rsdAllocation.cpp revision 9e2bda58db9e6427a1640ea302d4df079ffa0252 
1/*
2 * Copyright (C) 2011-2012 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
18#include "rsdCore.h"
19#include "rsdAllocation.h"
20#include "rsdFrameBufferObj.h"
21
22#include "rsAllocation.h"
23
24#include "system/window.h"
25#include "hardware/gralloc.h"
26#include "ui/Rect.h"
27#include "ui/GraphicBufferMapper.h"
28#include "gui/SurfaceTexture.h"
29
30#include <GLES/gl.h>
31#include <GLES2/gl2.h>
32#include <GLES/glext.h>
33
34using namespace android;
35using namespace android::renderscript;
36
37
38
39const static GLenum gFaceOrder[] = {
40    GL_TEXTURE_CUBE_MAP_POSITIVE_X,
41    GL_TEXTURE_CUBE_MAP_NEGATIVE_X,
42    GL_TEXTURE_CUBE_MAP_POSITIVE_Y,
43    GL_TEXTURE_CUBE_MAP_NEGATIVE_Y,
44    GL_TEXTURE_CUBE_MAP_POSITIVE_Z,
45    GL_TEXTURE_CUBE_MAP_NEGATIVE_Z
46};
47
48
49GLenum rsdTypeToGLType(RsDataType t) {
50    switch (t) {
51    case RS_TYPE_UNSIGNED_5_6_5:    return GL_UNSIGNED_SHORT_5_6_5;
52    case RS_TYPE_UNSIGNED_5_5_5_1:  return GL_UNSIGNED_SHORT_5_5_5_1;
53    case RS_TYPE_UNSIGNED_4_4_4_4:  return GL_UNSIGNED_SHORT_4_4_4_4;
54
55    //case RS_TYPE_FLOAT_16:      return GL_HALF_FLOAT;
56    case RS_TYPE_FLOAT_32:      return GL_FLOAT;
57    case RS_TYPE_UNSIGNED_8:    return GL_UNSIGNED_BYTE;
58    case RS_TYPE_UNSIGNED_16:   return GL_UNSIGNED_SHORT;
59    case RS_TYPE_SIGNED_8:      return GL_BYTE;
60    case RS_TYPE_SIGNED_16:     return GL_SHORT;
61    default:    break;
62    }
63    return 0;
64}
65
66GLenum rsdKindToGLFormat(RsDataKind k) {
67    switch (k) {
68    case RS_KIND_PIXEL_L: return GL_LUMINANCE;
69    case RS_KIND_PIXEL_A: return GL_ALPHA;
70    case RS_KIND_PIXEL_LA: return GL_LUMINANCE_ALPHA;
71    case RS_KIND_PIXEL_RGB: return GL_RGB;
72    case RS_KIND_PIXEL_RGBA: return GL_RGBA;
73    case RS_KIND_PIXEL_DEPTH: return GL_DEPTH_COMPONENT16;
74    default: break;
75    }
76    return 0;
77}
78
79uint8_t *GetOffsetPtr(const android::renderscript::Allocation *alloc,
80                      uint32_t xoff, uint32_t yoff, uint32_t lod,
81                      RsAllocationCubemapFace face) {
82    uint8_t *ptr = (uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
83    ptr += face * alloc->mHal.drvState.faceOffset;
84    ptr += yoff * alloc->mHal.drvState.lod[lod].stride;
85    ptr += xoff * alloc->mHal.state.elementSizeBytes;
86    return ptr;
87}
88
89
90static void Update2DTexture(const Context *rsc, const Allocation *alloc, const void *ptr,
91                            uint32_t xoff, uint32_t yoff, uint32_t lod,
92                            RsAllocationCubemapFace face, uint32_t w, uint32_t h) {
93    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
94
95    rsAssert(drv->textureID);
96    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
97    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
98    GLenum t = GL_TEXTURE_2D;
99    if (alloc->mHal.state.hasFaces) {
100        t = gFaceOrder[face];
101    }
102    RSD_CALL_GL(glTexSubImage2D, t, lod, xoff, yoff, w, h, drv->glFormat, drv->glType, ptr);
103}
104
105
106static void Upload2DTexture(const Context *rsc, const Allocation *alloc, bool isFirstUpload) {
107    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
108
109    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
110    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
111
112    uint32_t faceCount = 1;
113    if (alloc->mHal.state.hasFaces) {
114        faceCount = 6;
115    }
116
117    rsdGLCheckError(rsc, "Upload2DTexture 1 ");
118    for (uint32_t face = 0; face < faceCount; face ++) {
119        for (uint32_t lod = 0; lod < alloc->mHal.state.type->getLODCount(); lod++) {
120            const uint8_t *p = GetOffsetPtr(alloc, 0, 0, lod, (RsAllocationCubemapFace)face);
121
122            GLenum t = GL_TEXTURE_2D;
123            if (alloc->mHal.state.hasFaces) {
124                t = gFaceOrder[face];
125            }
126
127            if (isFirstUpload) {
128                RSD_CALL_GL(glTexImage2D, t, lod, drv->glFormat,
129                             alloc->mHal.state.type->getLODDimX(lod),
130                             alloc->mHal.state.type->getLODDimY(lod),
131                             0, drv->glFormat, drv->glType, p);
132            } else {
133                RSD_CALL_GL(glTexSubImage2D, t, lod, 0, 0,
134                                alloc->mHal.state.type->getLODDimX(lod),
135                                alloc->mHal.state.type->getLODDimY(lod),
136                                drv->glFormat, drv->glType, p);
137            }
138        }
139    }
140
141    if (alloc->mHal.state.mipmapControl == RS_ALLOCATION_MIPMAP_ON_SYNC_TO_TEXTURE) {
142        RSD_CALL_GL(glGenerateMipmap, drv->glTarget);
143    }
144    rsdGLCheckError(rsc, "Upload2DTexture");
145}
146
147static void UploadToTexture(const Context *rsc, const Allocation *alloc) {
148    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
149
150    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
151        if (!drv->textureID) {
152            RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
153        }
154        return;
155    }
156
157    if (!drv->glType || !drv->glFormat) {
158        return;
159    }
160
161    if (!alloc->mHal.drvState.lod[0].mallocPtr) {
162        return;
163    }
164
165    bool isFirstUpload = false;
166
167    if (!drv->textureID) {
168        RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
169        isFirstUpload = true;
170    }
171
172    Upload2DTexture(rsc, alloc, isFirstUpload);
173
174    if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
175        if (alloc->mHal.drvState.lod[0].mallocPtr) {
176            free(alloc->mHal.drvState.lod[0].mallocPtr);
177            alloc->mHal.drvState.lod[0].mallocPtr = NULL;
178        }
179    }
180    rsdGLCheckError(rsc, "UploadToTexture");
181}
182
183static void AllocateRenderTarget(const Context *rsc, const Allocation *alloc) {
184    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
185
186    if (!drv->glFormat) {
187        return;
188    }
189
190    if (!drv->renderTargetID) {
191        RSD_CALL_GL(glGenRenderbuffers, 1, &drv->renderTargetID);
192
193        if (!drv->renderTargetID) {
194            // This should generally not happen
195            ALOGE("allocateRenderTarget failed to gen mRenderTargetID");
196            rsc->dumpDebug();
197            return;
198        }
199        RSD_CALL_GL(glBindRenderbuffer, GL_RENDERBUFFER, drv->renderTargetID);
200        RSD_CALL_GL(glRenderbufferStorage, GL_RENDERBUFFER, drv->glFormat,
201                              alloc->mHal.state.dimensionX, alloc->mHal.state.dimensionY);
202    }
203    rsdGLCheckError(rsc, "AllocateRenderTarget");
204}
205
206static void UploadToBufferObject(const Context *rsc, const Allocation *alloc) {
207    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
208
209    rsAssert(!alloc->mHal.state.type->getDimY());
210    rsAssert(!alloc->mHal.state.type->getDimZ());
211
212    //alloc->mHal.state.usageFlags |= RS_ALLOCATION_USAGE_GRAPHICS_VERTEX;
213
214    if (!drv->bufferID) {
215        RSD_CALL_GL(glGenBuffers, 1, &drv->bufferID);
216    }
217    if (!drv->bufferID) {
218        ALOGE("Upload to buffer object failed");
219        drv->uploadDeferred = true;
220        return;
221    }
222    RSD_CALL_GL(glBindBuffer, drv->glTarget, drv->bufferID);
223    RSD_CALL_GL(glBufferData, drv->glTarget, alloc->mHal.state.type->getSizeBytes(),
224                 alloc->mHal.drvState.lod[0].mallocPtr, GL_DYNAMIC_DRAW);
225    RSD_CALL_GL(glBindBuffer, drv->glTarget, 0);
226    rsdGLCheckError(rsc, "UploadToBufferObject");
227}
228
229static size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc,
230        const Type *type, uint8_t *ptr) {
231    alloc->mHal.drvState.lod[0].dimX = type->getDimX();
232    alloc->mHal.drvState.lod[0].dimY = type->getDimY();
233    alloc->mHal.drvState.lod[0].mallocPtr = 0;
234    alloc->mHal.drvState.lod[0].stride = alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes();
235    alloc->mHal.drvState.lodCount = type->getLODCount();
236    alloc->mHal.drvState.faceCount = type->getDimFaces();
237
238    size_t offsets[Allocation::MAX_LOD];
239    memset(offsets, 0, sizeof(offsets));
240
241    size_t o = alloc->mHal.drvState.lod[0].stride * rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) *
242            rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u);
243    if(alloc->mHal.drvState.lodCount > 1) {
244        uint32_t tx = alloc->mHal.drvState.lod[0].dimX;
245        uint32_t ty = alloc->mHal.drvState.lod[0].dimY;
246        uint32_t tz = alloc->mHal.drvState.lod[0].dimZ;
247        for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
248            alloc->mHal.drvState.lod[lod].dimX = tx;
249            alloc->mHal.drvState.lod[lod].dimY = ty;
250            alloc->mHal.drvState.lod[lod].dimZ = tz;
251            alloc->mHal.drvState.lod[lod].stride = tx * type->getElementSizeBytes();
252            offsets[lod] = o;
253            o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u);
254            if (tx > 1) tx >>= 1;
255            if (ty > 1) ty >>= 1;
256            if (tz > 1) tz >>= 1;
257        }
258    }
259    alloc->mHal.drvState.faceOffset = o;
260
261    alloc->mHal.drvState.lod[0].mallocPtr = ptr;
262    for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
263        alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod];
264    }
265
266    size_t allocSize = alloc->mHal.drvState.faceOffset;
267    if(alloc->mHal.drvState.faceCount) {
268        allocSize *= 6;
269    }
270
271    return allocSize;
272}
273
274bool rsdAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) {
275    DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
276    if (!drv) {
277        return false;
278    }
279    alloc->mHal.drv = drv;
280
281    // Calculate the object size.
282    size_t allocSize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), NULL);
283
284    uint8_t * ptr = NULL;
285    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) {
286    } else if (alloc->mHal.state.userProvidedPtr != NULL) {
287        // user-provided allocation
288        // limitations: no faces, no LOD, USAGE_SCRIPT only
289        if (alloc->mHal.state.usageFlags != (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED)) {
290            ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT and USAGE_SHARED");
291            return false;
292        }
293        if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) {
294            ALOGE("User-allocated buffers must not have multiple faces or LODs");
295            return false;
296        }
297        ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr;
298    } else {
299        if (forceZero) {
300            ptr = (uint8_t *)calloc(1, allocSize);
301        } else {
302            ptr = (uint8_t *)malloc(allocSize);
303        }
304        if (!ptr) {
305            free(drv);
306            return false;
307        }
308    }
309    // Build the pointer tables
310    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr);
311    if(allocSize != verifySize) {
312        rsAssert(!"Size mismatch");
313    }
314
315    drv->glTarget = GL_NONE;
316    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
317        if (alloc->mHal.state.hasFaces) {
318            drv->glTarget = GL_TEXTURE_CUBE_MAP;
319        } else {
320            drv->glTarget = GL_TEXTURE_2D;
321        }
322    } else {
323        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
324            drv->glTarget = GL_ARRAY_BUFFER;
325        }
326    }
327
328    drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType());
329    drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind());
330
331    if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) {
332        drv->uploadDeferred = true;
333    }
334
335
336    drv->readBackFBO = NULL;
337
338    return true;
339}
340
341void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) {
342    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
343
344    if (drv->bufferID) {
345        // Causes a SW crash....
346        //ALOGV(" mBufferID %i", mBufferID);
347        //glDeleteBuffers(1, &mBufferID);
348        //mBufferID = 0;
349    }
350    if (drv->textureID) {
351        RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID);
352        drv->textureID = 0;
353    }
354    if (drv->renderTargetID) {
355        RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID);
356        drv->renderTargetID = 0;
357    }
358
359    if (alloc->mHal.drvState.lod[0].mallocPtr) {
360        // don't free user-allocated ptrs
361        if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED)) {
362            free(alloc->mHal.drvState.lod[0].mallocPtr);
363        }
364        alloc->mHal.drvState.lod[0].mallocPtr = NULL;
365    }
366    if (drv->readBackFBO != NULL) {
367        delete drv->readBackFBO;
368        drv->readBackFBO = NULL;
369    }
370    free(drv);
371    alloc->mHal.drv = NULL;
372}
373
374void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
375                         const Type *newType, bool zeroNew) {
376    // can't resize Allocations with user-allocated buffers
377    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
378        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
379        return;
380    }
381    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
382    // Calculate the object size
383    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL);
384    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
385    // Build the relative pointer tables.
386    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
387    if(s != verifySize) {
388        rsAssert(!"Size mismatch");
389    }
390
391    const uint32_t oldDimX = alloc->mHal.state.dimensionX;
392    const uint32_t dimX = newType->getDimX();
393
394    if (dimX > oldDimX) {
395        uint32_t stride = alloc->mHal.state.elementSizeBytes;
396        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
397                 0, stride * (dimX - oldDimX));
398    }
399}
400
401static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
402    if (!alloc->getIsScript()) {
403        return; // nothing to sync
404    }
405
406    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
407    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
408
409    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
410    if (!drv->textureID && !drv->renderTargetID) {
411        return; // nothing was rendered here yet, so nothing to sync
412    }
413    if (drv->readBackFBO == NULL) {
414        drv->readBackFBO = new RsdFrameBufferObj();
415        drv->readBackFBO->setColorTarget(drv, 0);
416        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
417                                        alloc->getType()->getDimY());
418    }
419
420    // Bind the framebuffer object so we can read back from it
421    drv->readBackFBO->setActive(rsc);
422
423    // Do the readback
424    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
425                alloc->mHal.drvState.lod[0].dimY,
426                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
427
428    // Revert framebuffer to its original
429    lastFbo->setActive(rsc);
430}
431
432
433void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
434                         RsAllocationUsageType src) {
435    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
436
437    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
438        if(!alloc->getIsRenderTarget()) {
439            rsc->setError(RS_ERROR_FATAL_DRIVER,
440                          "Attempting to sync allocation from render target, "
441                          "for non-render target allocation");
442        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
443            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
444                                                 "render target");
445        } else {
446            rsdAllocationSyncFromFBO(rsc, alloc);
447        }
448        return;
449    }
450
451    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT);
452
453    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
454        UploadToTexture(rsc, alloc);
455    } else {
456        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
457            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
458            AllocateRenderTarget(rsc, alloc);
459        }
460    }
461    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
462        UploadToBufferObject(rsc, alloc);
463    }
464
465    drv->uploadDeferred = false;
466}
467
468void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
469    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
470    drv->uploadDeferred = true;
471}
472
473int32_t rsdAllocationInitSurfaceTexture(const Context *rsc, const Allocation *alloc) {
474    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
475    UploadToTexture(rsc, alloc);
476    return drv->textureID;
477}
478
479static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
480    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
481
482    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
483    if (r) {
484        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
485        return false;
486    }
487
488    // Must lock the whole surface
489    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
490    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
491
492    void *dst = NULL;
493    mapper.lock(drv->wndBuffer->handle,
494            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
495            bounds, &dst);
496    alloc->mHal.drvState.lod[0].mallocPtr = dst;
497    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
498
499    return true;
500}
501
502void rsdAllocationSetSurfaceTexture(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
503    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
504
505    //ALOGE("rsdAllocationSetSurfaceTexture %p  %p", alloc, nw);
506
507    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
508        //TODO finish support for render target + script
509        drv->wnd = nw;
510        return;
511    }
512
513
514    // Cleanup old surface if there is one.
515    if (alloc->mHal.state.wndSurface) {
516        ANativeWindow *old = alloc->mHal.state.wndSurface;
517        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
518        mapper.unlock(drv->wndBuffer->handle);
519        old->queueBuffer(old, drv->wndBuffer, -1);
520    }
521
522    if (nw != NULL) {
523        int32_t r;
524        uint32_t flags = 0;
525        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
526            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
527        }
528        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
529            flags |= GRALLOC_USAGE_HW_RENDER;
530        }
531
532        r = native_window_set_usage(nw, flags);
533        if (r) {
534            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
535            return;
536        }
537
538        r = native_window_set_buffers_dimensions(nw, alloc->mHal.state.dimensionX,
539                                                 alloc->mHal.state.dimensionY);
540        if (r) {
541            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
542            return;
543        }
544
545        r = native_window_set_buffer_count(nw, 3);
546        if (r) {
547            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer count.");
548            return;
549        }
550
551        IoGetBuffer(rsc, alloc, nw);
552    }
553}
554
555void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
556    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
557    ANativeWindow *nw = alloc->mHal.state.wndSurface;
558
559    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
560        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
561        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
562        return;
563    }
564
565    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
566        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
567        mapper.unlock(drv->wndBuffer->handle);
568        int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
569        if (r) {
570            rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
571            return;
572        }
573
574        IoGetBuffer(rsc, alloc, nw);
575    }
576}
577
578void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
579    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
580    alloc->mHal.state.surfaceTexture->updateTexImage();
581}
582
583
584void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
585                         uint32_t xoff, uint32_t lod, uint32_t count,
586                         const void *data, size_t sizeBytes) {
587    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
588
589    const uint32_t eSize = alloc->mHal.state.type->getElementSizeBytes();
590    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
591    uint32_t size = count * eSize;
592
593    if (ptr != data) {
594        // Skip the copy if we are the same allocation. This can arise from
595        // our Bitmap optimization, where we share the same storage.
596        if (alloc->mHal.state.hasReferences) {
597            alloc->incRefs(data, count);
598            alloc->decRefs(ptr, count);
599        }
600        memcpy(ptr, data, size);
601    }
602    drv->uploadDeferred = true;
603}
604
605void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
606                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
607                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
608    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
609
610    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
611    uint32_t lineSize = eSize * w;
612    if (!stride) {
613        stride = lineSize;
614    }
615
616    if (alloc->mHal.drvState.lod[0].mallocPtr) {
617        const uint8_t *src = static_cast<const uint8_t *>(data);
618        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, lod, face);
619        if (dst == src) {
620            // Skip the copy if we are the same allocation. This can arise from
621            // our Bitmap optimization, where we share the same storage.
622            drv->uploadDeferred = true;
623            return;
624        }
625
626        for (uint32_t line=yoff; line < (yoff+h); line++) {
627            if (alloc->mHal.state.hasReferences) {
628                alloc->incRefs(src, w);
629                alloc->decRefs(dst, w);
630            }
631            memcpy(dst, src, lineSize);
632            src += stride;
633            dst += alloc->mHal.drvState.lod[lod].stride;
634        }
635        drv->uploadDeferred = true;
636    } else {
637        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
638    }
639}
640
641void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
642                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
643                         uint32_t lod, RsAllocationCubemapFace face,
644                         uint32_t w, uint32_t h, uint32_t d, const void *data, uint32_t sizeBytes) {
645
646}
647
648void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
649                         uint32_t xoff, uint32_t lod, uint32_t count,
650                         void *data, size_t sizeBytes) {
651    const uint32_t eSize = alloc->mHal.state.type->getElementSizeBytes();
652    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
653    if (data != ptr) {
654        // Skip the copy if we are the same allocation. This can arise from
655        // our Bitmap optimization, where we share the same storage.
656        memcpy(data, ptr, count * eSize);
657    }
658}
659
660void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
661                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
662                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
663    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
664    uint32_t lineSize = eSize * w;
665    if (!stride) {
666        stride = lineSize;
667    }
668
669    if (alloc->mHal.drvState.lod[0].mallocPtr) {
670        uint8_t *dst = static_cast<uint8_t *>(data);
671        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, lod, face);
672        if (dst == src) {
673            // Skip the copy if we are the same allocation. This can arise from
674            // our Bitmap optimization, where we share the same storage.
675            return;
676        }
677
678        for (uint32_t line=yoff; line < (yoff+h); line++) {
679            memcpy(dst, src, lineSize);
680            dst += stride;
681            src += alloc->mHal.drvState.lod[lod].stride;
682        }
683    } else {
684        ALOGE("Add code to readback from non-script memory");
685    }
686}
687
688
689void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
690                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
691                         uint32_t lod, RsAllocationCubemapFace face,
692                         uint32_t w, uint32_t h, uint32_t d, void *data, uint32_t sizeBytes) {
693
694}
695
696void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
697                          const android::renderscript::Allocation *alloc) {
698    return alloc->mHal.drvState.lod[0].mallocPtr;
699}
700
701void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
702                          const android::renderscript::Allocation *alloc) {
703
704}
705
706void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
707                               const android::renderscript::Allocation *dstAlloc,
708                               uint32_t dstXoff, uint32_t dstLod, uint32_t count,
709                               const android::renderscript::Allocation *srcAlloc,
710                               uint32_t srcXoff, uint32_t srcLod) {
711}
712
713
714void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
715                                      const android::renderscript::Allocation *dstAlloc,
716                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
717                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
718                                      const android::renderscript::Allocation *srcAlloc,
719                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
720                                      RsAllocationCubemapFace srcFace) {
721    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
722    for (uint32_t i = 0; i < h; i ++) {
723        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstLod, dstFace);
724        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcLod, srcFace);
725        memcpy(dstPtr, srcPtr, w * elementSize);
726
727        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
728        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
729    }
730}
731
732void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
733                               const android::renderscript::Allocation *dstAlloc,
734                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
735                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
736                               const android::renderscript::Allocation *srcAlloc,
737                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
738                               RsAllocationCubemapFace srcFace) {
739    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
740        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
741                                             "yet implemented.");
742        return;
743    }
744    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
745                                     dstLod, dstFace, w, h, srcAlloc,
746                                     srcXoff, srcYoff, srcLod, srcFace);
747}
748
749void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
750                               const android::renderscript::Allocation *dstAlloc,
751                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
752                               uint32_t dstLod, RsAllocationCubemapFace dstFace,
753                               uint32_t w, uint32_t h, uint32_t d,
754                               const android::renderscript::Allocation *srcAlloc,
755                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
756                               uint32_t srcLod, RsAllocationCubemapFace srcFace) {
757}
758
759void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc,
760                                uint32_t x,
761                                const void *data, uint32_t cIdx, uint32_t sizeBytes) {
762    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
763
764    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
765    uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
766
767    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
768    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
769
770    if (alloc->mHal.state.hasReferences) {
771        e->incRefs(data);
772        e->decRefs(ptr);
773    }
774
775    memcpy(ptr, data, sizeBytes);
776    drv->uploadDeferred = true;
777}
778
779void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc,
780                                uint32_t x, uint32_t y,
781                                const void *data, uint32_t cIdx, uint32_t sizeBytes) {
782    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
783
784    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
785    uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
786
787    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
788    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
789
790    if (alloc->mHal.state.hasReferences) {
791        e->incRefs(data);
792        e->decRefs(ptr);
793    }
794
795    memcpy(ptr, data, sizeBytes);
796    drv->uploadDeferred = true;
797}
798
799static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
800    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
801    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
802
803    for (uint32_t y=0; y < h; y++) {
804        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, lod + 1, face);
805        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, y*2, lod, face);
806        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, y*2+1, lod, face);
807
808        for (uint32_t x=0; x < w; x++) {
809            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
810            oPtr ++;
811            i1 += 2;
812            i2 += 2;
813        }
814    }
815}
816
817static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
818    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
819    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
820
821    for (uint32_t y=0; y < h; y++) {
822        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, lod + 1, face);
823        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, lod, face);
824        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, lod, face);
825
826        for (uint32_t x=0; x < w; x++) {
827            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
828            oPtr ++;
829            i1 += 2;
830            i2 += 2;
831        }
832    }
833}
834
835static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
836    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
837    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
838
839    for (uint32_t y=0; y < h; y++) {
840        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, lod + 1, face);
841        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, lod, face);
842        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, lod, face);
843
844        for (uint32_t x=0; x < w; x++) {
845            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
846            oPtr ++;
847            i1 += 2;
848            i2 += 2;
849        }
850    }
851}
852
853void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
854    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
855        return;
856    }
857    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
858    for (uint32_t face = 0; face < numFaces; face ++) {
859        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
860            switch (alloc->getType()->getElement()->getSizeBits()) {
861            case 32:
862                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
863                break;
864            case 16:
865                mip565(alloc, lod, (RsAllocationCubemapFace)face);
866                break;
867            case 8:
868                mip8(alloc, lod, (RsAllocationCubemapFace)face);
869                break;
870            }
871        }
872    }
873}
874
875
876