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 19 #include <GLES/gl.h> 20 #include <GLES2/gl2.h> 21 #include <GLES/glext.h> 22 23 using namespace android; 24 using namespace android::renderscript; 25 26 Allocation::Allocation(Context *rsc, const Type *type) : ObjectBase(rsc) 27 { 28 init(rsc, type); 29 30 mPtr = malloc(mType->getSizeBytes()); 31 if (!mPtr) { 32 LOGE("Allocation::Allocation, alloc failure"); 33 } 34 } 35 36 Allocation::Allocation(Context *rsc, const Type *type, void *bmp, 37 void *callbackData, RsBitmapCallback_t callback) 38 : ObjectBase(rsc) 39 { 40 init(rsc, type); 41 42 mPtr = bmp; 43 mUserBitmapCallback = callback; 44 mUserBitmapCallbackData = callbackData; 45 } 46 47 void Allocation::init(Context *rsc, const Type *type) 48 { 49 mAllocFile = __FILE__; 50 mAllocLine = __LINE__; 51 mPtr = NULL; 52 53 mCpuWrite = false; 54 mCpuRead = false; 55 mGpuWrite = false; 56 mGpuRead = false; 57 58 mReadWriteRatio = 0; 59 mUpdateSize = 0; 60 61 mIsTexture = false; 62 mTextureID = 0; 63 mIsVertexBuffer = false; 64 mBufferID = 0; 65 mUploadDefered = false; 66 67 mUserBitmapCallback = NULL; 68 mUserBitmapCallbackData = NULL; 69 70 mType.set(type); 71 rsAssert(type); 72 73 mPtr = NULL; 74 } 75 76 Allocation::~Allocation() 77 { 78 if (mUserBitmapCallback != NULL) { 79 mUserBitmapCallback(mUserBitmapCallbackData); 80 } else { 81 free(mPtr); 82 } 83 mPtr = NULL; 84 85 if (mBufferID) { 86 // Causes a SW crash.... 87 //LOGV(" mBufferID %i", mBufferID); 88 //glDeleteBuffers(1, &mBufferID); 89 //mBufferID = 0; 90 } 91 if (mTextureID) { 92 glDeleteTextures(1, &mTextureID); 93 mTextureID = 0; 94 } 95 } 96 97 void Allocation::setCpuWritable(bool) 98 { 99 } 100 101 void Allocation::setGpuWritable(bool) 102 { 103 } 104 105 void Allocation::setCpuReadable(bool) 106 { 107 } 108 109 void Allocation::setGpuReadable(bool) 110 { 111 } 112 113 bool Allocation::fixAllocation() 114 { 115 return false; 116 } 117 118 void Allocation::deferedUploadToTexture(const Context *rsc, bool genMipmap, uint32_t lodOffset) 119 { 120 rsAssert(lodOffset < mType->getLODCount()); 121 mIsTexture = true; 122 mTextureLOD = lodOffset; 123 mUploadDefered = true; 124 mTextureGenMipmap = !mType->getDimLOD() && genMipmap; 125 } 126 127 void Allocation::uploadToTexture(const Context *rsc) 128 { 129 //rsAssert(!mTextureId); 130 131 mIsTexture = true; 132 if (!rsc->checkDriver()) { 133 mUploadDefered = true; 134 return; 135 } 136 137 GLenum type = mType->getElement()->getComponent().getGLType(); 138 GLenum format = mType->getElement()->getComponent().getGLFormat(); 139 140 if (!type || !format) { 141 return; 142 } 143 144 if (!mTextureID) { 145 glGenTextures(1, &mTextureID); 146 147 if (!mTextureID) { 148 // This should not happen, however, its likely the cause of the 149 // white sqare bug. 150 // Force a crash to 1: restart the app, 2: make sure we get a bugreport. 151 LOGE("Upload to texture failed to gen mTextureID"); 152 rsc->dumpDebug(); 153 mUploadDefered = true; 154 return; 155 } 156 } 157 glBindTexture(GL_TEXTURE_2D, mTextureID); 158 glPixelStorei(GL_UNPACK_ALIGNMENT, 1); 159 160 Adapter2D adapt(getContext(), this); 161 for(uint32_t lod = 0; (lod + mTextureLOD) < mType->getLODCount(); lod++) { 162 adapt.setLOD(lod+mTextureLOD); 163 164 uint16_t * ptr = static_cast<uint16_t *>(adapt.getElement(0,0)); 165 glTexImage2D(GL_TEXTURE_2D, lod, format, 166 adapt.getDimX(), adapt.getDimY(), 167 0, format, type, ptr); 168 } 169 if (mTextureGenMipmap) { 170 glGenerateMipmap(GL_TEXTURE_2D); 171 } 172 173 } 174 175 void Allocation::deferedUploadToBufferObject(const Context *rsc) 176 { 177 mIsVertexBuffer = true; 178 mUploadDefered = true; 179 } 180 181 void Allocation::uploadToBufferObject(const Context *rsc) 182 { 183 rsAssert(!mType->getDimY()); 184 rsAssert(!mType->getDimZ()); 185 186 mIsVertexBuffer = true; 187 if (!rsc->checkDriver()) { 188 mUploadDefered = true; 189 return; 190 } 191 192 if (!mBufferID) { 193 glGenBuffers(1, &mBufferID); 194 } 195 if (!mBufferID) { 196 LOGE("Upload to buffer object failed"); 197 mUploadDefered = true; 198 return; 199 } 200 201 glBindBuffer(GL_ARRAY_BUFFER, mBufferID); 202 glBufferData(GL_ARRAY_BUFFER, mType->getSizeBytes(), getPtr(), GL_DYNAMIC_DRAW); 203 glBindBuffer(GL_ARRAY_BUFFER, 0); 204 } 205 206 void Allocation::uploadCheck(const Context *rsc) 207 { 208 if (mUploadDefered) { 209 mUploadDefered = false; 210 if (mIsVertexBuffer) { 211 uploadToBufferObject(rsc); 212 } 213 if (mIsTexture) { 214 uploadToTexture(rsc); 215 } 216 } 217 } 218 219 220 void Allocation::data(const void *data, uint32_t sizeBytes) 221 { 222 uint32_t size = mType->getSizeBytes(); 223 if (size != sizeBytes) { 224 LOGE("Allocation::data called with mismatched size expected %i, got %i", size, sizeBytes); 225 return; 226 } 227 memcpy(mPtr, data, size); 228 sendDirty(); 229 mUploadDefered = true; 230 } 231 232 void Allocation::read(void *data) 233 { 234 memcpy(data, mPtr, mType->getSizeBytes()); 235 } 236 237 void Allocation::subData(uint32_t xoff, uint32_t count, const void *data, uint32_t sizeBytes) 238 { 239 uint32_t eSize = mType->getElementSizeBytes(); 240 uint8_t * ptr = static_cast<uint8_t *>(mPtr); 241 ptr += eSize * xoff; 242 uint32_t size = count * eSize; 243 244 if (size != sizeBytes) { 245 LOGE("Allocation::subData called with mismatched size expected %i, got %i", size, sizeBytes); 246 mType->dumpLOGV("type info"); 247 return; 248 } 249 memcpy(ptr, data, size); 250 sendDirty(); 251 mUploadDefered = true; 252 } 253 254 void Allocation::subData(uint32_t xoff, uint32_t yoff, 255 uint32_t w, uint32_t h, const void *data, uint32_t sizeBytes) 256 { 257 uint32_t eSize = mType->getElementSizeBytes(); 258 uint32_t lineSize = eSize * w; 259 uint32_t destW = mType->getDimX(); 260 261 const uint8_t *src = static_cast<const uint8_t *>(data); 262 uint8_t *dst = static_cast<uint8_t *>(mPtr); 263 dst += eSize * (xoff + yoff * destW); 264 265 if ((lineSize * eSize * h) != sizeBytes) { 266 rsAssert(!"Allocation::subData called with mismatched size"); 267 return; 268 } 269 270 for (uint32_t line=yoff; line < (yoff+h); line++) { 271 uint8_t * ptr = static_cast<uint8_t *>(mPtr); 272 memcpy(dst, src, lineSize); 273 src += lineSize; 274 dst += destW * eSize; 275 } 276 sendDirty(); 277 mUploadDefered = true; 278 } 279 280 void Allocation::subData(uint32_t xoff, uint32_t yoff, uint32_t zoff, 281 uint32_t w, uint32_t h, uint32_t d, const void *data, uint32_t sizeBytes) 282 { 283 } 284 285 void Allocation::addProgramToDirty(const Program *p) 286 { 287 mToDirtyList.add(p); 288 } 289 290 void Allocation::removeProgramToDirty(const Program *p) 291 { 292 for (size_t ct=0; ct < mToDirtyList.size(); ct++) { 293 if (mToDirtyList[ct] == p) { 294 mToDirtyList.removeAt(ct); 295 return; 296 } 297 } 298 rsAssert(0); 299 } 300 301 void Allocation::dumpLOGV(const char *prefix) const 302 { 303 ObjectBase::dumpLOGV(prefix); 304 305 String8 s(prefix); 306 s.append(" type "); 307 if (mType.get()) { 308 mType->dumpLOGV(s.string()); 309 } 310 311 LOGV("%s allocation ptr=%p mCpuWrite=%i, mCpuRead=%i, mGpuWrite=%i, mGpuRead=%i", 312 prefix, mPtr, mCpuWrite, mCpuRead, mGpuWrite, mGpuRead); 313 314 LOGV("%s allocation mIsTexture=%i mTextureID=%i, mIsVertexBuffer=%i, mBufferID=%i", 315 prefix, mIsTexture, mTextureID, mIsVertexBuffer, mBufferID); 316 317 } 318 319 void Allocation::sendDirty() const 320 { 321 for (size_t ct=0; ct < mToDirtyList.size(); ct++) { 322 mToDirtyList[ct]->forceDirty(); 323 } 324 } 325 326 ///////////////// 327 // 328 329 330 namespace android { 331 namespace renderscript { 332 333 RsAllocation rsi_AllocationCreateTyped(Context *rsc, RsType vtype) 334 { 335 const Type * type = static_cast<const Type *>(vtype); 336 337 Allocation * alloc = new Allocation(rsc, type); 338 alloc->incUserRef(); 339 return alloc; 340 } 341 342 RsAllocation rsi_AllocationCreateSized(Context *rsc, RsElement e, size_t count) 343 { 344 Type * type = new Type(rsc); 345 type->setDimX(count); 346 type->setElement(static_cast<Element *>(e)); 347 type->compute(); 348 return rsi_AllocationCreateTyped(rsc, type); 349 } 350 351 void rsi_AllocationUploadToTexture(Context *rsc, RsAllocation va, bool genmip, uint32_t baseMipLevel) 352 { 353 Allocation *alloc = static_cast<Allocation *>(va); 354 alloc->deferedUploadToTexture(rsc, genmip, baseMipLevel); 355 } 356 357 void rsi_AllocationUploadToBufferObject(Context *rsc, RsAllocation va) 358 { 359 Allocation *alloc = static_cast<Allocation *>(va); 360 alloc->deferedUploadToBufferObject(rsc); 361 } 362 363 static void mip565(const Adapter2D &out, const Adapter2D &in) 364 { 365 uint32_t w = out.getDimX(); 366 uint32_t h = out.getDimY(); 367 368 for (uint32_t y=0; y < h; y++) { 369 uint16_t *oPtr = static_cast<uint16_t *>(out.getElement(0, y)); 370 const uint16_t *i1 = static_cast<uint16_t *>(in.getElement(0, y*2)); 371 const uint16_t *i2 = static_cast<uint16_t *>(in.getElement(0, y*2+1)); 372 373 for (uint32_t x=0; x < w; x++) { 374 *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]); 375 oPtr ++; 376 i1 += 2; 377 i2 += 2; 378 } 379 } 380 } 381 382 static void mip8888(const Adapter2D &out, const Adapter2D &in) 383 { 384 uint32_t w = out.getDimX(); 385 uint32_t h = out.getDimY(); 386 387 for (uint32_t y=0; y < h; y++) { 388 uint32_t *oPtr = static_cast<uint32_t *>(out.getElement(0, y)); 389 const uint32_t *i1 = static_cast<uint32_t *>(in.getElement(0, y*2)); 390 const uint32_t *i2 = static_cast<uint32_t *>(in.getElement(0, y*2+1)); 391 392 for (uint32_t x=0; x < w; x++) { 393 *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]); 394 oPtr ++; 395 i1 += 2; 396 i2 += 2; 397 } 398 } 399 } 400 401 static void mip8(const Adapter2D &out, const Adapter2D &in) 402 { 403 uint32_t w = out.getDimX(); 404 uint32_t h = out.getDimY(); 405 406 for (uint32_t y=0; y < h; y++) { 407 uint8_t *oPtr = static_cast<uint8_t *>(out.getElement(0, y)); 408 const uint8_t *i1 = static_cast<uint8_t *>(in.getElement(0, y*2)); 409 const uint8_t *i2 = static_cast<uint8_t *>(in.getElement(0, y*2+1)); 410 411 for (uint32_t x=0; x < w; x++) { 412 *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f); 413 oPtr ++; 414 i1 += 2; 415 i2 += 2; 416 } 417 } 418 } 419 420 static void mip(const Adapter2D &out, const Adapter2D &in) 421 { 422 switch(out.getBaseType()->getElement()->getSizeBits()) { 423 case 32: 424 mip8888(out, in); 425 break; 426 case 16: 427 mip565(out, in); 428 break; 429 case 8: 430 mip8(out, in); 431 break; 432 433 } 434 435 } 436 437 typedef void (*ElementConverter_t)(void *dst, const void *src, uint32_t count); 438 439 static void elementConverter_cpy_16(void *dst, const void *src, uint32_t count) 440 { 441 memcpy(dst, src, count * 2); 442 } 443 static void elementConverter_cpy_8(void *dst, const void *src, uint32_t count) 444 { 445 memcpy(dst, src, count); 446 } 447 static void elementConverter_cpy_32(void *dst, const void *src, uint32_t count) 448 { 449 memcpy(dst, src, count * 4); 450 } 451 452 453 static void elementConverter_888_to_565(void *dst, const void *src, uint32_t count) 454 { 455 uint16_t *d = static_cast<uint16_t *>(dst); 456 const uint8_t *s = static_cast<const uint8_t *>(src); 457 458 while(count--) { 459 *d = rs888to565(s[0], s[1], s[2]); 460 d++; 461 s+= 3; 462 } 463 } 464 465 static void elementConverter_8888_to_565(void *dst, const void *src, uint32_t count) 466 { 467 uint16_t *d = static_cast<uint16_t *>(dst); 468 const uint8_t *s = static_cast<const uint8_t *>(src); 469 470 while(count--) { 471 *d = rs888to565(s[0], s[1], s[2]); 472 d++; 473 s+= 4; 474 } 475 } 476 477 static ElementConverter_t pickConverter(const Element *dst, const Element *src) 478 { 479 GLenum srcGLType = src->getComponent().getGLType(); 480 GLenum srcGLFmt = src->getComponent().getGLFormat(); 481 GLenum dstGLType = dst->getComponent().getGLType(); 482 GLenum dstGLFmt = dst->getComponent().getGLFormat(); 483 484 if (srcGLFmt == dstGLFmt && srcGLType == dstGLType) { 485 switch(dst->getSizeBytes()) { 486 case 4: 487 return elementConverter_cpy_32; 488 case 2: 489 return elementConverter_cpy_16; 490 case 1: 491 return elementConverter_cpy_8; 492 } 493 } 494 495 if (srcGLType == GL_UNSIGNED_BYTE && 496 srcGLFmt == GL_RGB && 497 dstGLType == GL_UNSIGNED_SHORT_5_6_5 && 498 dstGLType == GL_RGB) { 499 500 return elementConverter_888_to_565; 501 } 502 503 if (srcGLType == GL_UNSIGNED_BYTE && 504 srcGLFmt == GL_RGBA && 505 dstGLType == GL_UNSIGNED_SHORT_5_6_5 && 506 dstGLType == GL_RGB) { 507 508 return elementConverter_8888_to_565; 509 } 510 511 LOGE("pickConverter, unsuported combo, src %p, dst %p", src, dst); 512 return 0; 513 } 514 515 RsAllocation rsi_AllocationCreateBitmapRef(Context *rsc, RsType vtype, 516 void *bmp, void *callbackData, RsBitmapCallback_t callback) 517 { 518 const Type * type = static_cast<const Type *>(vtype); 519 Allocation * alloc = new Allocation(rsc, type, bmp, callbackData, callback); 520 alloc->incUserRef(); 521 return alloc; 522 } 523 524 RsAllocation rsi_AllocationCreateFromBitmap(Context *rsc, uint32_t w, uint32_t h, RsElement _dst, RsElement _src, bool genMips, const void *data) 525 { 526 const Element *src = static_cast<const Element *>(_src); 527 const Element *dst = static_cast<const Element *>(_dst); 528 529 // Check for pow2 on pre es 2.0 versions. 530 rsAssert(rsc->checkVersion2_0() || (!(w & (w-1)) && !(h & (h-1)))); 531 532 //LOGE("rsi_AllocationCreateFromBitmap %i %i %i %i %i", w, h, dstFmt, srcFmt, genMips); 533 rsi_TypeBegin(rsc, _dst); 534 rsi_TypeAdd(rsc, RS_DIMENSION_X, w); 535 rsi_TypeAdd(rsc, RS_DIMENSION_Y, h); 536 if (genMips) { 537 rsi_TypeAdd(rsc, RS_DIMENSION_LOD, 1); 538 } 539 RsType type = rsi_TypeCreate(rsc); 540 541 RsAllocation vTexAlloc = rsi_AllocationCreateTyped(rsc, type); 542 Allocation *texAlloc = static_cast<Allocation *>(vTexAlloc); 543 if (texAlloc == NULL) { 544 LOGE("Memory allocation failure"); 545 return NULL; 546 } 547 548 ElementConverter_t cvt = pickConverter(dst, src); 549 cvt(texAlloc->getPtr(), data, w * h); 550 551 if (genMips) { 552 Adapter2D adapt(rsc, texAlloc); 553 Adapter2D adapt2(rsc, texAlloc); 554 for(uint32_t lod=0; lod < (texAlloc->getType()->getLODCount() -1); lod++) { 555 adapt.setLOD(lod); 556 adapt2.setLOD(lod + 1); 557 mip(adapt2, adapt); 558 } 559 } 560 561 return texAlloc; 562 } 563 564 RsAllocation rsi_AllocationCreateFromBitmapBoxed(Context *rsc, uint32_t w, uint32_t h, RsElement _dst, RsElement _src, bool genMips, const void *data) 565 { 566 const Element *srcE = static_cast<const Element *>(_src); 567 const Element *dstE = static_cast<const Element *>(_dst); 568 uint32_t w2 = rsHigherPow2(w); 569 uint32_t h2 = rsHigherPow2(h); 570 571 if ((w2 == w) && (h2 == h)) { 572 return rsi_AllocationCreateFromBitmap(rsc, w, h, _dst, _src, genMips, data); 573 } 574 575 uint32_t bpp = srcE->getSizeBytes(); 576 size_t size = w2 * h2 * bpp; 577 uint8_t *tmp = static_cast<uint8_t *>(malloc(size)); 578 memset(tmp, 0, size); 579 580 const uint8_t * src = static_cast<const uint8_t *>(data); 581 for (uint32_t y = 0; y < h; y++) { 582 uint8_t * ydst = &tmp[(y + ((h2 - h) >> 1)) * w2 * bpp]; 583 memcpy(&ydst[((w2 - w) >> 1) * bpp], src, w * bpp); 584 src += w * bpp; 585 } 586 587 RsAllocation ret = rsi_AllocationCreateFromBitmap(rsc, w2, h2, _dst, _src, genMips, tmp); 588 free(tmp); 589 return ret; 590 } 591 592 void rsi_AllocationData(Context *rsc, RsAllocation va, const void *data, uint32_t sizeBytes) 593 { 594 Allocation *a = static_cast<Allocation *>(va); 595 a->data(data, sizeBytes); 596 } 597 598 void rsi_Allocation1DSubData(Context *rsc, RsAllocation va, uint32_t xoff, uint32_t count, const void *data, uint32_t sizeBytes) 599 { 600 Allocation *a = static_cast<Allocation *>(va); 601 a->subData(xoff, count, data, sizeBytes); 602 } 603 604 void rsi_Allocation2DSubData(Context *rsc, RsAllocation va, uint32_t xoff, uint32_t yoff, uint32_t w, uint32_t h, const void *data, uint32_t sizeBytes) 605 { 606 Allocation *a = static_cast<Allocation *>(va); 607 a->subData(xoff, yoff, w, h, data, sizeBytes); 608 } 609 610 void rsi_AllocationRead(Context *rsc, RsAllocation va, void *data) 611 { 612 Allocation *a = static_cast<Allocation *>(va); 613 a->read(data); 614 } 615 616 617 } 618 } 619