rsdAllocation.cpp revision 3bbc0fd40264ddae1592706d9023865b7b3e3195 
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[1].dimX = state->lod[0].dimX / 2;
260        state->lod[1].dimY = state->lod[0].dimY / 2;
261        state->lod[1].stride = rsRound(state->lod[0].stride >> 1, 16);
262        state->lod[1].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
263                (state->lod[0].stride * state->lod[0].dimY);
264        uvSize += state->lod[1].stride * state->lod[1].dimY;
265
266        state->lod[2].dimX = state->lod[1].dimX;
267        state->lod[2].dimY = state->lod[1].dimY;
268        state->lod[2].stride = state->lod[1].stride;
269        state->lod[2].mallocPtr = ((uint8_t *)state->lod[1].mallocPtr) +
270                (state->lod[1].stride * state->lod[1].dimY);
271        uvSize += state->lod[2].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 only
379        if (alloc->mHal.state.usageFlags != (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED)) {
380            ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT and USAGE_SHARED");
381            return false;
382        }
383        if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) {
384            ALOGE("User-allocated buffers must not have multiple faces or LODs");
385            return false;
386        }
387
388        // rows must be 16-byte aligned
389        // validate that here, otherwise fall back to not use the user-backed allocation
390        if (((alloc->getType()->getDimX() * alloc->getType()->getElement()->getSizeBytes()) % 16) != 0) {
391            ALOGV("User-backed allocation failed stride requirement, falling back to separate allocation");
392            drv->useUserProvidedPtr = false;
393
394            ptr = allocAlignedMemory(allocSize, forceZero);
395            if (!ptr) {
396                alloc->mHal.drv = NULL;
397                free(drv);
398                return false;
399            }
400
401        } else {
402            drv->useUserProvidedPtr = true;
403            ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr;
404        }
405    } else {
406        ptr = allocAlignedMemory(allocSize, forceZero);
407        if (!ptr) {
408            alloc->mHal.drv = NULL;
409            free(drv);
410            return false;
411        }
412    }
413    // Build the pointer tables
414    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr);
415    if(allocSize != verifySize) {
416        rsAssert(!"Size mismatch");
417    }
418
419#ifndef RS_SERVER
420    drv->glTarget = GL_NONE;
421    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
422        if (alloc->mHal.state.hasFaces) {
423            drv->glTarget = GL_TEXTURE_CUBE_MAP;
424        } else {
425            drv->glTarget = GL_TEXTURE_2D;
426        }
427    } else {
428        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
429            drv->glTarget = GL_ARRAY_BUFFER;
430        }
431    }
432#endif
433
434#ifndef RS_COMPATIBILITY_LIB
435    drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType());
436    drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind());
437#else
438    drv->glType = 0;
439    drv->glFormat = 0;
440#endif
441
442    if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) {
443        drv->uploadDeferred = true;
444    }
445
446
447    drv->readBackFBO = NULL;
448
449    // fill out the initial state of the buffer if we couldn't use the user-provided ptr and USAGE_SHARED was accepted
450    if ((alloc->mHal.state.userProvidedPtr != 0) && (drv->useUserProvidedPtr == false)) {
451        rsdAllocationData2D(rsc, alloc, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X, alloc->getType()->getDimX(), alloc->getType()->getDimY(), alloc->mHal.state.userProvidedPtr, allocSize, 0);
452    }
453
454    return true;
455}
456
457void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) {
458    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
459
460#ifndef RS_COMPATIBILITY_LIB
461    if (drv->bufferID) {
462        // Causes a SW crash....
463        //ALOGV(" mBufferID %i", mBufferID);
464        //glDeleteBuffers(1, &mBufferID);
465        //mBufferID = 0;
466    }
467    if (drv->textureID) {
468        RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID);
469        drv->textureID = 0;
470    }
471    if (drv->renderTargetID) {
472        RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID);
473        drv->renderTargetID = 0;
474    }
475#endif
476
477    if (alloc->mHal.drvState.lod[0].mallocPtr) {
478        // don't free user-allocated ptrs
479        if (!(drv->useUserProvidedPtr)) {
480            free(alloc->mHal.drvState.lod[0].mallocPtr);
481        }
482        alloc->mHal.drvState.lod[0].mallocPtr = NULL;
483    }
484
485#ifndef RS_COMPATIBILITY_LIB
486    if (drv->readBackFBO != NULL) {
487        delete drv->readBackFBO;
488        drv->readBackFBO = NULL;
489    }
490
491    if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) &&
492        (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
493
494        DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
495        ANativeWindow *nw = drv->wndSurface;
496        if (nw) {
497            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
498            mapper.unlock(drv->wndBuffer->handle);
499            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
500        }
501    }
502#endif
503
504    free(drv);
505    alloc->mHal.drv = NULL;
506}
507
508void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
509                         const Type *newType, bool zeroNew) {
510    const uint32_t oldDimX = alloc->mHal.drvState.lod[0].dimX;
511    const uint32_t dimX = newType->getDimX();
512
513    // can't resize Allocations with user-allocated buffers
514    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
515        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
516        return;
517    }
518    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
519    // Calculate the object size
520    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL);
521    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
522    // Build the relative pointer tables.
523    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
524    if(s != verifySize) {
525        rsAssert(!"Size mismatch");
526    }
527
528
529    if (dimX > oldDimX) {
530        size_t stride = alloc->mHal.state.elementSizeBytes;
531        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
532                 0, stride * (dimX - oldDimX));
533    }
534}
535
536static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
537#ifndef RS_COMPATIBILITY_LIB
538    if (!alloc->getIsScript()) {
539        return; // nothing to sync
540    }
541
542    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
543    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
544
545    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
546    if (!drv->textureID && !drv->renderTargetID) {
547        return; // nothing was rendered here yet, so nothing to sync
548    }
549    if (drv->readBackFBO == NULL) {
550        drv->readBackFBO = new RsdFrameBufferObj();
551        drv->readBackFBO->setColorTarget(drv, 0);
552        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
553                                        alloc->getType()->getDimY());
554    }
555
556    // Bind the framebuffer object so we can read back from it
557    drv->readBackFBO->setActive(rsc);
558
559    // Do the readback
560    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
561                alloc->mHal.drvState.lod[0].dimY,
562                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
563
564    // Revert framebuffer to its original
565    lastFbo->setActive(rsc);
566#endif
567}
568
569
570void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
571                         RsAllocationUsageType src) {
572    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
573
574    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
575        if(!alloc->getIsRenderTarget()) {
576            rsc->setError(RS_ERROR_FATAL_DRIVER,
577                          "Attempting to sync allocation from render target, "
578                          "for non-render target allocation");
579        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
580            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
581                                                 "render target");
582        } else {
583            rsdAllocationSyncFromFBO(rsc, alloc);
584        }
585        return;
586    }
587
588    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT);
589
590    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
591        UploadToTexture(rsc, alloc);
592    } else {
593        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
594            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
595            AllocateRenderTarget(rsc, alloc);
596        }
597    }
598    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
599        UploadToBufferObject(rsc, alloc);
600    }
601
602    drv->uploadDeferred = false;
603}
604
605void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
606    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
607    drv->uploadDeferred = true;
608}
609
610void* rsdAllocationGetSurface(const Context *rsc, const Allocation *alloc) {
611#ifndef RS_COMPATIBILITY_LIB
612    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
613
614    // Configure CpuConsumer to be in asynchronous mode
615    drv->cpuConsumer = new CpuConsumer(2, false);
616    sp<IGraphicBufferProducer> bp = drv->cpuConsumer->getProducerInterface();
617    bp->incStrong(NULL);
618    return bp.get();
619#else
620    return NULL;
621#endif
622}
623
624#ifndef RS_COMPATIBILITY_LIB
625static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
626    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
627
628    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
629    if (r) {
630        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
631        return false;
632    }
633
634    // Must lock the whole surface
635    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
636    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
637
638    void *dst = NULL;
639    mapper.lock(drv->wndBuffer->handle,
640            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
641            bounds, &dst);
642    alloc->mHal.drvState.lod[0].mallocPtr = dst;
643    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
644    rsAssert((alloc->mHal.drvState.lod[0].stride & 0xf) == 0);
645
646    return true;
647}
648#endif
649
650void rsdAllocationSetSurface(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
651#ifndef RS_COMPATIBILITY_LIB
652    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
653    ANativeWindow *old = drv->wndSurface;
654
655    if (nw) {
656        nw->incStrong(NULL);
657    }
658
659    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
660        //TODO finish support for render target + script
661        drv->wnd = nw;
662        return;
663    }
664
665    // Cleanup old surface if there is one.
666    if (drv->wndSurface) {
667        ANativeWindow *old = drv->wndSurface;
668        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
669        mapper.unlock(drv->wndBuffer->handle);
670        old->cancelBuffer(old, drv->wndBuffer, -1);
671        drv->wndSurface = NULL;
672        old->decStrong(NULL);
673    }
674
675    if (nw != NULL) {
676        int32_t r;
677        uint32_t flags = 0;
678        r = native_window_set_buffer_count(nw, 3);
679        if (r) {
680            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer count.");
681            goto error;
682        }
683
684        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
685            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
686        }
687        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
688            flags |= GRALLOC_USAGE_HW_RENDER;
689        }
690
691        r = native_window_set_usage(nw, flags);
692        if (r) {
693            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
694            goto error;
695        }
696
697        r = native_window_set_buffers_dimensions(nw, alloc->mHal.drvState.lod[0].dimX,
698                                                 alloc->mHal.drvState.lod[0].dimY);
699        if (r) {
700            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
701            goto error;
702        }
703
704        int format = 0;
705        const Element *e = alloc->mHal.state.type->getElement();
706        switch(e->getType()) {
707        case RS_TYPE_UNSIGNED_8:
708            switch (e->getVectorSize()) {
709            case 1:
710                rsAssert(e->getKind() == RS_KIND_PIXEL_A);
711                format = PIXEL_FORMAT_A_8;
712                break;
713            case 4:
714                rsAssert(e->getKind() == RS_KIND_PIXEL_RGBA);
715                format = PIXEL_FORMAT_RGBA_8888;
716                break;
717            default:
718                rsAssert(0);
719            }
720            break;
721        default:
722            rsAssert(0);
723        }
724
725        r = native_window_set_buffers_format(nw, format);
726        if (r) {
727            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer format.");
728            goto error;
729        }
730
731        IoGetBuffer(rsc, alloc, nw);
732        drv->wndSurface = nw;
733    }
734
735    return;
736
737 error:
738
739    if (nw) {
740        nw->decStrong(NULL);
741    }
742
743
744#endif
745}
746
747void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
748#ifndef RS_COMPATIBILITY_LIB
749    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
750    ANativeWindow *nw = drv->wndSurface;
751    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
752        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
753        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
754        return;
755    }
756    if (nw) {
757        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
758            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
759            mapper.unlock(drv->wndBuffer->handle);
760            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
761            if (r) {
762                rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
763                return;
764            }
765
766            IoGetBuffer(rsc, alloc, nw);
767        }
768    } else {
769        rsc->setError(RS_ERROR_DRIVER, "Sent IO buffer with no attached surface.");
770        return;
771    }
772#endif
773}
774
775void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
776#ifndef RS_COMPATIBILITY_LIB
777    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
778
779    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
780        CpuConsumer::LockedBuffer lb;
781        status_t ret = drv->cpuConsumer->lockNextBuffer(&lb);
782        if (ret == OK) {
783            if (drv->lb.data != NULL) {
784                drv->cpuConsumer->unlockBuffer(drv->lb);
785            }
786            drv->lb = lb;
787            alloc->mHal.drvState.lod[0].mallocPtr = drv->lb.data;
788            alloc->mHal.drvState.lod[0].stride = drv->lb.stride *
789                    alloc->mHal.state.elementSizeBytes;
790
791            if (alloc->mHal.state.yuv) {
792                DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
793            }
794        } else if (ret == BAD_VALUE) {
795            // No new frame, don't do anything
796        } else {
797            rsc->setError(RS_ERROR_DRIVER, "Error receiving IO input buffer.");
798        }
799
800    } else {
801        drv->surfaceTexture->updateTexImage();
802    }
803
804
805#endif
806}
807
808
809void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
810                         uint32_t xoff, uint32_t lod, size_t count,
811                         const void *data, size_t sizeBytes) {
812    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
813
814    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
815    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
816    size_t size = count * eSize;
817
818    if (ptr != data) {
819        // Skip the copy if we are the same allocation. This can arise from
820        // our Bitmap optimization, where we share the same storage.
821        if (alloc->mHal.state.hasReferences) {
822            alloc->incRefs(data, count);
823            alloc->decRefs(ptr, count);
824        }
825        memcpy(ptr, data, size);
826    }
827    drv->uploadDeferred = true;
828}
829
830void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
831                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
832                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
833    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
834
835    size_t eSize = alloc->mHal.state.elementSizeBytes;
836    size_t lineSize = eSize * w;
837    if (!stride) {
838        stride = lineSize;
839    }
840
841    if (alloc->mHal.drvState.lod[0].mallocPtr) {
842        const uint8_t *src = static_cast<const uint8_t *>(data);
843        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
844        if (dst == src) {
845            // Skip the copy if we are the same allocation. This can arise from
846            // our Bitmap optimization, where we share the same storage.
847            drv->uploadDeferred = true;
848            return;
849        }
850
851        for (uint32_t line=yoff; line < (yoff+h); line++) {
852            if (alloc->mHal.state.hasReferences) {
853                alloc->incRefs(src, w);
854                alloc->decRefs(dst, w);
855            }
856            memcpy(dst, src, lineSize);
857            src += stride;
858            dst += alloc->mHal.drvState.lod[lod].stride;
859        }
860        if (alloc->mHal.state.yuv) {
861            int lod = 1;
862            while (alloc->mHal.drvState.lod[lod].mallocPtr) {
863                size_t lineSize = alloc->mHal.drvState.lod[lod].dimX;
864                uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
865
866                for (uint32_t line=(yoff >> 1); line < ((yoff+h)>>1); line++) {
867                    memcpy(dst, src, lineSize);
868                    src += lineSize;
869                    dst += alloc->mHal.drvState.lod[lod].stride;
870                }
871                lod++;
872            }
873
874        }
875        drv->uploadDeferred = true;
876    } else {
877        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
878    }
879}
880
881void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
882                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
883                         uint32_t lod,
884                         uint32_t w, uint32_t h, uint32_t d, const void *data,
885                         size_t sizeBytes, size_t stride) {
886    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
887
888    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
889    uint32_t lineSize = eSize * w;
890    if (!stride) {
891        stride = lineSize;
892    }
893
894    if (alloc->mHal.drvState.lod[0].mallocPtr) {
895        const uint8_t *src = static_cast<const uint8_t *>(data);
896        for (uint32_t z = zoff; z < d; z++) {
897            uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, z, lod,
898                                        RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
899            if (dst == src) {
900                // Skip the copy if we are the same allocation. This can arise from
901                // our Bitmap optimization, where we share the same storage.
902                drv->uploadDeferred = true;
903                return;
904            }
905
906            for (uint32_t line=yoff; line < (yoff+h); line++) {
907                if (alloc->mHal.state.hasReferences) {
908                    alloc->incRefs(src, w);
909                    alloc->decRefs(dst, w);
910                }
911                memcpy(dst, src, lineSize);
912                src += stride;
913                dst += alloc->mHal.drvState.lod[lod].stride;
914            }
915        }
916        drv->uploadDeferred = true;
917    }
918}
919
920void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
921                         uint32_t xoff, uint32_t lod, size_t count,
922                         void *data, size_t sizeBytes) {
923    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
924    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
925    if (data != ptr) {
926        // Skip the copy if we are the same allocation. This can arise from
927        // our Bitmap optimization, where we share the same storage.
928        memcpy(data, ptr, count * eSize);
929    }
930}
931
932void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
933                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
934                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
935    size_t eSize = alloc->mHal.state.elementSizeBytes;
936    size_t lineSize = eSize * w;
937    if (!stride) {
938        stride = lineSize;
939    }
940
941    if (alloc->mHal.drvState.lod[0].mallocPtr) {
942        uint8_t *dst = static_cast<uint8_t *>(data);
943        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
944        if (dst == src) {
945            // Skip the copy if we are the same allocation. This can arise from
946            // our Bitmap optimization, where we share the same storage.
947            return;
948        }
949
950        for (uint32_t line=yoff; line < (yoff+h); line++) {
951            memcpy(dst, src, lineSize);
952            dst += stride;
953            src += alloc->mHal.drvState.lod[lod].stride;
954        }
955    } else {
956        ALOGE("Add code to readback from non-script memory");
957    }
958}
959
960
961void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
962                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
963                         uint32_t lod,
964                         uint32_t w, uint32_t h, uint32_t d, void *data, size_t sizeBytes, size_t stride) {
965    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
966    uint32_t lineSize = eSize * w;
967    if (!stride) {
968        stride = lineSize;
969    }
970
971    if (alloc->mHal.drvState.lod[0].mallocPtr) {
972        uint8_t *dst = static_cast<uint8_t *>(data);
973        for (uint32_t z = zoff; z < d; z++) {
974            const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, z, lod,
975                                              RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
976            if (dst == src) {
977                // Skip the copy if we are the same allocation. This can arise from
978                // our Bitmap optimization, where we share the same storage.
979                return;
980            }
981
982            for (uint32_t line=yoff; line < (yoff+h); line++) {
983                memcpy(dst, src, lineSize);
984                dst += stride;
985                src += alloc->mHal.drvState.lod[lod].stride;
986            }
987        }
988    }
989}
990
991void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
992                          const android::renderscript::Allocation *alloc) {
993    return alloc->mHal.drvState.lod[0].mallocPtr;
994}
995
996void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
997                          const android::renderscript::Allocation *alloc) {
998
999}
1000
1001void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
1002                               const android::renderscript::Allocation *dstAlloc,
1003                               uint32_t dstXoff, uint32_t dstLod, size_t count,
1004                               const android::renderscript::Allocation *srcAlloc,
1005                               uint32_t srcXoff, uint32_t srcLod) {
1006}
1007
1008
1009void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
1010                                      const android::renderscript::Allocation *dstAlloc,
1011                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1012                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1013                                      const android::renderscript::Allocation *srcAlloc,
1014                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1015                                      RsAllocationCubemapFace srcFace) {
1016    size_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1017    for (uint32_t i = 0; i < h; i ++) {
1018        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, 0, dstLod, dstFace);
1019        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, 0, srcLod, srcFace);
1020        memcpy(dstPtr, srcPtr, w * elementSize);
1021
1022        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1023        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1024    }
1025}
1026
1027void rsdAllocationData3D_alloc_script(const android::renderscript::Context *rsc,
1028                                      const android::renderscript::Allocation *dstAlloc,
1029                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff, uint32_t dstLod,
1030                                      uint32_t w, uint32_t h, uint32_t d,
1031                                      const android::renderscript::Allocation *srcAlloc,
1032                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff, uint32_t srcLod) {
1033    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1034    for (uint32_t j = 0; j < d; j++) {
1035        for (uint32_t i = 0; i < h; i ++) {
1036            uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstZoff + j,
1037                                           dstLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1038            uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcZoff + j,
1039                                           srcLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1040            memcpy(dstPtr, srcPtr, w * elementSize);
1041
1042            //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1043            //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1044        }
1045    }
1046}
1047
1048void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
1049                               const android::renderscript::Allocation *dstAlloc,
1050                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1051                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1052                               const android::renderscript::Allocation *srcAlloc,
1053                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1054                               RsAllocationCubemapFace srcFace) {
1055    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1056        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1057                                             "yet implemented.");
1058        return;
1059    }
1060    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
1061                                     dstLod, dstFace, w, h, srcAlloc,
1062                                     srcXoff, srcYoff, srcLod, srcFace);
1063}
1064
1065void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
1066                               const android::renderscript::Allocation *dstAlloc,
1067                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
1068                               uint32_t dstLod,
1069                               uint32_t w, uint32_t h, uint32_t d,
1070                               const android::renderscript::Allocation *srcAlloc,
1071                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
1072                               uint32_t srcLod) {
1073    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1074        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1075                                             "yet implemented.");
1076        return;
1077    }
1078    rsdAllocationData3D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstZoff,
1079                                     dstLod, w, h, d, srcAlloc,
1080                                     srcXoff, srcYoff, srcZoff, srcLod);
1081}
1082
1083void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc,
1084                                uint32_t x,
1085                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1086    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1087
1088    size_t eSize = alloc->mHal.state.elementSizeBytes;
1089    uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1090
1091    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1092    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1093
1094    if (alloc->mHal.state.hasReferences) {
1095        e->incRefs(data);
1096        e->decRefs(ptr);
1097    }
1098
1099    memcpy(ptr, data, sizeBytes);
1100    drv->uploadDeferred = true;
1101}
1102
1103void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc,
1104                                uint32_t x, uint32_t y,
1105                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1106    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1107
1108    size_t eSize = alloc->mHal.state.elementSizeBytes;
1109    uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1110
1111    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1112    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1113
1114    if (alloc->mHal.state.hasReferences) {
1115        e->incRefs(data);
1116        e->decRefs(ptr);
1117    }
1118
1119    memcpy(ptr, data, sizeBytes);
1120    drv->uploadDeferred = true;
1121}
1122
1123static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1124    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1125    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1126
1127    for (uint32_t y=0; y < h; y++) {
1128        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1129        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2, lod, face);
1130        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2+1, lod, face);
1131
1132        for (uint32_t x=0; x < w; x++) {
1133            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
1134            oPtr ++;
1135            i1 += 2;
1136            i2 += 2;
1137        }
1138    }
1139}
1140
1141static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1142    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1143    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1144
1145    for (uint32_t y=0; y < h; y++) {
1146        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1147        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1148        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1149
1150        for (uint32_t x=0; x < w; x++) {
1151            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
1152            oPtr ++;
1153            i1 += 2;
1154            i2 += 2;
1155        }
1156    }
1157}
1158
1159static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1160    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1161    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1162
1163    for (uint32_t y=0; y < h; y++) {
1164        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1165        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1166        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1167
1168        for (uint32_t x=0; x < w; x++) {
1169            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
1170            oPtr ++;
1171            i1 += 2;
1172            i2 += 2;
1173        }
1174    }
1175}
1176
1177void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
1178    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
1179        return;
1180    }
1181    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
1182    for (uint32_t face = 0; face < numFaces; face ++) {
1183        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
1184            switch (alloc->getType()->getElement()->getSizeBits()) {
1185            case 32:
1186                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
1187                break;
1188            case 16:
1189                mip565(alloc, lod, (RsAllocationCubemapFace)face);
1190                break;
1191            case 8:
1192                mip8(alloc, lod, (RsAllocationCubemapFace)face);
1193                break;
1194            }
1195        }
1196    }
1197}
1198