rsdAllocation.cpp revision 28847289a9052522d4f8d25b6fc47c3ca51286c3 
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#endif
27
28#ifdef RS_COMPATIBILITY_LIB
29#include "rsCompatibilityLib.h"
30#else
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#ifndef RS_COMPATIBILITY_LIB
51const static GLenum gFaceOrder[] = {
52    GL_TEXTURE_CUBE_MAP_POSITIVE_X,
53    GL_TEXTURE_CUBE_MAP_NEGATIVE_X,
54    GL_TEXTURE_CUBE_MAP_POSITIVE_Y,
55    GL_TEXTURE_CUBE_MAP_NEGATIVE_Y,
56    GL_TEXTURE_CUBE_MAP_POSITIVE_Z,
57    GL_TEXTURE_CUBE_MAP_NEGATIVE_Z
58};
59
60GLenum rsdTypeToGLType(RsDataType t) {
61    switch (t) {
62    case RS_TYPE_UNSIGNED_5_6_5:    return GL_UNSIGNED_SHORT_5_6_5;
63    case RS_TYPE_UNSIGNED_5_5_5_1:  return GL_UNSIGNED_SHORT_5_5_5_1;
64    case RS_TYPE_UNSIGNED_4_4_4_4:  return GL_UNSIGNED_SHORT_4_4_4_4;
65
66    //case RS_TYPE_FLOAT_16:      return GL_HALF_FLOAT;
67    case RS_TYPE_FLOAT_32:      return GL_FLOAT;
68    case RS_TYPE_UNSIGNED_8:    return GL_UNSIGNED_BYTE;
69    case RS_TYPE_UNSIGNED_16:   return GL_UNSIGNED_SHORT;
70    case RS_TYPE_SIGNED_8:      return GL_BYTE;
71    case RS_TYPE_SIGNED_16:     return GL_SHORT;
72    default:    break;
73    }
74    return 0;
75}
76
77GLenum rsdKindToGLFormat(RsDataKind k) {
78    switch (k) {
79    case RS_KIND_PIXEL_L: return GL_LUMINANCE;
80    case RS_KIND_PIXEL_A: return GL_ALPHA;
81    case RS_KIND_PIXEL_LA: return GL_LUMINANCE_ALPHA;
82    case RS_KIND_PIXEL_RGB: return GL_RGB;
83    case RS_KIND_PIXEL_RGBA: return GL_RGBA;
84    case RS_KIND_PIXEL_DEPTH: return GL_DEPTH_COMPONENT16;
85    default: break;
86    }
87    return 0;
88}
89#endif
90
91uint8_t *GetOffsetPtr(const android::renderscript::Allocation *alloc,
92                      uint32_t xoff, uint32_t yoff, uint32_t zoff,
93                      uint32_t lod, RsAllocationCubemapFace face) {
94    uint8_t *ptr = (uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
95    ptr += face * alloc->mHal.drvState.faceOffset;
96    ptr += zoff * alloc->mHal.drvState.lod[lod].dimY * alloc->mHal.drvState.lod[lod].stride;
97    ptr += yoff * alloc->mHal.drvState.lod[lod].stride;
98    ptr += xoff * alloc->mHal.state.elementSizeBytes;
99    return ptr;
100}
101
102
103static void Update2DTexture(const Context *rsc, const Allocation *alloc, const void *ptr,
104                            uint32_t xoff, uint32_t yoff, uint32_t lod,
105                            RsAllocationCubemapFace face, uint32_t w, uint32_t h) {
106#ifndef RS_COMPATIBILITY_LIB
107    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
108
109    rsAssert(drv->textureID);
110    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
111    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
112    GLenum t = GL_TEXTURE_2D;
113    if (alloc->mHal.state.hasFaces) {
114        t = gFaceOrder[face];
115    }
116    RSD_CALL_GL(glTexSubImage2D, t, lod, xoff, yoff, w, h, drv->glFormat, drv->glType, ptr);
117#endif
118}
119
120
121#ifndef RS_COMPATIBILITY_LIB
122static void Upload2DTexture(const Context *rsc, const Allocation *alloc, bool isFirstUpload) {
123    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
124
125    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
126    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
127
128    uint32_t faceCount = 1;
129    if (alloc->mHal.state.hasFaces) {
130        faceCount = 6;
131    }
132
133    rsdGLCheckError(rsc, "Upload2DTexture 1 ");
134    for (uint32_t face = 0; face < faceCount; face ++) {
135        for (uint32_t lod = 0; lod < alloc->mHal.state.type->getLODCount(); lod++) {
136            const uint8_t *p = GetOffsetPtr(alloc, 0, 0, 0, lod, (RsAllocationCubemapFace)face);
137
138            GLenum t = GL_TEXTURE_2D;
139            if (alloc->mHal.state.hasFaces) {
140                t = gFaceOrder[face];
141            }
142
143            if (isFirstUpload) {
144                RSD_CALL_GL(glTexImage2D, t, lod, drv->glFormat,
145                             alloc->mHal.state.type->getLODDimX(lod),
146                             alloc->mHal.state.type->getLODDimY(lod),
147                             0, drv->glFormat, drv->glType, p);
148            } else {
149                RSD_CALL_GL(glTexSubImage2D, t, lod, 0, 0,
150                                alloc->mHal.state.type->getLODDimX(lod),
151                                alloc->mHal.state.type->getLODDimY(lod),
152                                drv->glFormat, drv->glType, p);
153            }
154        }
155    }
156
157    if (alloc->mHal.state.mipmapControl == RS_ALLOCATION_MIPMAP_ON_SYNC_TO_TEXTURE) {
158        RSD_CALL_GL(glGenerateMipmap, drv->glTarget);
159    }
160    rsdGLCheckError(rsc, "Upload2DTexture");
161}
162#endif
163
164static void UploadToTexture(const Context *rsc, const Allocation *alloc) {
165#ifndef RS_COMPATIBILITY_LIB
166    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
167
168    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
169        if (!drv->textureID) {
170            RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
171        }
172        return;
173    }
174
175    if (!drv->glType || !drv->glFormat) {
176        return;
177    }
178
179    if (!alloc->mHal.drvState.lod[0].mallocPtr) {
180        return;
181    }
182
183    bool isFirstUpload = false;
184
185    if (!drv->textureID) {
186        RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
187        isFirstUpload = true;
188    }
189
190    Upload2DTexture(rsc, alloc, isFirstUpload);
191
192    if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
193        if (alloc->mHal.drvState.lod[0].mallocPtr) {
194            free(alloc->mHal.drvState.lod[0].mallocPtr);
195            alloc->mHal.drvState.lod[0].mallocPtr = nullptr;
196        }
197    }
198    rsdGLCheckError(rsc, "UploadToTexture");
199#endif
200}
201
202static void AllocateRenderTarget(const Context *rsc, const Allocation *alloc) {
203#ifndef RS_COMPATIBILITY_LIB
204    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
205
206    if (!drv->glFormat) {
207        return;
208    }
209
210    if (!drv->renderTargetID) {
211        RSD_CALL_GL(glGenRenderbuffers, 1, &drv->renderTargetID);
212
213        if (!drv->renderTargetID) {
214            // This should generally not happen
215            ALOGE("allocateRenderTarget failed to gen mRenderTargetID");
216            rsc->dumpDebug();
217            return;
218        }
219        RSD_CALL_GL(glBindRenderbuffer, GL_RENDERBUFFER, drv->renderTargetID);
220        RSD_CALL_GL(glRenderbufferStorage, GL_RENDERBUFFER, drv->glFormat,
221                    alloc->mHal.drvState.lod[0].dimX, alloc->mHal.drvState.lod[0].dimY);
222    }
223    rsdGLCheckError(rsc, "AllocateRenderTarget");
224#endif
225}
226
227static void UploadToBufferObject(const Context *rsc, const Allocation *alloc) {
228#ifndef RS_COMPATIBILITY_LIB
229    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
230
231    rsAssert(!alloc->mHal.state.type->getDimY());
232    rsAssert(!alloc->mHal.state.type->getDimZ());
233
234    //alloc->mHal.state.usageFlags |= RS_ALLOCATION_USAGE_GRAPHICS_VERTEX;
235
236    if (!drv->bufferID) {
237        RSD_CALL_GL(glGenBuffers, 1, &drv->bufferID);
238    }
239    if (!drv->bufferID) {
240        ALOGE("Upload to buffer object failed");
241        drv->uploadDeferred = true;
242        return;
243    }
244    RSD_CALL_GL(glBindBuffer, drv->glTarget, drv->bufferID);
245    RSD_CALL_GL(glBufferData, drv->glTarget,
246                alloc->mHal.state.type->getPackedSizeBytes(),
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    state->lod[1].dimX = state->lod[0].dimX / 2;
259    state->lod[1].dimY = state->lod[0].dimY / 2;
260    state->lod[2].dimX = state->lod[0].dimX / 2;
261    state->lod[2].dimY = state->lod[0].dimY / 2;
262    state->yuv.shift = 1;
263    state->yuv.step = 1;
264    state->lodCount = 3;
265
266#ifndef RS_SERVER
267    switch(yuv) {
268    case HAL_PIXEL_FORMAT_YV12:
269        state->lod[2].stride = rsRound(state->lod[0].stride >> 1, 16);
270        state->lod[2].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
271                (state->lod[0].stride * state->lod[0].dimY);
272        uvSize += state->lod[2].stride * state->lod[2].dimY;
273
274        state->lod[1].stride = state->lod[2].stride;
275        state->lod[1].mallocPtr = ((uint8_t *)state->lod[2].mallocPtr) +
276                (state->lod[2].stride * state->lod[2].dimY);
277        uvSize += state->lod[1].stride * state->lod[2].dimY;
278        break;
279    case HAL_PIXEL_FORMAT_YCrCb_420_SP:  // NV21
280        //state->lod[1].dimX = state->lod[0].dimX;
281        state->lod[1].stride = state->lod[0].stride;
282        state->lod[2].stride = state->lod[0].stride;
283        state->lod[2].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
284                (state->lod[0].stride * state->lod[0].dimY);
285        state->lod[1].mallocPtr = ((uint8_t *)state->lod[2].mallocPtr) + 1;
286        uvSize += state->lod[1].stride * state->lod[1].dimY;
287        state->yuv.step = 2;
288        break;
289#ifndef RS_COMPATIBILITY_LIB
290    case HAL_PIXEL_FORMAT_YCbCr_420_888:
291        // This will be filled in by ioReceive()
292        break;
293#endif
294    default:
295        rsAssert(0);
296    }
297#endif
298    return uvSize;
299}
300
301
302static size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc,
303        const Type *type, uint8_t *ptr) {
304    alloc->mHal.drvState.lod[0].dimX = type->getDimX();
305    alloc->mHal.drvState.lod[0].dimY = type->getDimY();
306    alloc->mHal.drvState.lod[0].dimZ = type->getDimZ();
307    alloc->mHal.drvState.lod[0].mallocPtr = 0;
308    // Stride needs to be 16-byte aligned too!
309    size_t stride = alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes();
310    alloc->mHal.drvState.lod[0].stride = rsRound(stride, 16);
311    alloc->mHal.drvState.lodCount = type->getLODCount();
312    alloc->mHal.drvState.faceCount = type->getDimFaces();
313
314    size_t offsets[Allocation::MAX_LOD];
315    memset(offsets, 0, sizeof(offsets));
316
317    size_t o = alloc->mHal.drvState.lod[0].stride * rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) *
318            rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u);
319    if (alloc->mHal.state.yuv) {
320        o += DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
321
322        for (uint32_t ct = 1; ct < alloc->mHal.drvState.lodCount; ct++) {
323            offsets[ct] = (size_t)alloc->mHal.drvState.lod[ct].mallocPtr;
324        }
325    } else if(alloc->mHal.drvState.lodCount > 1) {
326        uint32_t tx = alloc->mHal.drvState.lod[0].dimX;
327        uint32_t ty = alloc->mHal.drvState.lod[0].dimY;
328        uint32_t tz = alloc->mHal.drvState.lod[0].dimZ;
329        for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
330            alloc->mHal.drvState.lod[lod].dimX = tx;
331            alloc->mHal.drvState.lod[lod].dimY = ty;
332            alloc->mHal.drvState.lod[lod].dimZ = tz;
333            alloc->mHal.drvState.lod[lod].stride =
334                    rsRound(tx * type->getElementSizeBytes(), 16);
335            offsets[lod] = o;
336            o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u);
337            if (tx > 1) tx >>= 1;
338            if (ty > 1) ty >>= 1;
339            if (tz > 1) tz >>= 1;
340        }
341    }
342
343    alloc->mHal.drvState.faceOffset = o;
344
345    alloc->mHal.drvState.lod[0].mallocPtr = ptr;
346    for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
347        alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod];
348    }
349
350    size_t allocSize = alloc->mHal.drvState.faceOffset;
351    if(alloc->mHal.drvState.faceCount) {
352        allocSize *= 6;
353    }
354
355    return allocSize;
356}
357
358static uint8_t* allocAlignedMemory(size_t allocSize, bool forceZero) {
359    // We align all allocations to a 16-byte boundary.
360    uint8_t* ptr = (uint8_t *)memalign(16, allocSize);
361    if (!ptr) {
362        return nullptr;
363    }
364    if (forceZero) {
365        memset(ptr, 0, allocSize);
366    }
367    return ptr;
368}
369
370bool rsdAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) {
371    DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
372    if (!drv) {
373        return false;
374    }
375    alloc->mHal.drv = drv;
376
377    // Calculate the object size.
378    size_t allocSize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), nullptr);
379
380    uint8_t * ptr = nullptr;
381    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) {
382
383    } else if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
384        // Allocation is allocated when the surface is created
385        // in getSurface
386    } else if (alloc->mHal.state.userProvidedPtr != nullptr) {
387        // user-provided allocation
388        // limitations: no faces, no LOD, USAGE_SCRIPT or SCRIPT+TEXTURE only
389        if (!(alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED) ||
390              alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED | RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE))) {
391            ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT | USAGE_SHARED or USAGE_SCRIPT | USAGE_SHARED | USAGE_GRAPHICS_TEXTURE");
392            return false;
393        }
394        if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) {
395            ALOGE("User-allocated buffers must not have multiple faces or LODs");
396            return false;
397        }
398
399        // rows must be 16-byte aligned
400        // validate that here, otherwise fall back to not use the user-backed allocation
401        if (((alloc->getType()->getDimX() * alloc->getType()->getElement()->getSizeBytes()) % 16) != 0) {
402            ALOGV("User-backed allocation failed stride requirement, falling back to separate allocation");
403            drv->useUserProvidedPtr = false;
404
405            ptr = allocAlignedMemory(allocSize, forceZero);
406            if (!ptr) {
407                alloc->mHal.drv = nullptr;
408                free(drv);
409                return false;
410            }
411
412        } else {
413            drv->useUserProvidedPtr = true;
414            ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr;
415        }
416    } else {
417        ptr = allocAlignedMemory(allocSize, forceZero);
418        if (!ptr) {
419            alloc->mHal.drv = nullptr;
420            free(drv);
421            return false;
422        }
423    }
424    // Build the pointer tables
425    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr);
426    if(allocSize != verifySize) {
427        rsAssert(!"Size mismatch");
428    }
429
430#ifndef RS_SERVER
431    drv->glTarget = GL_NONE;
432    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
433        if (alloc->mHal.state.hasFaces) {
434            drv->glTarget = GL_TEXTURE_CUBE_MAP;
435        } else {
436            drv->glTarget = GL_TEXTURE_2D;
437        }
438    } else {
439        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
440            drv->glTarget = GL_ARRAY_BUFFER;
441        }
442    }
443#endif
444
445#ifndef RS_COMPATIBILITY_LIB
446    drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType());
447    drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind());
448#else
449    drv->glType = 0;
450    drv->glFormat = 0;
451#endif
452
453    if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) {
454        drv->uploadDeferred = true;
455    }
456
457
458    drv->readBackFBO = nullptr;
459
460    // fill out the initial state of the buffer if we couldn't use the user-provided ptr and USAGE_SHARED was accepted
461    if ((alloc->mHal.state.userProvidedPtr != 0) && (drv->useUserProvidedPtr == false)) {
462        rsdAllocationData2D(rsc, alloc, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X, alloc->getType()->getDimX(), alloc->getType()->getDimY(), alloc->mHal.state.userProvidedPtr, allocSize, 0);
463    }
464
465
466#ifdef RS_FIND_OFFSETS
467    ALOGE("pointer for allocation: %p", alloc);
468    ALOGE("pointer for allocation.drv: %p", &alloc->mHal.drv);
469#endif
470
471
472    return true;
473}
474
475void rsdAllocationAdapterOffset(const Context *rsc, const Allocation *alloc) {
476    //ALOGE("rsdAllocationAdapterOffset");
477
478    // Get a base pointer to the new LOD
479    const Allocation *base = alloc->mHal.state.baseAlloc;
480    const Type *type = alloc->mHal.state.type;
481    if (base == nullptr) {
482        return;
483    }
484
485    uint8_t * ptrA = (uint8_t *)base->getPointerUnchecked(alloc->mHal.state.originX, alloc->mHal.state.originY);
486    uint8_t * ptrB = (uint8_t *)base->getPointerUnchecked(0, 0);
487
488    //ALOGE("rsdAllocationAdapterOffset  %p  %p", ptrA, ptrB);
489    //ALOGE("rsdAllocationAdapterOffset  lodCount %i", alloc->mHal.drvState.lodCount);
490
491    const int lodBias = alloc->mHal.state.originLOD;
492    uint32_t lodCount = rsMax(alloc->mHal.drvState.lodCount, (uint32_t)1);
493    for (uint32_t lod=0; lod < lodCount; lod++) {
494        alloc->mHal.drvState.lod[lod] = base->mHal.drvState.lod[lod + lodBias];
495        alloc->mHal.drvState.lod[lod].mallocPtr =
496                ((uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr + (ptrA - ptrB));
497        //ALOGE("rsdAllocationAdapterOffset  lod  %p  %i %i", alloc->mHal.drvState.lod[lod].mallocPtr, alloc->mHal.drvState.lod[lod].dimX, alloc->mHal.drvState.lod[lod].dimY);
498    }
499}
500
501bool rsdAllocationAdapterInit(const Context *rsc, Allocation *alloc) {
502    DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
503    if (!drv) {
504        return false;
505    }
506    alloc->mHal.drv = drv;
507
508    // We need to build an allocation that looks like a subset of the parent allocation
509    rsdAllocationAdapterOffset(rsc, alloc);
510
511    return true;
512}
513
514void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) {
515    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
516
517    if (alloc->mHal.state.baseAlloc == nullptr) {
518#ifndef RS_COMPATIBILITY_LIB
519        if (drv->bufferID) {
520            // Causes a SW crash....
521            //ALOGV(" mBufferID %i", mBufferID);
522            //glDeleteBuffers(1, &mBufferID);
523            //mBufferID = 0;
524        }
525        if (drv->textureID) {
526            RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID);
527            drv->textureID = 0;
528        }
529        if (drv->renderTargetID) {
530            RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID);
531            drv->renderTargetID = 0;
532        }
533#endif
534
535        if (alloc->mHal.drvState.lod[0].mallocPtr) {
536            // don't free user-allocated ptrs or IO_OUTPUT buffers
537            if (!(drv->useUserProvidedPtr) &&
538                !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) &&
539                !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
540                    free(alloc->mHal.drvState.lod[0].mallocPtr);
541            }
542            alloc->mHal.drvState.lod[0].mallocPtr = nullptr;
543        }
544
545#ifndef RS_COMPATIBILITY_LIB
546        if (drv->readBackFBO != nullptr) {
547            delete drv->readBackFBO;
548            drv->readBackFBO = nullptr;
549        }
550
551        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) &&
552            (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
553
554            DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
555            ANativeWindow *nw = drv->wndSurface;
556            if (nw) {
557                GraphicBufferMapper &mapper = GraphicBufferMapper::get();
558                mapper.unlock(drv->wndBuffer->handle);
559                int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
560
561                drv->wndSurface = nullptr;
562                native_window_api_disconnect(nw, NATIVE_WINDOW_API_CPU);
563                nw->decStrong(nullptr);
564            }
565        }
566#endif
567    }
568
569    free(drv);
570    alloc->mHal.drv = nullptr;
571}
572
573void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
574                         const Type *newType, bool zeroNew) {
575    const uint32_t oldDimX = alloc->mHal.drvState.lod[0].dimX;
576    const uint32_t dimX = newType->getDimX();
577
578    // can't resize Allocations with user-allocated buffers
579    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
580        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
581        return;
582    }
583    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
584    // Calculate the object size
585    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, nullptr);
586    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
587    // Build the relative pointer tables.
588    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
589    if(s != verifySize) {
590        rsAssert(!"Size mismatch");
591    }
592
593
594    if (dimX > oldDimX) {
595        size_t stride = alloc->mHal.state.elementSizeBytes;
596        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
597                 0, stride * (dimX - oldDimX));
598    }
599}
600
601static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
602#ifndef RS_COMPATIBILITY_LIB
603    if (!alloc->getIsScript()) {
604        return; // nothing to sync
605    }
606
607    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
608    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
609
610    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
611    if (!drv->textureID && !drv->renderTargetID) {
612        return; // nothing was rendered here yet, so nothing to sync
613    }
614    if (drv->readBackFBO == nullptr) {
615        drv->readBackFBO = new RsdFrameBufferObj();
616        drv->readBackFBO->setColorTarget(drv, 0);
617        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
618                                        alloc->getType()->getDimY());
619    }
620
621    // Bind the framebuffer object so we can read back from it
622    drv->readBackFBO->setActive(rsc);
623
624    // Do the readback
625    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
626                alloc->mHal.drvState.lod[0].dimY,
627                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
628
629    // Revert framebuffer to its original
630    lastFbo->setActive(rsc);
631#endif
632}
633
634
635void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
636                         RsAllocationUsageType src) {
637    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
638
639    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
640        if(!alloc->getIsRenderTarget()) {
641            rsc->setError(RS_ERROR_FATAL_DRIVER,
642                          "Attempting to sync allocation from render target, "
643                          "for non-render target allocation");
644        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
645            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
646                                                 "render target");
647        } else {
648            rsdAllocationSyncFromFBO(rsc, alloc);
649        }
650        return;
651    }
652
653    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT || src == RS_ALLOCATION_USAGE_SHARED);
654
655    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
656        UploadToTexture(rsc, alloc);
657    } else {
658        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
659            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
660            AllocateRenderTarget(rsc, alloc);
661        }
662    }
663    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
664        UploadToBufferObject(rsc, alloc);
665    }
666
667    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
668
669        if (src == RS_ALLOCATION_USAGE_SHARED) {
670            // just a memory fence for the CPU driver
671            // vendor drivers probably want to flush any dirty cachelines for
672            // this particular Allocation
673            __sync_synchronize();
674        }
675    }
676
677    drv->uploadDeferred = false;
678}
679
680void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
681    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
682    drv->uploadDeferred = true;
683}
684
685#ifndef RS_COMPATIBILITY_LIB
686static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
687    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
688
689    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
690    if (r) {
691        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
692        return false;
693    }
694
695    // Must lock the whole surface
696    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
697    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
698
699    void *dst = nullptr;
700    mapper.lock(drv->wndBuffer->handle,
701            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
702            bounds, &dst);
703    alloc->mHal.drvState.lod[0].mallocPtr = dst;
704    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
705    rsAssert((alloc->mHal.drvState.lod[0].stride & 0xf) == 0);
706
707    return true;
708}
709#endif
710
711void rsdAllocationSetSurface(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
712#ifndef RS_COMPATIBILITY_LIB
713    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
714    ANativeWindow *old = drv->wndSurface;
715
716    if (nw) {
717        nw->incStrong(nullptr);
718    }
719
720    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
721        //TODO finish support for render target + script
722        drv->wnd = nw;
723        return;
724    }
725
726    // Cleanup old surface if there is one.
727    if (drv->wndSurface) {
728        ANativeWindow *old = drv->wndSurface;
729        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
730        mapper.unlock(drv->wndBuffer->handle);
731        old->cancelBuffer(old, drv->wndBuffer, -1);
732        drv->wndSurface = nullptr;
733
734        native_window_api_disconnect(old, NATIVE_WINDOW_API_CPU);
735        old->decStrong(nullptr);
736    }
737
738    if (nw != nullptr) {
739        int32_t r;
740        uint32_t flags = 0;
741
742        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
743            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
744        }
745        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
746            flags |= GRALLOC_USAGE_HW_RENDER;
747        }
748
749        r = native_window_api_connect(nw, NATIVE_WINDOW_API_CPU);
750        if (r) {
751            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
752            goto error;
753        }
754
755        r = native_window_set_usage(nw, flags);
756        if (r) {
757            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
758            goto error;
759        }
760
761        r = native_window_set_buffers_dimensions(nw, alloc->mHal.drvState.lod[0].dimX,
762                                                 alloc->mHal.drvState.lod[0].dimY);
763        if (r) {
764            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
765            goto error;
766        }
767
768        int format = 0;
769        const Element *e = alloc->mHal.state.type->getElement();
770        if ((e->getType() != RS_TYPE_UNSIGNED_8) ||
771            (e->getVectorSize() != 4)) {
772            // We do not check for RGBA, RGBx, to allow for interop with U8_4
773
774            rsc->setError(RS_ERROR_DRIVER, "Surface passed to setSurface is not U8_4, RGBA.");
775            goto error;
776        }
777        format = PIXEL_FORMAT_RGBA_8888;
778
779        r = native_window_set_buffers_format(nw, format);
780        if (r) {
781            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer format.");
782            goto error;
783        }
784
785        IoGetBuffer(rsc, alloc, nw);
786        drv->wndSurface = nw;
787    }
788
789    return;
790
791 error:
792
793    if (nw) {
794        nw->decStrong(nullptr);
795    }
796
797
798#endif
799}
800
801void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
802#ifndef RS_COMPATIBILITY_LIB
803    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
804    ANativeWindow *nw = drv->wndSurface;
805    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
806        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
807        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
808        return;
809    }
810    if (nw) {
811        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
812            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
813            mapper.unlock(drv->wndBuffer->handle);
814            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
815            if (r) {
816                rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
817                return;
818            }
819
820            IoGetBuffer(rsc, alloc, nw);
821        }
822    } else {
823        rsc->setError(RS_ERROR_DRIVER, "Sent IO buffer with no attached surface.");
824        return;
825    }
826#endif
827}
828
829void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
830#ifndef RS_COMPATIBILITY_LIB
831    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
832    if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
833        drv->surfaceTexture->updateTexImage();
834    }
835#endif
836    if (alloc->mHal.state.yuv) {
837        DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
838    }
839}
840
841
842void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
843                         uint32_t xoff, uint32_t lod, size_t count,
844                         const void *data, size_t sizeBytes) {
845    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
846
847    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
848    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
849    size_t size = count * eSize;
850    if (ptr != data) {
851        // Skip the copy if we are the same allocation. This can arise from
852        // our Bitmap optimization, where we share the same storage.
853        if (alloc->mHal.state.hasReferences) {
854            alloc->incRefs(data, count);
855            alloc->decRefs(ptr, count);
856        }
857        memcpy(ptr, data, size);
858    }
859    drv->uploadDeferred = true;
860}
861
862void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
863                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
864                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
865    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
866
867    size_t eSize = alloc->mHal.state.elementSizeBytes;
868    size_t lineSize = eSize * w;
869    if (!stride) {
870        stride = lineSize;
871    }
872
873    if (alloc->mHal.drvState.lod[0].mallocPtr) {
874        const uint8_t *src = static_cast<const uint8_t *>(data);
875        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
876        if (dst == src) {
877            // Skip the copy if we are the same allocation. This can arise from
878            // our Bitmap optimization, where we share the same storage.
879            drv->uploadDeferred = true;
880            return;
881        }
882
883        for (uint32_t line=yoff; line < (yoff+h); line++) {
884            if (alloc->mHal.state.hasReferences) {
885                alloc->incRefs(src, w);
886                alloc->decRefs(dst, w);
887            }
888            memcpy(dst, src, lineSize);
889            src += stride;
890            dst += alloc->mHal.drvState.lod[lod].stride;
891        }
892        if (alloc->mHal.state.yuv) {
893            size_t clineSize = lineSize;
894            int lod = 1;
895            int maxLod = 2;
896            if (alloc->mHal.state.yuv == HAL_PIXEL_FORMAT_YV12) {
897                maxLod = 3;
898                clineSize >>= 1;
899            } else if (alloc->mHal.state.yuv == HAL_PIXEL_FORMAT_YCrCb_420_SP) {
900                lod = 2;
901                maxLod = 3;
902            }
903
904            while (lod < maxLod) {
905                uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
906
907                for (uint32_t line=(yoff >> 1); line < ((yoff+h)>>1); line++) {
908                    memcpy(dst, src, clineSize);
909                    src += alloc->mHal.drvState.lod[lod].stride;
910                    dst += alloc->mHal.drvState.lod[lod].stride;
911                }
912                lod++;
913            }
914
915        }
916        drv->uploadDeferred = true;
917    } else {
918        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
919    }
920}
921
922void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
923                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
924                         uint32_t lod,
925                         uint32_t w, uint32_t h, uint32_t d, const void *data,
926                         size_t sizeBytes, size_t stride) {
927    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
928
929    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
930    uint32_t lineSize = eSize * w;
931    if (!stride) {
932        stride = lineSize;
933    }
934
935    if (alloc->mHal.drvState.lod[0].mallocPtr) {
936        const uint8_t *src = static_cast<const uint8_t *>(data);
937        for (uint32_t z = zoff; z < (d + zoff); z++) {
938            uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, z, lod,
939                                        RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
940            if (dst == src) {
941                // Skip the copy if we are the same allocation. This can arise from
942                // our Bitmap optimization, where we share the same storage.
943                drv->uploadDeferred = true;
944                return;
945            }
946
947            for (uint32_t line=yoff; line < (yoff+h); line++) {
948                if (alloc->mHal.state.hasReferences) {
949                    alloc->incRefs(src, w);
950                    alloc->decRefs(dst, w);
951                }
952                memcpy(dst, src, lineSize);
953                src += stride;
954                dst += alloc->mHal.drvState.lod[lod].stride;
955            }
956        }
957        drv->uploadDeferred = true;
958    }
959}
960
961void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
962                         uint32_t xoff, uint32_t lod, size_t count,
963                         void *data, size_t sizeBytes) {
964    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
965    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
966    if (data != ptr) {
967        // Skip the copy if we are the same allocation. This can arise from
968        // our Bitmap optimization, where we share the same storage.
969        memcpy(data, ptr, count * eSize);
970    }
971}
972
973void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
974                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
975                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
976    size_t eSize = alloc->mHal.state.elementSizeBytes;
977    size_t lineSize = eSize * w;
978    if (!stride) {
979        stride = lineSize;
980    }
981
982    if (alloc->mHal.drvState.lod[0].mallocPtr) {
983        uint8_t *dst = static_cast<uint8_t *>(data);
984        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
985        if (dst == src) {
986            // Skip the copy if we are the same allocation. This can arise from
987            // our Bitmap optimization, where we share the same storage.
988            return;
989        }
990
991        for (uint32_t line=yoff; line < (yoff+h); line++) {
992            memcpy(dst, src, lineSize);
993            dst += stride;
994            src += alloc->mHal.drvState.lod[lod].stride;
995        }
996    } else {
997        ALOGE("Add code to readback from non-script memory");
998    }
999}
1000
1001
1002void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
1003                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
1004                         uint32_t lod,
1005                         uint32_t w, uint32_t h, uint32_t d, void *data, size_t sizeBytes, size_t stride) {
1006    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
1007    uint32_t lineSize = eSize * w;
1008    if (!stride) {
1009        stride = lineSize;
1010    }
1011
1012    if (alloc->mHal.drvState.lod[0].mallocPtr) {
1013        uint8_t *dst = static_cast<uint8_t *>(data);
1014        for (uint32_t z = zoff; z < (d + zoff); z++) {
1015            const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, z, lod,
1016                                              RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1017            if (dst == src) {
1018                // Skip the copy if we are the same allocation. This can arise from
1019                // our Bitmap optimization, where we share the same storage.
1020                return;
1021            }
1022
1023            for (uint32_t line=yoff; line < (yoff+h); line++) {
1024                memcpy(dst, src, lineSize);
1025                dst += stride;
1026                src += alloc->mHal.drvState.lod[lod].stride;
1027            }
1028        }
1029    }
1030}
1031
1032void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
1033                          const android::renderscript::Allocation *alloc) {
1034    return alloc->mHal.drvState.lod[0].mallocPtr;
1035}
1036
1037void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
1038                          const android::renderscript::Allocation *alloc) {
1039
1040}
1041
1042void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
1043                               const android::renderscript::Allocation *dstAlloc,
1044                               uint32_t dstXoff, uint32_t dstLod, size_t count,
1045                               const android::renderscript::Allocation *srcAlloc,
1046                               uint32_t srcXoff, uint32_t srcLod) {
1047}
1048
1049
1050void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
1051                                      const android::renderscript::Allocation *dstAlloc,
1052                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1053                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1054                                      const android::renderscript::Allocation *srcAlloc,
1055                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1056                                      RsAllocationCubemapFace srcFace) {
1057    size_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1058    for (uint32_t i = 0; i < h; i ++) {
1059        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, 0, dstLod, dstFace);
1060        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, 0, srcLod, srcFace);
1061        memcpy(dstPtr, srcPtr, w * elementSize);
1062
1063        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1064        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1065    }
1066}
1067
1068void rsdAllocationData3D_alloc_script(const android::renderscript::Context *rsc,
1069                                      const android::renderscript::Allocation *dstAlloc,
1070                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff, uint32_t dstLod,
1071                                      uint32_t w, uint32_t h, uint32_t d,
1072                                      const android::renderscript::Allocation *srcAlloc,
1073                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff, uint32_t srcLod) {
1074    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
1075    for (uint32_t j = 0; j < d; j++) {
1076        for (uint32_t i = 0; i < h; i ++) {
1077            uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstZoff + j,
1078                                           dstLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1079            uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcZoff + j,
1080                                           srcLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1081            memcpy(dstPtr, srcPtr, w * elementSize);
1082
1083            //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
1084            //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
1085        }
1086    }
1087}
1088
1089void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
1090                               const android::renderscript::Allocation *dstAlloc,
1091                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1092                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1093                               const android::renderscript::Allocation *srcAlloc,
1094                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1095                               RsAllocationCubemapFace srcFace) {
1096    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1097        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1098                                             "yet implemented.");
1099        return;
1100    }
1101    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
1102                                     dstLod, dstFace, w, h, srcAlloc,
1103                                     srcXoff, srcYoff, srcLod, srcFace);
1104}
1105
1106void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
1107                               const android::renderscript::Allocation *dstAlloc,
1108                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
1109                               uint32_t dstLod,
1110                               uint32_t w, uint32_t h, uint32_t d,
1111                               const android::renderscript::Allocation *srcAlloc,
1112                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
1113                               uint32_t srcLod) {
1114    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1115        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1116                                             "yet implemented.");
1117        return;
1118    }
1119    rsdAllocationData3D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstZoff,
1120                                     dstLod, w, h, d, srcAlloc,
1121                                     srcXoff, srcYoff, srcZoff, srcLod);
1122}
1123
1124void rsdAllocationElementData(const Context *rsc, const Allocation *alloc,
1125                              uint32_t x, uint32_t y, uint32_t z,
1126                              const void *data, uint32_t cIdx, size_t sizeBytes) {
1127    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1128
1129    uint8_t * ptr = GetOffsetPtr(alloc, x, y, z, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1130
1131    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1132    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1133
1134    if (alloc->mHal.state.hasReferences) {
1135        e->incRefs(data);
1136        e->decRefs(ptr);
1137    }
1138
1139    memcpy(ptr, data, sizeBytes);
1140    drv->uploadDeferred = true;
1141}
1142
1143void rsdAllocationElementRead(const Context *rsc, const Allocation *alloc,
1144                              uint32_t x, uint32_t y, uint32_t z,
1145                              void *data, uint32_t cIdx, size_t sizeBytes) {
1146    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1147
1148    uint8_t * ptr = GetOffsetPtr(alloc, x, y, z, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1149
1150    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1151    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1152
1153    memcpy(data, ptr, sizeBytes);
1154}
1155
1156static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1157    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1158    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1159
1160    for (uint32_t y=0; y < h; y++) {
1161        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1162        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2, lod, face);
1163        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2+1, lod, face);
1164
1165        for (uint32_t x=0; x < w; x++) {
1166            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
1167            oPtr ++;
1168            i1 += 2;
1169            i2 += 2;
1170        }
1171    }
1172}
1173
1174static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1175    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1176    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1177
1178    for (uint32_t y=0; y < h; y++) {
1179        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1180        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1181        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1182
1183        for (uint32_t x=0; x < w; x++) {
1184            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
1185            oPtr ++;
1186            i1 += 2;
1187            i2 += 2;
1188        }
1189    }
1190}
1191
1192static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1193    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1194    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1195
1196    for (uint32_t y=0; y < h; y++) {
1197        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1198        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1199        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1200
1201        for (uint32_t x=0; x < w; x++) {
1202            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
1203            oPtr ++;
1204            i1 += 2;
1205            i2 += 2;
1206        }
1207    }
1208}
1209
1210void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
1211    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
1212        return;
1213    }
1214    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
1215    for (uint32_t face = 0; face < numFaces; face ++) {
1216        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
1217            switch (alloc->getType()->getElement()->getSizeBits()) {
1218            case 32:
1219                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
1220                break;
1221            case 16:
1222                mip565(alloc, lod, (RsAllocationCubemapFace)face);
1223                break;
1224            case 8:
1225                mip8(alloc, lod, (RsAllocationCubemapFace)face);
1226                break;
1227            }
1228        }
1229    }
1230}
1231
1232uint32_t rsdAllocationGrallocBits(const android::renderscript::Context *rsc,
1233                                  android::renderscript::Allocation *alloc)
1234{
1235    return 0;
1236}
1237
1238void rsdAllocationUpdateCachedObject(const Context *rsc,
1239                                     const Allocation *alloc,
1240                                     rs_allocation *obj)
1241{
1242    obj->p = alloc;
1243#ifdef __LP64__
1244    if (alloc != nullptr) {
1245        obj->r = alloc->mHal.drvState.lod[0].mallocPtr;
1246        obj->v1 = alloc->mHal.drv;
1247        obj->v2 = (void *)alloc->mHal.drvState.lod[0].stride;
1248    } else {
1249        obj->r = nullptr;
1250        obj->v1 = nullptr;
1251        obj->v2 = nullptr;
1252    }
1253#endif
1254}
1255