rsdAllocation.cpp revision 8b9b3aa8e0b92e677c0100ee402a7328fc3ce1c4 
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#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
23#include "system/window.h"
24#include "ui/Rect.h"
25#include "ui/GraphicBufferMapper.h"
26#else
27#include "system/graphics.h"
28#endif
29
30#ifndef RS_COMPATIBILITY_LIB
31#include "rsdFrameBufferObj.h"
32#include "gui/GLConsumer.h"
33#include "gui/CpuConsumer.h"
34#include "gui/Surface.h"
35#include "hardware/gralloc.h"
36
37#include <GLES/gl.h>
38#include <GLES2/gl2.h>
39#include <GLES/glext.h>
40#endif
41
42#ifdef RS_SERVER
43// server requires malloc.h for memalign
44#include <malloc.h>
45#endif
46
47using namespace android;
48using namespace android::renderscript;
49
50
51#ifndef RS_COMPATIBILITY_LIB
52const static GLenum gFaceOrder[] = {
53    GL_TEXTURE_CUBE_MAP_POSITIVE_X,
54    GL_TEXTURE_CUBE_MAP_NEGATIVE_X,
55    GL_TEXTURE_CUBE_MAP_POSITIVE_Y,
56    GL_TEXTURE_CUBE_MAP_NEGATIVE_Y,
57    GL_TEXTURE_CUBE_MAP_POSITIVE_Z,
58    GL_TEXTURE_CUBE_MAP_NEGATIVE_Z
59};
60
61GLenum rsdTypeToGLType(RsDataType t) {
62    switch (t) {
63    case RS_TYPE_UNSIGNED_5_6_5:    return GL_UNSIGNED_SHORT_5_6_5;
64    case RS_TYPE_UNSIGNED_5_5_5_1:  return GL_UNSIGNED_SHORT_5_5_5_1;
65    case RS_TYPE_UNSIGNED_4_4_4_4:  return GL_UNSIGNED_SHORT_4_4_4_4;
66
67    //case RS_TYPE_FLOAT_16:      return GL_HALF_FLOAT;
68    case RS_TYPE_FLOAT_32:      return GL_FLOAT;
69    case RS_TYPE_UNSIGNED_8:    return GL_UNSIGNED_BYTE;
70    case RS_TYPE_UNSIGNED_16:   return GL_UNSIGNED_SHORT;
71    case RS_TYPE_SIGNED_8:      return GL_BYTE;
72    case RS_TYPE_SIGNED_16:     return GL_SHORT;
73    default:    break;
74    }
75    return 0;
76}
77
78GLenum rsdKindToGLFormat(RsDataKind k) {
79    switch (k) {
80    case RS_KIND_PIXEL_L: return GL_LUMINANCE;
81    case RS_KIND_PIXEL_A: return GL_ALPHA;
82    case RS_KIND_PIXEL_LA: return GL_LUMINANCE_ALPHA;
83    case RS_KIND_PIXEL_RGB: return GL_RGB;
84    case RS_KIND_PIXEL_RGBA: return GL_RGBA;
85    case RS_KIND_PIXEL_DEPTH: return GL_DEPTH_COMPONENT16;
86    default: break;
87    }
88    return 0;
89}
90#endif
91
92uint8_t *GetOffsetPtr(const android::renderscript::Allocation *alloc,
93                      uint32_t xoff, uint32_t yoff, uint32_t zoff,
94                      uint32_t lod, RsAllocationCubemapFace face) {
95    uint8_t *ptr = (uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
96    ptr += face * alloc->mHal.drvState.faceOffset;
97    ptr += zoff * alloc->mHal.drvState.lod[lod].dimY * alloc->mHal.drvState.lod[lod].stride;
98    ptr += yoff * alloc->mHal.drvState.lod[lod].stride;
99    ptr += xoff * alloc->mHal.state.elementSizeBytes;
100    return ptr;
101}
102
103
104static void Update2DTexture(const Context *rsc, const Allocation *alloc, const void *ptr,
105                            uint32_t xoff, uint32_t yoff, uint32_t lod,
106                            RsAllocationCubemapFace face, uint32_t w, uint32_t h) {
107#ifndef RS_COMPATIBILITY_LIB
108    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
109
110    rsAssert(drv->textureID);
111    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
112    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
113    GLenum t = GL_TEXTURE_2D;
114    if (alloc->mHal.state.hasFaces) {
115        t = gFaceOrder[face];
116    }
117    RSD_CALL_GL(glTexSubImage2D, t, lod, xoff, yoff, w, h, drv->glFormat, drv->glType, ptr);
118#endif
119}
120
121
122#ifndef RS_COMPATIBILITY_LIB
123static void Upload2DTexture(const Context *rsc, const Allocation *alloc, bool isFirstUpload) {
124    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
125
126    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
127    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
128
129    uint32_t faceCount = 1;
130    if (alloc->mHal.state.hasFaces) {
131        faceCount = 6;
132    }
133
134    rsdGLCheckError(rsc, "Upload2DTexture 1 ");
135    for (uint32_t face = 0; face < faceCount; face ++) {
136        for (uint32_t lod = 0; lod < alloc->mHal.state.type->getLODCount(); lod++) {
137            const uint8_t *p = GetOffsetPtr(alloc, 0, 0, 0, lod, (RsAllocationCubemapFace)face);
138
139            GLenum t = GL_TEXTURE_2D;
140            if (alloc->mHal.state.hasFaces) {
141                t = gFaceOrder[face];
142            }
143
144            if (isFirstUpload) {
145                RSD_CALL_GL(glTexImage2D, t, lod, drv->glFormat,
146                             alloc->mHal.state.type->getLODDimX(lod),
147                             alloc->mHal.state.type->getLODDimY(lod),
148                             0, drv->glFormat, drv->glType, p);
149            } else {
150                RSD_CALL_GL(glTexSubImage2D, t, lod, 0, 0,
151                                alloc->mHal.state.type->getLODDimX(lod),
152                                alloc->mHal.state.type->getLODDimY(lod),
153                                drv->glFormat, drv->glType, p);
154            }
155        }
156    }
157
158    if (alloc->mHal.state.mipmapControl == RS_ALLOCATION_MIPMAP_ON_SYNC_TO_TEXTURE) {
159        RSD_CALL_GL(glGenerateMipmap, drv->glTarget);
160    }
161    rsdGLCheckError(rsc, "Upload2DTexture");
162}
163#endif
164
165static void UploadToTexture(const Context *rsc, const Allocation *alloc) {
166#ifndef RS_COMPATIBILITY_LIB
167    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
168
169    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
170        if (!drv->textureID) {
171            RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
172        }
173        return;
174    }
175
176    if (!drv->glType || !drv->glFormat) {
177        return;
178    }
179
180    if (!alloc->mHal.drvState.lod[0].mallocPtr) {
181        return;
182    }
183
184    bool isFirstUpload = false;
185
186    if (!drv->textureID) {
187        RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
188        isFirstUpload = true;
189    }
190
191    Upload2DTexture(rsc, alloc, isFirstUpload);
192
193    if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
194        if (alloc->mHal.drvState.lod[0].mallocPtr) {
195            free(alloc->mHal.drvState.lod[0].mallocPtr);
196            alloc->mHal.drvState.lod[0].mallocPtr = NULL;
197        }
198    }
199    rsdGLCheckError(rsc, "UploadToTexture");
200#endif
201}
202
203static void AllocateRenderTarget(const Context *rsc, const Allocation *alloc) {
204#ifndef RS_COMPATIBILITY_LIB
205    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
206
207    if (!drv->glFormat) {
208        return;
209    }
210
211    if (!drv->renderTargetID) {
212        RSD_CALL_GL(glGenRenderbuffers, 1, &drv->renderTargetID);
213
214        if (!drv->renderTargetID) {
215            // This should generally not happen
216            ALOGE("allocateRenderTarget failed to gen mRenderTargetID");
217            rsc->dumpDebug();
218            return;
219        }
220        RSD_CALL_GL(glBindRenderbuffer, GL_RENDERBUFFER, drv->renderTargetID);
221        RSD_CALL_GL(glRenderbufferStorage, GL_RENDERBUFFER, drv->glFormat,
222                    alloc->mHal.drvState.lod[0].dimX, alloc->mHal.drvState.lod[0].dimY);
223    }
224    rsdGLCheckError(rsc, "AllocateRenderTarget");
225#endif
226}
227
228static void UploadToBufferObject(const Context *rsc, const Allocation *alloc) {
229#ifndef RS_COMPATIBILITY_LIB
230    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
231
232    rsAssert(!alloc->mHal.state.type->getDimY());
233    rsAssert(!alloc->mHal.state.type->getDimZ());
234
235    //alloc->mHal.state.usageFlags |= RS_ALLOCATION_USAGE_GRAPHICS_VERTEX;
236
237    if (!drv->bufferID) {
238        RSD_CALL_GL(glGenBuffers, 1, &drv->bufferID);
239    }
240    if (!drv->bufferID) {
241        ALOGE("Upload to buffer object failed");
242        drv->uploadDeferred = true;
243        return;
244    }
245    RSD_CALL_GL(glBindBuffer, drv->glTarget, drv->bufferID);
246    RSD_CALL_GL(glBufferData, drv->glTarget, alloc->mHal.state.type->getSizeBytes(),
247                 alloc->mHal.drvState.lod[0].mallocPtr, GL_DYNAMIC_DRAW);
248    RSD_CALL_GL(glBindBuffer, drv->glTarget, 0);
249    rsdGLCheckError(rsc, "UploadToBufferObject");
250#endif
251}
252
253
254static size_t DeriveYUVLayout(int yuv, Allocation::Hal::DrvState *state) {
255    // YUV only supports basic 2d
256    // so we can stash the plane pointers in the mipmap levels.
257    size_t uvSize = 0;
258#ifndef RS_SERVER
259    switch(yuv) {
260    case HAL_PIXEL_FORMAT_YV12:
261        state->lod[2].dimX = state->lod[0].dimX / 2;
262        state->lod[2].dimY = state->lod[0].dimY / 2;
263        state->lod[2].stride = rsRound(state->lod[0].stride >> 1, 16);
264        state->lod[2].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
265                (state->lod[0].stride * state->lod[0].dimY);
266        uvSize += state->lod[2].stride * state->lod[2].dimY;
267
268        state->lod[1].dimX = state->lod[2].dimX;
269        state->lod[1].dimY = state->lod[2].dimY;
270        state->lod[1].stride = state->lod[2].stride;
271        state->lod[1].mallocPtr = ((uint8_t *)state->lod[2].mallocPtr) +
272                (state->lod[2].stride * state->lod[2].dimY);
273        uvSize += state->lod[1].stride * state->lod[2].dimY;
274
275        state->lodCount = 3;
276        break;
277    case HAL_PIXEL_FORMAT_YCrCb_420_SP:  // NV21
278        state->lod[1].dimX = state->lod[0].dimX;
279        state->lod[1].dimY = state->lod[0].dimY / 2;
280        state->lod[1].stride = state->lod[0].stride;
281        state->lod[1].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
282                (state->lod[0].stride * state->lod[0].dimY);
283        uvSize += state->lod[1].stride * state->lod[1].dimY;
284        state->lodCount = 2;
285        break;
286    default:
287        rsAssert(0);
288    }
289#endif
290    return uvSize;
291}
292
293
294static size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc,
295        const Type *type, uint8_t *ptr) {
296    alloc->mHal.drvState.lod[0].dimX = type->getDimX();
297    alloc->mHal.drvState.lod[0].dimY = type->getDimY();
298    alloc->mHal.drvState.lod[0].dimZ = type->getDimZ();
299    alloc->mHal.drvState.lod[0].mallocPtr = 0;
300    // Stride needs to be 16-byte aligned too!
301    size_t stride = alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes();
302    alloc->mHal.drvState.lod[0].stride = rsRound(stride, 16);
303    alloc->mHal.drvState.lodCount = type->getLODCount();
304    alloc->mHal.drvState.faceCount = type->getDimFaces();
305
306    size_t offsets[Allocation::MAX_LOD];
307    memset(offsets, 0, sizeof(offsets));
308
309    size_t o = alloc->mHal.drvState.lod[0].stride * rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) *
310            rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u);
311    if(alloc->mHal.drvState.lodCount > 1) {
312        uint32_t tx = alloc->mHal.drvState.lod[0].dimX;
313        uint32_t ty = alloc->mHal.drvState.lod[0].dimY;
314        uint32_t tz = alloc->mHal.drvState.lod[0].dimZ;
315        for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
316            alloc->mHal.drvState.lod[lod].dimX = tx;
317            alloc->mHal.drvState.lod[lod].dimY = ty;
318            alloc->mHal.drvState.lod[lod].dimZ = tz;
319            alloc->mHal.drvState.lod[lod].stride =
320                    rsRound(tx * type->getElementSizeBytes(), 16);
321            offsets[lod] = o;
322            o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u);
323            if (tx > 1) tx >>= 1;
324            if (ty > 1) ty >>= 1;
325            if (tz > 1) tz >>= 1;
326        }
327    } else if (alloc->mHal.state.yuv) {
328        o += DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
329
330        for (uint32_t ct = 1; ct < alloc->mHal.drvState.lodCount; ct++) {
331            offsets[ct] = (size_t)alloc->mHal.drvState.lod[ct].mallocPtr;
332        }
333    }
334
335    alloc->mHal.drvState.faceOffset = o;
336
337    alloc->mHal.drvState.lod[0].mallocPtr = ptr;
338    for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
339        alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod];
340    }
341
342    size_t allocSize = alloc->mHal.drvState.faceOffset;
343    if(alloc->mHal.drvState.faceCount) {
344        allocSize *= 6;
345    }
346
347    return allocSize;
348}
349
350static uint8_t* allocAlignedMemory(size_t allocSize, bool forceZero) {
351    // We align all allocations to a 16-byte boundary.
352    uint8_t* ptr = (uint8_t *)memalign(16, allocSize);
353    if (!ptr) {
354        return NULL;
355    }
356    if (forceZero) {
357        memset(ptr, 0, allocSize);
358    }
359    return ptr;
360}
361
362bool rsdAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) {
363    DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
364    if (!drv) {
365        return false;
366    }
367    alloc->mHal.drv = drv;
368
369    // Calculate the object size.
370    size_t allocSize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), NULL);
371
372    uint8_t * ptr = NULL;
373    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) {
374
375    } else if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
376        // Allocation is allocated when the surface is created
377        // in getSurface
378    } else if (alloc->mHal.state.userProvidedPtr != NULL) {
379        // user-provided allocation
380        // limitations: no faces, no LOD, USAGE_SCRIPT or SCRIPT+TEXTURE only
381        if (!(alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED) ||
382              alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED | RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE))) {
383            ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT | USAGE_SHARED or USAGE_SCRIPT | USAGE_SHARED | USAGE_GRAPHICS_TEXTURE");
384            return false;
385        }
386        if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) {
387            ALOGE("User-allocated buffers must not have multiple faces or LODs");
388            return false;
389        }
390
391        // rows must be 16-byte aligned
392        // validate that here, otherwise fall back to not use the user-backed allocation
393        if (((alloc->getType()->getDimX() * alloc->getType()->getElement()->getSizeBytes()) % 16) != 0) {
394            ALOGV("User-backed allocation failed stride requirement, falling back to separate allocation");
395            drv->useUserProvidedPtr = false;
396
397            ptr = allocAlignedMemory(allocSize, forceZero);
398            if (!ptr) {
399                alloc->mHal.drv = NULL;
400                free(drv);
401                return false;
402            }
403
404        } else {
405            drv->useUserProvidedPtr = true;
406            ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr;
407        }
408    } else {
409        ptr = allocAlignedMemory(allocSize, forceZero);
410        if (!ptr) {
411            alloc->mHal.drv = NULL;
412            free(drv);
413            return false;
414        }
415    }
416    // Build the pointer tables
417    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr);
418    if(allocSize != verifySize) {
419        rsAssert(!"Size mismatch");
420    }
421
422#ifndef RS_SERVER
423    drv->glTarget = GL_NONE;
424    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
425        if (alloc->mHal.state.hasFaces) {
426            drv->glTarget = GL_TEXTURE_CUBE_MAP;
427        } else {
428            drv->glTarget = GL_TEXTURE_2D;
429        }
430    } else {
431        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
432            drv->glTarget = GL_ARRAY_BUFFER;
433        }
434    }
435#endif
436
437#ifndef RS_COMPATIBILITY_LIB
438    drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType());
439    drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind());
440#else
441    drv->glType = 0;
442    drv->glFormat = 0;
443#endif
444
445    if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) {
446        drv->uploadDeferred = true;
447    }
448
449
450    drv->readBackFBO = NULL;
451
452    // fill out the initial state of the buffer if we couldn't use the user-provided ptr and USAGE_SHARED was accepted
453    if ((alloc->mHal.state.userProvidedPtr != 0) && (drv->useUserProvidedPtr == false)) {
454        rsdAllocationData2D(rsc, alloc, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X, alloc->getType()->getDimX(), alloc->getType()->getDimY(), alloc->mHal.state.userProvidedPtr, allocSize, 0);
455    }
456
457    return true;
458}
459
460void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) {
461    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
462
463#ifndef RS_COMPATIBILITY_LIB
464    if (drv->bufferID) {
465        // Causes a SW crash....
466        //ALOGV(" mBufferID %i", mBufferID);
467        //glDeleteBuffers(1, &mBufferID);
468        //mBufferID = 0;
469    }
470    if (drv->textureID) {
471        RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID);
472        drv->textureID = 0;
473    }
474    if (drv->renderTargetID) {
475        RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID);
476        drv->renderTargetID = 0;
477    }
478#endif
479
480    if (alloc->mHal.drvState.lod[0].mallocPtr) {
481        // don't free user-allocated ptrs or IO_OUTPUT buffers
482        if (!(drv->useUserProvidedPtr) &&
483            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) &&
484            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
485                free(alloc->mHal.drvState.lod[0].mallocPtr);
486        }
487        alloc->mHal.drvState.lod[0].mallocPtr = NULL;
488    }
489
490#ifndef RS_COMPATIBILITY_LIB
491    if (drv->readBackFBO != NULL) {
492        delete drv->readBackFBO;
493        drv->readBackFBO = NULL;
494    }
495
496    if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) &&
497        (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
498
499        DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
500        ANativeWindow *nw = drv->wndSurface;
501        if (nw) {
502            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
503            mapper.unlock(drv->wndBuffer->handle);
504            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
505        }
506    }
507#endif
508
509    free(drv);
510    alloc->mHal.drv = NULL;
511}
512
513void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
514                         const Type *newType, bool zeroNew) {
515    const uint32_t oldDimX = alloc->mHal.drvState.lod[0].dimX;
516    const uint32_t dimX = newType->getDimX();
517
518    // can't resize Allocations with user-allocated buffers
519    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
520        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
521        return;
522    }
523    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
524    // Calculate the object size
525    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL);
526    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
527    // Build the relative pointer tables.
528    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
529    if(s != verifySize) {
530        rsAssert(!"Size mismatch");
531    }
532
533
534    if (dimX > oldDimX) {
535        size_t stride = alloc->mHal.state.elementSizeBytes;
536        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
537                 0, stride * (dimX - oldDimX));
538    }
539}
540
541static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
542#ifndef RS_COMPATIBILITY_LIB
543    if (!alloc->getIsScript()) {
544        return; // nothing to sync
545    }
546
547    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
548    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
549
550    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
551    if (!drv->textureID && !drv->renderTargetID) {
552        return; // nothing was rendered here yet, so nothing to sync
553    }
554    if (drv->readBackFBO == NULL) {
555        drv->readBackFBO = new RsdFrameBufferObj();
556        drv->readBackFBO->setColorTarget(drv, 0);
557        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
558                                        alloc->getType()->getDimY());
559    }
560
561    // Bind the framebuffer object so we can read back from it
562    drv->readBackFBO->setActive(rsc);
563
564    // Do the readback
565    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
566                alloc->mHal.drvState.lod[0].dimY,
567                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
568
569    // Revert framebuffer to its original
570    lastFbo->setActive(rsc);
571#endif
572}
573
574
575void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
576                         RsAllocationUsageType src) {
577    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
578
579    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
580        if(!alloc->getIsRenderTarget()) {
581            rsc->setError(RS_ERROR_FATAL_DRIVER,
582                          "Attempting to sync allocation from render target, "
583                          "for non-render target allocation");
584        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
585            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
586                                                 "render target");
587        } else {
588            rsdAllocationSyncFromFBO(rsc, alloc);
589        }
590        return;
591    }
592
593    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT);
594
595    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
596        UploadToTexture(rsc, alloc);
597    } else {
598        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
599            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
600            AllocateRenderTarget(rsc, alloc);
601        }
602    }
603    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
604        UploadToBufferObject(rsc, alloc);
605    }
606
607    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
608        // NOP in CPU driver for now
609    }
610
611    drv->uploadDeferred = false;
612}
613
614void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
615    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
616    drv->uploadDeferred = true;
617}
618
619#ifndef RS_COMPATIBILITY_LIB
620void DrvAllocation::NewBufferListener::onFrameAvailable() {
621    intptr_t ip = (intptr_t)alloc;
622    rsc->sendMessageToClient(NULL, RS_MESSAGE_TO_CLIENT_NEW_BUFFER, ip, 0, true);
623}
624#endif
625
626void* rsdAllocationGetSurface(const Context *rsc, const Allocation *alloc) {
627#ifndef RS_COMPATIBILITY_LIB
628    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
629
630    // Configure CpuConsumer to be in asynchronous mode
631    sp<BufferQueue> bq = new BufferQueue();
632    drv->cpuConsumer = new CpuConsumer(bq, 2, false);
633    sp<IGraphicBufferProducer> bp = drv->cpuConsumer->getProducerInterface();
634    bp->incStrong(NULL);
635
636    drv->mBufferListener = new DrvAllocation::NewBufferListener();
637    drv->mBufferListener->rsc = rsc;
638    drv->mBufferListener->alloc = alloc;
639
640    drv->cpuConsumer->setFrameAvailableListener(drv->mBufferListener);
641
642    return bp.get();
643#else
644    return NULL;
645#endif
646}
647
648#ifndef RS_COMPATIBILITY_LIB
649static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
650    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
651
652    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
653    if (r) {
654        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
655        return false;
656    }
657
658    // Must lock the whole surface
659    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
660    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
661
662    void *dst = NULL;
663    mapper.lock(drv->wndBuffer->handle,
664            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
665            bounds, &dst);
666    alloc->mHal.drvState.lod[0].mallocPtr = dst;
667    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
668    rsAssert((alloc->mHal.drvState.lod[0].stride & 0xf) == 0);
669
670    return true;
671}
672#endif
673
674void rsdAllocationSetSurface(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
675#ifndef RS_COMPATIBILITY_LIB
676    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
677    ANativeWindow *old = drv->wndSurface;
678
679    if (nw) {
680        nw->incStrong(NULL);
681    }
682
683    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
684        //TODO finish support for render target + script
685        drv->wnd = nw;
686        return;
687    }
688
689    // Cleanup old surface if there is one.
690    if (drv->wndSurface) {
691        ANativeWindow *old = drv->wndSurface;
692        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
693        mapper.unlock(drv->wndBuffer->handle);
694        old->cancelBuffer(old, drv->wndBuffer, -1);
695        drv->wndSurface = NULL;
696        old->decStrong(NULL);
697    }
698
699    if (nw != NULL) {
700        int32_t r;
701        uint32_t flags = 0;
702
703        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
704            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
705        }
706        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
707            flags |= GRALLOC_USAGE_HW_RENDER;
708        }
709
710        r = native_window_set_usage(nw, flags);
711        if (r) {
712            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
713            goto error;
714        }
715
716        r = native_window_set_buffers_dimensions(nw, alloc->mHal.drvState.lod[0].dimX,
717                                                 alloc->mHal.drvState.lod[0].dimY);
718        if (r) {
719            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
720            goto error;
721        }
722
723        int format = 0;
724        const Element *e = alloc->mHal.state.type->getElement();
725        rsAssert(e->getType() == RS_TYPE_UNSIGNED_8);
726        rsAssert(e->getVectorSize() == 4);
727        rsAssert(e->getKind() == RS_KIND_PIXEL_RGBA);
728        format = PIXEL_FORMAT_RGBA_8888;
729
730        r = native_window_set_buffers_format(nw, format);
731        if (r) {
732            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer format.");
733            goto error;
734        }
735
736        IoGetBuffer(rsc, alloc, nw);
737        drv->wndSurface = nw;
738    }
739
740    return;
741
742 error:
743
744    if (nw) {
745        nw->decStrong(NULL);
746    }
747
748
749#endif
750}
751
752void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
753#ifndef RS_COMPATIBILITY_LIB
754    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
755    ANativeWindow *nw = drv->wndSurface;
756    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
757        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
758        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
759        return;
760    }
761    if (nw) {
762        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
763            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
764            mapper.unlock(drv->wndBuffer->handle);
765            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
766            if (r) {
767                rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
768                return;
769            }
770
771            IoGetBuffer(rsc, alloc, nw);
772        }
773    } else {
774        rsc->setError(RS_ERROR_DRIVER, "Sent IO buffer with no attached surface.");
775        return;
776    }
777#endif
778}
779
780void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
781#ifndef RS_COMPATIBILITY_LIB
782    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
783
784    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
785        CpuConsumer::LockedBuffer lb;
786        status_t ret = drv->cpuConsumer->lockNextBuffer(&lb);
787        if (ret == OK) {
788            if (drv->lb.data != NULL) {
789                drv->cpuConsumer->unlockBuffer(drv->lb);
790            }
791            drv->lb = lb;
792            alloc->mHal.drvState.lod[0].mallocPtr = drv->lb.data;
793            alloc->mHal.drvState.lod[0].stride = drv->lb.stride *
794                    alloc->mHal.state.elementSizeBytes;
795
796            if (alloc->mHal.state.yuv) {
797                DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
798            }
799        } else if (ret == BAD_VALUE) {
800            // No new frame, don't do anything
801        } else {
802            rsc->setError(RS_ERROR_DRIVER, "Error receiving IO input buffer.");
803        }
804
805    } else {
806        drv->surfaceTexture->updateTexImage();
807    }
808
809
810#endif
811}
812
813
814void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
815                         uint32_t xoff, uint32_t lod, size_t count,
816                         const void *data, size_t sizeBytes) {
817    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
818
819    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
820    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
821    size_t size = count * eSize;
822
823    if (ptr != data) {
824        // Skip the copy if we are the same allocation. This can arise from
825        // our Bitmap optimization, where we share the same storage.
826        if (alloc->mHal.state.hasReferences) {
827            alloc->incRefs(data, count);
828            alloc->decRefs(ptr, count);
829        }
830        memcpy(ptr, data, size);
831    }
832    drv->uploadDeferred = true;
833}
834
835void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
836                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
837                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
838    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
839
840    size_t eSize = alloc->mHal.state.elementSizeBytes;
841    size_t lineSize = eSize * w;
842    if (!stride) {
843        stride = lineSize;
844    }
845
846    if (alloc->mHal.drvState.lod[0].mallocPtr) {
847        const uint8_t *src = static_cast<const uint8_t *>(data);
848        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
849        if (dst == src) {
850            // Skip the copy if we are the same allocation. This can arise from
851            // our Bitmap optimization, where we share the same storage.
852            drv->uploadDeferred = true;
853            return;
854        }
855
856        for (uint32_t line=yoff; line < (yoff+h); line++) {
857            if (alloc->mHal.state.hasReferences) {
858                alloc->incRefs(src, w);
859                alloc->decRefs(dst, w);
860            }
861            memcpy(dst, src, lineSize);
862            src += stride;
863            dst += alloc->mHal.drvState.lod[lod].stride;
864        }
865        if (alloc->mHal.state.yuv) {
866            int lod = 1;
867            while (alloc->mHal.drvState.lod[lod].mallocPtr) {
868                size_t lineSize = alloc->mHal.drvState.lod[lod].dimX;
869                uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
870
871                for (uint32_t line=(yoff >> 1); line < ((yoff+h)>>1); line++) {
872                    memcpy(dst, src, lineSize);
873                    src += lineSize;
874                    dst += alloc->mHal.drvState.lod[lod].stride;
875                }
876                lod++;
877            }
878
879        }
880        drv->uploadDeferred = true;
881    } else {
882        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
883    }
884}
885
886void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
887                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
888                         uint32_t lod,
889                         uint32_t w, uint32_t h, uint32_t d, const void *data,
890                         size_t sizeBytes, size_t stride) {
891    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
892
893    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
894    uint32_t lineSize = eSize * w;
895    if (!stride) {
896        stride = lineSize;
897    }
898
899    if (alloc->mHal.drvState.lod[0].mallocPtr) {
900        const uint8_t *src = static_cast<const uint8_t *>(data);
901        for (uint32_t z = zoff; z < d; z++) {
902            uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, z, lod,
903                                        RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
904            if (dst == src) {
905                // Skip the copy if we are the same allocation. This can arise from
906                // our Bitmap optimization, where we share the same storage.
907                drv->uploadDeferred = true;
908                return;
909            }
910
911            for (uint32_t line=yoff; line < (yoff+h); line++) {
912                if (alloc->mHal.state.hasReferences) {
913                    alloc->incRefs(src, w);
914                    alloc->decRefs(dst, w);
915                }
916                memcpy(dst, src, lineSize);
917                src += stride;
918                dst += alloc->mHal.drvState.lod[lod].stride;
919            }
920        }
921        drv->uploadDeferred = true;
922    }
923}
924
925void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
926                         uint32_t xoff, uint32_t lod, size_t count,
927                         void *data, size_t sizeBytes) {
928    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
929    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
930    if (data != ptr) {
931        // Skip the copy if we are the same allocation. This can arise from
932        // our Bitmap optimization, where we share the same storage.
933        memcpy(data, ptr, count * eSize);
934    }
935}
936
937void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
938                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
939                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
940    size_t eSize = alloc->mHal.state.elementSizeBytes;
941    size_t lineSize = eSize * w;
942    if (!stride) {
943        stride = lineSize;
944    }
945
946    if (alloc->mHal.drvState.lod[0].mallocPtr) {
947        uint8_t *dst = static_cast<uint8_t *>(data);
948        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
949        if (dst == src) {
950            // Skip the copy if we are the same allocation. This can arise from
951            // our Bitmap optimization, where we share the same storage.
952            return;
953        }
954
955        for (uint32_t line=yoff; line < (yoff+h); line++) {
956            memcpy(dst, src, lineSize);
957            dst += stride;
958            src += alloc->mHal.drvState.lod[lod].stride;
959        }
960    } else {
961        ALOGE("Add code to readback from non-script memory");
962    }
963}
964
965
966void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
967                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
968                         uint32_t lod,
969                         uint32_t w, uint32_t h, uint32_t d, void *data, size_t sizeBytes, size_t stride) {
970    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
971    uint32_t lineSize = eSize * w;
972    if (!stride) {
973        stride = lineSize;
974    }
975
976    if (alloc->mHal.drvState.lod[0].mallocPtr) {
977        uint8_t *dst = static_cast<uint8_t *>(data);
978        for (uint32_t z = zoff; z < d; z++) {
979            const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, z, lod,
980                                              RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
981            if (dst == src) {
982                // Skip the copy if we are the same allocation. This can arise from
983                // our Bitmap optimization, where we share the same storage.
984                return;
985            }
986
987            for (uint32_t line=yoff; line < (yoff+h); line++) {
988                memcpy(dst, src, lineSize);
989                dst += stride;
990                src += alloc->mHal.drvState.lod[lod].stride;
991            }
992        }
993    }
994}
995
996void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
997                          const android::renderscript::Allocation *alloc) {
998    return alloc->mHal.drvState.lod[0].mallocPtr;
999}
1000
1001void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
1002                          const android::renderscript::Allocation *alloc) {
1003
1004}
1005
1006void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
1007                               const android::renderscript::Allocation *dstAlloc,
1008                               uint32_t dstXoff, uint32_t dstLod, size_t count,
1009                               const android::renderscript::Allocation *srcAlloc,
1010                               uint32_t srcXoff, uint32_t srcLod) {
1011}
1012
1013
1014void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
1015                                      const android::renderscript::Allocation *dstAlloc,
1016                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1017                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1018                                      const android::renderscript::Allocation *srcAlloc,
1019                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1020                                      RsAllocationCubemapFace srcFace) {
1021    size_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1022    for (uint32_t i = 0; i < h; i ++) {
1023        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, 0, dstLod, dstFace);
1024        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, 0, srcLod, srcFace);
1025        memcpy(dstPtr, srcPtr, w * elementSize);
1026
1027        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1028        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1029    }
1030}
1031
1032void rsdAllocationData3D_alloc_script(const android::renderscript::Context *rsc,
1033                                      const android::renderscript::Allocation *dstAlloc,
1034                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff, uint32_t dstLod,
1035                                      uint32_t w, uint32_t h, uint32_t d,
1036                                      const android::renderscript::Allocation *srcAlloc,
1037                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff, uint32_t srcLod) {
1038    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1039    for (uint32_t j = 0; j < d; j++) {
1040        for (uint32_t i = 0; i < h; i ++) {
1041            uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstZoff + j,
1042                                           dstLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1043            uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcZoff + j,
1044                                           srcLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1045            memcpy(dstPtr, srcPtr, w * elementSize);
1046
1047            //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1048            //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1049        }
1050    }
1051}
1052
1053void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
1054                               const android::renderscript::Allocation *dstAlloc,
1055                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1056                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1057                               const android::renderscript::Allocation *srcAlloc,
1058                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1059                               RsAllocationCubemapFace srcFace) {
1060    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1061        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1062                                             "yet implemented.");
1063        return;
1064    }
1065    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
1066                                     dstLod, dstFace, w, h, srcAlloc,
1067                                     srcXoff, srcYoff, srcLod, srcFace);
1068}
1069
1070void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
1071                               const android::renderscript::Allocation *dstAlloc,
1072                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
1073                               uint32_t dstLod,
1074                               uint32_t w, uint32_t h, uint32_t d,
1075                               const android::renderscript::Allocation *srcAlloc,
1076                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
1077                               uint32_t srcLod) {
1078    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1079        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1080                                             "yet implemented.");
1081        return;
1082    }
1083    rsdAllocationData3D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstZoff,
1084                                     dstLod, w, h, d, srcAlloc,
1085                                     srcXoff, srcYoff, srcZoff, srcLod);
1086}
1087
1088void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc,
1089                                uint32_t x,
1090                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1091    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1092
1093    size_t eSize = alloc->mHal.state.elementSizeBytes;
1094    uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1095
1096    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1097    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1098
1099    if (alloc->mHal.state.hasReferences) {
1100        e->incRefs(data);
1101        e->decRefs(ptr);
1102    }
1103
1104    memcpy(ptr, data, sizeBytes);
1105    drv->uploadDeferred = true;
1106}
1107
1108void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc,
1109                                uint32_t x, uint32_t y,
1110                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1111    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1112
1113    size_t eSize = alloc->mHal.state.elementSizeBytes;
1114    uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1115
1116    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1117    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1118
1119    if (alloc->mHal.state.hasReferences) {
1120        e->incRefs(data);
1121        e->decRefs(ptr);
1122    }
1123
1124    memcpy(ptr, data, sizeBytes);
1125    drv->uploadDeferred = true;
1126}
1127
1128static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1129    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1130    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1131
1132    for (uint32_t y=0; y < h; y++) {
1133        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1134        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2, lod, face);
1135        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2+1, lod, face);
1136
1137        for (uint32_t x=0; x < w; x++) {
1138            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
1139            oPtr ++;
1140            i1 += 2;
1141            i2 += 2;
1142        }
1143    }
1144}
1145
1146static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1147    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1148    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1149
1150    for (uint32_t y=0; y < h; y++) {
1151        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1152        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1153        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1154
1155        for (uint32_t x=0; x < w; x++) {
1156            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
1157            oPtr ++;
1158            i1 += 2;
1159            i2 += 2;
1160        }
1161    }
1162}
1163
1164static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1165    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1166    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1167
1168    for (uint32_t y=0; y < h; y++) {
1169        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1170        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1171        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1172
1173        for (uint32_t x=0; x < w; x++) {
1174            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
1175            oPtr ++;
1176            i1 += 2;
1177            i2 += 2;
1178        }
1179    }
1180}
1181
1182void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
1183    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
1184        return;
1185    }
1186    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
1187    for (uint32_t face = 0; face < numFaces; face ++) {
1188        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
1189            switch (alloc->getType()->getElement()->getSizeBits()) {
1190            case 32:
1191                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
1192                break;
1193            case 16:
1194                mip565(alloc, lod, (RsAllocationCubemapFace)face);
1195                break;
1196            case 8:
1197                mip8(alloc, lod, (RsAllocationCubemapFace)face);
1198                break;
1199            }
1200        }
1201    }
1202}
1203