rsdAllocation.cpp revision c88bcef407e94afe33f5d26a0644a7116926b5ee
1/* 2 * Copyright (C) 2013 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 "rsdCore.h" 18#include "rsdAllocation.h" 19 20#include "rsAllocation.h" 21 22#ifndef RS_SERVER 23#include "system/window.h" 24#include "ui/Rect.h" 25#include "ui/GraphicBufferMapper.h" 26#endif 27 28#ifndef RS_COMPATIBILITY_LIB 29#include "rsdFrameBufferObj.h" 30#include "gui/GLConsumer.h" 31#include "gui/CpuConsumer.h" 32#include "gui/Surface.h" 33#include "hardware/gralloc.h" 34 35#include <GLES/gl.h> 36#include <GLES2/gl2.h> 37#include <GLES/glext.h> 38#endif 39 40#ifdef RS_SERVER 41// server requires malloc.h for memalign 42#include <malloc.h> 43#endif 44 45using namespace android; 46using namespace android::renderscript; 47 48 49#ifndef RS_COMPATIBILITY_LIB 50const static GLenum gFaceOrder[] = { 51 GL_TEXTURE_CUBE_MAP_POSITIVE_X, 52 GL_TEXTURE_CUBE_MAP_NEGATIVE_X, 53 GL_TEXTURE_CUBE_MAP_POSITIVE_Y, 54 GL_TEXTURE_CUBE_MAP_NEGATIVE_Y, 55 GL_TEXTURE_CUBE_MAP_POSITIVE_Z, 56 GL_TEXTURE_CUBE_MAP_NEGATIVE_Z 57}; 58 59GLenum rsdTypeToGLType(RsDataType t) { 60 switch (t) { 61 case RS_TYPE_UNSIGNED_5_6_5: return GL_UNSIGNED_SHORT_5_6_5; 62 case RS_TYPE_UNSIGNED_5_5_5_1: return GL_UNSIGNED_SHORT_5_5_5_1; 63 case RS_TYPE_UNSIGNED_4_4_4_4: return GL_UNSIGNED_SHORT_4_4_4_4; 64 65 //case RS_TYPE_FLOAT_16: return GL_HALF_FLOAT; 66 case RS_TYPE_FLOAT_32: return GL_FLOAT; 67 case RS_TYPE_UNSIGNED_8: return GL_UNSIGNED_BYTE; 68 case RS_TYPE_UNSIGNED_16: return GL_UNSIGNED_SHORT; 69 case RS_TYPE_SIGNED_8: return GL_BYTE; 70 case RS_TYPE_SIGNED_16: return GL_SHORT; 71 default: break; 72 } 73 return 0; 74} 75 76GLenum rsdKindToGLFormat(RsDataKind k) { 77 switch (k) { 78 case RS_KIND_PIXEL_L: return GL_LUMINANCE; 79 case RS_KIND_PIXEL_A: return GL_ALPHA; 80 case RS_KIND_PIXEL_LA: return GL_LUMINANCE_ALPHA; 81 case RS_KIND_PIXEL_RGB: return GL_RGB; 82 case RS_KIND_PIXEL_RGBA: return GL_RGBA; 83 case RS_KIND_PIXEL_DEPTH: return GL_DEPTH_COMPONENT16; 84 default: break; 85 } 86 return 0; 87} 88#endif 89 90uint8_t *GetOffsetPtr(const android::renderscript::Allocation *alloc, 91 uint32_t xoff, uint32_t yoff, uint32_t zoff, 92 uint32_t lod, RsAllocationCubemapFace face) { 93 uint8_t *ptr = (uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr; 94 ptr += face * alloc->mHal.drvState.faceOffset; 95 ptr += zoff * alloc->mHal.drvState.lod[lod].dimY * alloc->mHal.drvState.lod[lod].stride; 96 ptr += yoff * alloc->mHal.drvState.lod[lod].stride; 97 ptr += xoff * alloc->mHal.state.elementSizeBytes; 98 return ptr; 99} 100 101 102static void Update2DTexture(const Context *rsc, const Allocation *alloc, const void *ptr, 103 uint32_t xoff, uint32_t yoff, uint32_t lod, 104 RsAllocationCubemapFace face, uint32_t w, uint32_t h) { 105#ifndef RS_COMPATIBILITY_LIB 106 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 107 108 rsAssert(drv->textureID); 109 RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID); 110 RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1); 111 GLenum t = GL_TEXTURE_2D; 112 if (alloc->mHal.state.hasFaces) { 113 t = gFaceOrder[face]; 114 } 115 RSD_CALL_GL(glTexSubImage2D, t, lod, xoff, yoff, w, h, drv->glFormat, drv->glType, ptr); 116#endif 117} 118 119 120#ifndef RS_COMPATIBILITY_LIB 121static void Upload2DTexture(const Context *rsc, const Allocation *alloc, bool isFirstUpload) { 122 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 123 124 RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID); 125 RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1); 126 127 uint32_t faceCount = 1; 128 if (alloc->mHal.state.hasFaces) { 129 faceCount = 6; 130 } 131 132 rsdGLCheckError(rsc, "Upload2DTexture 1 "); 133 for (uint32_t face = 0; face < faceCount; face ++) { 134 for (uint32_t lod = 0; lod < alloc->mHal.state.type->getLODCount(); lod++) { 135 const uint8_t *p = GetOffsetPtr(alloc, 0, 0, 0, lod, (RsAllocationCubemapFace)face); 136 137 GLenum t = GL_TEXTURE_2D; 138 if (alloc->mHal.state.hasFaces) { 139 t = gFaceOrder[face]; 140 } 141 142 if (isFirstUpload) { 143 RSD_CALL_GL(glTexImage2D, t, lod, drv->glFormat, 144 alloc->mHal.state.type->getLODDimX(lod), 145 alloc->mHal.state.type->getLODDimY(lod), 146 0, drv->glFormat, drv->glType, p); 147 } else { 148 RSD_CALL_GL(glTexSubImage2D, t, lod, 0, 0, 149 alloc->mHal.state.type->getLODDimX(lod), 150 alloc->mHal.state.type->getLODDimY(lod), 151 drv->glFormat, drv->glType, p); 152 } 153 } 154 } 155 156 if (alloc->mHal.state.mipmapControl == RS_ALLOCATION_MIPMAP_ON_SYNC_TO_TEXTURE) { 157 RSD_CALL_GL(glGenerateMipmap, drv->glTarget); 158 } 159 rsdGLCheckError(rsc, "Upload2DTexture"); 160} 161#endif 162 163static void UploadToTexture(const Context *rsc, const Allocation *alloc) { 164#ifndef RS_COMPATIBILITY_LIB 165 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 166 167 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) { 168 if (!drv->textureID) { 169 RSD_CALL_GL(glGenTextures, 1, &drv->textureID); 170 } 171 return; 172 } 173 174 if (!drv->glType || !drv->glFormat) { 175 return; 176 } 177 178 if (!alloc->mHal.drvState.lod[0].mallocPtr) { 179 return; 180 } 181 182 bool isFirstUpload = false; 183 184 if (!drv->textureID) { 185 RSD_CALL_GL(glGenTextures, 1, &drv->textureID); 186 isFirstUpload = true; 187 } 188 189 Upload2DTexture(rsc, alloc, isFirstUpload); 190 191 if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) { 192 if (alloc->mHal.drvState.lod[0].mallocPtr) { 193 free(alloc->mHal.drvState.lod[0].mallocPtr); 194 alloc->mHal.drvState.lod[0].mallocPtr = NULL; 195 } 196 } 197 rsdGLCheckError(rsc, "UploadToTexture"); 198#endif 199} 200 201static void AllocateRenderTarget(const Context *rsc, const Allocation *alloc) { 202#ifndef RS_COMPATIBILITY_LIB 203 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 204 205 if (!drv->glFormat) { 206 return; 207 } 208 209 if (!drv->renderTargetID) { 210 RSD_CALL_GL(glGenRenderbuffers, 1, &drv->renderTargetID); 211 212 if (!drv->renderTargetID) { 213 // This should generally not happen 214 ALOGE("allocateRenderTarget failed to gen mRenderTargetID"); 215 rsc->dumpDebug(); 216 return; 217 } 218 RSD_CALL_GL(glBindRenderbuffer, GL_RENDERBUFFER, drv->renderTargetID); 219 RSD_CALL_GL(glRenderbufferStorage, GL_RENDERBUFFER, drv->glFormat, 220 alloc->mHal.drvState.lod[0].dimX, alloc->mHal.drvState.lod[0].dimY); 221 } 222 rsdGLCheckError(rsc, "AllocateRenderTarget"); 223#endif 224} 225 226static void UploadToBufferObject(const Context *rsc, const Allocation *alloc) { 227#ifndef RS_COMPATIBILITY_LIB 228 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 229 230 rsAssert(!alloc->mHal.state.type->getDimY()); 231 rsAssert(!alloc->mHal.state.type->getDimZ()); 232 233 //alloc->mHal.state.usageFlags |= RS_ALLOCATION_USAGE_GRAPHICS_VERTEX; 234 235 if (!drv->bufferID) { 236 RSD_CALL_GL(glGenBuffers, 1, &drv->bufferID); 237 } 238 if (!drv->bufferID) { 239 ALOGE("Upload to buffer object failed"); 240 drv->uploadDeferred = true; 241 return; 242 } 243 RSD_CALL_GL(glBindBuffer, drv->glTarget, drv->bufferID); 244 RSD_CALL_GL(glBufferData, drv->glTarget, alloc->mHal.state.type->getSizeBytes(), 245 alloc->mHal.drvState.lod[0].mallocPtr, GL_DYNAMIC_DRAW); 246 RSD_CALL_GL(glBindBuffer, drv->glTarget, 0); 247 rsdGLCheckError(rsc, "UploadToBufferObject"); 248#endif 249} 250 251 252static size_t DeriveYUVLayout(int yuv, Allocation::Hal::DrvState *state) { 253 // YUV only supports basic 2d 254 // so we can stash the plane pointers in the mipmap levels. 255 size_t uvSize = 0; 256#ifndef RS_SERVER 257 switch(yuv) { 258 case HAL_PIXEL_FORMAT_YV12: 259 state->lod[2].dimX = state->lod[0].dimX / 2; 260 state->lod[2].dimY = state->lod[0].dimY / 2; 261 state->lod[2].stride = rsRound(state->lod[0].stride >> 1, 16); 262 state->lod[2].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) + 263 (state->lod[0].stride * state->lod[0].dimY); 264 uvSize += state->lod[2].stride * state->lod[2].dimY; 265 266 state->lod[1].dimX = state->lod[2].dimX; 267 state->lod[1].dimY = state->lod[2].dimY; 268 state->lod[1].stride = state->lod[2].stride; 269 state->lod[1].mallocPtr = ((uint8_t *)state->lod[2].mallocPtr) + 270 (state->lod[2].stride * state->lod[2].dimY); 271 uvSize += state->lod[1].stride * state->lod[2].dimY; 272 273 state->lodCount = 3; 274 break; 275 case HAL_PIXEL_FORMAT_YCrCb_420_SP: // NV21 276 state->lod[1].dimX = state->lod[0].dimX; 277 state->lod[1].dimY = state->lod[0].dimY / 2; 278 state->lod[1].stride = state->lod[0].stride; 279 state->lod[1].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) + 280 (state->lod[0].stride * state->lod[0].dimY); 281 uvSize += state->lod[1].stride * state->lod[1].dimY; 282 state->lodCount = 2; 283 break; 284 default: 285 rsAssert(0); 286 } 287#endif 288 return uvSize; 289} 290 291 292static size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc, 293 const Type *type, uint8_t *ptr) { 294 alloc->mHal.drvState.lod[0].dimX = type->getDimX(); 295 alloc->mHal.drvState.lod[0].dimY = type->getDimY(); 296 alloc->mHal.drvState.lod[0].dimZ = type->getDimZ(); 297 alloc->mHal.drvState.lod[0].mallocPtr = 0; 298 // Stride needs to be 16-byte aligned too! 299 size_t stride = alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes(); 300 alloc->mHal.drvState.lod[0].stride = rsRound(stride, 16); 301 alloc->mHal.drvState.lodCount = type->getLODCount(); 302 alloc->mHal.drvState.faceCount = type->getDimFaces(); 303 304 size_t offsets[Allocation::MAX_LOD]; 305 memset(offsets, 0, sizeof(offsets)); 306 307 size_t o = alloc->mHal.drvState.lod[0].stride * rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) * 308 rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u); 309 if(alloc->mHal.drvState.lodCount > 1) { 310 uint32_t tx = alloc->mHal.drvState.lod[0].dimX; 311 uint32_t ty = alloc->mHal.drvState.lod[0].dimY; 312 uint32_t tz = alloc->mHal.drvState.lod[0].dimZ; 313 for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) { 314 alloc->mHal.drvState.lod[lod].dimX = tx; 315 alloc->mHal.drvState.lod[lod].dimY = ty; 316 alloc->mHal.drvState.lod[lod].dimZ = tz; 317 alloc->mHal.drvState.lod[lod].stride = 318 rsRound(tx * type->getElementSizeBytes(), 16); 319 offsets[lod] = o; 320 o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u); 321 if (tx > 1) tx >>= 1; 322 if (ty > 1) ty >>= 1; 323 if (tz > 1) tz >>= 1; 324 } 325 } else if (alloc->mHal.state.yuv) { 326 o += DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState); 327 328 for (uint32_t ct = 1; ct < alloc->mHal.drvState.lodCount; ct++) { 329 offsets[ct] = (size_t)alloc->mHal.drvState.lod[ct].mallocPtr; 330 } 331 } 332 333 alloc->mHal.drvState.faceOffset = o; 334 335 alloc->mHal.drvState.lod[0].mallocPtr = ptr; 336 for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) { 337 alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod]; 338 } 339 340 size_t allocSize = alloc->mHal.drvState.faceOffset; 341 if(alloc->mHal.drvState.faceCount) { 342 allocSize *= 6; 343 } 344 345 return allocSize; 346} 347 348static uint8_t* allocAlignedMemory(size_t allocSize, bool forceZero) { 349 // We align all allocations to a 16-byte boundary. 350 uint8_t* ptr = (uint8_t *)memalign(16, allocSize); 351 if (!ptr) { 352 return NULL; 353 } 354 if (forceZero) { 355 memset(ptr, 0, allocSize); 356 } 357 return ptr; 358} 359 360bool rsdAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) { 361 DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation)); 362 if (!drv) { 363 return false; 364 } 365 alloc->mHal.drv = drv; 366 367 // Calculate the object size. 368 size_t allocSize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), NULL); 369 370 uint8_t * ptr = NULL; 371 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) { 372 373 } else if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) { 374 // Allocation is allocated when the surface is created 375 // in getSurface 376 } else if (alloc->mHal.state.userProvidedPtr != NULL) { 377 // user-provided allocation 378 // limitations: no faces, no LOD, USAGE_SCRIPT or SCRIPT+TEXTURE only 379 if (!(alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED) || 380 alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED | RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE))) { 381 ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT | USAGE_SHARED or USAGE_SCRIPT | USAGE_SHARED | USAGE_GRAPHICS_TEXTURE"); 382 return false; 383 } 384 if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) { 385 ALOGE("User-allocated buffers must not have multiple faces or LODs"); 386 return false; 387 } 388 389 // rows must be 16-byte aligned 390 // validate that here, otherwise fall back to not use the user-backed allocation 391 if (((alloc->getType()->getDimX() * alloc->getType()->getElement()->getSizeBytes()) % 16) != 0) { 392 ALOGV("User-backed allocation failed stride requirement, falling back to separate allocation"); 393 drv->useUserProvidedPtr = false; 394 395 ptr = allocAlignedMemory(allocSize, forceZero); 396 if (!ptr) { 397 alloc->mHal.drv = NULL; 398 free(drv); 399 return false; 400 } 401 402 } else { 403 drv->useUserProvidedPtr = true; 404 ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr; 405 } 406 } else { 407 ptr = allocAlignedMemory(allocSize, forceZero); 408 if (!ptr) { 409 alloc->mHal.drv = NULL; 410 free(drv); 411 return false; 412 } 413 } 414 // Build the pointer tables 415 size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr); 416 if(allocSize != verifySize) { 417 rsAssert(!"Size mismatch"); 418 } 419 420#ifndef RS_SERVER 421 drv->glTarget = GL_NONE; 422 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) { 423 if (alloc->mHal.state.hasFaces) { 424 drv->glTarget = GL_TEXTURE_CUBE_MAP; 425 } else { 426 drv->glTarget = GL_TEXTURE_2D; 427 } 428 } else { 429 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) { 430 drv->glTarget = GL_ARRAY_BUFFER; 431 } 432 } 433#endif 434 435#ifndef RS_COMPATIBILITY_LIB 436 drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType()); 437 drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind()); 438#else 439 drv->glType = 0; 440 drv->glFormat = 0; 441#endif 442 443 if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) { 444 drv->uploadDeferred = true; 445 } 446 447 448 drv->readBackFBO = NULL; 449 450 // fill out the initial state of the buffer if we couldn't use the user-provided ptr and USAGE_SHARED was accepted 451 if ((alloc->mHal.state.userProvidedPtr != 0) && (drv->useUserProvidedPtr == false)) { 452 rsdAllocationData2D(rsc, alloc, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X, alloc->getType()->getDimX(), alloc->getType()->getDimY(), alloc->mHal.state.userProvidedPtr, allocSize, 0); 453 } 454 455 return true; 456} 457 458void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) { 459 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 460 461#ifndef RS_COMPATIBILITY_LIB 462 if (drv->bufferID) { 463 // Causes a SW crash.... 464 //ALOGV(" mBufferID %i", mBufferID); 465 //glDeleteBuffers(1, &mBufferID); 466 //mBufferID = 0; 467 } 468 if (drv->textureID) { 469 RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID); 470 drv->textureID = 0; 471 } 472 if (drv->renderTargetID) { 473 RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID); 474 drv->renderTargetID = 0; 475 } 476#endif 477 478 if (alloc->mHal.drvState.lod[0].mallocPtr) { 479 // don't free user-allocated ptrs or IO_OUTPUT buffers 480 if (!(drv->useUserProvidedPtr) && 481 !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) && 482 !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) { 483 free(alloc->mHal.drvState.lod[0].mallocPtr); 484 } 485 alloc->mHal.drvState.lod[0].mallocPtr = NULL; 486 } 487 488#ifndef RS_COMPATIBILITY_LIB 489 if (drv->readBackFBO != NULL) { 490 delete drv->readBackFBO; 491 drv->readBackFBO = NULL; 492 } 493 494 if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) && 495 (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) { 496 497 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 498 ANativeWindow *nw = drv->wndSurface; 499 if (nw) { 500 GraphicBufferMapper &mapper = GraphicBufferMapper::get(); 501 mapper.unlock(drv->wndBuffer->handle); 502 int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1); 503 } 504 } 505#endif 506 507 free(drv); 508 alloc->mHal.drv = NULL; 509} 510 511void rsdAllocationResize(const Context *rsc, const Allocation *alloc, 512 const Type *newType, bool zeroNew) { 513 const uint32_t oldDimX = alloc->mHal.drvState.lod[0].dimX; 514 const uint32_t dimX = newType->getDimX(); 515 516 // can't resize Allocations with user-allocated buffers 517 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) { 518 ALOGE("Resize cannot be called on a USAGE_SHARED allocation"); 519 return; 520 } 521 void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr; 522 // Calculate the object size 523 size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL); 524 uint8_t *ptr = (uint8_t *)realloc(oldPtr, s); 525 // Build the relative pointer tables. 526 size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr); 527 if(s != verifySize) { 528 rsAssert(!"Size mismatch"); 529 } 530 531 532 if (dimX > oldDimX) { 533 size_t stride = alloc->mHal.state.elementSizeBytes; 534 memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX, 535 0, stride * (dimX - oldDimX)); 536 } 537} 538 539static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) { 540#ifndef RS_COMPATIBILITY_LIB 541 if (!alloc->getIsScript()) { 542 return; // nothing to sync 543 } 544 545 RsdHal *dc = (RsdHal *)rsc->mHal.drv; 546 RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer; 547 548 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 549 if (!drv->textureID && !drv->renderTargetID) { 550 return; // nothing was rendered here yet, so nothing to sync 551 } 552 if (drv->readBackFBO == NULL) { 553 drv->readBackFBO = new RsdFrameBufferObj(); 554 drv->readBackFBO->setColorTarget(drv, 0); 555 drv->readBackFBO->setDimensions(alloc->getType()->getDimX(), 556 alloc->getType()->getDimY()); 557 } 558 559 // Bind the framebuffer object so we can read back from it 560 drv->readBackFBO->setActive(rsc); 561 562 // Do the readback 563 RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX, 564 alloc->mHal.drvState.lod[0].dimY, 565 drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr); 566 567 // Revert framebuffer to its original 568 lastFbo->setActive(rsc); 569#endif 570} 571 572 573void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc, 574 RsAllocationUsageType src) { 575 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 576 577 if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) { 578 if(!alloc->getIsRenderTarget()) { 579 rsc->setError(RS_ERROR_FATAL_DRIVER, 580 "Attempting to sync allocation from render target, " 581 "for non-render target allocation"); 582 } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) { 583 rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA" 584 "render target"); 585 } else { 586 rsdAllocationSyncFromFBO(rsc, alloc); 587 } 588 return; 589 } 590 591 rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT); 592 593 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) { 594 UploadToTexture(rsc, alloc); 595 } else { 596 if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) && 597 !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) { 598 AllocateRenderTarget(rsc, alloc); 599 } 600 } 601 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) { 602 UploadToBufferObject(rsc, alloc); 603 } 604 605 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) { 606 // NOP in CPU driver for now 607 } 608 609 drv->uploadDeferred = false; 610} 611 612void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) { 613 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 614 drv->uploadDeferred = true; 615} 616 617#ifndef RS_COMPATIBILITY_LIB 618void DrvAllocation::NewBufferListener::onFrameAvailable() { 619 intptr_t ip = (intptr_t)alloc; 620 rsc->sendMessageToClient(NULL, RS_MESSAGE_TO_CLIENT_NEW_BUFFER, ip, 0, true); 621} 622#endif 623 624void* rsdAllocationGetSurface(const Context *rsc, const Allocation *alloc) { 625#ifndef RS_COMPATIBILITY_LIB 626 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 627 628 // Configure CpuConsumer to be in asynchronous mode 629 drv->cpuConsumer = new CpuConsumer(2, false); 630 sp<IGraphicBufferProducer> bp = drv->cpuConsumer->getProducerInterface(); 631 bp->incStrong(NULL); 632 633 drv->mBufferListener = new DrvAllocation::NewBufferListener(); 634 drv->mBufferListener->rsc = rsc; 635 drv->mBufferListener->alloc = alloc; 636 637 drv->cpuConsumer->setFrameAvailableListener(drv->mBufferListener); 638 639 return bp.get(); 640#else 641 return NULL; 642#endif 643} 644 645#ifndef RS_COMPATIBILITY_LIB 646static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) { 647 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 648 649 int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer); 650 if (r) { 651 rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer."); 652 return false; 653 } 654 655 // Must lock the whole surface 656 GraphicBufferMapper &mapper = GraphicBufferMapper::get(); 657 Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height); 658 659 void *dst = NULL; 660 mapper.lock(drv->wndBuffer->handle, 661 GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN, 662 bounds, &dst); 663 alloc->mHal.drvState.lod[0].mallocPtr = dst; 664 alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes; 665 rsAssert((alloc->mHal.drvState.lod[0].stride & 0xf) == 0); 666 667 return true; 668} 669#endif 670 671void rsdAllocationSetSurface(const Context *rsc, Allocation *alloc, ANativeWindow *nw) { 672#ifndef RS_COMPATIBILITY_LIB 673 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 674 ANativeWindow *old = drv->wndSurface; 675 676 if (nw) { 677 nw->incStrong(NULL); 678 } 679 680 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) { 681 //TODO finish support for render target + script 682 drv->wnd = nw; 683 return; 684 } 685 686 // Cleanup old surface if there is one. 687 if (drv->wndSurface) { 688 ANativeWindow *old = drv->wndSurface; 689 GraphicBufferMapper &mapper = GraphicBufferMapper::get(); 690 mapper.unlock(drv->wndBuffer->handle); 691 old->cancelBuffer(old, drv->wndBuffer, -1); 692 drv->wndSurface = NULL; 693 old->decStrong(NULL); 694 } 695 696 if (nw != NULL) { 697 int32_t r; 698 uint32_t flags = 0; 699 700 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) { 701 flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN; 702 } 703 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) { 704 flags |= GRALLOC_USAGE_HW_RENDER; 705 } 706 707 r = native_window_set_usage(nw, flags); 708 if (r) { 709 rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage."); 710 goto error; 711 } 712 713 r = native_window_set_buffers_dimensions(nw, alloc->mHal.drvState.lod[0].dimX, 714 alloc->mHal.drvState.lod[0].dimY); 715 if (r) { 716 rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions."); 717 goto error; 718 } 719 720 int format = 0; 721 const Element *e = alloc->mHal.state.type->getElement(); 722 switch(e->getType()) { 723 case RS_TYPE_UNSIGNED_8: 724 switch (e->getVectorSize()) { 725 case 1: 726 rsAssert(e->getKind() == RS_KIND_PIXEL_A); 727 format = PIXEL_FORMAT_A_8; 728 break; 729 case 4: 730 rsAssert(e->getKind() == RS_KIND_PIXEL_RGBA); 731 format = PIXEL_FORMAT_RGBA_8888; 732 break; 733 default: 734 rsAssert(0); 735 } 736 break; 737 default: 738 rsAssert(0); 739 } 740 741 r = native_window_set_buffers_format(nw, format); 742 if (r) { 743 rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer format."); 744 goto error; 745 } 746 747 IoGetBuffer(rsc, alloc, nw); 748 drv->wndSurface = nw; 749 } 750 751 return; 752 753 error: 754 755 if (nw) { 756 nw->decStrong(NULL); 757 } 758 759 760#endif 761} 762 763void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) { 764#ifndef RS_COMPATIBILITY_LIB 765 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 766 ANativeWindow *nw = drv->wndSurface; 767 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) { 768 RsdHal *dc = (RsdHal *)rsc->mHal.drv; 769 RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface); 770 return; 771 } 772 if (nw) { 773 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) { 774 GraphicBufferMapper &mapper = GraphicBufferMapper::get(); 775 mapper.unlock(drv->wndBuffer->handle); 776 int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1); 777 if (r) { 778 rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer."); 779 return; 780 } 781 782 IoGetBuffer(rsc, alloc, nw); 783 } 784 } else { 785 rsc->setError(RS_ERROR_DRIVER, "Sent IO buffer with no attached surface."); 786 return; 787 } 788#endif 789} 790 791void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) { 792#ifndef RS_COMPATIBILITY_LIB 793 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 794 795 if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) { 796 CpuConsumer::LockedBuffer lb; 797 status_t ret = drv->cpuConsumer->lockNextBuffer(&lb); 798 if (ret == OK) { 799 if (drv->lb.data != NULL) { 800 drv->cpuConsumer->unlockBuffer(drv->lb); 801 } 802 drv->lb = lb; 803 alloc->mHal.drvState.lod[0].mallocPtr = drv->lb.data; 804 alloc->mHal.drvState.lod[0].stride = drv->lb.stride * 805 alloc->mHal.state.elementSizeBytes; 806 807 if (alloc->mHal.state.yuv) { 808 DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState); 809 } 810 } else if (ret == BAD_VALUE) { 811 // No new frame, don't do anything 812 } else { 813 rsc->setError(RS_ERROR_DRIVER, "Error receiving IO input buffer."); 814 } 815 816 } else { 817 drv->surfaceTexture->updateTexImage(); 818 } 819 820 821#endif 822} 823 824 825void rsdAllocationData1D(const Context *rsc, const Allocation *alloc, 826 uint32_t xoff, uint32_t lod, size_t count, 827 const void *data, size_t sizeBytes) { 828 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 829 830 const size_t eSize = alloc->mHal.state.type->getElementSizeBytes(); 831 uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 832 size_t size = count * eSize; 833 834 if (ptr != data) { 835 // Skip the copy if we are the same allocation. This can arise from 836 // our Bitmap optimization, where we share the same storage. 837 if (alloc->mHal.state.hasReferences) { 838 alloc->incRefs(data, count); 839 alloc->decRefs(ptr, count); 840 } 841 memcpy(ptr, data, size); 842 } 843 drv->uploadDeferred = true; 844} 845 846void rsdAllocationData2D(const Context *rsc, const Allocation *alloc, 847 uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face, 848 uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) { 849 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 850 851 size_t eSize = alloc->mHal.state.elementSizeBytes; 852 size_t lineSize = eSize * w; 853 if (!stride) { 854 stride = lineSize; 855 } 856 857 if (alloc->mHal.drvState.lod[0].mallocPtr) { 858 const uint8_t *src = static_cast<const uint8_t *>(data); 859 uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face); 860 if (dst == src) { 861 // Skip the copy if we are the same allocation. This can arise from 862 // our Bitmap optimization, where we share the same storage. 863 drv->uploadDeferred = true; 864 return; 865 } 866 867 for (uint32_t line=yoff; line < (yoff+h); line++) { 868 if (alloc->mHal.state.hasReferences) { 869 alloc->incRefs(src, w); 870 alloc->decRefs(dst, w); 871 } 872 memcpy(dst, src, lineSize); 873 src += stride; 874 dst += alloc->mHal.drvState.lod[lod].stride; 875 } 876 if (alloc->mHal.state.yuv) { 877 int lod = 1; 878 while (alloc->mHal.drvState.lod[lod].mallocPtr) { 879 size_t lineSize = alloc->mHal.drvState.lod[lod].dimX; 880 uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face); 881 882 for (uint32_t line=(yoff >> 1); line < ((yoff+h)>>1); line++) { 883 memcpy(dst, src, lineSize); 884 src += lineSize; 885 dst += alloc->mHal.drvState.lod[lod].stride; 886 } 887 lod++; 888 } 889 890 } 891 drv->uploadDeferred = true; 892 } else { 893 Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h); 894 } 895} 896 897void rsdAllocationData3D(const Context *rsc, const Allocation *alloc, 898 uint32_t xoff, uint32_t yoff, uint32_t zoff, 899 uint32_t lod, 900 uint32_t w, uint32_t h, uint32_t d, const void *data, 901 size_t sizeBytes, size_t stride) { 902 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 903 904 uint32_t eSize = alloc->mHal.state.elementSizeBytes; 905 uint32_t lineSize = eSize * w; 906 if (!stride) { 907 stride = lineSize; 908 } 909 910 if (alloc->mHal.drvState.lod[0].mallocPtr) { 911 const uint8_t *src = static_cast<const uint8_t *>(data); 912 for (uint32_t z = zoff; z < d; z++) { 913 uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, z, lod, 914 RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 915 if (dst == src) { 916 // Skip the copy if we are the same allocation. This can arise from 917 // our Bitmap optimization, where we share the same storage. 918 drv->uploadDeferred = true; 919 return; 920 } 921 922 for (uint32_t line=yoff; line < (yoff+h); line++) { 923 if (alloc->mHal.state.hasReferences) { 924 alloc->incRefs(src, w); 925 alloc->decRefs(dst, w); 926 } 927 memcpy(dst, src, lineSize); 928 src += stride; 929 dst += alloc->mHal.drvState.lod[lod].stride; 930 } 931 } 932 drv->uploadDeferred = true; 933 } 934} 935 936void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc, 937 uint32_t xoff, uint32_t lod, size_t count, 938 void *data, size_t sizeBytes) { 939 const size_t eSize = alloc->mHal.state.type->getElementSizeBytes(); 940 const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 941 if (data != ptr) { 942 // Skip the copy if we are the same allocation. This can arise from 943 // our Bitmap optimization, where we share the same storage. 944 memcpy(data, ptr, count * eSize); 945 } 946} 947 948void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc, 949 uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face, 950 uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) { 951 size_t eSize = alloc->mHal.state.elementSizeBytes; 952 size_t lineSize = eSize * w; 953 if (!stride) { 954 stride = lineSize; 955 } 956 957 if (alloc->mHal.drvState.lod[0].mallocPtr) { 958 uint8_t *dst = static_cast<uint8_t *>(data); 959 const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face); 960 if (dst == src) { 961 // Skip the copy if we are the same allocation. This can arise from 962 // our Bitmap optimization, where we share the same storage. 963 return; 964 } 965 966 for (uint32_t line=yoff; line < (yoff+h); line++) { 967 memcpy(dst, src, lineSize); 968 dst += stride; 969 src += alloc->mHal.drvState.lod[lod].stride; 970 } 971 } else { 972 ALOGE("Add code to readback from non-script memory"); 973 } 974} 975 976 977void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc, 978 uint32_t xoff, uint32_t yoff, uint32_t zoff, 979 uint32_t lod, 980 uint32_t w, uint32_t h, uint32_t d, void *data, size_t sizeBytes, size_t stride) { 981 uint32_t eSize = alloc->mHal.state.elementSizeBytes; 982 uint32_t lineSize = eSize * w; 983 if (!stride) { 984 stride = lineSize; 985 } 986 987 if (alloc->mHal.drvState.lod[0].mallocPtr) { 988 uint8_t *dst = static_cast<uint8_t *>(data); 989 for (uint32_t z = zoff; z < d; z++) { 990 const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, z, lod, 991 RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 992 if (dst == src) { 993 // Skip the copy if we are the same allocation. This can arise from 994 // our Bitmap optimization, where we share the same storage. 995 return; 996 } 997 998 for (uint32_t line=yoff; line < (yoff+h); line++) { 999 memcpy(dst, src, lineSize); 1000 dst += stride; 1001 src += alloc->mHal.drvState.lod[lod].stride; 1002 } 1003 } 1004 } 1005} 1006 1007void * rsdAllocationLock1D(const android::renderscript::Context *rsc, 1008 const android::renderscript::Allocation *alloc) { 1009 return alloc->mHal.drvState.lod[0].mallocPtr; 1010} 1011 1012void rsdAllocationUnlock1D(const android::renderscript::Context *rsc, 1013 const android::renderscript::Allocation *alloc) { 1014 1015} 1016 1017void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc, 1018 const android::renderscript::Allocation *dstAlloc, 1019 uint32_t dstXoff, uint32_t dstLod, size_t count, 1020 const android::renderscript::Allocation *srcAlloc, 1021 uint32_t srcXoff, uint32_t srcLod) { 1022} 1023 1024 1025void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc, 1026 const android::renderscript::Allocation *dstAlloc, 1027 uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod, 1028 RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h, 1029 const android::renderscript::Allocation *srcAlloc, 1030 uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod, 1031 RsAllocationCubemapFace srcFace) { 1032 size_t elementSize = dstAlloc->getType()->getElementSizeBytes(); 1033 for (uint32_t i = 0; i < h; i ++) { 1034 uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, 0, dstLod, dstFace); 1035 uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, 0, srcLod, srcFace); 1036 memcpy(dstPtr, srcPtr, w * elementSize); 1037 1038 //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)", 1039 // dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace); 1040 } 1041} 1042 1043void rsdAllocationData3D_alloc_script(const android::renderscript::Context *rsc, 1044 const android::renderscript::Allocation *dstAlloc, 1045 uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff, uint32_t dstLod, 1046 uint32_t w, uint32_t h, uint32_t d, 1047 const android::renderscript::Allocation *srcAlloc, 1048 uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff, uint32_t srcLod) { 1049 uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes(); 1050 for (uint32_t j = 0; j < d; j++) { 1051 for (uint32_t i = 0; i < h; i ++) { 1052 uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstZoff + j, 1053 dstLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 1054 uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcZoff + j, 1055 srcLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 1056 memcpy(dstPtr, srcPtr, w * elementSize); 1057 1058 //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)", 1059 // dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace); 1060 } 1061 } 1062} 1063 1064void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc, 1065 const android::renderscript::Allocation *dstAlloc, 1066 uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod, 1067 RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h, 1068 const android::renderscript::Allocation *srcAlloc, 1069 uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod, 1070 RsAllocationCubemapFace srcFace) { 1071 if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) { 1072 rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not " 1073 "yet implemented."); 1074 return; 1075 } 1076 rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, 1077 dstLod, dstFace, w, h, srcAlloc, 1078 srcXoff, srcYoff, srcLod, srcFace); 1079} 1080 1081void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc, 1082 const android::renderscript::Allocation *dstAlloc, 1083 uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff, 1084 uint32_t dstLod, 1085 uint32_t w, uint32_t h, uint32_t d, 1086 const android::renderscript::Allocation *srcAlloc, 1087 uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff, 1088 uint32_t srcLod) { 1089 if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) { 1090 rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not " 1091 "yet implemented."); 1092 return; 1093 } 1094 rsdAllocationData3D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstZoff, 1095 dstLod, w, h, d, srcAlloc, 1096 srcXoff, srcYoff, srcZoff, srcLod); 1097} 1098 1099void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc, 1100 uint32_t x, 1101 const void *data, uint32_t cIdx, size_t sizeBytes) { 1102 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 1103 1104 size_t eSize = alloc->mHal.state.elementSizeBytes; 1105 uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 1106 1107 const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx); 1108 ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx); 1109 1110 if (alloc->mHal.state.hasReferences) { 1111 e->incRefs(data); 1112 e->decRefs(ptr); 1113 } 1114 1115 memcpy(ptr, data, sizeBytes); 1116 drv->uploadDeferred = true; 1117} 1118 1119void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc, 1120 uint32_t x, uint32_t y, 1121 const void *data, uint32_t cIdx, size_t sizeBytes) { 1122 DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv; 1123 1124 size_t eSize = alloc->mHal.state.elementSizeBytes; 1125 uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X); 1126 1127 const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx); 1128 ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx); 1129 1130 if (alloc->mHal.state.hasReferences) { 1131 e->incRefs(data); 1132 e->decRefs(ptr); 1133 } 1134 1135 memcpy(ptr, data, sizeBytes); 1136 drv->uploadDeferred = true; 1137} 1138 1139static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) { 1140 uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX; 1141 uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY; 1142 1143 for (uint32_t y=0; y < h; y++) { 1144 uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face); 1145 const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2, lod, face); 1146 const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2+1, lod, face); 1147 1148 for (uint32_t x=0; x < w; x++) { 1149 *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]); 1150 oPtr ++; 1151 i1 += 2; 1152 i2 += 2; 1153 } 1154 } 1155} 1156 1157static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) { 1158 uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX; 1159 uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY; 1160 1161 for (uint32_t y=0; y < h; y++) { 1162 uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face); 1163 const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, 0, lod, face); 1164 const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face); 1165 1166 for (uint32_t x=0; x < w; x++) { 1167 *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]); 1168 oPtr ++; 1169 i1 += 2; 1170 i2 += 2; 1171 } 1172 } 1173} 1174 1175static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) { 1176 uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX; 1177 uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY; 1178 1179 for (uint32_t y=0; y < h; y++) { 1180 uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, 0, lod + 1, face); 1181 const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, 0, lod, face); 1182 const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face); 1183 1184 for (uint32_t x=0; x < w; x++) { 1185 *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f); 1186 oPtr ++; 1187 i1 += 2; 1188 i2 += 2; 1189 } 1190 } 1191} 1192 1193void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) { 1194 if(!alloc->mHal.drvState.lod[0].mallocPtr) { 1195 return; 1196 } 1197 uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1; 1198 for (uint32_t face = 0; face < numFaces; face ++) { 1199 for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) { 1200 switch (alloc->getType()->getElement()->getSizeBits()) { 1201 case 32: 1202 mip8888(alloc, lod, (RsAllocationCubemapFace)face); 1203 break; 1204 case 16: 1205 mip565(alloc, lod, (RsAllocationCubemapFace)face); 1206 break; 1207 case 8: 1208 mip8(alloc, lod, (RsAllocationCubemapFace)face); 1209 break; 1210 } 1211 } 1212 } 1213} 1214