rsdAllocation.cpp revision aafa7a84e3158042ca456b1f3329c2890c72d53f 
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_OUTPUT)) {
482            free(alloc->mHal.drvState.lod[0].mallocPtr);
483        }
484        alloc->mHal.drvState.lod[0].mallocPtr = NULL;
485    }
486
487#ifndef RS_COMPATIBILITY_LIB
488    if (drv->readBackFBO != NULL) {
489        delete drv->readBackFBO;
490        drv->readBackFBO = NULL;
491    }
492
493    if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) &&
494        (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
495
496        DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
497        ANativeWindow *nw = drv->wndSurface;
498        if (nw) {
499            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
500            mapper.unlock(drv->wndBuffer->handle);
501            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
502        }
503    }
504#endif
505
506    free(drv);
507    alloc->mHal.drv = NULL;
508}
509
510void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
511                         const Type *newType, bool zeroNew) {
512    const uint32_t oldDimX = alloc->mHal.drvState.lod[0].dimX;
513    const uint32_t dimX = newType->getDimX();
514
515    // can't resize Allocations with user-allocated buffers
516    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
517        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
518        return;
519    }
520    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
521    // Calculate the object size
522    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL);
523    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
524    // Build the relative pointer tables.
525    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
526    if(s != verifySize) {
527        rsAssert(!"Size mismatch");
528    }
529
530
531    if (dimX > oldDimX) {
532        size_t stride = alloc->mHal.state.elementSizeBytes;
533        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
534                 0, stride * (dimX - oldDimX));
535    }
536}
537
538static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
539#ifndef RS_COMPATIBILITY_LIB
540    if (!alloc->getIsScript()) {
541        return; // nothing to sync
542    }
543
544    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
545    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
546
547    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
548    if (!drv->textureID && !drv->renderTargetID) {
549        return; // nothing was rendered here yet, so nothing to sync
550    }
551    if (drv->readBackFBO == NULL) {
552        drv->readBackFBO = new RsdFrameBufferObj();
553        drv->readBackFBO->setColorTarget(drv, 0);
554        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
555                                        alloc->getType()->getDimY());
556    }
557
558    // Bind the framebuffer object so we can read back from it
559    drv->readBackFBO->setActive(rsc);
560
561    // Do the readback
562    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
563                alloc->mHal.drvState.lod[0].dimY,
564                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
565
566    // Revert framebuffer to its original
567    lastFbo->setActive(rsc);
568#endif
569}
570
571
572void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
573                         RsAllocationUsageType src) {
574    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
575
576    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
577        if(!alloc->getIsRenderTarget()) {
578            rsc->setError(RS_ERROR_FATAL_DRIVER,
579                          "Attempting to sync allocation from render target, "
580                          "for non-render target allocation");
581        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
582            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
583                                                 "render target");
584        } else {
585            rsdAllocationSyncFromFBO(rsc, alloc);
586        }
587        return;
588    }
589
590    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT);
591
592    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
593        UploadToTexture(rsc, alloc);
594    } else {
595        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
596            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
597            AllocateRenderTarget(rsc, alloc);
598        }
599    }
600    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
601        UploadToBufferObject(rsc, alloc);
602    }
603
604    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
605        // NOP in CPU driver for now
606    }
607
608    drv->uploadDeferred = false;
609}
610
611void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
612    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
613    drv->uploadDeferred = true;
614}
615
616#ifndef RS_COMPATIBILITY_LIB
617void DrvAllocation::NewBufferListener::onFrameAvailable() {
618    intptr_t ip = (intptr_t)alloc;
619    rsc->sendMessageToClient(NULL, RS_MESSAGE_TO_CLIENT_NEW_BUFFER, ip, 0, true);
620}
621#endif
622
623void* rsdAllocationGetSurface(const Context *rsc, const Allocation *alloc) {
624#ifndef RS_COMPATIBILITY_LIB
625    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
626
627    // Configure CpuConsumer to be in asynchronous mode
628    drv->cpuConsumer = new CpuConsumer(2, false);
629    sp<IGraphicBufferProducer> bp = drv->cpuConsumer->getProducerInterface();
630    bp->incStrong(NULL);
631
632    drv->mBufferListener = new DrvAllocation::NewBufferListener();
633    drv->mBufferListener->rsc = rsc;
634    drv->mBufferListener->alloc = alloc;
635
636    drv->cpuConsumer->setFrameAvailableListener(drv->mBufferListener);
637
638    return bp.get();
639#else
640    return NULL;
641#endif
642}
643
644#ifndef RS_COMPATIBILITY_LIB
645static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
646    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
647
648    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
649    if (r) {
650        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
651        return false;
652    }
653
654    // Must lock the whole surface
655    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
656    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
657
658    void *dst = NULL;
659    mapper.lock(drv->wndBuffer->handle,
660            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
661            bounds, &dst);
662    alloc->mHal.drvState.lod[0].mallocPtr = dst;
663    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
664    rsAssert((alloc->mHal.drvState.lod[0].stride & 0xf) == 0);
665
666    return true;
667}
668#endif
669
670void rsdAllocationSetSurface(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
671#ifndef RS_COMPATIBILITY_LIB
672    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
673    ANativeWindow *old = drv->wndSurface;
674
675    if (nw) {
676        nw->incStrong(NULL);
677    }
678
679    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
680        //TODO finish support for render target + script
681        drv->wnd = nw;
682        return;
683    }
684
685    // Cleanup old surface if there is one.
686    if (drv->wndSurface) {
687        ANativeWindow *old = drv->wndSurface;
688        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
689        mapper.unlock(drv->wndBuffer->handle);
690        old->cancelBuffer(old, drv->wndBuffer, -1);
691        drv->wndSurface = NULL;
692        old->decStrong(NULL);
693    }
694
695    if (nw != NULL) {
696        int32_t r;
697        uint32_t flags = 0;
698
699        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
700            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
701        }
702        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
703            flags |= GRALLOC_USAGE_HW_RENDER;
704        }
705
706        r = native_window_set_usage(nw, flags);
707        if (r) {
708            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
709            goto error;
710        }
711
712        r = native_window_set_buffers_dimensions(nw, alloc->mHal.drvState.lod[0].dimX,
713                                                 alloc->mHal.drvState.lod[0].dimY);
714        if (r) {
715            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
716            goto error;
717        }
718
719        int format = 0;
720        const Element *e = alloc->mHal.state.type->getElement();
721        switch(e->getType()) {
722        case RS_TYPE_UNSIGNED_8:
723            switch (e->getVectorSize()) {
724            case 1:
725                rsAssert(e->getKind() == RS_KIND_PIXEL_A);
726                format = PIXEL_FORMAT_A_8;
727                break;
728            case 4:
729                rsAssert(e->getKind() == RS_KIND_PIXEL_RGBA);
730                format = PIXEL_FORMAT_RGBA_8888;
731                break;
732            default:
733                rsAssert(0);
734            }
735            break;
736        default:
737            rsAssert(0);
738        }
739
740        r = native_window_set_buffers_format(nw, format);
741        if (r) {
742            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer format.");
743            goto error;
744        }
745
746        IoGetBuffer(rsc, alloc, nw);
747        drv->wndSurface = nw;
748    }
749
750    return;
751
752 error:
753
754    if (nw) {
755        nw->decStrong(NULL);
756    }
757
758
759#endif
760}
761
762void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
763#ifndef RS_COMPATIBILITY_LIB
764    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
765    ANativeWindow *nw = drv->wndSurface;
766    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
767        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
768        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
769        return;
770    }
771    if (nw) {
772        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
773            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
774            mapper.unlock(drv->wndBuffer->handle);
775            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
776            if (r) {
777                rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
778                return;
779            }
780
781            IoGetBuffer(rsc, alloc, nw);
782        }
783    } else {
784        rsc->setError(RS_ERROR_DRIVER, "Sent IO buffer with no attached surface.");
785        return;
786    }
787#endif
788}
789
790void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
791#ifndef RS_COMPATIBILITY_LIB
792    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
793
794    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
795        CpuConsumer::LockedBuffer lb;
796        status_t ret = drv->cpuConsumer->lockNextBuffer(&lb);
797        if (ret == OK) {
798            if (drv->lb.data != NULL) {
799                drv->cpuConsumer->unlockBuffer(drv->lb);
800            }
801            drv->lb = lb;
802            alloc->mHal.drvState.lod[0].mallocPtr = drv->lb.data;
803            alloc->mHal.drvState.lod[0].stride = drv->lb.stride *
804                    alloc->mHal.state.elementSizeBytes;
805
806            if (alloc->mHal.state.yuv) {
807                DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
808            }
809        } else if (ret == BAD_VALUE) {
810            // No new frame, don't do anything
811        } else {
812            rsc->setError(RS_ERROR_DRIVER, "Error receiving IO input buffer.");
813        }
814
815    } else {
816        drv->surfaceTexture->updateTexImage();
817    }
818
819
820#endif
821}
822
823
824void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
825                         uint32_t xoff, uint32_t lod, size_t count,
826                         const void *data, size_t sizeBytes) {
827    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
828
829    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
830    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
831    size_t size = count * eSize;
832
833    if (ptr != data) {
834        // Skip the copy if we are the same allocation. This can arise from
835        // our Bitmap optimization, where we share the same storage.
836        if (alloc->mHal.state.hasReferences) {
837            alloc->incRefs(data, count);
838            alloc->decRefs(ptr, count);
839        }
840        memcpy(ptr, data, size);
841    }
842    drv->uploadDeferred = true;
843}
844
845void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
846                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
847                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
848    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
849
850    size_t eSize = alloc->mHal.state.elementSizeBytes;
851    size_t lineSize = eSize * w;
852    if (!stride) {
853        stride = lineSize;
854    }
855
856    if (alloc->mHal.drvState.lod[0].mallocPtr) {
857        const uint8_t *src = static_cast<const uint8_t *>(data);
858        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
859        if (dst == src) {
860            // Skip the copy if we are the same allocation. This can arise from
861            // our Bitmap optimization, where we share the same storage.
862            drv->uploadDeferred = true;
863            return;
864        }
865
866        for (uint32_t line=yoff; line < (yoff+h); line++) {
867            if (alloc->mHal.state.hasReferences) {
868                alloc->incRefs(src, w);
869                alloc->decRefs(dst, w);
870            }
871            memcpy(dst, src, lineSize);
872            src += stride;
873            dst += alloc->mHal.drvState.lod[lod].stride;
874        }
875        if (alloc->mHal.state.yuv) {
876            int lod = 1;
877            while (alloc->mHal.drvState.lod[lod].mallocPtr) {
878                size_t lineSize = alloc->mHal.drvState.lod[lod].dimX;
879                uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
880
881                for (uint32_t line=(yoff >> 1); line < ((yoff+h)>>1); line++) {
882                    memcpy(dst, src, lineSize);
883                    src += lineSize;
884                    dst += alloc->mHal.drvState.lod[lod].stride;
885                }
886                lod++;
887            }
888
889        }
890        drv->uploadDeferred = true;
891    } else {
892        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
893    }
894}
895
896void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
897                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
898                         uint32_t lod,
899                         uint32_t w, uint32_t h, uint32_t d, const void *data,
900                         size_t sizeBytes, size_t stride) {
901    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
902
903    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
904    uint32_t lineSize = eSize * w;
905    if (!stride) {
906        stride = lineSize;
907    }
908
909    if (alloc->mHal.drvState.lod[0].mallocPtr) {
910        const uint8_t *src = static_cast<const uint8_t *>(data);
911        for (uint32_t z = zoff; z < d; z++) {
912            uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, z, lod,
913                                        RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
914            if (dst == src) {
915                // Skip the copy if we are the same allocation. This can arise from
916                // our Bitmap optimization, where we share the same storage.
917                drv->uploadDeferred = true;
918                return;
919            }
920
921            for (uint32_t line=yoff; line < (yoff+h); line++) {
922                if (alloc->mHal.state.hasReferences) {
923                    alloc->incRefs(src, w);
924                    alloc->decRefs(dst, w);
925                }
926                memcpy(dst, src, lineSize);
927                src += stride;
928                dst += alloc->mHal.drvState.lod[lod].stride;
929            }
930        }
931        drv->uploadDeferred = true;
932    }
933}
934
935void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
936                         uint32_t xoff, uint32_t lod, size_t count,
937                         void *data, size_t sizeBytes) {
938    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
939    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
940    if (data != ptr) {
941        // Skip the copy if we are the same allocation. This can arise from
942        // our Bitmap optimization, where we share the same storage.
943        memcpy(data, ptr, count * eSize);
944    }
945}
946
947void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
948                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
949                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
950    size_t eSize = alloc->mHal.state.elementSizeBytes;
951    size_t lineSize = eSize * w;
952    if (!stride) {
953        stride = lineSize;
954    }
955
956    if (alloc->mHal.drvState.lod[0].mallocPtr) {
957        uint8_t *dst = static_cast<uint8_t *>(data);
958        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
959        if (dst == src) {
960            // Skip the copy if we are the same allocation. This can arise from
961            // our Bitmap optimization, where we share the same storage.
962            return;
963        }
964
965        for (uint32_t line=yoff; line < (yoff+h); line++) {
966            memcpy(dst, src, lineSize);
967            dst += stride;
968            src += alloc->mHal.drvState.lod[lod].stride;
969        }
970    } else {
971        ALOGE("Add code to readback from non-script memory");
972    }
973}
974
975
976void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
977                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
978                         uint32_t lod,
979                         uint32_t w, uint32_t h, uint32_t d, void *data, size_t sizeBytes, size_t stride) {
980    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
981    uint32_t lineSize = eSize * w;
982    if (!stride) {
983        stride = lineSize;
984    }
985
986    if (alloc->mHal.drvState.lod[0].mallocPtr) {
987        uint8_t *dst = static_cast<uint8_t *>(data);
988        for (uint32_t z = zoff; z < d; z++) {
989            const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, z, lod,
990                                              RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
991            if (dst == src) {
992                // Skip the copy if we are the same allocation. This can arise from
993                // our Bitmap optimization, where we share the same storage.
994                return;
995            }
996
997            for (uint32_t line=yoff; line < (yoff+h); line++) {
998                memcpy(dst, src, lineSize);
999                dst += stride;
1000                src += alloc->mHal.drvState.lod[lod].stride;
1001            }
1002        }
1003    }
1004}
1005
1006void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
1007                          const android::renderscript::Allocation *alloc) {
1008    return alloc->mHal.drvState.lod[0].mallocPtr;
1009}
1010
1011void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
1012                          const android::renderscript::Allocation *alloc) {
1013
1014}
1015
1016void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
1017                               const android::renderscript::Allocation *dstAlloc,
1018                               uint32_t dstXoff, uint32_t dstLod, size_t count,
1019                               const android::renderscript::Allocation *srcAlloc,
1020                               uint32_t srcXoff, uint32_t srcLod) {
1021}
1022
1023
1024void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
1025                                      const android::renderscript::Allocation *dstAlloc,
1026                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1027                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1028                                      const android::renderscript::Allocation *srcAlloc,
1029                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1030                                      RsAllocationCubemapFace srcFace) {
1031    size_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1032    for (uint32_t i = 0; i < h; i ++) {
1033        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, 0, dstLod, dstFace);
1034        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, 0, srcLod, srcFace);
1035        memcpy(dstPtr, srcPtr, w * elementSize);
1036
1037        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1038        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1039    }
1040}
1041
1042void rsdAllocationData3D_alloc_script(const android::renderscript::Context *rsc,
1043                                      const android::renderscript::Allocation *dstAlloc,
1044                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff, uint32_t dstLod,
1045                                      uint32_t w, uint32_t h, uint32_t d,
1046                                      const android::renderscript::Allocation *srcAlloc,
1047                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff, uint32_t srcLod) {
1048    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1049    for (uint32_t j = 0; j < d; j++) {
1050        for (uint32_t i = 0; i < h; i ++) {
1051            uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstZoff + j,
1052                                           dstLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1053            uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcZoff + j,
1054                                           srcLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1055            memcpy(dstPtr, srcPtr, w * elementSize);
1056
1057            //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1058            //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1059        }
1060    }
1061}
1062
1063void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
1064                               const android::renderscript::Allocation *dstAlloc,
1065                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1066                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1067                               const android::renderscript::Allocation *srcAlloc,
1068                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1069                               RsAllocationCubemapFace srcFace) {
1070    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1071        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1072                                             "yet implemented.");
1073        return;
1074    }
1075    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
1076                                     dstLod, dstFace, w, h, srcAlloc,
1077                                     srcXoff, srcYoff, srcLod, srcFace);
1078}
1079
1080void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
1081                               const android::renderscript::Allocation *dstAlloc,
1082                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
1083                               uint32_t dstLod,
1084                               uint32_t w, uint32_t h, uint32_t d,
1085                               const android::renderscript::Allocation *srcAlloc,
1086                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
1087                               uint32_t srcLod) {
1088    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1089        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1090                                             "yet implemented.");
1091        return;
1092    }
1093    rsdAllocationData3D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstZoff,
1094                                     dstLod, w, h, d, srcAlloc,
1095                                     srcXoff, srcYoff, srcZoff, srcLod);
1096}
1097
1098void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc,
1099                                uint32_t x,
1100                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1101    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1102
1103    size_t eSize = alloc->mHal.state.elementSizeBytes;
1104    uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1105
1106    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1107    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1108
1109    if (alloc->mHal.state.hasReferences) {
1110        e->incRefs(data);
1111        e->decRefs(ptr);
1112    }
1113
1114    memcpy(ptr, data, sizeBytes);
1115    drv->uploadDeferred = true;
1116}
1117
1118void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc,
1119                                uint32_t x, uint32_t y,
1120                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1121    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1122
1123    size_t eSize = alloc->mHal.state.elementSizeBytes;
1124    uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1125
1126    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1127    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1128
1129    if (alloc->mHal.state.hasReferences) {
1130        e->incRefs(data);
1131        e->decRefs(ptr);
1132    }
1133
1134    memcpy(ptr, data, sizeBytes);
1135    drv->uploadDeferred = true;
1136}
1137
1138static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1139    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1140    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1141
1142    for (uint32_t y=0; y < h; y++) {
1143        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1144        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2, lod, face);
1145        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2+1, lod, face);
1146
1147        for (uint32_t x=0; x < w; x++) {
1148            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
1149            oPtr ++;
1150            i1 += 2;
1151            i2 += 2;
1152        }
1153    }
1154}
1155
1156static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1157    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1158    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1159
1160    for (uint32_t y=0; y < h; y++) {
1161        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1162        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1163        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1164
1165        for (uint32_t x=0; x < w; x++) {
1166            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
1167            oPtr ++;
1168            i1 += 2;
1169            i2 += 2;
1170        }
1171    }
1172}
1173
1174static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1175    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1176    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1177
1178    for (uint32_t y=0; y < h; y++) {
1179        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1180        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1181        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1182
1183        for (uint32_t x=0; x < w; x++) {
1184            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
1185            oPtr ++;
1186            i1 += 2;
1187            i2 += 2;
1188        }
1189    }
1190}
1191
1192void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
1193    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
1194        return;
1195    }
1196    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
1197    for (uint32_t face = 0; face < numFaces; face ++) {
1198        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
1199            switch (alloc->getType()->getElement()->getSizeBits()) {
1200            case 32:
1201                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
1202                break;
1203            case 16:
1204                mip565(alloc, lod, (RsAllocationCubemapFace)face);
1205                break;
1206            case 8:
1207                mip8(alloc, lod, (RsAllocationCubemapFace)face);
1208                break;
1209            }
1210        }
1211    }
1212}
1213