarCount; ct++) { 459 if (drv->mFieldIsObject[ct]) { 460 // The field address can be NULL if the script-side has 461 // optimized the corresponding global variable away. 462 if (drv->mFieldAddress[ct]) { 463 rsrClearObject(dc, script, (ObjectBase **)drv->mFieldAddress[ct]); 464 } 465 } 466 } 467 free(drv->mFieldAddress); 468 drv->mFieldAddress = NULL; 469 free(drv->mFieldIsObject); 470 drv->mFieldIsObject = NULL; } 471 472 if (drv->mInvokeFunctions) { 473 free(drv->mInvokeFunctions); 474 drv->mInvokeFunctions = NULL; 475 } 476 477 delete drv->ME; 478 drv->ME = NULL; 479 480 free(drv); 481 script->mHal.drv = NULL; 482 483} 484 485 486