rsdAllocation.cpp revision ddceab9a001f07a3395226c5e06e3b420720af0f
1cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber/*
2cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * Copyright (C) 2013 The Android Open Source Project
3cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber *
4cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * Licensed under the Apache License, Version 2.0 (the "License");
5cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * you may not use this file except in compliance with the License.
6cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * You may obtain a copy of the License at
7cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber *
8cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber *      http://www.apache.org/licenses/LICENSE-2.0
9cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber *
10cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * Unless required by applicable law or agreed to in writing, software
11cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * distributed under the License is distributed on an "AS IS" BASIS,
12cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * See the License for the specific language governing permissions and
14cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber * limitations under the License.
15cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber */
16cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
17cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "rsdCore.h"
18cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "rsdAllocation.h"
19cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
20cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "rsAllocation.h"
21cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
22cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#if !defined(RS_SERVER) && !defined(RS_COMPATIBILITY_LIB)
23cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "system/window.h"
24cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "ui/Rect.h"
25cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "ui/GraphicBufferMapper.h"
265bc087c573c70c84c6a39946457590b42d392a33Andreas Huber#else
275bc087c573c70c84c6a39946457590b42d392a33Andreas Huber#include "system/graphics.h"
28cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#endif
29cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
30cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#ifndef RS_COMPATIBILITY_LIB
31cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "rsdFrameBufferObj.h"
32cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "gui/GLConsumer.h"
33cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "gui/CpuConsumer.h"
34cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include "gui/Surface.h"
355bc087c573c70c84c6a39946457590b42d392a33Andreas Huber#include "hardware/gralloc.h"
365bc087c573c70c84c6a39946457590b42d392a33Andreas Huber
37cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include <GLES/gl.h>
38cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include <GLES2/gl2.h>
39cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#include <GLES/glext.h>
40cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#endif
41cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
42cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#ifdef RS_SERVER
43cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber// server requires malloc.h for memalign
4414f7672b5d450ed26a06fd3bb3ce045ea78b11b2Andreas Huber#include <malloc.h>
4514f7672b5d450ed26a06fd3bb3ce045ea78b11b2Andreas Huber#endif
46a48d372833ccec13c96ece9efcc226e8beac7f59Chong Zhang
47cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huberusing namespace android;
48cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huberusing namespace android::renderscript;
49a48d372833ccec13c96ece9efcc226e8beac7f59Chong Zhang
50a48d372833ccec13c96ece9efcc226e8beac7f59Chong Zhang
51a48d372833ccec13c96ece9efcc226e8beac7f59Chong Zhang#ifndef RS_COMPATIBILITY_LIB
52bfd4d0d9fe0033abf3f55b94f30f6a58846a875eAndreas Huberconst static GLenum gFaceOrder[] = {
53bfd4d0d9fe0033abf3f55b94f30f6a58846a875eAndreas Huber    GL_TEXTURE_CUBE_MAP_POSITIVE_X,
54bfd4d0d9fe0033abf3f55b94f30f6a58846a875eAndreas Huber    GL_TEXTURE_CUBE_MAP_NEGATIVE_X,
55bfd4d0d9fe0033abf3f55b94f30f6a58846a875eAndreas Huber    GL_TEXTURE_CUBE_MAP_POSITIVE_Y,
56f933441648ef6a71dee783d733aac17b9508b452Andreas Huber    GL_TEXTURE_CUBE_MAP_NEGATIVE_Y,
57f933441648ef6a71dee783d733aac17b9508b452Andreas Huber    GL_TEXTURE_CUBE_MAP_POSITIVE_Z,
58cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    GL_TEXTURE_CUBE_MAP_NEGATIVE_Z
5932f3cefa373cd55e63deda36ca9d07c7fe22eaafAndreas Huber};
6032f3cefa373cd55e63deda36ca9d07c7fe22eaafAndreas Huber
61632740c58119a132ce19f6d498e39c5c3773971aChong ZhangGLenum rsdTypeToGLType(RsDataType t) {
62632740c58119a132ce19f6d498e39c5c3773971aChong Zhang    switch (t) {
63632740c58119a132ce19f6d498e39c5c3773971aChong Zhang    case RS_TYPE_UNSIGNED_5_6_5:    return GL_UNSIGNED_SHORT_5_6_5;
6432f3cefa373cd55e63deda36ca9d07c7fe22eaafAndreas Huber    case RS_TYPE_UNSIGNED_5_5_5_1:  return GL_UNSIGNED_SHORT_5_5_5_1;
65cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    case RS_TYPE_UNSIGNED_4_4_4_4:  return GL_UNSIGNED_SHORT_4_4_4_4;
66cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
67f933441648ef6a71dee783d733aac17b9508b452Andreas Huber    //case RS_TYPE_FLOAT_16:      return GL_HALF_FLOAT;
680852843d304006e3ab333081fddda13b07193de8Robert Shih    case RS_TYPE_FLOAT_32:      return GL_FLOAT;
69f933441648ef6a71dee783d733aac17b9508b452Andreas Huber    case RS_TYPE_UNSIGNED_8:    return GL_UNSIGNED_BYTE;
70b50e83eca302a12f0fced6e7bab1b8617d63deaaRoger Jönsson    case RS_TYPE_UNSIGNED_16:   return GL_UNSIGNED_SHORT;
71b50e83eca302a12f0fced6e7bab1b8617d63deaaRoger Jönsson    case RS_TYPE_SIGNED_8:      return GL_BYTE;
727c8708046117e03c0d38006bdd9685139df3ac6bChong Zhang    case RS_TYPE_SIGNED_16:     return GL_SHORT;
737c8708046117e03c0d38006bdd9685139df3ac6bChong Zhang    default:    break;
74309aa8bf5e4cd66fe988adf2654cac3fadc2a1c3Robert Shih    }
75309aa8bf5e4cd66fe988adf2654cac3fadc2a1c3Robert Shih    return 0;
767c8708046117e03c0d38006bdd9685139df3ac6bChong Zhang}
777c8708046117e03c0d38006bdd9685139df3ac6bChong Zhang
78d47dfcb5a2e5901c96fc92662cec7aa30f7f8843Chong ZhangGLenum rsdKindToGLFormat(RsDataKind k) {
79d47dfcb5a2e5901c96fc92662cec7aa30f7f8843Chong Zhang    switch (k) {
8069634506fbfe79605c37f337a8d6748cda4445b1Robert Shih    case RS_KIND_PIXEL_L: return GL_LUMINANCE;
81cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    case RS_KIND_PIXEL_A: return GL_ALPHA;
82cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    case RS_KIND_PIXEL_LA: return GL_LUMINANCE_ALPHA;
83cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    case RS_KIND_PIXEL_RGB: return GL_RGB;
84cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    case RS_KIND_PIXEL_RGBA: return GL_RGBA;
850dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    case RS_KIND_PIXEL_DEPTH: return GL_DEPTH_COMPONENT16;
860dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    default: break;
870dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    }
880dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    return 0;
890dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih}
900dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih#endif
910dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih
920dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shihuint8_t *GetOffsetPtr(const android::renderscript::Allocation *alloc,
930dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih                      uint32_t xoff, uint32_t yoff, uint32_t zoff,
940dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih                      uint32_t lod, RsAllocationCubemapFace face) {
950dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    uint8_t *ptr = (uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
960dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    ptr += face * alloc->mHal.drvState.faceOffset;
970dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    ptr += zoff * alloc->mHal.drvState.lod[lod].dimY * alloc->mHal.drvState.lod[lod].stride;
980dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    ptr += yoff * alloc->mHal.drvState.lod[lod].stride;
990dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    ptr += xoff * alloc->mHal.state.elementSizeBytes;
1000dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih    return ptr;
1010dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih}
1020dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih
1030dd229bb306df68b88e7419b3cc11fc6175be1daRobert Shih
104cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huberstatic void Update2DTexture(const Context *rsc, const Allocation *alloc, const void *ptr,
105cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber                            uint32_t xoff, uint32_t yoff, uint32_t lod,
106cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber                            RsAllocationCubemapFace face, uint32_t w, uint32_t h) {
1076e3d311b6631b12aac2879d1b08c3534aece78b1Andreas Huber#ifndef RS_COMPATIBILITY_LIB
1083423bbdb05f843a64356fa24e8657f03ecbb7d65Robert Shih    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1097c8708046117e03c0d38006bdd9685139df3ac6bChong Zhang
110cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    rsAssert(drv->textureID);
111b50e83eca302a12f0fced6e7bab1b8617d63deaaRoger Jönsson    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
112cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
113cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    GLenum t = GL_TEXTURE_2D;
11469634506fbfe79605c37f337a8d6748cda4445b1Robert Shih    if (alloc->mHal.state.hasFaces) {
115309aa8bf5e4cd66fe988adf2654cac3fadc2a1c3Robert Shih        t = gFaceOrder[face];
116cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    }
1176e3d311b6631b12aac2879d1b08c3534aece78b1Andreas Huber    RSD_CALL_GL(glTexSubImage2D, t, lod, xoff, yoff, w, h, drv->glFormat, drv->glType, ptr);
1186e3d311b6631b12aac2879d1b08c3534aece78b1Andreas Huber#endif
119cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber}
120cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
121cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
122cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber#ifndef RS_COMPATIBILITY_LIB
123cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huberstatic void Upload2DTexture(const Context *rsc, const Allocation *alloc, bool isFirstUpload) {
124cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
125cda17c606b0fe3ccda4dc68a6d43882410ea2462Andreas Huber
126    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
127    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
128
129    uint32_t faceCount = 1;
130    if (alloc->mHal.state.hasFaces) {
131        faceCount = 6;
132    }
133
134    rsdGLCheckError(rsc, "Upload2DTexture 1 ");
135    for (uint32_t face = 0; face < faceCount; face ++) {
136        for (uint32_t lod = 0; lod < alloc->mHal.state.type->getLODCount(); lod++) {
137            const uint8_t *p = GetOffsetPtr(alloc, 0, 0, 0, lod, (RsAllocationCubemapFace)face);
138
139            GLenum t = GL_TEXTURE_2D;
140            if (alloc->mHal.state.hasFaces) {
141                t = gFaceOrder[face];
142            }
143
144            if (isFirstUpload) {
145                RSD_CALL_GL(glTexImage2D, t, lod, drv->glFormat,
146                             alloc->mHal.state.type->getLODDimX(lod),
147                             alloc->mHal.state.type->getLODDimY(lod),
148                             0, drv->glFormat, drv->glType, p);
149            } else {
150                RSD_CALL_GL(glTexSubImage2D, t, lod, 0, 0,
151                                alloc->mHal.state.type->getLODDimX(lod),
152                                alloc->mHal.state.type->getLODDimY(lod),
153                                drv->glFormat, drv->glType, p);
154            }
155        }
156    }
157
158    if (alloc->mHal.state.mipmapControl == RS_ALLOCATION_MIPMAP_ON_SYNC_TO_TEXTURE) {
159        RSD_CALL_GL(glGenerateMipmap, drv->glTarget);
160    }
161    rsdGLCheckError(rsc, "Upload2DTexture");
162}
163#endif
164
165static void UploadToTexture(const Context *rsc, const Allocation *alloc) {
166#ifndef RS_COMPATIBILITY_LIB
167    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
168
169    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
170        if (!drv->textureID) {
171            RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
172        }
173        return;
174    }
175
176    if (!drv->glType || !drv->glFormat) {
177        return;
178    }
179
180    if (!alloc->mHal.drvState.lod[0].mallocPtr) {
181        return;
182    }
183
184    bool isFirstUpload = false;
185
186    if (!drv->textureID) {
187        RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
188        isFirstUpload = true;
189    }
190
191    Upload2DTexture(rsc, alloc, isFirstUpload);
192
193    if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
194        if (alloc->mHal.drvState.lod[0].mallocPtr) {
195            free(alloc->mHal.drvState.lod[0].mallocPtr);
196            alloc->mHal.drvState.lod[0].mallocPtr = NULL;
197        }
198    }
199    rsdGLCheckError(rsc, "UploadToTexture");
200#endif
201}
202
203static void AllocateRenderTarget(const Context *rsc, const Allocation *alloc) {
204#ifndef RS_COMPATIBILITY_LIB
205    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
206
207    if (!drv->glFormat) {
208        return;
209    }
210
211    if (!drv->renderTargetID) {
212        RSD_CALL_GL(glGenRenderbuffers, 1, &drv->renderTargetID);
213
214        if (!drv->renderTargetID) {
215            // This should generally not happen
216            ALOGE("allocateRenderTarget failed to gen mRenderTargetID");
217            rsc->dumpDebug();
218            return;
219        }
220        RSD_CALL_GL(glBindRenderbuffer, GL_RENDERBUFFER, drv->renderTargetID);
221        RSD_CALL_GL(glRenderbufferStorage, GL_RENDERBUFFER, drv->glFormat,
222                    alloc->mHal.drvState.lod[0].dimX, alloc->mHal.drvState.lod[0].dimY);
223    }
224    rsdGLCheckError(rsc, "AllocateRenderTarget");
225#endif
226}
227
228static void UploadToBufferObject(const Context *rsc, const Allocation *alloc) {
229#ifndef RS_COMPATIBILITY_LIB
230    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
231
232    rsAssert(!alloc->mHal.state.type->getDimY());
233    rsAssert(!alloc->mHal.state.type->getDimZ());
234
235    //alloc->mHal.state.usageFlags |= RS_ALLOCATION_USAGE_GRAPHICS_VERTEX;
236
237    if (!drv->bufferID) {
238        RSD_CALL_GL(glGenBuffers, 1, &drv->bufferID);
239    }
240    if (!drv->bufferID) {
241        ALOGE("Upload to buffer object failed");
242        drv->uploadDeferred = true;
243        return;
244    }
245    RSD_CALL_GL(glBindBuffer, drv->glTarget, drv->bufferID);
246    RSD_CALL_GL(glBufferData, drv->glTarget, alloc->mHal.state.type->getSizeBytes(),
247                 alloc->mHal.drvState.lod[0].mallocPtr, GL_DYNAMIC_DRAW);
248    RSD_CALL_GL(glBindBuffer, drv->glTarget, 0);
249    rsdGLCheckError(rsc, "UploadToBufferObject");
250#endif
251}
252
253
254static size_t DeriveYUVLayout(int yuv, Allocation::Hal::DrvState *state) {
255    // YUV only supports basic 2d
256    // so we can stash the plane pointers in the mipmap levels.
257    size_t uvSize = 0;
258#ifndef RS_SERVER
259    switch(yuv) {
260    case HAL_PIXEL_FORMAT_YV12:
261        state->lod[2].dimX = state->lod[0].dimX / 2;
262        state->lod[2].dimY = state->lod[0].dimY / 2;
263        state->lod[2].stride = rsRound(state->lod[0].stride >> 1, 16);
264        state->lod[2].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
265                (state->lod[0].stride * state->lod[0].dimY);
266        uvSize += state->lod[2].stride * state->lod[2].dimY;
267
268        state->lod[1].dimX = state->lod[2].dimX;
269        state->lod[1].dimY = state->lod[2].dimY;
270        state->lod[1].stride = state->lod[2].stride;
271        state->lod[1].mallocPtr = ((uint8_t *)state->lod[2].mallocPtr) +
272                (state->lod[2].stride * state->lod[2].dimY);
273        uvSize += state->lod[1].stride * state->lod[2].dimY;
274
275        state->lodCount = 3;
276        break;
277    case HAL_PIXEL_FORMAT_YCrCb_420_SP:  // NV21
278        state->lod[1].dimX = state->lod[0].dimX;
279        state->lod[1].dimY = state->lod[0].dimY / 2;
280        state->lod[1].stride = state->lod[0].stride;
281        state->lod[1].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
282                (state->lod[0].stride * state->lod[0].dimY);
283        uvSize += state->lod[1].stride * state->lod[1].dimY;
284        state->lodCount = 2;
285        break;
286    default:
287        rsAssert(0);
288    }
289#endif
290    return uvSize;
291}
292
293
294static size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc,
295        const Type *type, uint8_t *ptr) {
296    alloc->mHal.drvState.lod[0].dimX = type->getDimX();
297    alloc->mHal.drvState.lod[0].dimY = type->getDimY();
298    alloc->mHal.drvState.lod[0].dimZ = type->getDimZ();
299    alloc->mHal.drvState.lod[0].mallocPtr = 0;
300    // Stride needs to be 16-byte aligned too!
301    size_t stride = alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes();
302    alloc->mHal.drvState.lod[0].stride = rsRound(stride, 16);
303    alloc->mHal.drvState.lodCount = type->getLODCount();
304    alloc->mHal.drvState.faceCount = type->getDimFaces();
305
306    size_t offsets[Allocation::MAX_LOD];
307    memset(offsets, 0, sizeof(offsets));
308
309    size_t o = alloc->mHal.drvState.lod[0].stride * rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) *
310            rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u);
311    if(alloc->mHal.drvState.lodCount > 1) {
312        uint32_t tx = alloc->mHal.drvState.lod[0].dimX;
313        uint32_t ty = alloc->mHal.drvState.lod[0].dimY;
314        uint32_t tz = alloc->mHal.drvState.lod[0].dimZ;
315        for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
316            alloc->mHal.drvState.lod[lod].dimX = tx;
317            alloc->mHal.drvState.lod[lod].dimY = ty;
318            alloc->mHal.drvState.lod[lod].dimZ = tz;
319            alloc->mHal.drvState.lod[lod].stride =
320                    rsRound(tx * type->getElementSizeBytes(), 16);
321            offsets[lod] = o;
322            o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u);
323            if (tx > 1) tx >>= 1;
324            if (ty > 1) ty >>= 1;
325            if (tz > 1) tz >>= 1;
326        }
327    } else if (alloc->mHal.state.yuv) {
328        o += DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
329
330        for (uint32_t ct = 1; ct < alloc->mHal.drvState.lodCount; ct++) {
331            offsets[ct] = (size_t)alloc->mHal.drvState.lod[ct].mallocPtr;
332        }
333    }
334
335    alloc->mHal.drvState.faceOffset = o;
336
337    alloc->mHal.drvState.lod[0].mallocPtr = ptr;
338    for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
339        alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod];
340    }
341
342    size_t allocSize = alloc->mHal.drvState.faceOffset;
343    if(alloc->mHal.drvState.faceCount) {
344        allocSize *= 6;
345    }
346
347    return allocSize;
348}
349
350static uint8_t* allocAlignedMemory(size_t allocSize, bool forceZero) {
351    // We align all allocations to a 16-byte boundary.
352    uint8_t* ptr = (uint8_t *)memalign(16, allocSize);
353    if (!ptr) {
354        return NULL;
355    }
356    if (forceZero) {
357        memset(ptr, 0, allocSize);
358    }
359    return ptr;
360}
361
362bool rsdAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) {
363    DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
364    if (!drv) {
365        return false;
366    }
367    alloc->mHal.drv = drv;
368
369    // Calculate the object size.
370    size_t allocSize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), NULL);
371
372    uint8_t * ptr = NULL;
373    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) {
374
375    } else if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
376        // Allocation is allocated when the surface is created
377        // in getSurface
378    } else if (alloc->mHal.state.userProvidedPtr != NULL) {
379        // user-provided allocation
380        // limitations: no faces, no LOD, USAGE_SCRIPT or SCRIPT+TEXTURE only
381        if (!(alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED) ||
382              alloc->mHal.state.usageFlags == (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED | RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE))) {
383            ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT | USAGE_SHARED or USAGE_SCRIPT | USAGE_SHARED | USAGE_GRAPHICS_TEXTURE");
384            return false;
385        }
386        if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) {
387            ALOGE("User-allocated buffers must not have multiple faces or LODs");
388            return false;
389        }
390
391        // rows must be 16-byte aligned
392        // validate that here, otherwise fall back to not use the user-backed allocation
393        if (((alloc->getType()->getDimX() * alloc->getType()->getElement()->getSizeBytes()) % 16) != 0) {
394            ALOGV("User-backed allocation failed stride requirement, falling back to separate allocation");
395            drv->useUserProvidedPtr = false;
396
397            ptr = allocAlignedMemory(allocSize, forceZero);
398            if (!ptr) {
399                alloc->mHal.drv = NULL;
400                free(drv);
401                return false;
402            }
403
404        } else {
405            drv->useUserProvidedPtr = true;
406            ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr;
407        }
408    } else {
409        ptr = allocAlignedMemory(allocSize, forceZero);
410        if (!ptr) {
411            alloc->mHal.drv = NULL;
412            free(drv);
413            return false;
414        }
415    }
416    // Build the pointer tables
417    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr);
418    if(allocSize != verifySize) {
419        rsAssert(!"Size mismatch");
420    }
421
422#ifndef RS_SERVER
423    drv->glTarget = GL_NONE;
424    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
425        if (alloc->mHal.state.hasFaces) {
426            drv->glTarget = GL_TEXTURE_CUBE_MAP;
427        } else {
428            drv->glTarget = GL_TEXTURE_2D;
429        }
430    } else {
431        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
432            drv->glTarget = GL_ARRAY_BUFFER;
433        }
434    }
435#endif
436
437#ifndef RS_COMPATIBILITY_LIB
438    drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType());
439    drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind());
440#else
441    drv->glType = 0;
442    drv->glFormat = 0;
443#endif
444
445    if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) {
446        drv->uploadDeferred = true;
447    }
448
449
450    drv->readBackFBO = NULL;
451
452    // fill out the initial state of the buffer if we couldn't use the user-provided ptr and USAGE_SHARED was accepted
453    if ((alloc->mHal.state.userProvidedPtr != 0) && (drv->useUserProvidedPtr == false)) {
454        rsdAllocationData2D(rsc, alloc, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X, alloc->getType()->getDimX(), alloc->getType()->getDimY(), alloc->mHal.state.userProvidedPtr, allocSize, 0);
455    }
456
457    return true;
458}
459
460void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) {
461    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
462
463#ifndef RS_COMPATIBILITY_LIB
464    if (drv->bufferID) {
465        // Causes a SW crash....
466        //ALOGV(" mBufferID %i", mBufferID);
467        //glDeleteBuffers(1, &mBufferID);
468        //mBufferID = 0;
469    }
470    if (drv->textureID) {
471        RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID);
472        drv->textureID = 0;
473    }
474    if (drv->renderTargetID) {
475        RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID);
476        drv->renderTargetID = 0;
477    }
478#endif
479
480    if (alloc->mHal.drvState.lod[0].mallocPtr) {
481        // don't free user-allocated ptrs or IO_OUTPUT buffers
482        if (!(drv->useUserProvidedPtr) &&
483            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) &&
484            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
485                free(alloc->mHal.drvState.lod[0].mallocPtr);
486        }
487        alloc->mHal.drvState.lod[0].mallocPtr = NULL;
488    }
489
490#ifndef RS_COMPATIBILITY_LIB
491    if (drv->readBackFBO != NULL) {
492        delete drv->readBackFBO;
493        drv->readBackFBO = NULL;
494    }
495
496    if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) &&
497        (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
498
499        DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
500        ANativeWindow *nw = drv->wndSurface;
501        if (nw) {
502            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
503            mapper.unlock(drv->wndBuffer->handle);
504            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
505        }
506    }
507#endif
508
509    free(drv);
510    alloc->mHal.drv = NULL;
511}
512
513void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
514                         const Type *newType, bool zeroNew) {
515    const uint32_t oldDimX = alloc->mHal.drvState.lod[0].dimX;
516    const uint32_t dimX = newType->getDimX();
517
518    // can't resize Allocations with user-allocated buffers
519    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
520        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
521        return;
522    }
523    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
524    // Calculate the object size
525    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL);
526    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
527    // Build the relative pointer tables.
528    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
529    if(s != verifySize) {
530        rsAssert(!"Size mismatch");
531    }
532
533
534    if (dimX > oldDimX) {
535        size_t stride = alloc->mHal.state.elementSizeBytes;
536        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
537                 0, stride * (dimX - oldDimX));
538    }
539}
540
541static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
542#ifndef RS_COMPATIBILITY_LIB
543    if (!alloc->getIsScript()) {
544        return; // nothing to sync
545    }
546
547    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
548    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
549
550    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
551    if (!drv->textureID && !drv->renderTargetID) {
552        return; // nothing was rendered here yet, so nothing to sync
553    }
554    if (drv->readBackFBO == NULL) {
555        drv->readBackFBO = new RsdFrameBufferObj();
556        drv->readBackFBO->setColorTarget(drv, 0);
557        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
558                                        alloc->getType()->getDimY());
559    }
560
561    // Bind the framebuffer object so we can read back from it
562    drv->readBackFBO->setActive(rsc);
563
564    // Do the readback
565    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
566                alloc->mHal.drvState.lod[0].dimY,
567                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
568
569    // Revert framebuffer to its original
570    lastFbo->setActive(rsc);
571#endif
572}
573
574
575void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
576                         RsAllocationUsageType src) {
577    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
578
579    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
580        if(!alloc->getIsRenderTarget()) {
581            rsc->setError(RS_ERROR_FATAL_DRIVER,
582                          "Attempting to sync allocation from render target, "
583                          "for non-render target allocation");
584        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
585            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
586                                                 "render target");
587        } else {
588            rsdAllocationSyncFromFBO(rsc, alloc);
589        }
590        return;
591    }
592
593    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT);
594
595    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
596        UploadToTexture(rsc, alloc);
597    } else {
598        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
599            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
600            AllocateRenderTarget(rsc, alloc);
601        }
602    }
603    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
604        UploadToBufferObject(rsc, alloc);
605    }
606
607    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
608        // NOP in CPU driver for now
609    }
610
611    drv->uploadDeferred = false;
612}
613
614void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
615    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
616    drv->uploadDeferred = true;
617}
618
619#ifndef RS_COMPATIBILITY_LIB
620static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
621    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
622
623    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
624    if (r) {
625        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
626        return false;
627    }
628
629    // Must lock the whole surface
630    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
631    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
632
633    void *dst = NULL;
634    mapper.lock(drv->wndBuffer->handle,
635            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
636            bounds, &dst);
637    alloc->mHal.drvState.lod[0].mallocPtr = dst;
638    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
639    rsAssert((alloc->mHal.drvState.lod[0].stride & 0xf) == 0);
640
641    return true;
642}
643#endif
644
645void rsdAllocationSetSurface(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
646#ifndef RS_COMPATIBILITY_LIB
647    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
648    ANativeWindow *old = drv->wndSurface;
649
650    if (nw) {
651        nw->incStrong(NULL);
652    }
653
654    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
655        //TODO finish support for render target + script
656        drv->wnd = nw;
657        return;
658    }
659
660    // Cleanup old surface if there is one.
661    if (drv->wndSurface) {
662        ANativeWindow *old = drv->wndSurface;
663        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
664        mapper.unlock(drv->wndBuffer->handle);
665        old->cancelBuffer(old, drv->wndBuffer, -1);
666        drv->wndSurface = NULL;
667        old->decStrong(NULL);
668    }
669
670    if (nw != NULL) {
671        int32_t r;
672        uint32_t flags = 0;
673
674        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
675            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
676        }
677        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
678            flags |= GRALLOC_USAGE_HW_RENDER;
679        }
680
681        r = native_window_set_usage(nw, flags);
682        if (r) {
683            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
684            goto error;
685        }
686
687        r = native_window_set_buffers_dimensions(nw, alloc->mHal.drvState.lod[0].dimX,
688                                                 alloc->mHal.drvState.lod[0].dimY);
689        if (r) {
690            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
691            goto error;
692        }
693
694        int format = 0;
695        const Element *e = alloc->mHal.state.type->getElement();
696        rsAssert(e->getType() == RS_TYPE_UNSIGNED_8);
697        rsAssert(e->getVectorSize() == 4);
698        rsAssert(e->getKind() == RS_KIND_PIXEL_RGBA);
699        format = PIXEL_FORMAT_RGBA_8888;
700
701        r = native_window_set_buffers_format(nw, format);
702        if (r) {
703            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer format.");
704            goto error;
705        }
706
707        IoGetBuffer(rsc, alloc, nw);
708        drv->wndSurface = nw;
709    }
710
711    return;
712
713 error:
714
715    if (nw) {
716        nw->decStrong(NULL);
717    }
718
719
720#endif
721}
722
723void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
724#ifndef RS_COMPATIBILITY_LIB
725    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
726    ANativeWindow *nw = drv->wndSurface;
727    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
728        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
729        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
730        return;
731    }
732    if (nw) {
733        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
734            GraphicBufferMapper &mapper = GraphicBufferMapper::get();
735            mapper.unlock(drv->wndBuffer->handle);
736            int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
737            if (r) {
738                rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
739                return;
740            }
741
742            IoGetBuffer(rsc, alloc, nw);
743        }
744    } else {
745        rsc->setError(RS_ERROR_DRIVER, "Sent IO buffer with no attached surface.");
746        return;
747    }
748#endif
749}
750
751void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
752#ifndef RS_COMPATIBILITY_LIB
753    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
754    if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
755        drv->surfaceTexture->updateTexImage();
756    }
757#endif
758}
759
760
761void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
762                         uint32_t xoff, uint32_t lod, size_t count,
763                         const void *data, size_t sizeBytes) {
764    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
765
766    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
767    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
768    size_t size = count * eSize;
769
770    if (ptr != data) {
771        // Skip the copy if we are the same allocation. This can arise from
772        // our Bitmap optimization, where we share the same storage.
773        if (alloc->mHal.state.hasReferences) {
774            alloc->incRefs(data, count);
775            alloc->decRefs(ptr, count);
776        }
777        memcpy(ptr, data, size);
778    }
779    drv->uploadDeferred = true;
780}
781
782void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
783                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
784                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
785    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
786
787    size_t eSize = alloc->mHal.state.elementSizeBytes;
788    size_t lineSize = eSize * w;
789    if (!stride) {
790        stride = lineSize;
791    }
792
793    if (alloc->mHal.drvState.lod[0].mallocPtr) {
794        const uint8_t *src = static_cast<const uint8_t *>(data);
795        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
796        if (dst == src) {
797            // Skip the copy if we are the same allocation. This can arise from
798            // our Bitmap optimization, where we share the same storage.
799            drv->uploadDeferred = true;
800            return;
801        }
802
803        for (uint32_t line=yoff; line < (yoff+h); line++) {
804            if (alloc->mHal.state.hasReferences) {
805                alloc->incRefs(src, w);
806                alloc->decRefs(dst, w);
807            }
808            memcpy(dst, src, lineSize);
809            src += stride;
810            dst += alloc->mHal.drvState.lod[lod].stride;
811        }
812        if (alloc->mHal.state.yuv) {
813            int lod = 1;
814            while (alloc->mHal.drvState.lod[lod].mallocPtr) {
815                size_t lineSize = alloc->mHal.drvState.lod[lod].dimX;
816                uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
817
818                for (uint32_t line=(yoff >> 1); line < ((yoff+h)>>1); line++) {
819                    memcpy(dst, src, lineSize);
820                    src += lineSize;
821                    dst += alloc->mHal.drvState.lod[lod].stride;
822                }
823                lod++;
824            }
825
826        }
827        drv->uploadDeferred = true;
828    } else {
829        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
830    }
831}
832
833void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
834                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
835                         uint32_t lod,
836                         uint32_t w, uint32_t h, uint32_t d, const void *data,
837                         size_t sizeBytes, size_t stride) {
838    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
839
840    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
841    uint32_t lineSize = eSize * w;
842    if (!stride) {
843        stride = lineSize;
844    }
845
846    if (alloc->mHal.drvState.lod[0].mallocPtr) {
847        const uint8_t *src = static_cast<const uint8_t *>(data);
848        for (uint32_t z = zoff; z < d; z++) {
849            uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, z, lod,
850                                        RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
851            if (dst == src) {
852                // Skip the copy if we are the same allocation. This can arise from
853                // our Bitmap optimization, where we share the same storage.
854                drv->uploadDeferred = true;
855                return;
856            }
857
858            for (uint32_t line=yoff; line < (yoff+h); line++) {
859                if (alloc->mHal.state.hasReferences) {
860                    alloc->incRefs(src, w);
861                    alloc->decRefs(dst, w);
862                }
863                memcpy(dst, src, lineSize);
864                src += stride;
865                dst += alloc->mHal.drvState.lod[lod].stride;
866            }
867        }
868        drv->uploadDeferred = true;
869    }
870}
871
872void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
873                         uint32_t xoff, uint32_t lod, size_t count,
874                         void *data, size_t sizeBytes) {
875    const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
876    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
877    if (data != ptr) {
878        // Skip the copy if we are the same allocation. This can arise from
879        // our Bitmap optimization, where we share the same storage.
880        memcpy(data, ptr, count * eSize);
881    }
882}
883
884void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
885                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
886                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
887    size_t eSize = alloc->mHal.state.elementSizeBytes;
888    size_t lineSize = eSize * w;
889    if (!stride) {
890        stride = lineSize;
891    }
892
893    if (alloc->mHal.drvState.lod[0].mallocPtr) {
894        uint8_t *dst = static_cast<uint8_t *>(data);
895        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
896        if (dst == src) {
897            // Skip the copy if we are the same allocation. This can arise from
898            // our Bitmap optimization, where we share the same storage.
899            return;
900        }
901
902        for (uint32_t line=yoff; line < (yoff+h); line++) {
903            memcpy(dst, src, lineSize);
904            dst += stride;
905            src += alloc->mHal.drvState.lod[lod].stride;
906        }
907    } else {
908        ALOGE("Add code to readback from non-script memory");
909    }
910}
911
912
913void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
914                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
915                         uint32_t lod,
916                         uint32_t w, uint32_t h, uint32_t d, void *data, size_t sizeBytes, size_t stride) {
917    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
918    uint32_t lineSize = eSize * w;
919    if (!stride) {
920        stride = lineSize;
921    }
922
923    if (alloc->mHal.drvState.lod[0].mallocPtr) {
924        uint8_t *dst = static_cast<uint8_t *>(data);
925        for (uint32_t z = zoff; z < d; z++) {
926            const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, z, lod,
927                                              RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
928            if (dst == src) {
929                // Skip the copy if we are the same allocation. This can arise from
930                // our Bitmap optimization, where we share the same storage.
931                return;
932            }
933
934            for (uint32_t line=yoff; line < (yoff+h); line++) {
935                memcpy(dst, src, lineSize);
936                dst += stride;
937                src += alloc->mHal.drvState.lod[lod].stride;
938            }
939        }
940    }
941}
942
943void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
944                          const android::renderscript::Allocation *alloc) {
945    return alloc->mHal.drvState.lod[0].mallocPtr;
946}
947
948void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
949                          const android::renderscript::Allocation *alloc) {
950
951}
952
953void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
954                               const android::renderscript::Allocation *dstAlloc,
955                               uint32_t dstXoff, uint32_t dstLod, size_t count,
956                               const android::renderscript::Allocation *srcAlloc,
957                               uint32_t srcXoff, uint32_t srcLod) {
958}
959
960
961void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
962                                      const android::renderscript::Allocation *dstAlloc,
963                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
964                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
965                                      const android::renderscript::Allocation *srcAlloc,
966                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
967                                      RsAllocationCubemapFace srcFace) {
968    size_t elementSize = dstAlloc->getType()->getElementSizeBytes();
969    for (uint32_t i = 0; i < h; i ++) {
970        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, 0, dstLod, dstFace);
971        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, 0, srcLod, srcFace);
972        memcpy(dstPtr, srcPtr, w * elementSize);
973
974        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
975        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
976    }
977}
978
979void rsdAllocationData3D_alloc_script(const android::renderscript::Context *rsc,
980                                      const android::renderscript::Allocation *dstAlloc,
981                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff, uint32_t dstLod,
982                                      uint32_t w, uint32_t h, uint32_t d,
983                                      const android::renderscript::Allocation *srcAlloc,
984                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff, uint32_t srcLod) {
985    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
986    for (uint32_t j = 0; j < d; j++) {
987        for (uint32_t i = 0; i < h; i ++) {
988            uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstZoff + j,
989                                           dstLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
990            uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcZoff + j,
991                                           srcLod, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
992            memcpy(dstPtr, srcPtr, w * elementSize);
993
994            //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
995            //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
996        }
997    }
998}
999
1000void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
1001                               const android::renderscript::Allocation *dstAlloc,
1002                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
1003                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
1004                               const android::renderscript::Allocation *srcAlloc,
1005                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
1006                               RsAllocationCubemapFace srcFace) {
1007    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1008        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1009                                             "yet implemented.");
1010        return;
1011    }
1012    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
1013                                     dstLod, dstFace, w, h, srcAlloc,
1014                                     srcXoff, srcYoff, srcLod, srcFace);
1015}
1016
1017void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
1018                               const android::renderscript::Allocation *dstAlloc,
1019                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
1020                               uint32_t dstLod,
1021                               uint32_t w, uint32_t h, uint32_t d,
1022                               const android::renderscript::Allocation *srcAlloc,
1023                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
1024                               uint32_t srcLod) {
1025    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
1026        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
1027                                             "yet implemented.");
1028        return;
1029    }
1030    rsdAllocationData3D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstZoff,
1031                                     dstLod, w, h, d, srcAlloc,
1032                                     srcXoff, srcYoff, srcZoff, srcLod);
1033}
1034
1035void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc,
1036                                uint32_t x,
1037                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1038    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1039
1040    size_t eSize = alloc->mHal.state.elementSizeBytes;
1041    uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1042
1043    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1044    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1045
1046    if (alloc->mHal.state.hasReferences) {
1047        e->incRefs(data);
1048        e->decRefs(ptr);
1049    }
1050
1051    memcpy(ptr, data, sizeBytes);
1052    drv->uploadDeferred = true;
1053}
1054
1055void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc,
1056                                uint32_t x, uint32_t y,
1057                                const void *data, uint32_t cIdx, size_t sizeBytes) {
1058    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
1059
1060    size_t eSize = alloc->mHal.state.elementSizeBytes;
1061    uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
1062
1063    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
1064    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
1065
1066    if (alloc->mHal.state.hasReferences) {
1067        e->incRefs(data);
1068        e->decRefs(ptr);
1069    }
1070
1071    memcpy(ptr, data, sizeBytes);
1072    drv->uploadDeferred = true;
1073}
1074
1075static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1076    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1077    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1078
1079    for (uint32_t y=0; y < h; y++) {
1080        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1081        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2, lod, face);
1082        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, 0, y*2+1, lod, face);
1083
1084        for (uint32_t x=0; x < w; x++) {
1085            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
1086            oPtr ++;
1087            i1 += 2;
1088            i2 += 2;
1089        }
1090    }
1091}
1092
1093static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1094    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1095    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1096
1097    for (uint32_t y=0; y < h; y++) {
1098        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1099        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1100        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1101
1102        for (uint32_t x=0; x < w; x++) {
1103            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
1104            oPtr ++;
1105            i1 += 2;
1106            i2 += 2;
1107        }
1108    }
1109}
1110
1111static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
1112    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
1113    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
1114
1115    for (uint32_t y=0; y < h; y++) {
1116        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
1117        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, 0, lod, face);
1118        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, 0, lod, face);
1119
1120        for (uint32_t x=0; x < w; x++) {
1121            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
1122            oPtr ++;
1123            i1 += 2;
1124            i2 += 2;
1125        }
1126    }
1127}
1128
1129void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
1130    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
1131        return;
1132    }
1133    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
1134    for (uint32_t face = 0; face < numFaces; face ++) {
1135        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
1136            switch (alloc->getType()->getElement()->getSizeBits()) {
1137            case 32:
1138                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
1139                break;
1140            case 16:
1141                mip565(alloc, lod, (RsAllocationCubemapFace)face);
1142                break;
1143            case 8:
1144                mip8(alloc, lod, (RsAllocationCubemapFace)face);
1145                break;
1146            }
1147        }
1148    }
1149}
1150
1151uint32_t rsdAllocationGrallocBits(const android::renderscript::Context *rsc,
1152                                  android::renderscript::Allocation *alloc)
1153{
1154    return 0;
1155}
1156
1157