rsdAllocation.cpp revision 94999c3c7e3dcb6b6c0a9fe5785a0d3216357b0e 
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.drvState.lod[0].dimX, alloc->mHal.drvState.lod[0].dimY);
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].dimZ = type->getDimZ();
247    alloc->mHal.drvState.lod[0].mallocPtr = 0;
248    // Stride needs to be 16-byte aligned too!
249    size_t stride = alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes();
250    alloc->mHal.drvState.lod[0].stride = rsRound(stride, 16);
251    alloc->mHal.drvState.lodCount = type->getLODCount();
252    alloc->mHal.drvState.faceCount = type->getDimFaces();
253
254    size_t offsets[Allocation::MAX_LOD];
255    memset(offsets, 0, sizeof(offsets));
256
257    size_t o = alloc->mHal.drvState.lod[0].stride * rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) *
258            rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u);
259    if(alloc->mHal.drvState.lodCount > 1) {
260        uint32_t tx = alloc->mHal.drvState.lod[0].dimX;
261        uint32_t ty = alloc->mHal.drvState.lod[0].dimY;
262        uint32_t tz = alloc->mHal.drvState.lod[0].dimZ;
263        for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
264            alloc->mHal.drvState.lod[lod].dimX = tx;
265            alloc->mHal.drvState.lod[lod].dimY = ty;
266            alloc->mHal.drvState.lod[lod].dimZ = tz;
267            alloc->mHal.drvState.lod[lod].stride =
268                    rsRound(tx * type->getElementSizeBytes(), 16);
269            offsets[lod] = o;
270            o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u);
271            if (tx > 1) tx >>= 1;
272            if (ty > 1) ty >>= 1;
273            if (tz > 1) tz >>= 1;
274        }
275    }
276    alloc->mHal.drvState.faceOffset = o;
277
278    alloc->mHal.drvState.lod[0].mallocPtr = ptr;
279    for (uint32_t lod=1; lod < alloc->mHal.drvState.lodCount; lod++) {
280        alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod];
281    }
282
283    size_t allocSize = alloc->mHal.drvState.faceOffset;
284    if(alloc->mHal.drvState.faceCount) {
285        allocSize *= 6;
286    }
287
288    return allocSize;
289}
290
291bool rsdAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) {
292    DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
293    if (!drv) {
294        return false;
295    }
296    alloc->mHal.drv = drv;
297
298    // Calculate the object size.
299    size_t allocSize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), NULL);
300
301    uint8_t * ptr = NULL;
302    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) {
303    } else if (alloc->mHal.state.userProvidedPtr != NULL) {
304        // user-provided allocation
305        // limitations: no faces, no LOD, USAGE_SCRIPT only
306        if (alloc->mHal.state.usageFlags != (RS_ALLOCATION_USAGE_SCRIPT | RS_ALLOCATION_USAGE_SHARED)) {
307            ALOGE("Can't use user-allocated buffers if usage is not USAGE_SCRIPT and USAGE_SHARED");
308            return false;
309        }
310        if (alloc->getType()->getDimLOD() || alloc->getType()->getDimFaces()) {
311            ALOGE("User-allocated buffers must not have multiple faces or LODs");
312            return false;
313        }
314        ptr = (uint8_t*)alloc->mHal.state.userProvidedPtr;
315    } else {
316        // We align all allocations to a 16-byte boundary.
317        ptr = (uint8_t *)memalign(16, allocSize);
318        if (!ptr) {
319            alloc->mHal.drv = NULL;
320            free(drv);
321            return false;
322        }
323        if (forceZero) {
324            memset(ptr, 0, allocSize);
325        }
326    }
327    // Build the pointer tables
328    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, alloc->getType(), ptr);
329    if(allocSize != verifySize) {
330        rsAssert(!"Size mismatch");
331    }
332
333    drv->glTarget = GL_NONE;
334    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
335        if (alloc->mHal.state.hasFaces) {
336            drv->glTarget = GL_TEXTURE_CUBE_MAP;
337        } else {
338            drv->glTarget = GL_TEXTURE_2D;
339        }
340    } else {
341        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
342            drv->glTarget = GL_ARRAY_BUFFER;
343        }
344    }
345
346#ifndef RS_COMPATIBILITY_LIB
347    drv->glType = rsdTypeToGLType(alloc->mHal.state.type->getElement()->getComponent().getType());
348    drv->glFormat = rsdKindToGLFormat(alloc->mHal.state.type->getElement()->getComponent().getKind());
349#else
350    drv->glType = 0;
351    drv->glFormat = 0;
352#endif
353
354    if (alloc->mHal.state.usageFlags & ~RS_ALLOCATION_USAGE_SCRIPT) {
355        drv->uploadDeferred = true;
356    }
357
358
359    drv->readBackFBO = NULL;
360
361    return true;
362}
363
364void rsdAllocationDestroy(const Context *rsc, Allocation *alloc) {
365    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
366
367#ifndef RS_COMPATIBILITY_LIB
368    if (drv->bufferID) {
369        // Causes a SW crash....
370        //ALOGV(" mBufferID %i", mBufferID);
371        //glDeleteBuffers(1, &mBufferID);
372        //mBufferID = 0;
373    }
374    if (drv->textureID) {
375        RSD_CALL_GL(glDeleteTextures, 1, &drv->textureID);
376        drv->textureID = 0;
377    }
378    if (drv->renderTargetID) {
379        RSD_CALL_GL(glDeleteRenderbuffers, 1, &drv->renderTargetID);
380        drv->renderTargetID = 0;
381    }
382#endif
383
384    if (alloc->mHal.drvState.lod[0].mallocPtr) {
385        // don't free user-allocated ptrs
386        if (!(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED)) {
387            free(alloc->mHal.drvState.lod[0].mallocPtr);
388        }
389        alloc->mHal.drvState.lod[0].mallocPtr = NULL;
390    }
391
392#ifndef RS_COMPATIBILITY_LIB
393    if (drv->readBackFBO != NULL) {
394        delete drv->readBackFBO;
395        drv->readBackFBO = NULL;
396    }
397
398    if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT) &&
399        (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT)) {
400
401        DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
402        ANativeWindow *nw = alloc->mHal.state.wndSurface;
403
404        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
405        mapper.unlock(drv->wndBuffer->handle);
406        int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
407    }
408#endif
409
410    free(drv);
411    alloc->mHal.drv = NULL;
412}
413
414void rsdAllocationResize(const Context *rsc, const Allocation *alloc,
415                         const Type *newType, bool zeroNew) {
416    const uint32_t oldDimX = alloc->mHal.drvState.lod[0].dimX;
417    const uint32_t dimX = newType->getDimX();
418
419    // can't resize Allocations with user-allocated buffers
420    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SHARED) {
421        ALOGE("Resize cannot be called on a USAGE_SHARED allocation");
422        return;
423    }
424    void * oldPtr = alloc->mHal.drvState.lod[0].mallocPtr;
425    // Calculate the object size
426    size_t s = AllocationBuildPointerTable(rsc, alloc, newType, NULL);
427    uint8_t *ptr = (uint8_t *)realloc(oldPtr, s);
428    // Build the relative pointer tables.
429    size_t verifySize = AllocationBuildPointerTable(rsc, alloc, newType, ptr);
430    if(s != verifySize) {
431        rsAssert(!"Size mismatch");
432    }
433
434
435    if (dimX > oldDimX) {
436        uint32_t stride = alloc->mHal.state.elementSizeBytes;
437        memset(((uint8_t *)alloc->mHal.drvState.lod[0].mallocPtr) + stride * oldDimX,
438                 0, stride * (dimX - oldDimX));
439    }
440}
441
442static void rsdAllocationSyncFromFBO(const Context *rsc, const Allocation *alloc) {
443#ifndef RS_COMPATIBILITY_LIB
444    if (!alloc->getIsScript()) {
445        return; // nothing to sync
446    }
447
448    RsdHal *dc = (RsdHal *)rsc->mHal.drv;
449    RsdFrameBufferObj *lastFbo = dc->gl.currentFrameBuffer;
450
451    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
452    if (!drv->textureID && !drv->renderTargetID) {
453        return; // nothing was rendered here yet, so nothing to sync
454    }
455    if (drv->readBackFBO == NULL) {
456        drv->readBackFBO = new RsdFrameBufferObj();
457        drv->readBackFBO->setColorTarget(drv, 0);
458        drv->readBackFBO->setDimensions(alloc->getType()->getDimX(),
459                                        alloc->getType()->getDimY());
460    }
461
462    // Bind the framebuffer object so we can read back from it
463    drv->readBackFBO->setActive(rsc);
464
465    // Do the readback
466    RSD_CALL_GL(glReadPixels, 0, 0, alloc->mHal.drvState.lod[0].dimX,
467                alloc->mHal.drvState.lod[0].dimY,
468                drv->glFormat, drv->glType, alloc->mHal.drvState.lod[0].mallocPtr);
469
470    // Revert framebuffer to its original
471    lastFbo->setActive(rsc);
472#endif
473}
474
475
476void rsdAllocationSyncAll(const Context *rsc, const Allocation *alloc,
477                         RsAllocationUsageType src) {
478    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
479
480    if (src == RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
481        if(!alloc->getIsRenderTarget()) {
482            rsc->setError(RS_ERROR_FATAL_DRIVER,
483                          "Attempting to sync allocation from render target, "
484                          "for non-render target allocation");
485        } else if (alloc->getType()->getElement()->getKind() != RS_KIND_PIXEL_RGBA) {
486            rsc->setError(RS_ERROR_FATAL_DRIVER, "Cannot only sync from RGBA"
487                                                 "render target");
488        } else {
489            rsdAllocationSyncFromFBO(rsc, alloc);
490        }
491        return;
492    }
493
494    rsAssert(src == RS_ALLOCATION_USAGE_SCRIPT);
495
496    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_TEXTURE) {
497        UploadToTexture(rsc, alloc);
498    } else {
499        if ((alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) &&
500            !(alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_IO_OUTPUT)) {
501            AllocateRenderTarget(rsc, alloc);
502        }
503    }
504    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_VERTEX) {
505        UploadToBufferObject(rsc, alloc);
506    }
507
508    drv->uploadDeferred = false;
509}
510
511void rsdAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
512    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
513    drv->uploadDeferred = true;
514}
515
516int32_t rsdAllocationInitSurfaceTexture(const Context *rsc, const Allocation *alloc) {
517#ifndef RS_COMPATIBILITY_LIB
518    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
519    UploadToTexture(rsc, alloc);
520    return drv->textureID;
521#else
522    return 0;
523#endif
524}
525
526#ifndef RS_COMPATIBILITY_LIB
527static bool IoGetBuffer(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
528    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
529
530    int32_t r = native_window_dequeue_buffer_and_wait(nw, &drv->wndBuffer);
531    if (r) {
532        rsc->setError(RS_ERROR_DRIVER, "Error getting next IO output buffer.");
533        return false;
534    }
535
536    // Must lock the whole surface
537    GraphicBufferMapper &mapper = GraphicBufferMapper::get();
538    Rect bounds(drv->wndBuffer->width, drv->wndBuffer->height);
539
540    void *dst = NULL;
541    mapper.lock(drv->wndBuffer->handle,
542            GRALLOC_USAGE_SW_READ_NEVER | GRALLOC_USAGE_SW_WRITE_OFTEN,
543            bounds, &dst);
544    alloc->mHal.drvState.lod[0].mallocPtr = dst;
545    alloc->mHal.drvState.lod[0].stride = drv->wndBuffer->stride * alloc->mHal.state.elementSizeBytes;
546    rsAssert((alloc->mHal.drvState.lod[0].stride & 0xf) == 0);
547
548    return true;
549}
550#endif
551
552void rsdAllocationSetSurfaceTexture(const Context *rsc, Allocation *alloc, ANativeWindow *nw) {
553#ifndef RS_COMPATIBILITY_LIB
554    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
555
556    //ALOGE("rsdAllocationSetSurfaceTexture %p  %p", alloc, nw);
557
558    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
559        //TODO finish support for render target + script
560        drv->wnd = nw;
561        return;
562    }
563
564
565    // Cleanup old surface if there is one.
566    if (alloc->mHal.state.wndSurface) {
567        ANativeWindow *old = alloc->mHal.state.wndSurface;
568        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
569        mapper.unlock(drv->wndBuffer->handle);
570        old->queueBuffer(old, drv->wndBuffer, -1);
571    }
572
573    if (nw != NULL) {
574        int32_t r;
575        uint32_t flags = 0;
576        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
577            flags |= GRALLOC_USAGE_SW_READ_RARELY | GRALLOC_USAGE_SW_WRITE_OFTEN;
578        }
579        if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
580            flags |= GRALLOC_USAGE_HW_RENDER;
581        }
582
583        r = native_window_set_usage(nw, flags);
584        if (r) {
585            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer usage.");
586            return;
587        }
588
589        r = native_window_set_buffers_dimensions(nw, alloc->mHal.drvState.lod[0].dimX,
590                                                 alloc->mHal.drvState.lod[0].dimY);
591        if (r) {
592            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer dimensions.");
593            return;
594        }
595
596        r = native_window_set_buffer_count(nw, 3);
597        if (r) {
598            rsc->setError(RS_ERROR_DRIVER, "Error setting IO output buffer count.");
599            return;
600        }
601
602        IoGetBuffer(rsc, alloc, nw);
603    }
604#endif
605}
606
607void rsdAllocationIoSend(const Context *rsc, Allocation *alloc) {
608#ifndef RS_COMPATIBILITY_LIB
609    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
610    ANativeWindow *nw = alloc->mHal.state.wndSurface;
611
612    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_GRAPHICS_RENDER_TARGET) {
613        RsdHal *dc = (RsdHal *)rsc->mHal.drv;
614        RSD_CALL_GL(eglSwapBuffers, dc->gl.egl.display, dc->gl.egl.surface);
615        return;
616    }
617
618    if (alloc->mHal.state.usageFlags & RS_ALLOCATION_USAGE_SCRIPT) {
619        GraphicBufferMapper &mapper = GraphicBufferMapper::get();
620        mapper.unlock(drv->wndBuffer->handle);
621        int32_t r = nw->queueBuffer(nw, drv->wndBuffer, -1);
622        if (r) {
623            rsc->setError(RS_ERROR_DRIVER, "Error sending IO output buffer.");
624            return;
625        }
626
627        IoGetBuffer(rsc, alloc, nw);
628    }
629#endif
630}
631
632void rsdAllocationIoReceive(const Context *rsc, Allocation *alloc) {
633#ifndef RS_COMPATIBILITY_LIB
634    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
635    alloc->mHal.state.surfaceTexture->updateTexImage();
636#endif
637}
638
639
640void rsdAllocationData1D(const Context *rsc, const Allocation *alloc,
641                         uint32_t xoff, uint32_t lod, uint32_t count,
642                         const void *data, size_t sizeBytes) {
643    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
644
645    const uint32_t eSize = alloc->mHal.state.type->getElementSizeBytes();
646    uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
647    uint32_t size = count * eSize;
648
649    if (ptr != data) {
650        // Skip the copy if we are the same allocation. This can arise from
651        // our Bitmap optimization, where we share the same storage.
652        if (alloc->mHal.state.hasReferences) {
653            alloc->incRefs(data, count);
654            alloc->decRefs(ptr, count);
655        }
656        memcpy(ptr, data, size);
657    }
658    drv->uploadDeferred = true;
659}
660
661void rsdAllocationData2D(const Context *rsc, const Allocation *alloc,
662                         uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
663                         uint32_t w, uint32_t h, const void *data, size_t sizeBytes, size_t stride) {
664    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
665
666    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
667    uint32_t lineSize = eSize * w;
668    if (!stride) {
669        stride = lineSize;
670    }
671
672    if (alloc->mHal.drvState.lod[0].mallocPtr) {
673        const uint8_t *src = static_cast<const uint8_t *>(data);
674        uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, lod, face);
675        if (dst == src) {
676            // Skip the copy if we are the same allocation. This can arise from
677            // our Bitmap optimization, where we share the same storage.
678            drv->uploadDeferred = true;
679            return;
680        }
681
682        for (uint32_t line=yoff; line < (yoff+h); line++) {
683            if (alloc->mHal.state.hasReferences) {
684                alloc->incRefs(src, w);
685                alloc->decRefs(dst, w);
686            }
687            memcpy(dst, src, lineSize);
688            src += stride;
689            dst += alloc->mHal.drvState.lod[lod].stride;
690        }
691        drv->uploadDeferred = true;
692    } else {
693        Update2DTexture(rsc, alloc, data, xoff, yoff, lod, face, w, h);
694    }
695}
696
697void rsdAllocationData3D(const Context *rsc, const Allocation *alloc,
698                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
699                         uint32_t lod, RsAllocationCubemapFace face,
700                         uint32_t w, uint32_t h, uint32_t d, const void *data, uint32_t sizeBytes) {
701
702}
703
704void rsdAllocationRead1D(const Context *rsc, const Allocation *alloc,
705                         uint32_t xoff, uint32_t lod, uint32_t count,
706                         void *data, size_t sizeBytes) {
707    const uint32_t eSize = alloc->mHal.state.type->getElementSizeBytes();
708    const uint8_t * ptr = GetOffsetPtr(alloc, xoff, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
709    if (data != ptr) {
710        // Skip the copy if we are the same allocation. This can arise from
711        // our Bitmap optimization, where we share the same storage.
712        memcpy(data, ptr, count * eSize);
713    }
714}
715
716void rsdAllocationRead2D(const Context *rsc, const Allocation *alloc,
717                                uint32_t xoff, uint32_t yoff, uint32_t lod, RsAllocationCubemapFace face,
718                                uint32_t w, uint32_t h, void *data, size_t sizeBytes, size_t stride) {
719    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
720    uint32_t lineSize = eSize * w;
721    if (!stride) {
722        stride = lineSize;
723    }
724
725    if (alloc->mHal.drvState.lod[0].mallocPtr) {
726        uint8_t *dst = static_cast<uint8_t *>(data);
727        const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, lod, face);
728        if (dst == src) {
729            // Skip the copy if we are the same allocation. This can arise from
730            // our Bitmap optimization, where we share the same storage.
731            return;
732        }
733
734        for (uint32_t line=yoff; line < (yoff+h); line++) {
735            memcpy(dst, src, lineSize);
736            dst += stride;
737            src += alloc->mHal.drvState.lod[lod].stride;
738        }
739    } else {
740        ALOGE("Add code to readback from non-script memory");
741    }
742}
743
744
745void rsdAllocationRead3D(const Context *rsc, const Allocation *alloc,
746                         uint32_t xoff, uint32_t yoff, uint32_t zoff,
747                         uint32_t lod, RsAllocationCubemapFace face,
748                         uint32_t w, uint32_t h, uint32_t d, void *data, uint32_t sizeBytes) {
749
750}
751
752void * rsdAllocationLock1D(const android::renderscript::Context *rsc,
753                          const android::renderscript::Allocation *alloc) {
754    return alloc->mHal.drvState.lod[0].mallocPtr;
755}
756
757void rsdAllocationUnlock1D(const android::renderscript::Context *rsc,
758                          const android::renderscript::Allocation *alloc) {
759
760}
761
762void rsdAllocationData1D_alloc(const android::renderscript::Context *rsc,
763                               const android::renderscript::Allocation *dstAlloc,
764                               uint32_t dstXoff, uint32_t dstLod, uint32_t count,
765                               const android::renderscript::Allocation *srcAlloc,
766                               uint32_t srcXoff, uint32_t srcLod) {
767}
768
769
770void rsdAllocationData2D_alloc_script(const android::renderscript::Context *rsc,
771                                      const android::renderscript::Allocation *dstAlloc,
772                                      uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
773                                      RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
774                                      const android::renderscript::Allocation *srcAlloc,
775                                      uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
776                                      RsAllocationCubemapFace srcFace) {
777    uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
778    for (uint32_t i = 0; i < h; i ++) {
779        uint8_t *dstPtr = GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstLod, dstFace);
780        uint8_t *srcPtr = GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcLod, srcFace);
781        memcpy(dstPtr, srcPtr, w * elementSize);
782
783        //ALOGE("COPIED dstXoff(%u), dstYoff(%u), dstLod(%u), dstFace(%u), w(%u), h(%u), srcXoff(%u), srcYoff(%u), srcLod(%u), srcFace(%u)",
784        //     dstXoff, dstYoff, dstLod, dstFace, w, h, srcXoff, srcYoff, srcLod, srcFace);
785    }
786}
787
788void rsdAllocationData2D_alloc(const android::renderscript::Context *rsc,
789                               const android::renderscript::Allocation *dstAlloc,
790                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstLod,
791                               RsAllocationCubemapFace dstFace, uint32_t w, uint32_t h,
792                               const android::renderscript::Allocation *srcAlloc,
793                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcLod,
794                               RsAllocationCubemapFace srcFace) {
795    if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
796        rsc->setError(RS_ERROR_FATAL_DRIVER, "Non-script allocation copies not "
797                                             "yet implemented.");
798        return;
799    }
800    rsdAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff,
801                                     dstLod, dstFace, w, h, srcAlloc,
802                                     srcXoff, srcYoff, srcLod, srcFace);
803}
804
805void rsdAllocationData3D_alloc(const android::renderscript::Context *rsc,
806                               const android::renderscript::Allocation *dstAlloc,
807                               uint32_t dstXoff, uint32_t dstYoff, uint32_t dstZoff,
808                               uint32_t dstLod, RsAllocationCubemapFace dstFace,
809                               uint32_t w, uint32_t h, uint32_t d,
810                               const android::renderscript::Allocation *srcAlloc,
811                               uint32_t srcXoff, uint32_t srcYoff, uint32_t srcZoff,
812                               uint32_t srcLod, RsAllocationCubemapFace srcFace) {
813}
814
815void rsdAllocationElementData1D(const Context *rsc, const Allocation *alloc,
816                                uint32_t x,
817                                const void *data, uint32_t cIdx, uint32_t sizeBytes) {
818    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
819
820    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
821    uint8_t * ptr = GetOffsetPtr(alloc, x, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
822
823    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
824    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
825
826    if (alloc->mHal.state.hasReferences) {
827        e->incRefs(data);
828        e->decRefs(ptr);
829    }
830
831    memcpy(ptr, data, sizeBytes);
832    drv->uploadDeferred = true;
833}
834
835void rsdAllocationElementData2D(const Context *rsc, const Allocation *alloc,
836                                uint32_t x, uint32_t y,
837                                const void *data, uint32_t cIdx, uint32_t sizeBytes) {
838    DrvAllocation *drv = (DrvAllocation *)alloc->mHal.drv;
839
840    uint32_t eSize = alloc->mHal.state.elementSizeBytes;
841    uint8_t * ptr = GetOffsetPtr(alloc, x, y, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
842
843    const Element * e = alloc->mHal.state.type->getElement()->getField(cIdx);
844    ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
845
846    if (alloc->mHal.state.hasReferences) {
847        e->incRefs(data);
848        e->decRefs(ptr);
849    }
850
851    memcpy(ptr, data, sizeBytes);
852    drv->uploadDeferred = true;
853}
854
855static void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
856    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
857    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
858
859    for (uint32_t y=0; y < h; y++) {
860        uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, lod + 1, face);
861        const uint16_t *i1 = (uint16_t *)GetOffsetPtr(alloc, 0, y*2, lod, face);
862        const uint16_t *i2 = (uint16_t *)GetOffsetPtr(alloc, 0, y*2+1, lod, face);
863
864        for (uint32_t x=0; x < w; x++) {
865            *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
866            oPtr ++;
867            i1 += 2;
868            i2 += 2;
869        }
870    }
871}
872
873static void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
874    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
875    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
876
877    for (uint32_t y=0; y < h; y++) {
878        uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, lod + 1, face);
879        const uint32_t *i1 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2, lod, face);
880        const uint32_t *i2 = (uint32_t *)GetOffsetPtr(alloc, 0, y*2+1, lod, face);
881
882        for (uint32_t x=0; x < w; x++) {
883            *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
884            oPtr ++;
885            i1 += 2;
886            i2 += 2;
887        }
888    }
889}
890
891static void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
892    uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
893    uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
894
895    for (uint32_t y=0; y < h; y++) {
896        uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, lod + 1, face);
897        const uint8_t *i1 = GetOffsetPtr(alloc, 0, y*2, lod, face);
898        const uint8_t *i2 = GetOffsetPtr(alloc, 0, y*2+1, lod, face);
899
900        for (uint32_t x=0; x < w; x++) {
901            *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
902            oPtr ++;
903            i1 += 2;
904            i2 += 2;
905        }
906    }
907}
908
909void rsdAllocationGenerateMipmaps(const Context *rsc, const Allocation *alloc) {
910    if(!alloc->mHal.drvState.lod[0].mallocPtr) {
911        return;
912    }
913    uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
914    for (uint32_t face = 0; face < numFaces; face ++) {
915        for (uint32_t lod=0; lod < (alloc->getType()->getLODCount() -1); lod++) {
916            switch (alloc->getType()->getElement()->getSizeBits()) {
917            case 32:
918                mip8888(alloc, lod, (RsAllocationCubemapFace)face);
919                break;
920            case 16:
921                mip565(alloc, lod, (RsAllocationCubemapFace)face);
922                break;
923            case 8:
924                mip8(alloc, lod, (RsAllocationCubemapFace)face);
925                break;
926            }
927        }
928    }
929}
930
931
932