rsdAllocation.cpp revision 93eacc7ce0aad4314b4cb41a281f59ce54bb3286
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
21#include "rsAllocation.h"
22
23#include "system/window.h"
24#include "ui/Rect.h"
25#include "ui/GraphicBufferMapper.h"
26
27#ifndef RS_COMPATIBILITY_LIB
28#include "rsdFrameBufferObj.h"
29#include "gui/GLConsumer.h"
30#include "hardware/gralloc.h"
31
32#include <GLES/gl.h>
33#include <GLES2/gl2.h>
34#include <GLES/glext.h>
35#endif
36
37using namespace android;
38using namespace android::renderscript;
39
40
41#ifndef RS_COMPATIBILITY_LIB
42const static GLenum gFaceOrder[] = {
43    GL_TEXTURE_CUBE_MAP_POSITIVE_X,
44    GL_TEXTURE_CUBE_MAP_NEGATIVE_X,
45    GL_TEXTURE_CUBE_MAP_POSITIVE_Y,
46    GL_TEXTURE_CUBE_MAP_NEGATIVE_Y,
47    GL_TEXTURE_CUBE_MAP_POSITIVE_Z,
48    GL_TEXTURE_CUBE_MAP_NEGATIVE_Z
49};
50
51GLenum rsdTypeToGLType(RsDataType t) {
52    switch (t) {
53    case RS_TYPE_UNSIGNED_5_6_5:    return GL_UNSIGNED_SHORT_5_6_5;
54    case RS_TYPE_UNSIGNED_5_5_5_1:  return GL_UNSIGNED_SHORT_5_5_5_1;
55    case RS_TYPE_UNSIGNED_4_4_4_4:  return GL_UNSIGNED_SHORT_4_4_4_4;
56
57    //case RS_TYPE_FLOAT_16:      return GL_HALF_FLOAT;
58    case RS_TYPE_FLOAT_32:      return GL_FLOAT;
59    case RS_TYPE_UNSIGNED_8:    return GL_UNSIGNED_BYTE;
60    case RS_TYPE_UNSIGNED_16:   return GL_UNSIGNED_SHORT;
61    case RS_TYPE_SIGNED_8:      return GL_BYTE;
62    case RS_TYPE_SIGNED_16:     return GL_SHORT;
63    default:    break;
64    }
65    return 0;
66}
67
68GLenum rsdKindToGLFormat(RsDataKind k) {
69    switch (k) {
70    case RS_KIND_PIXEL_L: return GL_LUMINANCE;
71    case RS_KIND_PIXEL_A: return GL_ALPHA;
72    case RS_KIND_PIXEL_LA: return GL_LUMINANCE_ALPHA;
73    case RS_KIND_PIXEL_RGB: return GL_RGB;
74    case RS_KIND_PIXEL_RGBA: return GL_RGBA;
75    case RS_KIND_PIXEL_DEPTH: return GL_DEPTH_COMPONENT16;
76    default: break;
77    }
78    return 0;
79}
80#endif
81
82uint8_t *GetOffsetPtr(const android::renderscript::Allocation *alloc,
83                      uint32_t xoff, uint32_t yoff, uint32_t lod,
84                      RsAllocationCubemapFace face) {
85    uint8_t *ptr = (uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
86    ptr += face * alloc->mHal.drvState.faceOffset;
87    ptr += yoff * alloc->mHal.drvState.lod[lod].stride;
88    ptr += xoff * alloc->mHal.state.elementSizeBytes;
89    return ptr;
90}
91
92
93static void Update2DTexture(const Context *rsc, const Allocation *alloc, const void *ptr,
94                            uint32_t xoff, uint32_t yoff, uint32_t lod,
95                            RsAllocationCubemapFace face, uint32_t w, uint32_t h) {
96#ifndef RS_COMPATIBILITY_LIB
97    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
98
99    rsAssert(drv->textureID);
100    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
101    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
102    GLenum t = GL_TEXTURE_2D;
103    if (alloc->mHal.state.hasFaces) {
104        t = gFaceOrder[face];
105    }
106    RSD_CALL_GL(glTexSubImage2D, t, lod, xoff, yoff, w, h, drv->glFormat, drv->glType, ptr);
107#endif
108}
109
110
111#ifndef RS_COMPATIBILITY_LIB
112static void Upload2DTexture(const Context *rsc, const Allocation *alloc, bool isFirstUpload) {
113    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
114
115    RSD_CALL_GL(glBindTexture, drv->glTarget, drv->textureID);
116    RSD_CALL_GL(glPixelStorei, GL_UNPACK_ALIGNMENT, 1);
117
118    uint32_t faceCount = 1;
119    if (alloc->mHal.state.hasFaces) {
120        faceCount = 6;
121    }
122
123    rsdGLCheckError(rsc, "Upload2DTexture 1 ");
124    for (uint32_t face = 0; face < faceCount; face ++) {
125        for (uint32_t lod = 0; lod < alloc->mHal.state.type->getLODCount(); lod++) {
126            const uint8_t *p = GetOffsetPtr(alloc, 0, 0, lod, (RsAllocationCubemapFace)face);
127
128            GLenum t = GL_TEXTURE_2D;
129            if (alloc->mHal.state.hasFaces) {
130                t = gFaceOrder[face];
131            }
132
133            if (isFirstUpload) {
134                RSD_CALL_GL(glTexImage2D, t, lod, drv->glFormat,
135                             alloc->mHal.state.type->getLODDimX(lod),
136                             alloc->mHal.state.type->getLODDimY(lod),
137                             0, drv->glFormat, drv->glType, p);
138            } else {
139                RSD_CALL_GL(glTexSubImage2D, t, lod, 0, 0,
140                                alloc->mHal.state.type->getLODDimX(lod),
141                                alloc->mHal.state.type->getLODDimY(lod),
142                                drv->glFormat, drv->glType, p);
143            }
144        }
145    }
146
147    if (alloc->mHal.state.mipmapControl == RS_ALLOCATION_MIPMAP_ON_SYNC_TO_TEXTURE) {
148        RSD_CALL_GL(glGenerateMipmap, drv->glTarget);
149    }
150    rsdGLCheckError(rsc, "Upload2DTexture");
151}
152#endif
153
154static void UploadToTexture(const Context *rsc, const Allocation *alloc) {
155#ifndef RS_COMPATIBILITY_LIB
156    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
157
158    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_INPUT) {
159        if (!drv->textureID) {
160            RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
161        }
162        return;
163    }
164
165    if (!drv->glType || !drv->glFormat) {
166        return;
167    }
168
169    if (!alloc->mHal.drvState.lod[0].mallocPtr) {
170        return;
171    }
172
173    bool isFirstUpload = false;
174
175    if (!drv->textureID) {
176        RSD_CALL_GL(glGenTextures, 1, &drv->textureID);
177        isFirstUpload = true;
178    }
179
180    Upload2DTexture(rsc, alloc, isFirstUpload);
181
182    if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
183        if (alloc->mHal.drvState.lod[0].mallocPtr) {
184            free(alloc->mHal.drvState.lod[0].mallocPtr);
185            alloc->mHal.drvState.lod[0].mallocPtr = NULL;
186        }
187    }
188    rsdGLCheckError(rsc, "UploadToTexture");
189#endif
190}
191
192static void AllocateRenderTarget(const Context *rsc, const Allocation *alloc) {
193#ifndef RS_COMPATIBILITY_LIB
194    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
195
196    if (!drv->glFormat) {
197        return;
198    }
199
200    if (!drv->renderTargetID) {
201        RSD_CALL_GL(glGenRenderbuffers, 1, &drv->renderTargetID);
202
203        if (!drv->renderTargetID) {
204            // This should generally not happen
205            ALOGE("allocateRenderTarget failed to gen mRenderTargetID");
206            rsc->dumpDebug();
207            return;
208        }
209        RSD_CALL_GL(glBindRenderbuffer, GL_RENDERBUFFER, drv->renderTargetID);
210        RSD_CALL_GL(glRenderbufferStorage, GL_RENDERBUFFER, drv->glFormat,
211                              alloc->mHal.state.dimensionX, alloc->mHal.state.dimensionY);
212    }
213    rsdGLCheckError(rsc, "AllocateRenderTarget");
214#endif
215}
216
217static void UploadToBufferObject(const Context *rsc, const Allocation *alloc) {
218#ifndef RS_COMPATIBILITY_LIB
219    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
220
221    rsAssert(!alloc->mHal.state.type->getDimY());
222    rsAssert(!alloc->mHal.state.type->getDimZ());
223
224    //alloc->mHal.state.usageFlags |= RS_ALLOCATION_USAGE_GRAPHICS_VERTEX;
225
226    if (!drv->bufferID) {
227        RSD_CALL_GL(glGenBuffers, 1, &drv->bufferID);
228    }
229    if (!drv->bufferID) {
230        ALOGE("Upload to buffer object failed");
231        drv->uploadDeferred = true;
232        return;
233    }
234    RSD_CALL_GL(glBindBuffer, drv->glTarget, drv->bufferID);
235    RSD_CALL_GL(glBufferData, drv->glTarget, alloc->mHal.state.type->getSizeBytes(),
236                 alloc->mHal.drvState.lod[0].mallocPtr, GL_DYNAMIC_DRAW);
237    RSD_CALL_GL(glBindBuffer, drv->glTarget, 0);
238    rsdGLCheckError(rsc, "UploadToBufferObject");
239#endif
240}
241
242static size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc,
243        const Type *type, uint8_t *ptr) {
244    alloc->mHal.drvState.lod[0].dimX = type->getDimX();
245    alloc->mHal.drvState.lod[0].dimY = type->getDimY();
246    alloc->mHal.drvState.lod[0].mallocPtr = 0;
247    alloc->mHal.drvState.lod[0].stride = alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes();
248    alloc->mHal.drvState.lodCount = type->getLODCount();
249    alloc->mHal.drvState.faceCount = type->getDimFaces();
250
251    size_t offsets[Allocation::MAX_LOD];
252    memset(offsets, 0, sizeof(offsets));
253
254    size_t o = alloc->mHal.drvState.lod[0].stride * rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) *
255            rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u);
256    if(alloc->mHal.drvState.lodCount > 1) {
257        uint32_t tx = alloc->mHal.drvState.lod[0].dimX;
258        uint32_t ty = alloc->mHal.drvState.lod[0].dimY;
259        uint32_t tz = alloc->mHal.drvState.lod[0].dimZ;
260        for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
261            alloc->mHal.drvState.lod[lod].dimX = tx;
262            alloc->mHal.drvState.lod[lod].dimY = ty;
263            alloc->mHal.drvState.lod[lod].dimZ = tz;
264            alloc->mHal.drvState.lod[lod].stride = tx * type->getElementSizeBytes();
265            offsets[lod] = o;
266            o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u);
267            if (tx > 1) tx >>= 1;
268            if (ty > 1) ty >>= 1;
269            if (tz > 1) tz >>= 1;
270        }
271    }
272    alloc->mHal.drvState.faceOffset = o;
273
274    alloc->mHal.drvState.lod[0].mallocPtr = ptr;
275    for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
276        alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod];
277    }
278
279    size_t allocSize = alloc->mHal.drvState.faceOffset;
280    if(alloc->mHal.drvState.faceCount) {
281        allocSize *= 6;
282    }
283
284    return allocSize;
285}
286
287bool rsdAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) {
288    DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
289    if (!drv) {
290        return false;
291    }
292    alloc->mHal.drv = drv;
293
294    // Calculate the object size.
295    size_t allocSize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), NULL);
296
297    uint8_t * ptr = NULL;
298    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) {
299    } else if (alloc->mHal.state.userProvidedPtr != NULL) {
300        // user-provided allocation
301        // limitations: no faces, no LOD, USAGE_SCRIPT only
302        if (alloc->mHal.state.usageFlags != (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED)) {
303            ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT and USAGE_SHARED");
304            return false;
305        }
306        if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) {
307            ALOGE("User-allocated buffers must not have multiple faces or LODs");
308            return false;
309        }
310        ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr;
311    } else {
312        if (forceZero) {
313            ptr = (uint8_t *)calloc(1, allocSize);
314        } else {
315            ptr = (uint8_t *)malloc(allocSize);
316        }
317        if (!ptr) {
318            alloc->mHal.drv = NULL;
319            free(drv);
320            return false;
321        }
322    }
323    // Build the pointer tables
324    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr);
325    if(allocSize != verifySize) {
326        rsAssert(!"Size mismatch");
327    }
328
329    drv->glTarget = GL_NONE;
330    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
331        if (alloc->mHal.state.hasFaces) {
332            drv->glTarget = GL_TEXTURE_CUBE_MAP;
333        } else {
334            drv->glTarget = GL_TEXTURE_2D;
335        }
336    } else {
337        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
338            drv->glTarget = GL_ARRAY_BUFFER;
339        }
340    }
341
342#ifndef RS_COMPATIBILITY_LIB
343    drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType());
344    drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind());
345#else
346    drv->glType = 0;
347    drv->glFormat = 0;
348#endif
349
350    if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) {
351        drv->uploadDeferred = true;
352    }
353
354
355    drv->readBackFBO = NULL;
356
357    return true;
358}
359
360void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) {
361    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
362
363#ifndef RS_COMPATIBILITY_LIB
364    if (drv->bufferID) {
365        // Causes a SW crash....
366        //ALOGV(" mBufferID %i", mBufferID);
367        //glDeleteBuffers(1, &mBufferID);
368        //mBufferID = 0;
369    }
370    if (drv->textureID) {
371        RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID);
372        drv->textureID = 0;
373    }
374    if (drv->renderTargetID) {
375        RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID);
376        drv->renderTargetID = 0;
377    }
378#endif
379
380    if (alloc->mHal.drvState.lod[0].mallocPtr) {
381        // don't free user-allocated ptrs
382        if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED)) {
383            free(alloc->mHal.drvState.lod[0].mallocPtr);
384        }
385        alloc->mHal.drvState.lod[0].mallocPtr = NULL;
386    }
387
388#ifndef RS_COMPATIBILITY_LIB
389    if (drv->readBackFBO != NULL) {
390        delete drv->readBackFBO;
391        drv->readBackFBO = NULL;
392    }
393#endif
394
395    free(drv);
396    alloc->mHal.drv = NULL;
397}
398
399void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
400                         const Type *newType, bool zeroNew) {
401    // can't resize Allocations with user-allocated buffers
402    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
403        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
404        return;
405    }
406    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
407    // Calculate the object size
408    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL);
409    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
410    // Build the relative pointer tables.
411    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
412    if(s != verifySize) {
413        rsAssert(!"Size mismatch");
414    }
415
416    const uint32_t oldDimX = alloc->mHal.state.dimensionX;
417    const uint32_t dimX = newType->getDimX();
418
419    if (dimX > oldDimX) {
420        uint32_t stride = alloc->mHal.state.elementSizeBytes;
421        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
422                 0, stride * (dimX - oldDimX));
423    }
424}
425
426static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
427#ifndef RS_COMPATIBILITY_LIB
428    if (!alloc->getIsScript()) {
429        return; // nothing to sync
430    }
431
432    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
433    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
434
435    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
436    if (!drv->textureID && !drv->renderTargetID) {
437        return; // nothing was rendered here yet, so nothing to sync
438    }
439    if (drv->readBackFBO == NULL) {
440        drv->readBackFBO = new RsdFrameBufferObj();
441        drv->readBackFBO->setColorTarget(drv, 0);
442        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
443                                        alloc->getType()->getDimY());
444    }
445
446    // Bind the framebuffer object so we can read back from it
447    drv->readBackFBO->setActive(rsc);
448
449    // Do the readback
450    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
451                alloc->mHal.drvState.lod[0].dimY,
452                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
453
454    // Revert framebuffer to its original
455    lastFbo->setActive(rsc);
456#endif
457}
458
459
460void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
461                         RsAllocationUsageType src) {
462    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
463
464    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
465        if(!alloc->getIsRenderTarget()) {
466            rsc->setError(RS_ERROR_FATAL_DRIVER,
467                          "Attempting to sync allocation from render target, "
468                          "for non-render target allocation");
469        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
470            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
471                                                 "render target");
472        } else {
473            rsdAllocationSyncFromFBO(rsc, alloc);
474        }
475        return;
476    }
477
478    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT);
479
480    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
481        UploadToTexture(rsc, alloc);
482    } else {
483        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
484            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
485            AllocateRenderTarget(rsc, alloc);
486        }
487    }
488    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
489        UploadToBufferObject(rsc, alloc);
490    }
491
492    drv->uploadDeferred = false;
493}
494
495void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
496    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
497    drv->uploadDeferred = true;
498}
499
500int32_t rsdAllocationInitSurfaceTexture(const Context *rsc, const Allocation *alloc) {
501#ifndef RS_COMPATIBILITY_LIB
502    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
503    UploadToTexture(rsc, alloc);
504    return drv->textureID;
505#else
506    return 0;
507#endif
508}
509
510#ifndef RS_COMPATIBILITY_LIB
511static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
512    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
513
514    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
515    if (r) {
516        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
517        return false;
518    }
519
520    // Must lock the whole surface
521    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
522    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
523
524    void *dst = NULL;
525    mapper.lock(drv->wndBuffer->handle,
526            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
527            bounds, &dst);
528    alloc->mHal.drvState.lod[0].mallocPtr = dst;
529    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
530
531    return true;
532}
533#endif
534
535void rsdAllocationSetSurfaceTexture(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
536#ifndef RS_COMPATIBILITY_LIB
537    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
538
539    //ALOGE("rsdAllocationSetSurfaceTexture %p  %p", alloc, nw);
540
541    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
542        //TODO finish support for render target + script
543        drv->wnd = nw;
544        return;
545    }
546
547
548    // Cleanup old surface if there is one.
549    if (alloc->mHal.state.wndSurface) {
550        ANativeWindow *old = alloc->mHal.state.wndSurface;
551        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
552        mapper.unlock(drv->wndBuffer->handle);
553        old->queueBuffer(old, drv->wndBuffer, -1);
554    }
555
556    if (nw != NULL) {
557        int32_t r;
558        uint32_t flags = 0;
559        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
560            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
561        }
562        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
563            flags |= GRALLOC_USAGE_HW_RENDER;
564        }
565
566        r = native_window_set_usage(nw, flags);
567        if (r) {
568            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
569            return;
570        }
571
572        r = native_window_set_buffers_dimensions(nw, alloc->mHal.state.dimensionX,
573                                                 alloc->mHal.state.dimensionY);
574        if (r) {
575            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
576            return;
577        }
578
579        r = native_window_set_buffer_count(nw, 3);
580        if (r) {
581            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer count.");
582            return;
583        }
584
585        IoGetBuffer(rsc, alloc, nw);
586    }
587#endif
588}
589
590void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
591#ifndef RS_COMPATIBILITY_LIB
592    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
593    ANativeWindow *nw = alloc->mHal.state.wndSurface;
594
595    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
596        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
597        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
598        return;
599    }
600
601    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
602        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
603        mapper.unlock(drv->wndBuffer->handle);
604        int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
605        if (r) {
606            rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
607            return;
608        }
609
610        IoGetBuffer(rsc, alloc, nw);
611    }
612#endif
613}
614
615void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
616#ifndef RS_COMPATIBILITY_LIB
617    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
618    alloc->mHal.state.surfaceTexture->updateTexImage();
619#endif
620}
621
622
623void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
624                         uint32_t xoff, uint32_t lod, uint32_t count,
625                         const void *data, size_t sizeBytes) {
626    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
627
628    const uint32_t eSize = alloc->mHal.state.type->getElementSizeBytes();
629    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
630    uint32_t size = count * eSize;
631
632    if (ptr != data) {
633        // Skip the copy if we are the same allocation. This can arise from
634        // our Bitmap optimization, where we share the same storage.
635        if (alloc->mHal.state.hasReferences) {
636            alloc->incRefs(data, count);
637            alloc->decRefs(ptr, count);
638        }
639        memcpy(ptr, data, size);
640    }
641    drv->uploadDeferred = true;
642}
643
644void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
645                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
646                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
647    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
648
649    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
650    uint32_t lineSize = eSize * w;
651    if (!stride) {
652        stride = lineSize;
653    }
654
655    if (alloc->mHal.drvState.lod[0].mallocPtr) {
656        const uint8_t *src = static_cast<const uint8_t *>(data);
657        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, lod, face);
658        if (dst == src) {
659            // Skip the copy if we are the same allocation. This can arise from
660            // our Bitmap optimization, where we share the same storage.
661            drv->uploadDeferred = true;
662            return;
663        }
664
665        for (uint32_t line=yoff; line < (yoff+h); line++) {
666            if (alloc->mHal.state.hasReferences) {
667                alloc->incRefs(src, w);
668                alloc->decRefs(dst, w);
669            }
670            memcpy(dst, src, lineSize);
671            src += stride;
672            dst += alloc->mHal.drvState.lod[lod].stride;
673        }
674        drv->uploadDeferred = true;
675    } else {
676        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
677    }
678}
679
680void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
681                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
682                         uint32_t lod, RsAllocationCubemapFace face,
683                         uint32_t w, uint32_t h, uint32_t d, const void *data, uint32_t sizeBytes) {
684
685}
686
687void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
688                         uint32_t xoff, uint32_t lod, uint32_t count,
689                         void *data, size_t sizeBytes) {
690    const uint32_t eSize = alloc->mHal.state.type->getElementSizeBytes();
691    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
692    if (data != ptr) {
693        // Skip the copy if we are the same allocation. This can arise from
694        // our Bitmap optimization, where we share the same storage.
695        memcpy(data, ptr, count * eSize);
696    }
697}
698
699void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
700                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
701                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
702    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
703    uint32_t lineSize = eSize * w;
704    if (!stride) {
705        stride = lineSize;
706    }
707
708    if (alloc->mHal.drvState.lod[0].mallocPtr) {
709        uint8_t *dst = static_cast<uint8_t *>(data);
710        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, lod, face);
711        if (dst == src) {
712            // Skip the copy if we are the same allocation. This can arise from
713            // our Bitmap optimization, where we share the same storage.
714            return;
715        }
716
717        for (uint32_t line=yoff; line < (yoff+h); line++) {
718            memcpy(dst, src, lineSize);
719            dst += stride;
720            src += alloc->mHal.drvState.lod[lod].stride;
721        }
722    } else {
723        ALOGE("Add code to readback from non-script memory");
724    }
725}
726
727
728void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
729                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
730                         uint32_t lod, RsAllocationCubemapFace face,
731                         uint32_t w, uint32_t h, uint32_t d, void *data, uint32_t sizeBytes) {
732
733}
734
735void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
736                          const android::renderscript::Allocation *alloc) {
737    return alloc->mHal.drvState.lod[0].mallocPtr;
738}
739
740void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
741                          const android::renderscript::Allocation *alloc) {
742
743}
744
745void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
746                               const android::renderscript::Allocation *dstAlloc,
747                               uint32_t dstXoff, uint32_t dstLod, uint32_t count,
748                               const android::renderscript::Allocation *srcAlloc,
749                               uint32_t srcXoff, uint32_t srcLod) {
750}
751
752
753void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
754                                      const android::renderscript::Allocation *dstAlloc,
755                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
756                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
757                                      const android::renderscript::Allocation *srcAlloc,
758                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
759                                      RsAllocationCubemapFace srcFace) {
760    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
761    for (uint32_t i = 0; i < h; i ++) {
762        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstLod, dstFace);
763        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcLod, srcFace);
764        memcpy(dstPtr, srcPtr, w * elementSize);
765
766        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
767        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
768    }
769}
770
771void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
772                               const android::renderscript::Allocation *dstAlloc,
773                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
774                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
775                               const android::renderscript::Allocation *srcAlloc,
776                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
777                               RsAllocationCubemapFace srcFace) {
778    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
779        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
780                                             "yet implemented.");
781        return;
782    }
783    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
784                                     dstLod, dstFace, w, h, srcAlloc,
785                                     srcXoff, srcYoff, srcLod, srcFace);
786}
787
788void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
789                               const android::renderscript::Allocation *dstAlloc,
790                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
791                               uint32_t dstLod, RsAllocationCubemapFace dstFace,
792                               uint32_t w, uint32_t h, uint32_t d,
793                               const android::renderscript::Allocation *srcAlloc,
794                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
795                               uint32_t srcLod, RsAllocationCubemapFace srcFace) {
796}
797
798void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc,
799                                uint32_t x,
800                                const void *data, uint32_t cIdx, uint32_t sizeBytes) {
801    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
802
803    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
804    uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
805
806    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
807    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
808
809    if (alloc->mHal.state.hasReferences) {
810        e->incRefs(data);
811        e->decRefs(ptr);
812    }
813
814    memcpy(ptr, data, sizeBytes);
815    drv->uploadDeferred = true;
816}
817
818void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc,
819                                uint32_t x, uint32_t y,
820                                const void *data, uint32_t cIdx, uint32_t sizeBytes) {
821    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
822
823    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
824    uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
825
826    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
827    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
828
829    if (alloc->mHal.state.hasReferences) {
830        e->incRefs(data);
831        e->decRefs(ptr);
832    }
833
834    memcpy(ptr, data, sizeBytes);
835    drv->uploadDeferred = true;
836}
837
838static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
839    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
840    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
841
842    for (uint32_t y=0; y < h; y++) {
843        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, lod + 1, face);
844        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, y*2, lod, face);
845        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, y*2+1, lod, face);
846
847        for (uint32_t x=0; x < w; x++) {
848            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
849            oPtr ++;
850            i1 += 2;
851            i2 += 2;
852        }
853    }
854}
855
856static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
857    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
858    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
859
860    for (uint32_t y=0; y < h; y++) {
861        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, lod + 1, face);
862        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, lod, face);
863        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, lod, face);
864
865        for (uint32_t x=0; x < w; x++) {
866            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
867            oPtr ++;
868            i1 += 2;
869            i2 += 2;
870        }
871    }
872}
873
874static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
875    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
876    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
877
878    for (uint32_t y=0; y < h; y++) {
879        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, lod + 1, face);
880        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, lod, face);
881        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, lod, face);
882
883        for (uint32_t x=0; x < w; x++) {
884            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
885            oPtr ++;
886            i1 += 2;
887            i2 += 2;
888        }
889    }
890}
891
892void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
893    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
894        return;
895    }
896    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
897    for (uint32_t face = 0; face < numFaces; face ++) {
898        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
899            switch (alloc->getType()->getElement()->getSizeBits()) {
900            case 32:
901                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
902                break;
903            case 16:
904                mip565(alloc, lod, (RsAllocationCubemapFace)face);
905                break;
906            case 8:
907                mip8(alloc, lod, (RsAllocationCubemapFace)face);
908                break;
909            }
910        }
911    }
912}
913
914
915