1 /*
2  * Copyright (C) 2016 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include "rsovAllocation.h"
18 
19 #include "rsAllocation.h"
20 #include "rsContext.h"
21 #include "rsCppUtils.h"
22 #include "rsElement.h"
23 #include "rsType.h"
24 #include "rsovContext.h"
25 #include "rsovCore.h"
26 
27 namespace android {
28 namespace renderscript {
29 namespace rsov {
30 
31 namespace {
32 
DeriveYUVLayout(int yuv,Allocation::Hal::DrvState * state)33 size_t DeriveYUVLayout(int yuv, Allocation::Hal::DrvState *state) {
34   // For the flexible YCbCr format, layout is initialized during call to
35   // Allocation::ioReceive.  Return early and avoid clobberring any
36   // pre-existing layout.
37   if (yuv == HAL_PIXEL_FORMAT_YCbCr_420_888) {
38     return 0;
39   }
40 
41   // YUV only supports basic 2d
42   // so we can stash the plane pointers in the mipmap levels.
43   size_t uvSize = 0;
44   state->lod[1].dimX = state->lod[0].dimX / 2;
45   state->lod[1].dimY = state->lod[0].dimY / 2;
46   state->lod[2].dimX = state->lod[0].dimX / 2;
47   state->lod[2].dimY = state->lod[0].dimY / 2;
48   state->yuv.shift = 1;
49   state->yuv.step = 1;
50   state->lodCount = 3;
51 
52   switch (yuv) {
53     case HAL_PIXEL_FORMAT_YV12:
54       state->lod[2].stride = rsRound(state->lod[0].stride >> 1, 16);
55       state->lod[2].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
56                                 (state->lod[0].stride * state->lod[0].dimY);
57       uvSize += state->lod[2].stride * state->lod[2].dimY;
58 
59       state->lod[1].stride = state->lod[2].stride;
60       state->lod[1].mallocPtr = ((uint8_t *)state->lod[2].mallocPtr) +
61                                 (state->lod[2].stride * state->lod[2].dimY);
62       uvSize += state->lod[1].stride * state->lod[2].dimY;
63       break;
64     case HAL_PIXEL_FORMAT_YCrCb_420_SP:  // NV21
65       // state->lod[1].dimX = state->lod[0].dimX;
66       state->lod[1].stride = state->lod[0].stride;
67       state->lod[2].stride = state->lod[0].stride;
68       state->lod[2].mallocPtr = ((uint8_t *)state->lod[0].mallocPtr) +
69                                 (state->lod[0].stride * state->lod[0].dimY);
70       state->lod[1].mallocPtr = ((uint8_t *)state->lod[2].mallocPtr) + 1;
71       uvSize += state->lod[1].stride * state->lod[1].dimY;
72       state->yuv.step = 2;
73       break;
74     default:
75       rsAssert(0);
76   }
77 
78   return uvSize;
79 }
80 
81 // TODO: Dedup this with the same code under frameworks/rs/driver
AllocationBuildPointerTable(const Context * rsc,const Allocation * alloc,const Type * type,uint8_t * ptr,size_t requiredAlignment)82 size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc,
83                                    const Type *type, uint8_t *ptr,
84                                    size_t requiredAlignment) {
85   alloc->mHal.drvState.lod[0].dimX = type->getDimX();
86   alloc->mHal.drvState.lod[0].dimY = type->getDimY();
87   alloc->mHal.drvState.lod[0].dimZ = type->getDimZ();
88   alloc->mHal.drvState.lod[0].mallocPtr = 0;
89   // Stride needs to be aligned to a boundary defined by requiredAlignment!
90   size_t stride =
91       alloc->mHal.drvState.lod[0].dimX * type->getElementSizeBytes();
92   alloc->mHal.drvState.lod[0].stride = rsRound(stride, requiredAlignment);
93   alloc->mHal.drvState.lodCount = type->getLODCount();
94   alloc->mHal.drvState.faceCount = type->getDimFaces();
95 
96   size_t offsets[Allocation::MAX_LOD];
97   memset(offsets, 0, sizeof(offsets));
98 
99   size_t o = alloc->mHal.drvState.lod[0].stride *
100              rsMax(alloc->mHal.drvState.lod[0].dimY, 1u) *
101              rsMax(alloc->mHal.drvState.lod[0].dimZ, 1u);
102   if (alloc->mHal.state.yuv) {
103     o += DeriveYUVLayout(alloc->mHal.state.yuv, &alloc->mHal.drvState);
104 
105     for (uint32_t ct = 1; ct < alloc->mHal.drvState.lodCount; ct++) {
106       offsets[ct] = (size_t)alloc->mHal.drvState.lod[ct].mallocPtr;
107     }
108   } else if (alloc->mHal.drvState.lodCount > 1) {
109     uint32_t tx = alloc->mHal.drvState.lod[0].dimX;
110     uint32_t ty = alloc->mHal.drvState.lod[0].dimY;
111     uint32_t tz = alloc->mHal.drvState.lod[0].dimZ;
112     for (uint32_t lod = 1; lod < alloc->mHal.drvState.lodCount; lod++) {
113       alloc->mHal.drvState.lod[lod].dimX = tx;
114       alloc->mHal.drvState.lod[lod].dimY = ty;
115       alloc->mHal.drvState.lod[lod].dimZ = tz;
116       alloc->mHal.drvState.lod[lod].stride =
117           rsRound(tx * type->getElementSizeBytes(), requiredAlignment);
118       offsets[lod] = o;
119       o += alloc->mHal.drvState.lod[lod].stride * rsMax(ty, 1u) * rsMax(tz, 1u);
120       if (tx > 1) tx >>= 1;
121       if (ty > 1) ty >>= 1;
122       if (tz > 1) tz >>= 1;
123     }
124   }
125 
126   alloc->mHal.drvState.faceOffset = o;
127 
128   alloc->mHal.drvState.lod[0].mallocPtr = ptr;
129   for (uint32_t lod = 1; lod < alloc->mHal.drvState.lodCount; lod++) {
130     alloc->mHal.drvState.lod[lod].mallocPtr = ptr + offsets[lod];
131   }
132 
133   size_t allocSize = alloc->mHal.drvState.faceOffset;
134   if (alloc->mHal.drvState.faceCount) {
135     allocSize *= 6;
136   }
137 
138   return allocSize;
139 }
140 
AllocationBuildPointerTable(const Context * rsc,const Allocation * alloc,const Type * type,uint8_t * ptr)141 size_t AllocationBuildPointerTable(const Context *rsc, const Allocation *alloc,
142                                    const Type *type, uint8_t *ptr) {
143   return AllocationBuildPointerTable(rsc, alloc, type, ptr,
144                                      Allocation::kMinimumRSAlignment);
145 }
146 
GetOffsetPtr(const Allocation * alloc,uint32_t xoff,uint32_t yoff,uint32_t zoff,uint32_t lod,RsAllocationCubemapFace face)147 uint8_t *GetOffsetPtr(const Allocation *alloc, uint32_t xoff, uint32_t yoff,
148                       uint32_t zoff, uint32_t lod,
149                       RsAllocationCubemapFace face) {
150   uint8_t *ptr = (uint8_t *)alloc->mHal.drvState.lod[lod].mallocPtr;
151   ptr += face * alloc->mHal.drvState.faceOffset;
152   ptr += zoff * alloc->mHal.drvState.lod[lod].dimY *
153          alloc->mHal.drvState.lod[lod].stride;
154   ptr += yoff * alloc->mHal.drvState.lod[lod].stride;
155   ptr += xoff * alloc->mHal.state.elementSizeBytes;
156   return ptr;
157 }
158 
mip565(const Allocation * alloc,int lod,RsAllocationCubemapFace face)159 void mip565(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
160   uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
161   uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
162 
163   for (uint32_t y = 0; y < h; y++) {
164     uint16_t *oPtr = (uint16_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
165     const uint16_t *i1 =
166         (uint16_t *)GetOffsetPtr(alloc, 0, 0, y * 2, lod, face);
167     const uint16_t *i2 =
168         (uint16_t *)GetOffsetPtr(alloc, 0, 0, y * 2 + 1, lod, face);
169 
170     for (uint32_t x = 0; x < w; x++) {
171       *oPtr = rsBoxFilter565(i1[0], i1[1], i2[0], i2[1]);
172       oPtr++;
173       i1 += 2;
174       i2 += 2;
175     }
176   }
177 }
178 
mip8888(const Allocation * alloc,int lod,RsAllocationCubemapFace face)179 void mip8888(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
180   uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
181   uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
182 
183   for (uint32_t y = 0; y < h; y++) {
184     uint32_t *oPtr = (uint32_t *)GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
185     const uint32_t *i1 =
186         (uint32_t *)GetOffsetPtr(alloc, 0, y * 2, 0, lod, face);
187     const uint32_t *i2 =
188         (uint32_t *)GetOffsetPtr(alloc, 0, y * 2 + 1, 0, lod, face);
189 
190     for (uint32_t x = 0; x < w; x++) {
191       *oPtr = rsBoxFilter8888(i1[0], i1[1], i2[0], i2[1]);
192       oPtr++;
193       i1 += 2;
194       i2 += 2;
195     }
196   }
197 }
198 
mip8(const Allocation * alloc,int lod,RsAllocationCubemapFace face)199 void mip8(const Allocation *alloc, int lod, RsAllocationCubemapFace face) {
200   uint32_t w = alloc->mHal.drvState.lod[lod + 1].dimX;
201   uint32_t h = alloc->mHal.drvState.lod[lod + 1].dimY;
202 
203   for (uint32_t y = 0; y < h; y++) {
204     uint8_t *oPtr = GetOffsetPtr(alloc, 0, y, 0, lod + 1, face);
205     const uint8_t *i1 = GetOffsetPtr(alloc, 0, y * 2, 0, lod, face);
206     const uint8_t *i2 = GetOffsetPtr(alloc, 0, y * 2 + 1, 0, lod, face);
207 
208     for (uint32_t x = 0; x < w; x++) {
209       *oPtr = (uint8_t)(((uint32_t)i1[0] + i1[1] + i2[0] + i2[1]) * 0.25f);
210       oPtr++;
211       i1 += 2;
212       i2 += 2;
213     }
214   }
215 }
216 
217 }  // anonymous namespace
218 
RSoVAllocation(RSoVContext * context,const Type * type,size_t bufferSize)219 RSoVAllocation::RSoVAllocation(RSoVContext *context, const Type *type,
220                                size_t bufferSize)
221     : mBuffer(new RSoVBuffer(context, bufferSize)),
222       mType(type),
223       mWidth(type->getDimX()),
224       mHeight(type->getDimY()),
225       mDepth(type->getDimZ()) {}
226 
RSoVBuffer(RSoVContext * context,size_t size)227 RSoVBuffer::RSoVBuffer(RSoVContext *context, size_t size)
228     : mRSoV(context), mDevice(context->getDevice()) {
229   InitBuffer(size);
230 }
231 
~RSoVBuffer()232 RSoVBuffer::~RSoVBuffer() {
233   vkUnmapMemory(mDevice, mMem);
234   vkDestroyBuffer(mDevice, mBuf, nullptr);
235   vkFreeMemory(mDevice, mMem, nullptr);
236 }
237 
InitBuffer(size_t bufferSize)238 void RSoVBuffer::InitBuffer(size_t bufferSize) {
239   VkResult res;
240 
241   VkBufferCreateInfo buf_info = {
242       .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
243       .pNext = nullptr,
244       .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
245       .size = bufferSize,
246       .queueFamilyIndexCount = 0,
247       .pQueueFamilyIndices = nullptr,
248       .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
249       .flags = 0,
250   };
251   res = vkCreateBuffer(mDevice, &buf_info, nullptr, &mBuf);
252   rsAssert(res == VK_SUCCESS);
253 
254   VkMemoryRequirements mem_reqs;
255   vkGetBufferMemoryRequirements(mDevice, mBuf, &mem_reqs);
256 
257   VkMemoryAllocateInfo allocateInfo = {
258       .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
259       .pNext = nullptr,
260       .memoryTypeIndex = 0,
261       .allocationSize = mem_reqs.size,
262   };
263 
264   bool pass;
265   pass =
266       mRSoV->MemoryTypeFromProperties(mem_reqs.memoryTypeBits,
267                                       VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
268                                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
269                                       &allocateInfo.memoryTypeIndex);
270   ALOGV("TypeBits = 0x%08X", mem_reqs.memoryTypeBits);
271   rsAssert(pass);
272 
273   // TODO: Make this aligned
274   res = vkAllocateMemory(mDevice, &allocateInfo, nullptr, &mMem);
275   rsAssert(res == VK_SUCCESS);
276 
277   res = vkBindBufferMemory(mDevice, mBuf, mMem, 0);
278   rsAssert(res == VK_SUCCESS);
279 
280   mBufferInfo.buffer = mBuf;
281   mBufferInfo.offset = 0;
282   mBufferInfo.range = bufferSize;
283 
284   res = vkMapMemory(mDevice, mMem, 0, mem_reqs.size, 0, (void **)&mPtr);
285   rsAssert(res == VK_SUCCESS);
286 }
287 
288 }  // namespace rsov
289 }  // namespace renderscript
290 }  // namespace android
291 
292 using android::renderscript::Allocation;
293 using android::renderscript::Context;
294 using android::renderscript::Element;
295 using android::renderscript::Type;
296 using android::renderscript::rs_allocation;
297 using android::renderscript::rsMax;
298 using namespace android::renderscript::rsov;
299 
rsovAllocationInit(const Context * rsc,Allocation * alloc,bool forceZero)300 bool rsovAllocationInit(const Context *rsc, Allocation *alloc, bool forceZero) {
301   RSoVHal *hal = static_cast<RSoVHal *>(rsc->mHal.drv);
302   RSoVContext *rsov = hal->mRSoV;
303   const Type *type = alloc->getType();
304 
305   // Calculate the object size.
306   size_t allocSize = AllocationBuildPointerTable(rsc, alloc, type, nullptr);
307   RSoVAllocation *rsovAlloc = new RSoVAllocation(rsov, type, allocSize);
308   alloc->mHal.drv = rsovAlloc;
309   AllocationBuildPointerTable(rsc, alloc, type,
310                               (uint8_t *)rsovAlloc->getHostPtr());
311   return true;
312 }
313 
rsovAllocationDestroy(const Context * rsc,Allocation * alloc)314 void rsovAllocationDestroy(const Context *rsc, Allocation *alloc) {
315   RSoVAllocation *rsovAlloc = static_cast<RSoVAllocation *>(alloc->mHal.drv);
316   delete rsovAlloc;
317   alloc->mHal.drv = nullptr;
318 }
319 
rsovAllocationData1D(const Context * rsc,const Allocation * alloc,uint32_t xoff,uint32_t lod,size_t count,const void * data,size_t sizeBytes)320 void rsovAllocationData1D(const Context *rsc, const Allocation *alloc,
321                           uint32_t xoff, uint32_t lod, size_t count,
322                           const void *data, size_t sizeBytes) {
323   const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
324   uint8_t *ptr =
325       GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
326   size_t size = count * eSize;
327   if (ptr != data) {
328     // Skip the copy if we are the same allocation. This can arise from
329     // our Bitmap optimization, where we share the same storage.
330     if (alloc->mHal.state.hasReferences) {
331       alloc->incRefs(data, count);
332       alloc->decRefs(ptr, count);
333     }
334     memcpy(ptr, data, size);
335   }
336 }
337 
rsovAllocationData2D(const Context * rsc,const Allocation * alloc,uint32_t xoff,uint32_t yoff,uint32_t lod,RsAllocationCubemapFace face,uint32_t w,uint32_t h,const void * data,size_t sizeBytes,size_t stride)338 void rsovAllocationData2D(const Context *rsc, const Allocation *alloc,
339                           uint32_t xoff, uint32_t yoff, uint32_t lod,
340                           RsAllocationCubemapFace face, uint32_t w, uint32_t h,
341                           const void *data, size_t sizeBytes, size_t stride) {
342   size_t eSize = alloc->mHal.state.elementSizeBytes;
343   size_t lineSize = eSize * w;
344   if (!stride) {
345     stride = lineSize;
346   }
347 
348   if (alloc->mHal.drvState.lod[0].mallocPtr) {
349     const uint8_t *src = static_cast<const uint8_t *>(data);
350     uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
351 
352     for (uint32_t line = yoff; line < (yoff + h); line++) {
353       if (alloc->mHal.state.hasReferences) {
354         alloc->incRefs(src, w);
355         alloc->decRefs(dst, w);
356       }
357       memcpy(dst, src, lineSize);
358       src += stride;
359       dst += alloc->mHal.drvState.lod[lod].stride;
360     }
361     // TODO: handle YUV Allocations
362     if (alloc->mHal.state.yuv) {
363       size_t clineSize = lineSize;
364       int lod = 1;
365       int maxLod = 2;
366       if (alloc->mHal.state.yuv == HAL_PIXEL_FORMAT_YV12) {
367         maxLod = 3;
368         clineSize >>= 1;
369       } else if (alloc->mHal.state.yuv == HAL_PIXEL_FORMAT_YCrCb_420_SP) {
370         lod = 2;
371         maxLod = 3;
372       }
373 
374       while (lod < maxLod) {
375         uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
376 
377         for (uint32_t line = (yoff >> 1); line < ((yoff + h) >> 1); line++) {
378           memcpy(dst, src, clineSize);
379           // When copying from an array to an Allocation, the src pointer
380           // to the array should just move by the number of bytes copied.
381           src += clineSize;
382           dst += alloc->mHal.drvState.lod[lod].stride;
383         }
384         lod++;
385       }
386     }
387   }
388 }
389 
rsovAllocationData3D(const Context * rsc,const Allocation * alloc,uint32_t xoff,uint32_t yoff,uint32_t zoff,uint32_t lod,uint32_t w,uint32_t h,uint32_t d,const void * data,size_t sizeBytes,size_t stride)390 void rsovAllocationData3D(const Context *rsc, const Allocation *alloc,
391                           uint32_t xoff, uint32_t yoff, uint32_t zoff,
392                           uint32_t lod, uint32_t w, uint32_t h, uint32_t d,
393                           const void *data, size_t sizeBytes, size_t stride) {
394   uint32_t eSize = alloc->mHal.state.elementSizeBytes;
395   uint32_t lineSize = eSize * w;
396   if (!stride) {
397     stride = lineSize;
398   }
399 
400   if (alloc->mHal.drvState.lod[0].mallocPtr) {
401     const uint8_t *src = static_cast<const uint8_t *>(data);
402     for (uint32_t z = zoff; z < (d + zoff); z++) {
403       uint8_t *dst = GetOffsetPtr(alloc, xoff, yoff, z, lod,
404                                   RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
405       for (uint32_t line = yoff; line < (yoff + h); line++) {
406         if (alloc->mHal.state.hasReferences) {
407           alloc->incRefs(src, w);
408           alloc->decRefs(dst, w);
409         }
410         memcpy(dst, src, lineSize);
411         src += stride;
412         dst += alloc->mHal.drvState.lod[lod].stride;
413       }
414     }
415   }
416 }
417 
rsovAllocationRead1D(const Context * rsc,const Allocation * alloc,uint32_t xoff,uint32_t lod,size_t count,void * data,size_t sizeBytes)418 void rsovAllocationRead1D(const Context *rsc, const Allocation *alloc,
419                           uint32_t xoff, uint32_t lod, size_t count, void *data,
420                           size_t sizeBytes) {
421   const size_t eSize = alloc->mHal.state.type->getElementSizeBytes();
422   const uint8_t *ptr =
423       GetOffsetPtr(alloc, xoff, 0, 0, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
424   if (data != ptr) {
425     // Skip the copy if we are the same allocation. This can arise from
426     // our Bitmap optimization, where we share the same storage.
427     memcpy(data, ptr, count * eSize);
428   }
429 }
430 
rsovAllocationRead2D(const Context * rsc,const Allocation * alloc,uint32_t xoff,uint32_t yoff,uint32_t lod,RsAllocationCubemapFace face,uint32_t w,uint32_t h,void * data,size_t sizeBytes,size_t stride)431 void rsovAllocationRead2D(const Context *rsc, const Allocation *alloc,
432                           uint32_t xoff, uint32_t yoff, uint32_t lod,
433                           RsAllocationCubemapFace face, uint32_t w, uint32_t h,
434                           void *data, size_t sizeBytes, size_t stride) {
435   size_t eSize = alloc->mHal.state.elementSizeBytes;
436   size_t lineSize = eSize * w;
437   if (!stride) {
438     stride = lineSize;
439   }
440 
441   if (alloc->mHal.drvState.lod[0].mallocPtr) {
442     uint8_t *dst = static_cast<uint8_t *>(data);
443     const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, 0, lod, face);
444     if (dst == src) {
445       // Skip the copy if we are the same allocation. This can arise from
446       // our Bitmap optimization, where we share the same storage.
447       return;
448     }
449 
450     for (uint32_t line = yoff; line < (yoff + h); line++) {
451       memcpy(dst, src, lineSize);
452       dst += stride;
453       src += alloc->mHal.drvState.lod[lod].stride;
454     }
455   } else {
456     ALOGE("Add code to readback from non-script memory");
457   }
458 }
459 
rsovAllocationRead3D(const Context * rsc,const Allocation * alloc,uint32_t xoff,uint32_t yoff,uint32_t zoff,uint32_t lod,uint32_t w,uint32_t h,uint32_t d,void * data,size_t sizeBytes,size_t stride)460 void rsovAllocationRead3D(const Context *rsc, const Allocation *alloc,
461                           uint32_t xoff, uint32_t yoff, uint32_t zoff,
462                           uint32_t lod, uint32_t w, uint32_t h, uint32_t d,
463                           void *data, size_t sizeBytes, size_t stride) {
464   uint32_t eSize = alloc->mHal.state.elementSizeBytes;
465   uint32_t lineSize = eSize * w;
466   if (!stride) {
467     stride = lineSize;
468   }
469 
470   if (alloc->mHal.drvState.lod[0].mallocPtr) {
471     uint8_t *dst = static_cast<uint8_t *>(data);
472     for (uint32_t z = zoff; z < (d + zoff); z++) {
473       const uint8_t *src = GetOffsetPtr(alloc, xoff, yoff, z, lod,
474                                         RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
475       if (dst == src) {
476         // Skip the copy if we are the same allocation. This can arise from
477         // our Bitmap optimization, where we share the same storage.
478         return;
479       }
480 
481       for (uint32_t line = yoff; line < (yoff + h); line++) {
482         memcpy(dst, src, lineSize);
483         dst += stride;
484         src += alloc->mHal.drvState.lod[lod].stride;
485       }
486     }
487   }
488 }
489 
rsovAllocationLock1D(const Context * rsc,const Allocation * alloc)490 void *rsovAllocationLock1D(const Context *rsc, const Allocation *alloc) {
491   return alloc->mHal.drvState.lod[0].mallocPtr;
492 }
493 
rsovAllocationUnlock1D(const Context * rsc,const Allocation * alloc)494 void rsovAllocationUnlock1D(const Context *rsc, const Allocation *alloc) {}
495 
rsovAllocationData1D_alloc(const Context * rsc,const Allocation * dstAlloc,uint32_t dstXoff,uint32_t dstLod,size_t count,const Allocation * srcAlloc,uint32_t srcXoff,uint32_t srcLod)496 void rsovAllocationData1D_alloc(const Context *rsc, const Allocation *dstAlloc,
497                                 uint32_t dstXoff, uint32_t dstLod, size_t count,
498                                 const Allocation *srcAlloc, uint32_t srcXoff,
499                                 uint32_t srcLod) {}
500 
rsovAllocationData2D_alloc_script(const Context * rsc,const Allocation * dstAlloc,uint32_t dstXoff,uint32_t dstYoff,uint32_t dstLod,RsAllocationCubemapFace dstFace,uint32_t w,uint32_t h,const Allocation * srcAlloc,uint32_t srcXoff,uint32_t srcYoff,uint32_t srcLod,RsAllocationCubemapFace srcFace)501 void rsovAllocationData2D_alloc_script(
502     const Context *rsc, const Allocation *dstAlloc, uint32_t dstXoff,
503     uint32_t dstYoff, uint32_t dstLod, RsAllocationCubemapFace dstFace,
504     uint32_t w, uint32_t h, const Allocation *srcAlloc, uint32_t srcXoff,
505     uint32_t srcYoff, uint32_t srcLod, RsAllocationCubemapFace srcFace) {
506   size_t elementSize = dstAlloc->getType()->getElementSizeBytes();
507   for (uint32_t i = 0; i < h; i++) {
508     uint8_t *dstPtr =
509         GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, 0, dstLod, dstFace);
510     uint8_t *srcPtr =
511         GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, 0, srcLod, srcFace);
512     memcpy(dstPtr, srcPtr, w * elementSize);
513   }
514 }
515 
rsovAllocationData3D_alloc_script(const Context * rsc,const Allocation * dstAlloc,uint32_t dstXoff,uint32_t dstYoff,uint32_t dstZoff,uint32_t dstLod,uint32_t w,uint32_t h,uint32_t d,const Allocation * srcAlloc,uint32_t srcXoff,uint32_t srcYoff,uint32_t srcZoff,uint32_t srcLod)516 void rsovAllocationData3D_alloc_script(
517     const Context *rsc, const Allocation *dstAlloc, uint32_t dstXoff,
518     uint32_t dstYoff, uint32_t dstZoff, uint32_t dstLod, uint32_t w, uint32_t h,
519     uint32_t d, const Allocation *srcAlloc, uint32_t srcXoff, uint32_t srcYoff,
520     uint32_t srcZoff, uint32_t srcLod) {
521   uint32_t elementSize = dstAlloc->getType()->getElementSizeBytes();
522   for (uint32_t j = 0; j < d; j++) {
523     for (uint32_t i = 0; i < h; i++) {
524       uint8_t *dstPtr =
525           GetOffsetPtr(dstAlloc, dstXoff, dstYoff + i, dstZoff + j, dstLod,
526                        RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
527       uint8_t *srcPtr =
528           GetOffsetPtr(srcAlloc, srcXoff, srcYoff + i, srcZoff + j, srcLod,
529                        RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
530       memcpy(dstPtr, srcPtr, w * elementSize);
531     }
532   }
533 }
534 
rsovAllocationData2D_alloc(const Context * rsc,const Allocation * dstAlloc,uint32_t dstXoff,uint32_t dstYoff,uint32_t dstLod,RsAllocationCubemapFace dstFace,uint32_t w,uint32_t h,const Allocation * srcAlloc,uint32_t srcXoff,uint32_t srcYoff,uint32_t srcLod,RsAllocationCubemapFace srcFace)535 void rsovAllocationData2D_alloc(
536     const Context *rsc, const Allocation *dstAlloc, uint32_t dstXoff,
537     uint32_t dstYoff, uint32_t dstLod, RsAllocationCubemapFace dstFace,
538     uint32_t w, uint32_t h, const Allocation *srcAlloc, uint32_t srcXoff,
539     uint32_t srcYoff, uint32_t srcLod, RsAllocationCubemapFace srcFace) {
540   if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
541     rsc->setError(RS_ERROR_FATAL_DRIVER,
542                   "Non-script allocation copies not "
543                   "yet implemented.");
544     return;
545   }
546   rsovAllocationData2D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstLod,
547                                     dstFace, w, h, srcAlloc, srcXoff, srcYoff,
548                                     srcLod, srcFace);
549 }
550 
rsovAllocationData3D_alloc(const Context * rsc,const Allocation * dstAlloc,uint32_t dstXoff,uint32_t dstYoff,uint32_t dstZoff,uint32_t dstLod,uint32_t w,uint32_t h,uint32_t d,const Allocation * srcAlloc,uint32_t srcXoff,uint32_t srcYoff,uint32_t srcZoff,uint32_t srcLod)551 void rsovAllocationData3D_alloc(const Context *rsc, const Allocation *dstAlloc,
552                                 uint32_t dstXoff, uint32_t dstYoff,
553                                 uint32_t dstZoff, uint32_t dstLod, uint32_t w,
554                                 uint32_t h, uint32_t d,
555                                 const Allocation *srcAlloc, uint32_t srcXoff,
556                                 uint32_t srcYoff, uint32_t srcZoff,
557                                 uint32_t srcLod) {
558   if (!dstAlloc->getIsScript() && !srcAlloc->getIsScript()) {
559     rsc->setError(RS_ERROR_FATAL_DRIVER,
560                   "Non-script allocation copies not "
561                   "yet implemented.");
562     return;
563   }
564   rsovAllocationData3D_alloc_script(rsc, dstAlloc, dstXoff, dstYoff, dstZoff,
565                                     dstLod, w, h, d, srcAlloc, srcXoff, srcYoff,
566                                     srcZoff, srcLod);
567 }
568 
rsovAllocationAdapterOffset(const Context * rsc,const Allocation * alloc)569 void rsovAllocationAdapterOffset(const Context *rsc, const Allocation *alloc) {
570   // Get a base pointer to the new LOD
571   const Allocation *base = alloc->mHal.state.baseAlloc;
572   const Type *type = alloc->mHal.state.type;
573   if (base == nullptr) {
574     return;
575   }
576 
577   const int lodBias = alloc->mHal.state.originLOD;
578   uint32_t lodCount = rsMax(alloc->mHal.drvState.lodCount, (uint32_t)1);
579   for (uint32_t lod = 0; lod < lodCount; lod++) {
580     alloc->mHal.drvState.lod[lod] = base->mHal.drvState.lod[lod + lodBias];
581     alloc->mHal.drvState.lod[lod].mallocPtr = GetOffsetPtr(
582         alloc, alloc->mHal.state.originX, alloc->mHal.state.originY,
583         alloc->mHal.state.originZ, lodBias,
584         (RsAllocationCubemapFace)alloc->mHal.state.originFace);
585   }
586 }
587 
rsovAllocationAdapterInit(const Context * rsc,Allocation * alloc)588 bool rsovAllocationAdapterInit(const Context *rsc, Allocation *alloc) {
589 // TODO: may need a RSoV Allocation here
590 #if 0
591     DrvAllocation *drv = (DrvAllocation *)calloc(1, sizeof(DrvAllocation));
592     if (!drv) {
593         return false;
594     }
595     alloc->mHal.drv = drv;
596 #endif
597   // We need to build an allocation that looks like a subset of the parent
598   // allocation
599   rsovAllocationAdapterOffset(rsc, alloc);
600 
601   return true;
602 }
603 
rsovAllocationSyncAll(const Context * rsc,const Allocation * alloc,RsAllocationUsageType src)604 void rsovAllocationSyncAll(const Context *rsc, const Allocation *alloc,
605                            RsAllocationUsageType src) {
606   // TODO: anything to do here?
607 }
608 
rsovAllocationMarkDirty(const Context * rsc,const Allocation * alloc)609 void rsovAllocationMarkDirty(const Context *rsc, const Allocation *alloc) {
610   // TODO: anything to do here?
611 }
612 
rsovAllocationResize(const Context * rsc,const Allocation * alloc,const Type * newType,bool zeroNew)613 void rsovAllocationResize(const Context *rsc, const Allocation *alloc,
614                           const Type *newType, bool zeroNew) {
615   // TODO: implement this
616   // can this be done without copying, if the new size is greater than the
617   // original?
618 }
619 
rsovAllocationGenerateMipmaps(const Context * rsc,const Allocation * alloc)620 void rsovAllocationGenerateMipmaps(const Context *rsc,
621                                    const Allocation *alloc) {
622   if (!alloc->mHal.drvState.lod[0].mallocPtr) {
623     return;
624   }
625   uint32_t numFaces = alloc->getType()->getDimFaces() ? 6 : 1;
626   for (uint32_t face = 0; face < numFaces; face++) {
627     for (uint32_t lod = 0; lod < (alloc->getType()->getLODCount() - 1); lod++) {
628       switch (alloc->getType()->getElement()->getSizeBits()) {
629         case 32:
630           mip8888(alloc, lod, (RsAllocationCubemapFace)face);
631           break;
632         case 16:
633           mip565(alloc, lod, (RsAllocationCubemapFace)face);
634           break;
635         case 8:
636           mip8(alloc, lod, (RsAllocationCubemapFace)face);
637           break;
638       }
639     }
640   }
641 }
642 
rsovAllocationGrallocBits(const Context * rsc,Allocation * alloc)643 uint32_t rsovAllocationGrallocBits(const Context *rsc, Allocation *alloc) {
644   return 0;
645 }
646 
rsovAllocationUpdateCachedObject(const Context * rsc,const Allocation * alloc,rs_allocation * obj)647 void rsovAllocationUpdateCachedObject(const Context *rsc,
648                                       const Allocation *alloc,
649                                       rs_allocation *obj) {
650   obj->p = alloc;
651 #ifdef __LP64__
652   obj->unused1 = nullptr;
653   obj->unused2 = nullptr;
654   obj->unused3 = nullptr;
655 #endif
656 }
657 
rsovAllocationSetSurface(const Context * rsc,Allocation * alloc,ANativeWindow * nw)658 void rsovAllocationSetSurface(const Context *rsc, Allocation *alloc,
659                               ANativeWindow *nw) {
660   // TODO: implement this
661 }
662 
rsovAllocationIoSend(const Context * rsc,Allocation * alloc)663 void rsovAllocationIoSend(const Context *rsc, Allocation *alloc) {
664   // TODO: implement this
665 }
666 
rsovAllocationIoReceive(const Context * rsc,Allocation * alloc)667 void rsovAllocationIoReceive(const Context *rsc, Allocation *alloc) {
668   // TODO: implement this
669 }
670 
rsovAllocationElementData(const Context * rsc,const Allocation * alloc,uint32_t x,uint32_t y,uint32_t z,const void * data,uint32_t cIdx,size_t sizeBytes)671 void rsovAllocationElementData(const Context *rsc, const Allocation *alloc,
672                                uint32_t x, uint32_t y, uint32_t z,
673                                const void *data, uint32_t cIdx,
674                                size_t sizeBytes) {
675   uint8_t *ptr =
676       GetOffsetPtr(alloc, x, y, z, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
677 
678   const Element *e = alloc->mHal.state.type->getElement()->getField(cIdx);
679   ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
680 
681   if (alloc->mHal.state.hasReferences) {
682     e->incRefs(data);
683     e->decRefs(ptr);
684   }
685 
686   memcpy(ptr, data, sizeBytes);
687 }
688 
rsovAllocationElementRead(const Context * rsc,const Allocation * alloc,uint32_t x,uint32_t y,uint32_t z,void * data,uint32_t cIdx,size_t sizeBytes)689 void rsovAllocationElementRead(const Context *rsc, const Allocation *alloc,
690                                uint32_t x, uint32_t y, uint32_t z, void *data,
691                                uint32_t cIdx, size_t sizeBytes) {
692   uint8_t *ptr =
693       GetOffsetPtr(alloc, x, y, z, 0, RS_ALLOCATION_CUBEMAP_FACE_POSITIVE_X);
694 
695   const Element *e = alloc->mHal.state.type->getElement()->getField(cIdx);
696   ptr += alloc->mHal.state.type->getElement()->getFieldOffsetBytes(cIdx);
697 
698   memcpy(data, ptr, sizeBytes);
699 }
700