rsScriptC.cpp revision 2353ae303868d04e3a26002b2f2dc456c15e8170
1/* 2 * Copyright (C) 2009 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#include "rsContext.h" 18#include "rsScriptC.h" 19#include "rsMatrix.h" 20#include "../../compile/libbcc/include/bcc/bcc.h" 21#include "utils/Timers.h" 22#include "utils/StopWatch.h" 23 24#include <GLES/gl.h> 25#include <GLES/glext.h> 26 27using namespace android; 28using namespace android::renderscript; 29 30#define GET_TLS() Context::ScriptTLSStruct * tls = \ 31 (Context::ScriptTLSStruct *)pthread_getspecific(Context::gThreadTLSKey); \ 32 Context * rsc = tls->mContext; \ 33 ScriptC * sc = (ScriptC *) tls->mScript 34 35 36ScriptC::ScriptC(Context *rsc) : Script(rsc) 37{ 38 mAllocFile = __FILE__; 39 mAllocLine = __LINE__; 40 mBccScript = NULL; 41 memset(&mProgram, 0, sizeof(mProgram)); 42} 43 44ScriptC::~ScriptC() 45{ 46 if (mBccScript) { 47 bccDeleteScript(mBccScript); 48 } 49 free(mEnviroment.mScriptText); 50 mEnviroment.mScriptText = NULL; 51} 52 53void ScriptC::setupScript(Context *rsc) 54{ 55 setupGLState(rsc); 56 mEnviroment.mStartTimeMillis 57 = nanoseconds_to_milliseconds(systemTime(SYSTEM_TIME_MONOTONIC)); 58 59 for (uint32_t ct=0; ct < mEnviroment.mFieldCount; ct++) { 60 if (mSlots[ct].get() && !mTypes[ct].get()) { 61 mTypes[ct].set(mSlots[ct]->getType()); 62 } 63 64 if (!mTypes[ct].get()) 65 continue; 66 void *ptr = NULL; 67 if (mSlots[ct].get()) { 68 ptr = mSlots[ct]->getPtr(); 69 } 70 void **dest = ((void ***)mEnviroment.mFieldAddress)[ct]; 71 72 if (rsc->props.mLogScripts) { 73 LOGV("%p ScriptC::setupScript slot=%i dst=%p src=%p type=%p", rsc, ct, dest, ptr, mSlots[ct]->getType()); 74 75 //const uint32_t *p32 = (const uint32_t *)ptr; 76 //for (uint32_t ct2=0; ct2 < mSlots[ct]->getType()->getDimX(); ct2++) { 77 //LOGE(" %i = 0x%08x ", ct2, p32[ct2]); 78 //} 79 } 80 81 if (dest) { 82 *dest = ptr; 83 } else { 84 if (rsc->props.mLogScripts) { 85 LOGV("ScriptC::setupScript, NULL var binding address."); 86 } 87 } 88 } 89} 90 91const Allocation *ScriptC::ptrToAllocation(const void *ptr) const 92{ 93 if (!ptr) { 94 return NULL; 95 } 96 for (uint32_t ct=0; ct < mEnviroment.mFieldCount; ct++) { 97 if (!mSlots[ct].get()) 98 continue; 99 if (mSlots[ct]->getPtr() == ptr) { 100 return mSlots[ct].get(); 101 } 102 } 103 LOGE("ScriptC::ptrToAllocation, failed to find %p", ptr); 104 return NULL; 105} 106 107Script * ScriptC::setTLS(Script *sc) 108{ 109 Context::ScriptTLSStruct * tls = (Context::ScriptTLSStruct *) 110 pthread_getspecific(Context::gThreadTLSKey); 111 rsAssert(tls); 112 Script *old = tls->mScript; 113 tls->mScript = sc; 114 return old; 115} 116 117 118void ScriptC::setupGLState(Context *rsc) 119{ 120 if (mEnviroment.mFragmentStore.get()) { 121 rsc->setFragmentStore(mEnviroment.mFragmentStore.get()); 122 } 123 if (mEnviroment.mFragment.get()) { 124 rsc->setFragment(mEnviroment.mFragment.get()); 125 } 126 if (mEnviroment.mVertex.get()) { 127 rsc->setVertex(mEnviroment.mVertex.get()); 128 } 129 if (mEnviroment.mRaster.get()) { 130 rsc->setRaster(mEnviroment.mRaster.get()); 131 } 132} 133 134uint32_t ScriptC::run(Context *rsc) 135{ 136 if (mProgram.mRoot == NULL) { 137 rsc->setError(RS_ERROR_BAD_SCRIPT, "Attempted to run bad script"); 138 return 0; 139 } 140 141 setupScript(rsc); 142 143 uint32_t ret = 0; 144 Script * oldTLS = setTLS(this); 145 146 if (rsc->props.mLogScripts) { 147 LOGV("%p ScriptC::run invoking root, ptr %p", rsc, mProgram.mRoot); 148 } 149 150 ret = mProgram.mRoot(); 151 152 if (rsc->props.mLogScripts) { 153 LOGV("%p ScriptC::run invoking complete, ret=%i", rsc, ret); 154 } 155 156 setTLS(oldTLS); 157 return ret; 158} 159 160 161typedef struct { 162 Context *rsc; 163 ScriptC *script; 164 const Allocation * ain; 165 Allocation * aout; 166 const void * usr; 167 168 uint32_t mSliceSize; 169 volatile int mSliceNum; 170 171 const uint8_t *ptrIn; 172 uint32_t eStrideIn; 173 uint8_t *ptrOut; 174 uint32_t eStrideOut; 175 176 uint32_t xStart; 177 uint32_t xEnd; 178 uint32_t yStart; 179 uint32_t yEnd; 180 uint32_t zStart; 181 uint32_t zEnd; 182 uint32_t arrayStart; 183 uint32_t arrayEnd; 184 185 uint32_t dimX; 186 uint32_t dimY; 187 uint32_t dimZ; 188 uint32_t dimArray; 189} MTLaunchStruct; 190typedef int (*rs_t)(const void *, void *, const void *, uint32_t, uint32_t, uint32_t, uint32_t); 191 192static void wc_xy(void *usr, uint32_t idx) 193{ 194 MTLaunchStruct *mtls = (MTLaunchStruct *)usr; 195 196 while (1) { 197 uint32_t slice = (uint32_t)android_atomic_inc(&mtls->mSliceNum); 198 uint32_t yStart = mtls->yStart + slice * mtls->mSliceSize; 199 uint32_t yEnd = yStart + mtls->mSliceSize; 200 yEnd = rsMin(yEnd, mtls->yEnd); 201 if (yEnd <= yStart) { 202 return; 203 } 204 205 //LOGE("usr idx %i, x %i,%i y %i,%i", idx, mtls->xStart, mtls->xEnd, yStart, yEnd); 206 //LOGE("usr ptr in %p, out %p", mtls->ptrIn, mtls->ptrOut); 207 for (uint32_t y = yStart; y < yEnd; y++) { 208 uint32_t offset = mtls->dimX * y; 209 uint8_t *xPtrOut = mtls->ptrOut + (mtls->eStrideOut * offset); 210 const uint8_t *xPtrIn = mtls->ptrIn + (mtls->eStrideIn * offset); 211 212 for (uint32_t x = mtls->xStart; x < mtls->xEnd; x++) { 213 ((rs_t)mtls->script->mProgram.mRoot) (xPtrIn, xPtrOut, mtls->usr, x, y, 0, 0); 214 xPtrIn += mtls->eStrideIn; 215 xPtrOut += mtls->eStrideOut; 216 } 217 } 218 } 219 220} 221 222void ScriptC::runForEach(Context *rsc, 223 const Allocation * ain, 224 Allocation * aout, 225 const void * usr, 226 const RsScriptCall *sc) 227{ 228 MTLaunchStruct mtls; 229 memset(&mtls, 0, sizeof(mtls)); 230 231 if (ain) { 232 mtls.dimX = ain->getType()->getDimX(); 233 mtls.dimY = ain->getType()->getDimY(); 234 mtls.dimZ = ain->getType()->getDimZ(); 235 //mtls.dimArray = ain->getType()->getDimArray(); 236 } else if (aout) { 237 mtls.dimX = aout->getType()->getDimX(); 238 mtls.dimY = aout->getType()->getDimY(); 239 mtls.dimZ = aout->getType()->getDimZ(); 240 //mtls.dimArray = aout->getType()->getDimArray(); 241 } else { 242 rsc->setError(RS_ERROR_BAD_SCRIPT, "rsForEach called with null allocations"); 243 return; 244 } 245 246 if (!sc || (sc->xEnd == 0)) { 247 mtls.xEnd = mtls.dimX; 248 } else { 249 rsAssert(sc->xStart < mtls.dimX); 250 rsAssert(sc->xEnd <= mtls.dimX); 251 rsAssert(sc->xStart < sc->xEnd); 252 mtls.xStart = rsMin(mtls.dimX, sc->xStart); 253 mtls.xEnd = rsMin(mtls.dimX, sc->xEnd); 254 if (mtls.xStart >= mtls.xEnd) return; 255 } 256 257 if (!sc || (sc->yEnd == 0)) { 258 mtls.yEnd = mtls.dimY; 259 } else { 260 rsAssert(sc->yStart < mtls.dimY); 261 rsAssert(sc->yEnd <= mtls.dimY); 262 rsAssert(sc->yStart < sc->yEnd); 263 mtls.yStart = rsMin(mtls.dimY, sc->yStart); 264 mtls.yEnd = rsMin(mtls.dimY, sc->yEnd); 265 if (mtls.yStart >= mtls.yEnd) return; 266 } 267 268 mtls.xEnd = rsMax((uint32_t)1, mtls.xEnd); 269 mtls.yEnd = rsMax((uint32_t)1, mtls.yEnd); 270 mtls.zEnd = rsMax((uint32_t)1, mtls.zEnd); 271 mtls.arrayEnd = rsMax((uint32_t)1, mtls.arrayEnd); 272 273 rsAssert(ain->getType()->getDimZ() == 0); 274 275 setupScript(rsc); 276 Script * oldTLS = setTLS(this); 277 278 279 mtls.rsc = rsc; 280 mtls.ain = ain; 281 mtls.aout = aout; 282 mtls.script = this; 283 mtls.usr = usr; 284 mtls.mSliceSize = 10; 285 mtls.mSliceNum = 0; 286 287 mtls.ptrIn = NULL; 288 mtls.eStrideIn = 0; 289 if (ain) { 290 mtls.ptrIn = (const uint8_t *)ain->getPtr(); 291 mtls.eStrideIn = ain->getType()->getElementSizeBytes(); 292 } 293 294 mtls.ptrOut = NULL; 295 mtls.eStrideOut = 0; 296 if (aout) { 297 mtls.ptrOut = (uint8_t *)aout->getPtr(); 298 mtls.eStrideOut = aout->getType()->getElementSizeBytes(); 299 } 300 301 if ((rsc->getWorkerPoolSize() > 1) && mEnviroment.mIsThreadable && (mtls.dimY > 1)) { 302 303 //LOGE("launch 1"); 304 rsc->launchThreads(wc_xy, &mtls); 305 } else { 306 //LOGE("launch 3"); 307 for (uint32_t ar = mtls.arrayStart; ar < mtls.arrayEnd; ar++) { 308 for (uint32_t z = mtls.zStart; z < mtls.zEnd; z++) { 309 for (uint32_t y = mtls.yStart; y < mtls.yEnd; y++) { 310 uint32_t offset = mtls.dimX * mtls.dimY * mtls.dimZ * ar + 311 mtls.dimX * mtls.dimY * z + 312 mtls.dimX * y; 313 uint8_t *xPtrOut = mtls.ptrOut + (mtls.eStrideOut * offset); 314 const uint8_t *xPtrIn = mtls.ptrIn + (mtls.eStrideIn * offset); 315 316 for (uint32_t x = mtls.xStart; x < mtls.xEnd; x++) { 317 ((rs_t)mProgram.mRoot) (xPtrIn, xPtrOut, usr, x, y, z, ar); 318 xPtrIn += mtls.eStrideIn; 319 xPtrOut += mtls.eStrideOut; 320 } 321 } 322 } 323 } 324 } 325 326 setTLS(oldTLS); 327} 328 329void ScriptC::Invoke(Context *rsc, uint32_t slot, const void *data, uint32_t len) 330{ 331 //LOGE("rsi_ScriptInvoke %i", slot); 332 if ((slot >= mEnviroment.mInvokeFunctionCount) || 333 (mEnviroment.mInvokeFunctions[slot] == NULL)) { 334 rsc->setError(RS_ERROR_BAD_SCRIPT, "Calling invoke on bad script"); 335 return; 336 } 337 setupScript(rsc); 338 Script * oldTLS = setTLS(this); 339 340 if (rsc->props.mLogScripts) { 341 LOGV("%p ScriptC::Invoke invoking slot %i, ptr %p", rsc, slot, mEnviroment.mInvokeFunctions[slot]); 342 } 343 ((void (*)(const void *, uint32_t)) 344 mEnviroment.mInvokeFunctions[slot])(data, len); 345 if (rsc->props.mLogScripts) { 346 LOGV("%p ScriptC::Invoke complete", rsc); 347 } 348 349 setTLS(oldTLS); 350} 351 352ScriptCState::ScriptCState() 353{ 354 mScript.clear(); 355} 356 357ScriptCState::~ScriptCState() 358{ 359 mScript.clear(); 360} 361 362void ScriptCState::init(Context *rsc) 363{ 364 clear(rsc); 365} 366 367void ScriptCState::clear(Context *rsc) 368{ 369 rsAssert(rsc); 370 mScript.clear(); 371 mScript.set(new ScriptC(rsc)); 372} 373 374static BCCvoid* symbolLookup(BCCvoid* pContext, const BCCchar* name) 375{ 376 const ScriptCState::SymbolTable_t *sym; 377 ScriptC *s = (ScriptC *)pContext; 378 sym = ScriptCState::lookupSymbol(name); 379 if (sym) { 380 return sym->mPtr; 381 } 382 sym = ScriptCState::lookupSymbolCL(name); 383 if (sym) { 384 return sym->mPtr; 385 } 386 s->mEnviroment.mIsThreadable = false; 387 sym = ScriptCState::lookupSymbolGL(name); 388 if (sym) { 389 return sym->mPtr; 390 } 391 LOGE("ScriptC sym lookup failed for %s", name); 392 return NULL; 393} 394 395void ScriptCState::runCompiler(Context *rsc, ScriptC *s) 396{ 397 LOGV("%p ScriptCState::runCompiler ", rsc); 398 { 399 StopWatch compileTimer("RenderScript compile time"); 400 s->mBccScript = bccCreateScript(); 401 s->mEnviroment.mIsThreadable = true; 402 bccScriptBitcode(s->mBccScript, s->mEnviroment.mScriptText, s->mEnviroment.mScriptTextLength); 403 bccRegisterSymbolCallback(s->mBccScript, symbolLookup, s); 404 bccCompileScript(s->mBccScript); 405 bccGetScriptLabel(s->mBccScript, "root", (BCCvoid**) &s->mProgram.mRoot); 406 bccGetScriptLabel(s->mBccScript, "init", (BCCvoid**) &s->mProgram.mInit); 407 } 408 LOGV("%p ScriptCState::runCompiler root %p, init %p", rsc, s->mProgram.mRoot, s->mProgram.mInit); 409 410 if (s->mProgram.mInit) { 411 s->mProgram.mInit(); 412 } 413 414 bccGetExportFuncs(s->mBccScript, (BCCsizei*) &s->mEnviroment.mInvokeFunctionCount, 0, NULL); 415 if(s->mEnviroment.mInvokeFunctionCount <= 0) 416 s->mEnviroment.mInvokeFunctions = NULL; 417 else { 418 s->mEnviroment.mInvokeFunctions = (Script::InvokeFunc_t*) calloc(s->mEnviroment.mInvokeFunctionCount, sizeof(Script::InvokeFunc_t)); 419 bccGetExportFuncs(s->mBccScript, NULL, s->mEnviroment.mInvokeFunctionCount, (BCCvoid **) s->mEnviroment.mInvokeFunctions); 420 } 421 422 bccGetExportVars(s->mBccScript, (BCCsizei*) &s->mEnviroment.mFieldCount, 0, NULL); 423 if(s->mEnviroment.mFieldCount <= 0) 424 s->mEnviroment.mFieldAddress = NULL; 425 else { 426 s->mEnviroment.mFieldAddress = (void **) calloc(s->mEnviroment.mFieldCount, sizeof(void *)); 427 bccGetExportVars(s->mBccScript, NULL, s->mEnviroment.mFieldCount, (BCCvoid **) s->mEnviroment.mFieldAddress); 428 s->initSlots(); 429 } 430 431 s->mEnviroment.mFragment.set(rsc->getDefaultProgramFragment()); 432 s->mEnviroment.mVertex.set(rsc->getDefaultProgramVertex()); 433 s->mEnviroment.mFragmentStore.set(rsc->getDefaultProgramStore()); 434 s->mEnviroment.mRaster.set(rsc->getDefaultProgramRaster()); 435 436 if (s->mProgram.mRoot) { 437 const static int pragmaMax = 16; 438 BCCsizei pragmaCount; 439 BCCchar * str[pragmaMax]; 440 bccGetPragmas(s->mBccScript, &pragmaCount, pragmaMax, &str[0]); 441 442 for (int ct=0; ct < pragmaCount; ct+=2) { 443 //LOGE("pragme %s %s", str[ct], str[ct+1]); 444 if (!strcmp(str[ct], "version")) { 445 continue; 446 } 447 448 if (!strcmp(str[ct], "stateVertex")) { 449 if (!strcmp(str[ct+1], "default")) { 450 continue; 451 } 452 if (!strcmp(str[ct+1], "parent")) { 453 s->mEnviroment.mVertex.clear(); 454 continue; 455 } 456 LOGE("Unreconized value %s passed to stateVertex", str[ct+1]); 457 } 458 459 if (!strcmp(str[ct], "stateRaster")) { 460 if (!strcmp(str[ct+1], "default")) { 461 continue; 462 } 463 if (!strcmp(str[ct+1], "parent")) { 464 s->mEnviroment.mRaster.clear(); 465 continue; 466 } 467 LOGE("Unreconized value %s passed to stateRaster", str[ct+1]); 468 } 469 470 if (!strcmp(str[ct], "stateFragment")) { 471 if (!strcmp(str[ct+1], "default")) { 472 continue; 473 } 474 if (!strcmp(str[ct+1], "parent")) { 475 s->mEnviroment.mFragment.clear(); 476 continue; 477 } 478 LOGE("Unreconized value %s passed to stateFragment", str[ct+1]); 479 } 480 481 if (!strcmp(str[ct], "stateStore")) { 482 if (!strcmp(str[ct+1], "default")) { 483 continue; 484 } 485 if (!strcmp(str[ct+1], "parent")) { 486 s->mEnviroment.mFragmentStore.clear(); 487 continue; 488 } 489 LOGE("Unreconized value %s passed to stateStore", str[ct+1]); 490 } 491 492 } 493 494 495 } else { 496 // Deal with an error. 497 } 498} 499 500 501 502namespace android { 503namespace renderscript { 504 505void rsi_ScriptCBegin(Context * rsc) 506{ 507 ScriptCState *ss = &rsc->mScriptC; 508 ss->clear(rsc); 509} 510 511void rsi_ScriptCSetText(Context *rsc, const char *text, uint32_t len) 512{ 513 ScriptCState *ss = &rsc->mScriptC; 514 515 char *t = (char *)malloc(len + 1); 516 memcpy(t, text, len); 517 t[len] = 0; 518 ss->mScript->mEnviroment.mScriptText = t; 519 ss->mScript->mEnviroment.mScriptTextLength = len; 520} 521 522 523RsScript rsi_ScriptCCreate(Context * rsc) 524{ 525 ScriptCState *ss = &rsc->mScriptC; 526 527 ObjectBaseRef<ScriptC> s = ss->mScript.get(); 528 ss->mScript.clear(); 529 530 ss->runCompiler(rsc, s.get()); 531 s->incUserRef(); 532 ss->clear(rsc); 533 return s.get(); 534} 535 536} 537} 538 539 540