1 // Copyright 2018 The SwiftShader Authors. All Rights Reserved.
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 //    http://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 
15 #include "VkImage.hpp"
16 
17 #include "VkBuffer.hpp"
18 #include "VkDevice.hpp"
19 #include "VkDeviceMemory.hpp"
20 #include "Device/ASTC_Decoder.hpp"
21 #include "Device/BC_Decoder.hpp"
22 #include "Device/Blitter.hpp"
23 #include "Device/ETC_Decoder.hpp"
24 
25 #ifdef __ANDROID__
26 #	include "System/GrallocAndroid.hpp"
27 #endif
28 
29 #include <cstring>
30 
31 namespace {
32 
GetInputType(const vk::Format & format)33 ETC_Decoder::InputType GetInputType(const vk::Format &format)
34 {
35 	switch(format)
36 	{
37 		case VK_FORMAT_EAC_R11_UNORM_BLOCK:
38 			return ETC_Decoder::ETC_R_UNSIGNED;
39 		case VK_FORMAT_EAC_R11_SNORM_BLOCK:
40 			return ETC_Decoder::ETC_R_SIGNED;
41 		case VK_FORMAT_EAC_R11G11_UNORM_BLOCK:
42 			return ETC_Decoder::ETC_RG_UNSIGNED;
43 		case VK_FORMAT_EAC_R11G11_SNORM_BLOCK:
44 			return ETC_Decoder::ETC_RG_SIGNED;
45 		case VK_FORMAT_ETC2_R8G8B8_UNORM_BLOCK:
46 		case VK_FORMAT_ETC2_R8G8B8_SRGB_BLOCK:
47 			return ETC_Decoder::ETC_RGB;
48 		case VK_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK:
49 		case VK_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK:
50 			return ETC_Decoder::ETC_RGB_PUNCHTHROUGH_ALPHA;
51 		case VK_FORMAT_ETC2_R8G8B8A8_UNORM_BLOCK:
52 		case VK_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK:
53 			return ETC_Decoder::ETC_RGBA;
54 		default:
55 			UNSUPPORTED("format: %d", int(format));
56 			return ETC_Decoder::ETC_RGBA;
57 	}
58 }
59 
GetBCn(const vk::Format & format)60 int GetBCn(const vk::Format &format)
61 {
62 	switch(format)
63 	{
64 		case VK_FORMAT_BC1_RGB_UNORM_BLOCK:
65 		case VK_FORMAT_BC1_RGBA_UNORM_BLOCK:
66 		case VK_FORMAT_BC1_RGB_SRGB_BLOCK:
67 		case VK_FORMAT_BC1_RGBA_SRGB_BLOCK:
68 			return 1;
69 		case VK_FORMAT_BC2_UNORM_BLOCK:
70 		case VK_FORMAT_BC2_SRGB_BLOCK:
71 			return 2;
72 		case VK_FORMAT_BC3_UNORM_BLOCK:
73 		case VK_FORMAT_BC3_SRGB_BLOCK:
74 			return 3;
75 		case VK_FORMAT_BC4_UNORM_BLOCK:
76 		case VK_FORMAT_BC4_SNORM_BLOCK:
77 			return 4;
78 		case VK_FORMAT_BC5_UNORM_BLOCK:
79 		case VK_FORMAT_BC5_SNORM_BLOCK:
80 			return 5;
81 		case VK_FORMAT_BC6H_UFLOAT_BLOCK:
82 		case VK_FORMAT_BC6H_SFLOAT_BLOCK:
83 			return 6;
84 		case VK_FORMAT_BC7_UNORM_BLOCK:
85 		case VK_FORMAT_BC7_SRGB_BLOCK:
86 			return 7;
87 		default:
88 			UNSUPPORTED("format: %d", int(format));
89 			return 0;
90 	}
91 }
92 
93 // Returns true for BC1 if we have an RGB format, false for RGBA
94 // Returns true for BC4 and BC5 if we have an unsigned format, false for signed
95 // Ignored by BC2, BC3, BC6 and BC7
GetNoAlphaOrUnsigned(const vk::Format & format)96 bool GetNoAlphaOrUnsigned(const vk::Format &format)
97 {
98 	switch(format)
99 	{
100 		case VK_FORMAT_BC1_RGB_UNORM_BLOCK:
101 		case VK_FORMAT_BC1_RGB_SRGB_BLOCK:
102 		case VK_FORMAT_BC4_UNORM_BLOCK:
103 		case VK_FORMAT_BC5_UNORM_BLOCK:
104 			return true;
105 		case VK_FORMAT_BC1_RGBA_UNORM_BLOCK:
106 		case VK_FORMAT_BC1_RGBA_SRGB_BLOCK:
107 		case VK_FORMAT_BC2_UNORM_BLOCK:
108 		case VK_FORMAT_BC2_SRGB_BLOCK:
109 		case VK_FORMAT_BC3_UNORM_BLOCK:
110 		case VK_FORMAT_BC3_SRGB_BLOCK:
111 		case VK_FORMAT_BC4_SNORM_BLOCK:
112 		case VK_FORMAT_BC5_SNORM_BLOCK:
113 		case VK_FORMAT_BC6H_UFLOAT_BLOCK:
114 		case VK_FORMAT_BC6H_SFLOAT_BLOCK:
115 		case VK_FORMAT_BC7_SRGB_BLOCK:
116 		case VK_FORMAT_BC7_UNORM_BLOCK:
117 			return false;
118 		default:
119 			UNSUPPORTED("format: %d", int(format));
120 			return false;
121 	}
122 }
123 
124 }  // anonymous namespace
125 
126 namespace vk {
127 
Image(const VkImageCreateInfo * pCreateInfo,void * mem,Device * device)128 Image::Image(const VkImageCreateInfo *pCreateInfo, void *mem, Device *device)
129     : device(device)
130     , flags(pCreateInfo->flags)
131     , imageType(pCreateInfo->imageType)
132     , format(pCreateInfo->format)
133     , extent(pCreateInfo->extent)
134     , mipLevels(pCreateInfo->mipLevels)
135     , arrayLayers(pCreateInfo->arrayLayers)
136     , samples(pCreateInfo->samples)
137     , tiling(pCreateInfo->tiling)
138     , usage(pCreateInfo->usage)
139 {
140 	if(format.isCompressed())
141 	{
142 		VkImageCreateInfo compressedImageCreateInfo = *pCreateInfo;
143 		compressedImageCreateInfo.format = format.getDecompressedFormat();
144 		decompressedImage = new(mem) Image(&compressedImageCreateInfo, nullptr, device);
145 	}
146 
147 	const auto *nextInfo = reinterpret_cast<const VkBaseInStructure *>(pCreateInfo->pNext);
148 	for(; nextInfo != nullptr; nextInfo = nextInfo->pNext)
149 	{
150 		if(nextInfo->sType == VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO)
151 		{
152 			const auto *externalInfo = reinterpret_cast<const VkExternalMemoryImageCreateInfo *>(nextInfo);
153 			supportedExternalMemoryHandleTypes = externalInfo->handleTypes;
154 		}
155 	}
156 }
157 
destroy(const VkAllocationCallbacks * pAllocator)158 void Image::destroy(const VkAllocationCallbacks *pAllocator)
159 {
160 	if(decompressedImage)
161 	{
162 		vk::deallocate(decompressedImage, pAllocator);
163 	}
164 }
165 
ComputeRequiredAllocationSize(const VkImageCreateInfo * pCreateInfo)166 size_t Image::ComputeRequiredAllocationSize(const VkImageCreateInfo *pCreateInfo)
167 {
168 	return Format(pCreateInfo->format).isCompressed() ? sizeof(Image) : 0;
169 }
170 
getMemoryRequirements() const171 const VkMemoryRequirements Image::getMemoryRequirements() const
172 {
173 	VkMemoryRequirements memoryRequirements;
174 	memoryRequirements.alignment = vk::REQUIRED_MEMORY_ALIGNMENT;
175 	memoryRequirements.memoryTypeBits = vk::MEMORY_TYPE_GENERIC_BIT;
176 	memoryRequirements.size = getStorageSize(format.getAspects()) +
177 	                          (decompressedImage ? decompressedImage->getStorageSize(decompressedImage->format.getAspects()) : 0);
178 	return memoryRequirements;
179 }
180 
getSizeInBytes(const VkImageSubresourceRange & subresourceRange) const181 size_t Image::getSizeInBytes(const VkImageSubresourceRange &subresourceRange) const
182 {
183 	size_t size = 0;
184 	uint32_t lastLayer = getLastLayerIndex(subresourceRange);
185 	uint32_t lastMipLevel = getLastMipLevel(subresourceRange);
186 	uint32_t layerCount = lastLayer - subresourceRange.baseArrayLayer + 1;
187 	uint32_t mipLevelCount = lastMipLevel - subresourceRange.baseMipLevel + 1;
188 
189 	auto aspect = static_cast<VkImageAspectFlagBits>(subresourceRange.aspectMask);
190 
191 	if(layerCount > 1)
192 	{
193 		if(mipLevelCount < mipLevels)  // Compute size for all layers except the last one, then add relevant mip level sizes only for last layer
194 		{
195 			size = (layerCount - 1) * getLayerSize(aspect);
196 			for(uint32_t mipLevel = subresourceRange.baseMipLevel; mipLevel <= lastMipLevel; ++mipLevel)
197 			{
198 				size += getMultiSampledLevelSize(aspect, mipLevel);
199 			}
200 		}
201 		else  // All mip levels used, compute full layer sizes
202 		{
203 			size = layerCount * getLayerSize(aspect);
204 		}
205 	}
206 	else  // Single layer, add all mip levels in the subresource range
207 	{
208 		for(uint32_t mipLevel = subresourceRange.baseMipLevel; mipLevel <= lastMipLevel; ++mipLevel)
209 		{
210 			size += getMultiSampledLevelSize(aspect, mipLevel);
211 		}
212 	}
213 
214 	return size;
215 }
216 
canBindToMemory(DeviceMemory * pDeviceMemory) const217 bool Image::canBindToMemory(DeviceMemory *pDeviceMemory) const
218 {
219 	return pDeviceMemory->checkExternalMemoryHandleType(supportedExternalMemoryHandleTypes);
220 }
221 
bind(DeviceMemory * pDeviceMemory,VkDeviceSize pMemoryOffset)222 void Image::bind(DeviceMemory *pDeviceMemory, VkDeviceSize pMemoryOffset)
223 {
224 	deviceMemory = pDeviceMemory;
225 	memoryOffset = pMemoryOffset;
226 	if(decompressedImage)
227 	{
228 		decompressedImage->deviceMemory = deviceMemory;
229 		decompressedImage->memoryOffset = memoryOffset + getStorageSize(format.getAspects());
230 	}
231 }
232 
233 #ifdef __ANDROID__
prepareForExternalUseANDROID() const234 VkResult Image::prepareForExternalUseANDROID() const
235 {
236 	void *nativeBuffer = nullptr;
237 	VkExtent3D extent = getMipLevelExtent(VK_IMAGE_ASPECT_COLOR_BIT, 0);
238 
239 	if(GrallocModule::getInstance()->lock(backingMemory.nativeHandle, GRALLOC_USAGE_SW_WRITE_OFTEN, 0, 0, extent.width, extent.height, &nativeBuffer) != 0)
240 	{
241 		return VK_ERROR_OUT_OF_DATE_KHR;
242 	}
243 
244 	if(!nativeBuffer)
245 	{
246 		return VK_ERROR_OUT_OF_DATE_KHR;
247 	}
248 
249 	int imageRowBytes = rowPitchBytes(VK_IMAGE_ASPECT_COLOR_BIT, 0);
250 	int bufferRowBytes = backingMemory.stride * getFormat().bytes();
251 	ASSERT(imageRowBytes <= bufferRowBytes);
252 
253 	uint8_t *srcBuffer = static_cast<uint8_t *>(deviceMemory->getOffsetPointer(0));
254 	uint8_t *dstBuffer = static_cast<uint8_t *>(nativeBuffer);
255 	for(uint32_t i = 0; i < extent.height; i++)
256 	{
257 		memcpy(dstBuffer + (i * bufferRowBytes), srcBuffer + (i * imageRowBytes), imageRowBytes);
258 	}
259 
260 	if(GrallocModule::getInstance()->unlock(backingMemory.nativeHandle) != 0)
261 	{
262 		return VK_ERROR_OUT_OF_DATE_KHR;
263 	}
264 
265 	return VK_SUCCESS;
266 }
267 
getExternalMemory() const268 VkDeviceMemory Image::getExternalMemory() const
269 {
270 	return backingMemory.externalMemory ? *deviceMemory : VkDeviceMemory{ VK_NULL_HANDLE };
271 }
272 #endif
273 
getSubresourceLayout(const VkImageSubresource * pSubresource,VkSubresourceLayout * pLayout) const274 void Image::getSubresourceLayout(const VkImageSubresource *pSubresource, VkSubresourceLayout *pLayout) const
275 {
276 	// By spec, aspectMask has a single bit set.
277 	if(!((pSubresource->aspectMask == VK_IMAGE_ASPECT_COLOR_BIT) ||
278 	     (pSubresource->aspectMask == VK_IMAGE_ASPECT_DEPTH_BIT) ||
279 	     (pSubresource->aspectMask == VK_IMAGE_ASPECT_STENCIL_BIT) ||
280 	     (pSubresource->aspectMask == VK_IMAGE_ASPECT_PLANE_0_BIT) ||
281 	     (pSubresource->aspectMask == VK_IMAGE_ASPECT_PLANE_1_BIT) ||
282 	     (pSubresource->aspectMask == VK_IMAGE_ASPECT_PLANE_2_BIT)))
283 	{
284 		UNSUPPORTED("aspectMask %X", pSubresource->aspectMask);
285 	}
286 
287 	auto aspect = static_cast<VkImageAspectFlagBits>(pSubresource->aspectMask);
288 	pLayout->offset = getMemoryOffset(aspect, pSubresource->mipLevel, pSubresource->arrayLayer);
289 	pLayout->size = getMultiSampledLevelSize(aspect, pSubresource->mipLevel);
290 	pLayout->rowPitch = rowPitchBytes(aspect, pSubresource->mipLevel);
291 	pLayout->depthPitch = slicePitchBytes(aspect, pSubresource->mipLevel);
292 	pLayout->arrayPitch = getLayerSize(aspect);
293 }
294 
copyTo(Image * dstImage,const VkImageCopy & region) const295 void Image::copyTo(Image *dstImage, const VkImageCopy &region) const
296 {
297 	// Image copy does not perform any conversion, it simply copies memory from
298 	// an image to another image that has the same number of bytes per pixel.
299 
300 	if(!((region.srcSubresource.aspectMask == VK_IMAGE_ASPECT_COLOR_BIT) ||
301 	     (region.srcSubresource.aspectMask == VK_IMAGE_ASPECT_DEPTH_BIT) ||
302 	     (region.srcSubresource.aspectMask == VK_IMAGE_ASPECT_STENCIL_BIT) ||
303 	     (region.srcSubresource.aspectMask == VK_IMAGE_ASPECT_PLANE_0_BIT) ||
304 	     (region.srcSubresource.aspectMask == VK_IMAGE_ASPECT_PLANE_1_BIT) ||
305 	     (region.srcSubresource.aspectMask == VK_IMAGE_ASPECT_PLANE_2_BIT)))
306 	{
307 		UNSUPPORTED("srcSubresource.aspectMask %X", region.srcSubresource.aspectMask);
308 	}
309 
310 	if(!((region.dstSubresource.aspectMask == VK_IMAGE_ASPECT_COLOR_BIT) ||
311 	     (region.dstSubresource.aspectMask == VK_IMAGE_ASPECT_DEPTH_BIT) ||
312 	     (region.dstSubresource.aspectMask == VK_IMAGE_ASPECT_STENCIL_BIT) ||
313 	     (region.dstSubresource.aspectMask == VK_IMAGE_ASPECT_PLANE_0_BIT) ||
314 	     (region.dstSubresource.aspectMask == VK_IMAGE_ASPECT_PLANE_1_BIT) ||
315 	     (region.dstSubresource.aspectMask == VK_IMAGE_ASPECT_PLANE_2_BIT)))
316 	{
317 		UNSUPPORTED("dstSubresource.aspectMask %X", region.dstSubresource.aspectMask);
318 	}
319 
320 	VkImageAspectFlagBits srcAspect = static_cast<VkImageAspectFlagBits>(region.srcSubresource.aspectMask);
321 	VkImageAspectFlagBits dstAspect = static_cast<VkImageAspectFlagBits>(region.dstSubresource.aspectMask);
322 
323 	Format srcFormat = getFormat(srcAspect);
324 	Format dstFormat = dstImage->getFormat(dstAspect);
325 
326 	if((samples > VK_SAMPLE_COUNT_1_BIT) && (imageType == VK_IMAGE_TYPE_2D) && !format.isUnnormalizedInteger())
327 	{
328 		// Requires multisampling resolve
329 		VkImageBlit blitRegion;
330 		blitRegion.srcSubresource = region.srcSubresource;
331 		blitRegion.srcOffsets[0] = region.srcOffset;
332 		blitRegion.srcOffsets[1].x = blitRegion.srcOffsets[0].x + region.extent.width;
333 		blitRegion.srcOffsets[1].y = blitRegion.srcOffsets[0].y + region.extent.height;
334 		blitRegion.srcOffsets[1].z = blitRegion.srcOffsets[0].z + region.extent.depth;
335 
336 		blitRegion.dstSubresource = region.dstSubresource;
337 		blitRegion.dstOffsets[0] = region.dstOffset;
338 		blitRegion.dstOffsets[1].x = blitRegion.dstOffsets[0].x + region.extent.width;
339 		blitRegion.dstOffsets[1].y = blitRegion.dstOffsets[0].y + region.extent.height;
340 		blitRegion.dstOffsets[1].z = blitRegion.dstOffsets[0].z + region.extent.depth;
341 
342 		return device->getBlitter()->blit(this, dstImage, blitRegion, VK_FILTER_NEAREST);
343 	}
344 
345 	int srcBytesPerBlock = srcFormat.bytesPerBlock();
346 	ASSERT(srcBytesPerBlock == dstFormat.bytesPerBlock());
347 
348 	const uint8_t *srcMem = static_cast<const uint8_t *>(getTexelPointer(region.srcOffset, region.srcSubresource));
349 	uint8_t *dstMem = static_cast<uint8_t *>(dstImage->getTexelPointer(region.dstOffset, region.dstSubresource));
350 
351 	int srcRowPitchBytes = rowPitchBytes(srcAspect, region.srcSubresource.mipLevel);
352 	int srcSlicePitchBytes = slicePitchBytes(srcAspect, region.srcSubresource.mipLevel);
353 	int dstRowPitchBytes = dstImage->rowPitchBytes(dstAspect, region.dstSubresource.mipLevel);
354 	int dstSlicePitchBytes = dstImage->slicePitchBytes(dstAspect, region.dstSubresource.mipLevel);
355 
356 	VkExtent3D srcExtent = getMipLevelExtent(srcAspect, region.srcSubresource.mipLevel);
357 	VkExtent3D dstExtent = dstImage->getMipLevelExtent(dstAspect, region.dstSubresource.mipLevel);
358 	VkExtent3D copyExtent = imageExtentInBlocks(region.extent, srcAspect);
359 
360 	bool isSinglePlane = (copyExtent.depth == 1);
361 	bool isSingleLine = (copyExtent.height == 1) && isSinglePlane;
362 	// In order to copy multiple lines using a single memcpy call, we
363 	// have to make sure that we need to copy the entire line and that
364 	// both source and destination lines have the same length in bytes
365 	bool isEntireLine = (region.extent.width == srcExtent.width) &&
366 	                    (region.extent.width == dstExtent.width) &&
367 	                    // For non compressed formats, blockWidth is 1. For compressed
368 	                    // formats, rowPitchBytes returns the number of bytes for a row of
369 	                    // blocks, so we have to divide by the block height, which means:
370 	                    // srcRowPitchBytes / srcBlockWidth == dstRowPitchBytes / dstBlockWidth
371 	                    // And, to avoid potential non exact integer division, for example if a
372 	                    // block has 16 bytes and represents 5 lines, we change the equation to:
373 	                    // srcRowPitchBytes * dstBlockWidth == dstRowPitchBytes * srcBlockWidth
374 	                    ((srcRowPitchBytes * dstFormat.blockWidth()) ==
375 	                     (dstRowPitchBytes * srcFormat.blockWidth()));
376 	// In order to copy multiple planes using a single memcpy call, we
377 	// have to make sure that we need to copy the entire plane and that
378 	// both source and destination planes have the same length in bytes
379 	bool isEntirePlane = isEntireLine &&
380 	                     (copyExtent.height == srcExtent.height) &&
381 	                     (copyExtent.height == dstExtent.height) &&
382 	                     (srcSlicePitchBytes == dstSlicePitchBytes);
383 
384 	if(isSingleLine)  // Copy one line
385 	{
386 		size_t copySize = copyExtent.width * srcBytesPerBlock;
387 		ASSERT((srcMem + copySize) < end());
388 		ASSERT((dstMem + copySize) < dstImage->end());
389 		memcpy(dstMem, srcMem, copySize);
390 	}
391 	else if(isEntireLine && isSinglePlane)  // Copy one plane
392 	{
393 		size_t copySize = copyExtent.height * srcRowPitchBytes;
394 		ASSERT((srcMem + copySize) < end());
395 		ASSERT((dstMem + copySize) < dstImage->end());
396 		memcpy(dstMem, srcMem, copySize);
397 	}
398 	else if(isEntirePlane)  // Copy multiple planes
399 	{
400 		size_t copySize = copyExtent.depth * srcSlicePitchBytes;
401 		ASSERT((srcMem + copySize) < end());
402 		ASSERT((dstMem + copySize) < dstImage->end());
403 		memcpy(dstMem, srcMem, copySize);
404 	}
405 	else if(isEntireLine)  // Copy plane by plane
406 	{
407 		size_t copySize = copyExtent.height * srcRowPitchBytes;
408 
409 		for(uint32_t z = 0; z < copyExtent.depth; z++, dstMem += dstSlicePitchBytes, srcMem += srcSlicePitchBytes)
410 		{
411 			ASSERT((srcMem + copySize) < end());
412 			ASSERT((dstMem + copySize) < dstImage->end());
413 			memcpy(dstMem, srcMem, copySize);
414 		}
415 	}
416 	else  // Copy line by line
417 	{
418 		size_t copySize = copyExtent.width * srcBytesPerBlock;
419 
420 		for(uint32_t z = 0; z < copyExtent.depth; z++, dstMem += dstSlicePitchBytes, srcMem += srcSlicePitchBytes)
421 		{
422 			const uint8_t *srcSlice = srcMem;
423 			uint8_t *dstSlice = dstMem;
424 			for(uint32_t y = 0; y < copyExtent.height; y++, dstSlice += dstRowPitchBytes, srcSlice += srcRowPitchBytes)
425 			{
426 				ASSERT((srcSlice + copySize) < end());
427 				ASSERT((dstSlice + copySize) < dstImage->end());
428 				memcpy(dstSlice, srcSlice, copySize);
429 			}
430 		}
431 	}
432 
433 	dstImage->prepareForSampling({ region.dstSubresource.aspectMask, region.dstSubresource.mipLevel, 1,
434 	                               region.dstSubresource.baseArrayLayer, region.dstSubresource.layerCount });
435 }
436 
copy(Buffer * buffer,const VkBufferImageCopy & region,bool bufferIsSource)437 void Image::copy(Buffer *buffer, const VkBufferImageCopy &region, bool bufferIsSource)
438 {
439 	switch(region.imageSubresource.aspectMask)
440 	{
441 		case VK_IMAGE_ASPECT_COLOR_BIT:
442 		case VK_IMAGE_ASPECT_DEPTH_BIT:
443 		case VK_IMAGE_ASPECT_STENCIL_BIT:
444 		case VK_IMAGE_ASPECT_PLANE_0_BIT:
445 		case VK_IMAGE_ASPECT_PLANE_1_BIT:
446 		case VK_IMAGE_ASPECT_PLANE_2_BIT:
447 			break;
448 		default:
449 			UNSUPPORTED("aspectMask %x", int(region.imageSubresource.aspectMask));
450 			break;
451 	}
452 
453 	auto aspect = static_cast<VkImageAspectFlagBits>(region.imageSubresource.aspectMask);
454 	Format copyFormat = getFormat(aspect);
455 
456 	VkExtent3D imageExtent = imageExtentInBlocks(region.imageExtent, aspect);
457 	VkExtent2D bufferExtent = bufferExtentInBlocks({ imageExtent.width, imageExtent.height }, region);
458 	int bytesPerBlock = copyFormat.bytesPerBlock();
459 	int bufferRowPitchBytes = bufferExtent.width * bytesPerBlock;
460 	int bufferSlicePitchBytes = bufferExtent.height * bufferRowPitchBytes;
461 
462 	uint8_t *bufferMemory = static_cast<uint8_t *>(buffer->getOffsetPointer(region.bufferOffset));
463 	uint8_t *imageMemory = static_cast<uint8_t *>(getTexelPointer(region.imageOffset, region.imageSubresource));
464 	uint8_t *srcMemory = bufferIsSource ? bufferMemory : imageMemory;
465 	uint8_t *dstMemory = bufferIsSource ? imageMemory : bufferMemory;
466 	int imageRowPitchBytes = rowPitchBytes(aspect, region.imageSubresource.mipLevel);
467 	int imageSlicePitchBytes = slicePitchBytes(aspect, region.imageSubresource.mipLevel);
468 
469 	int srcSlicePitchBytes = bufferIsSource ? bufferSlicePitchBytes : imageSlicePitchBytes;
470 	int dstSlicePitchBytes = bufferIsSource ? imageSlicePitchBytes : bufferSlicePitchBytes;
471 	int srcRowPitchBytes = bufferIsSource ? bufferRowPitchBytes : imageRowPitchBytes;
472 	int dstRowPitchBytes = bufferIsSource ? imageRowPitchBytes : bufferRowPitchBytes;
473 
474 	VkExtent3D mipLevelExtent = getMipLevelExtent(aspect, region.imageSubresource.mipLevel);
475 	bool isSinglePlane = (imageExtent.depth == 1);
476 	bool isSingleLine = (imageExtent.height == 1) && isSinglePlane;
477 	bool isEntireLine = (imageExtent.width == mipLevelExtent.width) &&
478 	                    (imageRowPitchBytes == bufferRowPitchBytes);
479 	bool isEntirePlane = isEntireLine && (imageExtent.height == mipLevelExtent.height) &&
480 	                     (imageSlicePitchBytes == bufferSlicePitchBytes);
481 
482 	VkDeviceSize copySize = 0;
483 	VkDeviceSize bufferLayerSize = 0;
484 	if(isSingleLine)
485 	{
486 		copySize = imageExtent.width * bytesPerBlock;
487 		bufferLayerSize = copySize;
488 	}
489 	else if(isEntireLine && isSinglePlane)
490 	{
491 		copySize = imageExtent.height * imageRowPitchBytes;
492 		bufferLayerSize = copySize;
493 	}
494 	else if(isEntirePlane)
495 	{
496 		copySize = imageExtent.depth * imageSlicePitchBytes;  // Copy multiple planes
497 		bufferLayerSize = copySize;
498 	}
499 	else if(isEntireLine)  // Copy plane by plane
500 	{
501 		copySize = imageExtent.height * imageRowPitchBytes;
502 		bufferLayerSize = copySize * imageExtent.depth;
503 	}
504 	else  // Copy line by line
505 	{
506 		copySize = imageExtent.width * bytesPerBlock;
507 		bufferLayerSize = copySize * imageExtent.depth * imageExtent.height;
508 	}
509 
510 	VkDeviceSize imageLayerSize = getLayerSize(aspect);
511 	VkDeviceSize srcLayerSize = bufferIsSource ? bufferLayerSize : imageLayerSize;
512 	VkDeviceSize dstLayerSize = bufferIsSource ? imageLayerSize : bufferLayerSize;
513 
514 	for(uint32_t i = 0; i < region.imageSubresource.layerCount; i++)
515 	{
516 		if(isSingleLine || (isEntireLine && isSinglePlane) || isEntirePlane)
517 		{
518 			ASSERT(((bufferIsSource ? dstMemory : srcMemory) + copySize) < end());
519 			ASSERT(((bufferIsSource ? srcMemory : dstMemory) + copySize) < buffer->end());
520 			memcpy(dstMemory, srcMemory, copySize);
521 		}
522 		else if(isEntireLine)  // Copy plane by plane
523 		{
524 			uint8_t *srcPlaneMemory = srcMemory;
525 			uint8_t *dstPlaneMemory = dstMemory;
526 			for(uint32_t z = 0; z < imageExtent.depth; z++)
527 			{
528 				ASSERT(((bufferIsSource ? dstPlaneMemory : srcPlaneMemory) + copySize) < end());
529 				ASSERT(((bufferIsSource ? srcPlaneMemory : dstPlaneMemory) + copySize) < buffer->end());
530 				memcpy(dstPlaneMemory, srcPlaneMemory, copySize);
531 				srcPlaneMemory += srcSlicePitchBytes;
532 				dstPlaneMemory += dstSlicePitchBytes;
533 			}
534 		}
535 		else  // Copy line by line
536 		{
537 			uint8_t *srcLayerMemory = srcMemory;
538 			uint8_t *dstLayerMemory = dstMemory;
539 			for(uint32_t z = 0; z < imageExtent.depth; z++)
540 			{
541 				uint8_t *srcPlaneMemory = srcLayerMemory;
542 				uint8_t *dstPlaneMemory = dstLayerMemory;
543 				for(uint32_t y = 0; y < imageExtent.height; y++)
544 				{
545 					ASSERT(((bufferIsSource ? dstPlaneMemory : srcPlaneMemory) + copySize) < end());
546 					ASSERT(((bufferIsSource ? srcPlaneMemory : dstPlaneMemory) + copySize) < buffer->end());
547 					memcpy(dstPlaneMemory, srcPlaneMemory, copySize);
548 					srcPlaneMemory += srcRowPitchBytes;
549 					dstPlaneMemory += dstRowPitchBytes;
550 				}
551 				srcLayerMemory += srcSlicePitchBytes;
552 				dstLayerMemory += dstSlicePitchBytes;
553 			}
554 		}
555 
556 		srcMemory += srcLayerSize;
557 		dstMemory += dstLayerSize;
558 	}
559 
560 	if(bufferIsSource)
561 	{
562 		prepareForSampling({ region.imageSubresource.aspectMask, region.imageSubresource.mipLevel, 1,
563 		                     region.imageSubresource.baseArrayLayer, region.imageSubresource.layerCount });
564 	}
565 }
566 
copyTo(Buffer * dstBuffer,const VkBufferImageCopy & region)567 void Image::copyTo(Buffer *dstBuffer, const VkBufferImageCopy &region)
568 {
569 	copy(dstBuffer, region, false);
570 }
571 
copyFrom(Buffer * srcBuffer,const VkBufferImageCopy & region)572 void Image::copyFrom(Buffer *srcBuffer, const VkBufferImageCopy &region)
573 {
574 	copy(srcBuffer, region, true);
575 }
576 
getTexelPointer(const VkOffset3D & offset,const VkImageSubresourceLayers & subresource) const577 void *Image::getTexelPointer(const VkOffset3D &offset, const VkImageSubresourceLayers &subresource) const
578 {
579 	VkImageAspectFlagBits aspect = static_cast<VkImageAspectFlagBits>(subresource.aspectMask);
580 	return deviceMemory->getOffsetPointer(texelOffsetBytesInStorage(offset, subresource) +
581 	                                      getMemoryOffset(aspect, subresource.mipLevel, subresource.baseArrayLayer));
582 }
583 
imageExtentInBlocks(const VkExtent3D & extent,VkImageAspectFlagBits aspect) const584 VkExtent3D Image::imageExtentInBlocks(const VkExtent3D &extent, VkImageAspectFlagBits aspect) const
585 {
586 	VkExtent3D adjustedExtent = extent;
587 	Format usedFormat = getFormat(aspect);
588 	if(usedFormat.isCompressed())
589 	{
590 		// When using a compressed format, we use the block as the base unit, instead of the texel
591 		int blockWidth = usedFormat.blockWidth();
592 		int blockHeight = usedFormat.blockHeight();
593 
594 		// Mip level allocations will round up to the next block for compressed texture
595 		adjustedExtent.width = ((adjustedExtent.width + blockWidth - 1) / blockWidth);
596 		adjustedExtent.height = ((adjustedExtent.height + blockHeight - 1) / blockHeight);
597 	}
598 	return adjustedExtent;
599 }
600 
imageOffsetInBlocks(const VkOffset3D & offset,VkImageAspectFlagBits aspect) const601 VkOffset3D Image::imageOffsetInBlocks(const VkOffset3D &offset, VkImageAspectFlagBits aspect) const
602 {
603 	VkOffset3D adjustedOffset = offset;
604 	Format usedFormat = getFormat(aspect);
605 	if(usedFormat.isCompressed())
606 	{
607 		// When using a compressed format, we use the block as the base unit, instead of the texel
608 		int blockWidth = usedFormat.blockWidth();
609 		int blockHeight = usedFormat.blockHeight();
610 
611 		ASSERT(((offset.x % blockWidth) == 0) && ((offset.y % blockHeight) == 0));  // We can't offset within a block
612 
613 		adjustedOffset.x /= blockWidth;
614 		adjustedOffset.y /= blockHeight;
615 	}
616 	return adjustedOffset;
617 }
618 
bufferExtentInBlocks(const VkExtent2D & extent,const VkBufferImageCopy & region) const619 VkExtent2D Image::bufferExtentInBlocks(const VkExtent2D &extent, const VkBufferImageCopy &region) const
620 {
621 	VkExtent2D adjustedExtent = extent;
622 	VkImageAspectFlagBits aspect = static_cast<VkImageAspectFlagBits>(region.imageSubresource.aspectMask);
623 	Format usedFormat = getFormat(aspect);
624 	if(region.bufferRowLength != 0)
625 	{
626 		adjustedExtent.width = region.bufferRowLength;
627 
628 		if(usedFormat.isCompressed())
629 		{
630 			int blockWidth = usedFormat.blockWidth();
631 			ASSERT((adjustedExtent.width % blockWidth) == 0);
632 			adjustedExtent.width /= blockWidth;
633 		}
634 	}
635 	if(region.bufferImageHeight != 0)
636 	{
637 		adjustedExtent.height = region.bufferImageHeight;
638 
639 		if(usedFormat.isCompressed())
640 		{
641 			int blockHeight = usedFormat.blockHeight();
642 			ASSERT((adjustedExtent.height % blockHeight) == 0);
643 			adjustedExtent.height /= blockHeight;
644 		}
645 	}
646 	return adjustedExtent;
647 }
648 
borderSize() const649 int Image::borderSize() const
650 {
651 	// We won't add a border to compressed cube textures, we'll add it when we decompress the texture
652 	return (isCube() && !format.isCompressed()) ? 1 : 0;
653 }
654 
texelOffsetBytesInStorage(const VkOffset3D & offset,const VkImageSubresourceLayers & subresource) const655 VkDeviceSize Image::texelOffsetBytesInStorage(const VkOffset3D &offset, const VkImageSubresourceLayers &subresource) const
656 {
657 	VkImageAspectFlagBits aspect = static_cast<VkImageAspectFlagBits>(subresource.aspectMask);
658 	VkOffset3D adjustedOffset = imageOffsetInBlocks(offset, aspect);
659 	int border = borderSize();
660 	return adjustedOffset.z * slicePitchBytes(aspect, subresource.mipLevel) +
661 	       (adjustedOffset.y + border) * rowPitchBytes(aspect, subresource.mipLevel) +
662 	       (adjustedOffset.x + border) * getFormat(aspect).bytesPerBlock();
663 }
664 
getMipLevelExtent(VkImageAspectFlagBits aspect,uint32_t mipLevel) const665 VkExtent3D Image::getMipLevelExtent(VkImageAspectFlagBits aspect, uint32_t mipLevel) const
666 {
667 	VkExtent3D mipLevelExtent;
668 	mipLevelExtent.width = extent.width >> mipLevel;
669 	mipLevelExtent.height = extent.height >> mipLevel;
670 	mipLevelExtent.depth = extent.depth >> mipLevel;
671 
672 	if(mipLevelExtent.width == 0) { mipLevelExtent.width = 1; }
673 	if(mipLevelExtent.height == 0) { mipLevelExtent.height = 1; }
674 	if(mipLevelExtent.depth == 0) { mipLevelExtent.depth = 1; }
675 
676 	switch(aspect)
677 	{
678 		case VK_IMAGE_ASPECT_COLOR_BIT:
679 		case VK_IMAGE_ASPECT_DEPTH_BIT:
680 		case VK_IMAGE_ASPECT_STENCIL_BIT:
681 		case VK_IMAGE_ASPECT_PLANE_0_BIT:  // Vulkan 1.1 Table 31. Plane Format Compatibility Table: plane 0 of all defined formats is full resolution.
682 			break;
683 		case VK_IMAGE_ASPECT_PLANE_1_BIT:
684 		case VK_IMAGE_ASPECT_PLANE_2_BIT:
685 			switch(format)
686 			{
687 				case VK_FORMAT_G8_B8_R8_3PLANE_420_UNORM:
688 				case VK_FORMAT_G8_B8R8_2PLANE_420_UNORM:
689 					ASSERT(mipLevelExtent.width % 2 == 0 && mipLevelExtent.height % 2 == 0);  // Vulkan 1.1: "Images in this format must be defined with a width and height that is a multiple of two."
690 					// Vulkan 1.1 Table 31. Plane Format Compatibility Table:
691 					// Half-resolution U and V planes.
692 					mipLevelExtent.width /= 2;
693 					mipLevelExtent.height /= 2;
694 					break;
695 				default:
696 					UNSUPPORTED("format %d", int(format));
697 			}
698 			break;
699 		default:
700 			UNSUPPORTED("aspect %x", int(aspect));
701 	}
702 
703 	return mipLevelExtent;
704 }
705 
rowPitchBytes(VkImageAspectFlagBits aspect,uint32_t mipLevel) const706 int Image::rowPitchBytes(VkImageAspectFlagBits aspect, uint32_t mipLevel) const
707 {
708 	// Depth and Stencil pitch should be computed separately
709 	ASSERT((aspect & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) !=
710 	       (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT));
711 
712 	VkExtent3D mipLevelExtent = getMipLevelExtent(aspect, mipLevel);
713 	Format usedFormat = getFormat(aspect);
714 	if(usedFormat.isCompressed())
715 	{
716 		VkExtent3D extentInBlocks = imageExtentInBlocks(mipLevelExtent, aspect);
717 		return extentInBlocks.width * usedFormat.bytesPerBlock();
718 	}
719 
720 	return usedFormat.pitchB(mipLevelExtent.width, borderSize(), true);
721 }
722 
slicePitchBytes(VkImageAspectFlagBits aspect,uint32_t mipLevel) const723 int Image::slicePitchBytes(VkImageAspectFlagBits aspect, uint32_t mipLevel) const
724 {
725 	// Depth and Stencil slice should be computed separately
726 	ASSERT((aspect & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) !=
727 	       (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT));
728 
729 	VkExtent3D mipLevelExtent = getMipLevelExtent(aspect, mipLevel);
730 	Format usedFormat = getFormat(aspect);
731 	if(usedFormat.isCompressed())
732 	{
733 		VkExtent3D extentInBlocks = imageExtentInBlocks(mipLevelExtent, aspect);
734 		return extentInBlocks.height * extentInBlocks.width * usedFormat.bytesPerBlock();
735 	}
736 
737 	return usedFormat.sliceB(mipLevelExtent.width, mipLevelExtent.height, borderSize(), true);
738 }
739 
getFormat(VkImageAspectFlagBits aspect) const740 Format Image::getFormat(VkImageAspectFlagBits aspect) const
741 {
742 	return format.getAspectFormat(aspect);
743 }
744 
isCube() const745 bool Image::isCube() const
746 {
747 	return (flags & VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT) && (imageType == VK_IMAGE_TYPE_2D);
748 }
749 
end() const750 uint8_t *Image::end() const
751 {
752 	return reinterpret_cast<uint8_t *>(deviceMemory->getOffsetPointer(deviceMemory->getCommittedMemoryInBytes() + 1));
753 }
754 
getMemoryOffset(VkImageAspectFlagBits aspect) const755 VkDeviceSize Image::getMemoryOffset(VkImageAspectFlagBits aspect) const
756 {
757 	switch(format)
758 	{
759 		case VK_FORMAT_D16_UNORM_S8_UINT:
760 		case VK_FORMAT_D24_UNORM_S8_UINT:
761 		case VK_FORMAT_D32_SFLOAT_S8_UINT:
762 			if(aspect == VK_IMAGE_ASPECT_STENCIL_BIT)
763 			{
764 				// Offset by depth buffer to get to stencil buffer
765 				return memoryOffset + getStorageSize(VK_IMAGE_ASPECT_DEPTH_BIT);
766 			}
767 			break;
768 
769 		case VK_FORMAT_G8_B8_R8_3PLANE_420_UNORM:
770 			if(aspect == VK_IMAGE_ASPECT_PLANE_2_BIT)
771 			{
772 				return memoryOffset + getStorageSize(VK_IMAGE_ASPECT_PLANE_1_BIT) + getStorageSize(VK_IMAGE_ASPECT_PLANE_0_BIT);
773 			}
774 			// Fall through to 2PLANE case:
775 		case VK_FORMAT_G8_B8R8_2PLANE_420_UNORM:
776 			if(aspect == VK_IMAGE_ASPECT_PLANE_1_BIT)
777 			{
778 				return memoryOffset + getStorageSize(VK_IMAGE_ASPECT_PLANE_0_BIT);
779 			}
780 			else
781 			{
782 				ASSERT(aspect == VK_IMAGE_ASPECT_PLANE_0_BIT);
783 
784 				return memoryOffset;
785 			}
786 			break;
787 
788 		default:
789 			break;
790 	}
791 
792 	return memoryOffset;
793 }
794 
getMemoryOffset(VkImageAspectFlagBits aspect,uint32_t mipLevel) const795 VkDeviceSize Image::getMemoryOffset(VkImageAspectFlagBits aspect, uint32_t mipLevel) const
796 {
797 	VkDeviceSize offset = getMemoryOffset(aspect);
798 	for(uint32_t i = 0; i < mipLevel; ++i)
799 	{
800 		offset += getMultiSampledLevelSize(aspect, i);
801 	}
802 	return offset;
803 }
804 
getMemoryOffset(VkImageAspectFlagBits aspect,uint32_t mipLevel,uint32_t layer) const805 VkDeviceSize Image::getMemoryOffset(VkImageAspectFlagBits aspect, uint32_t mipLevel, uint32_t layer) const
806 {
807 	return layer * getLayerOffset(aspect, mipLevel) + getMemoryOffset(aspect, mipLevel);
808 }
809 
getMipLevelSize(VkImageAspectFlagBits aspect,uint32_t mipLevel) const810 VkDeviceSize Image::getMipLevelSize(VkImageAspectFlagBits aspect, uint32_t mipLevel) const
811 {
812 	return getMipLevelExtent(aspect, mipLevel).depth * slicePitchBytes(aspect, mipLevel);
813 }
814 
getMultiSampledLevelSize(VkImageAspectFlagBits aspect,uint32_t mipLevel) const815 VkDeviceSize Image::getMultiSampledLevelSize(VkImageAspectFlagBits aspect, uint32_t mipLevel) const
816 {
817 	return getMipLevelSize(aspect, mipLevel) * samples;
818 }
819 
is3DSlice() const820 bool Image::is3DSlice() const
821 {
822 	return ((imageType == VK_IMAGE_TYPE_3D) && (flags & VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT));
823 }
824 
getLayerOffset(VkImageAspectFlagBits aspect,uint32_t mipLevel) const825 VkDeviceSize Image::getLayerOffset(VkImageAspectFlagBits aspect, uint32_t mipLevel) const
826 {
827 	if(is3DSlice())
828 	{
829 		// When the VkImageSubresourceRange structure is used to select a subset of the slices of a 3D
830 		// image's mip level in order to create a 2D or 2D array image view of a 3D image created with
831 		// VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT, baseArrayLayer and layerCount specify the first
832 		// slice index and the number of slices to include in the created image view.
833 		ASSERT(samples == VK_SAMPLE_COUNT_1_BIT);
834 
835 		// Offset to the proper slice of the 3D image's mip level
836 		return slicePitchBytes(aspect, mipLevel);
837 	}
838 
839 	return getLayerSize(aspect);
840 }
841 
getLayerSize(VkImageAspectFlagBits aspect) const842 VkDeviceSize Image::getLayerSize(VkImageAspectFlagBits aspect) const
843 {
844 	VkDeviceSize layerSize = 0;
845 
846 	for(uint32_t mipLevel = 0; mipLevel < mipLevels; ++mipLevel)
847 	{
848 		layerSize += getMultiSampledLevelSize(aspect, mipLevel);
849 	}
850 
851 	return layerSize;
852 }
853 
getStorageSize(VkImageAspectFlags aspectMask) const854 VkDeviceSize Image::getStorageSize(VkImageAspectFlags aspectMask) const
855 {
856 	if((aspectMask & ~(VK_IMAGE_ASPECT_COLOR_BIT | VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT |
857 	                   VK_IMAGE_ASPECT_PLANE_0_BIT | VK_IMAGE_ASPECT_PLANE_1_BIT | VK_IMAGE_ASPECT_PLANE_2_BIT)) != 0)
858 	{
859 		UNSUPPORTED("aspectMask %x", int(aspectMask));
860 	}
861 
862 	VkDeviceSize storageSize = 0;
863 
864 	if(aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) storageSize += getLayerSize(VK_IMAGE_ASPECT_COLOR_BIT);
865 	if(aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT) storageSize += getLayerSize(VK_IMAGE_ASPECT_DEPTH_BIT);
866 	if(aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT) storageSize += getLayerSize(VK_IMAGE_ASPECT_STENCIL_BIT);
867 	if(aspectMask & VK_IMAGE_ASPECT_PLANE_0_BIT) storageSize += getLayerSize(VK_IMAGE_ASPECT_PLANE_0_BIT);
868 	if(aspectMask & VK_IMAGE_ASPECT_PLANE_1_BIT) storageSize += getLayerSize(VK_IMAGE_ASPECT_PLANE_1_BIT);
869 	if(aspectMask & VK_IMAGE_ASPECT_PLANE_2_BIT) storageSize += getLayerSize(VK_IMAGE_ASPECT_PLANE_2_BIT);
870 
871 	return arrayLayers * storageSize;
872 }
873 
getSampledImage(const vk::Format & imageViewFormat) const874 const Image *Image::getSampledImage(const vk::Format &imageViewFormat) const
875 {
876 	bool isImageViewCompressed = imageViewFormat.isCompressed();
877 	if(decompressedImage && !isImageViewCompressed)
878 	{
879 		ASSERT(flags & VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT);
880 		ASSERT(format.bytesPerBlock() == imageViewFormat.bytesPerBlock());
881 	}
882 	// If the ImageView's format is compressed, then we do need to decompress the image so that
883 	// it may be sampled properly by texture sampling functions, which don't support compressed
884 	// textures. If the ImageView's format is NOT compressed, then we reinterpret cast the
885 	// compressed image into the ImageView's format, so we must return the compressed image as is.
886 	return (decompressedImage && isImageViewCompressed) ? decompressedImage : this;
887 }
888 
blit(Image * dstImage,const VkImageBlit & region,VkFilter filter) const889 void Image::blit(Image *dstImage, const VkImageBlit &region, VkFilter filter) const
890 {
891 	device->getBlitter()->blit(this, dstImage, region, filter);
892 }
893 
blitToBuffer(VkImageSubresourceLayers subresource,VkOffset3D offset,VkExtent3D extent,uint8_t * dst,int bufferRowPitch,int bufferSlicePitch) const894 void Image::blitToBuffer(VkImageSubresourceLayers subresource, VkOffset3D offset, VkExtent3D extent, uint8_t *dst, int bufferRowPitch, int bufferSlicePitch) const
895 {
896 	device->getBlitter()->blitToBuffer(this, subresource, offset, extent, dst, bufferRowPitch, bufferSlicePitch);
897 }
898 
resolve(Image * dstImage,const VkImageResolve & region) const899 void Image::resolve(Image *dstImage, const VkImageResolve &region) const
900 {
901 	VkImageBlit blitRegion;
902 
903 	blitRegion.srcOffsets[0] = blitRegion.srcOffsets[1] = region.srcOffset;
904 	blitRegion.srcOffsets[1].x += region.extent.width;
905 	blitRegion.srcOffsets[1].y += region.extent.height;
906 	blitRegion.srcOffsets[1].z += region.extent.depth;
907 
908 	blitRegion.dstOffsets[0] = blitRegion.dstOffsets[1] = region.dstOffset;
909 	blitRegion.dstOffsets[1].x += region.extent.width;
910 	blitRegion.dstOffsets[1].y += region.extent.height;
911 	blitRegion.dstOffsets[1].z += region.extent.depth;
912 
913 	blitRegion.srcSubresource = region.srcSubresource;
914 	blitRegion.dstSubresource = region.dstSubresource;
915 
916 	device->getBlitter()->blit(this, dstImage, blitRegion, VK_FILTER_NEAREST);
917 }
918 
getClearFormat() const919 VkFormat Image::getClearFormat() const
920 {
921 	// Set the proper format for the clear value, as described here:
922 	// https://www.khronos.org/registry/vulkan/specs/1.1-extensions/html/vkspec.html#clears-values
923 	if(format.isSignedUnnormalizedInteger())
924 	{
925 		return VK_FORMAT_R32G32B32A32_SINT;
926 	}
927 	else if(format.isUnsignedUnnormalizedInteger())
928 	{
929 		return VK_FORMAT_R32G32B32A32_UINT;
930 	}
931 
932 	return VK_FORMAT_R32G32B32A32_SFLOAT;
933 }
934 
getLastLayerIndex(const VkImageSubresourceRange & subresourceRange) const935 uint32_t Image::getLastLayerIndex(const VkImageSubresourceRange &subresourceRange) const
936 {
937 	return ((subresourceRange.layerCount == VK_REMAINING_ARRAY_LAYERS) ? arrayLayers : (subresourceRange.baseArrayLayer + subresourceRange.layerCount)) - 1;
938 }
939 
getLastMipLevel(const VkImageSubresourceRange & subresourceRange) const940 uint32_t Image::getLastMipLevel(const VkImageSubresourceRange &subresourceRange) const
941 {
942 	return ((subresourceRange.levelCount == VK_REMAINING_MIP_LEVELS) ? mipLevels : (subresourceRange.baseMipLevel + subresourceRange.levelCount)) - 1;
943 }
944 
clear(void * pixelData,VkFormat pixelFormat,const vk::Format & viewFormat,const VkImageSubresourceRange & subresourceRange,const VkRect2D & renderArea)945 void Image::clear(void *pixelData, VkFormat pixelFormat, const vk::Format &viewFormat, const VkImageSubresourceRange &subresourceRange, const VkRect2D &renderArea)
946 {
947 	device->getBlitter()->clear(pixelData, pixelFormat, this, viewFormat, subresourceRange, &renderArea);
948 }
949 
clear(const VkClearColorValue & color,const VkImageSubresourceRange & subresourceRange)950 void Image::clear(const VkClearColorValue &color, const VkImageSubresourceRange &subresourceRange)
951 {
952 	ASSERT(subresourceRange.aspectMask == VK_IMAGE_ASPECT_COLOR_BIT);
953 
954 	device->getBlitter()->clear((void *)color.float32, getClearFormat(), this, format, subresourceRange);
955 }
956 
clear(const VkClearDepthStencilValue & color,const VkImageSubresourceRange & subresourceRange)957 void Image::clear(const VkClearDepthStencilValue &color, const VkImageSubresourceRange &subresourceRange)
958 {
959 	ASSERT((subresourceRange.aspectMask & ~(VK_IMAGE_ASPECT_DEPTH_BIT |
960 	                                        VK_IMAGE_ASPECT_STENCIL_BIT)) == 0);
961 
962 	if(subresourceRange.aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT)
963 	{
964 		VkImageSubresourceRange depthSubresourceRange = subresourceRange;
965 		depthSubresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
966 		device->getBlitter()->clear((void *)(&color.depth), VK_FORMAT_D32_SFLOAT, this, format, depthSubresourceRange);
967 	}
968 
969 	if(subresourceRange.aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT)
970 	{
971 		VkImageSubresourceRange stencilSubresourceRange = subresourceRange;
972 		stencilSubresourceRange.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
973 		device->getBlitter()->clear((void *)(&color.stencil), VK_FORMAT_S8_UINT, this, format, stencilSubresourceRange);
974 	}
975 }
976 
clear(const VkClearValue & clearValue,const vk::Format & viewFormat,const VkRect2D & renderArea,const VkImageSubresourceRange & subresourceRange)977 void Image::clear(const VkClearValue &clearValue, const vk::Format &viewFormat, const VkRect2D &renderArea, const VkImageSubresourceRange &subresourceRange)
978 {
979 	ASSERT((subresourceRange.aspectMask == VK_IMAGE_ASPECT_COLOR_BIT) ||
980 	       (subresourceRange.aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT |
981 	                                       VK_IMAGE_ASPECT_STENCIL_BIT)));
982 
983 	if(subresourceRange.aspectMask == VK_IMAGE_ASPECT_COLOR_BIT)
984 	{
985 		clear((void *)(clearValue.color.float32), getClearFormat(), viewFormat, subresourceRange, renderArea);
986 	}
987 	else
988 	{
989 		if(subresourceRange.aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT)
990 		{
991 			VkImageSubresourceRange depthSubresourceRange = subresourceRange;
992 			depthSubresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
993 			clear((void *)(&clearValue.depthStencil.depth), VK_FORMAT_D32_SFLOAT, viewFormat, depthSubresourceRange, renderArea);
994 		}
995 
996 		if(subresourceRange.aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT)
997 		{
998 			VkImageSubresourceRange stencilSubresourceRange = subresourceRange;
999 			stencilSubresourceRange.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
1000 			clear((void *)(&clearValue.depthStencil.stencil), VK_FORMAT_S8_UINT, viewFormat, stencilSubresourceRange, renderArea);
1001 		}
1002 	}
1003 }
1004 
prepareForSampling(const VkImageSubresourceRange & subresourceRange)1005 void Image::prepareForSampling(const VkImageSubresourceRange &subresourceRange)
1006 {
1007 	if(decompressedImage)
1008 	{
1009 		switch(format)
1010 		{
1011 			case VK_FORMAT_EAC_R11_UNORM_BLOCK:
1012 			case VK_FORMAT_EAC_R11_SNORM_BLOCK:
1013 			case VK_FORMAT_EAC_R11G11_UNORM_BLOCK:
1014 			case VK_FORMAT_EAC_R11G11_SNORM_BLOCK:
1015 			case VK_FORMAT_ETC2_R8G8B8_UNORM_BLOCK:
1016 			case VK_FORMAT_ETC2_R8G8B8_SRGB_BLOCK:
1017 			case VK_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK:
1018 			case VK_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK:
1019 			case VK_FORMAT_ETC2_R8G8B8A8_UNORM_BLOCK:
1020 			case VK_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK:
1021 				decodeETC2(subresourceRange);
1022 				break;
1023 			case VK_FORMAT_BC1_RGB_UNORM_BLOCK:
1024 			case VK_FORMAT_BC1_RGB_SRGB_BLOCK:
1025 			case VK_FORMAT_BC1_RGBA_UNORM_BLOCK:
1026 			case VK_FORMAT_BC1_RGBA_SRGB_BLOCK:
1027 			case VK_FORMAT_BC2_UNORM_BLOCK:
1028 			case VK_FORMAT_BC2_SRGB_BLOCK:
1029 			case VK_FORMAT_BC3_UNORM_BLOCK:
1030 			case VK_FORMAT_BC3_SRGB_BLOCK:
1031 			case VK_FORMAT_BC4_UNORM_BLOCK:
1032 			case VK_FORMAT_BC4_SNORM_BLOCK:
1033 			case VK_FORMAT_BC5_UNORM_BLOCK:
1034 			case VK_FORMAT_BC5_SNORM_BLOCK:
1035 			case VK_FORMAT_BC6H_UFLOAT_BLOCK:
1036 			case VK_FORMAT_BC6H_SFLOAT_BLOCK:
1037 			case VK_FORMAT_BC7_UNORM_BLOCK:
1038 			case VK_FORMAT_BC7_SRGB_BLOCK:
1039 				decodeBC(subresourceRange);
1040 				break;
1041 			case VK_FORMAT_ASTC_4x4_UNORM_BLOCK:
1042 			case VK_FORMAT_ASTC_5x4_UNORM_BLOCK:
1043 			case VK_FORMAT_ASTC_5x5_UNORM_BLOCK:
1044 			case VK_FORMAT_ASTC_6x5_UNORM_BLOCK:
1045 			case VK_FORMAT_ASTC_6x6_UNORM_BLOCK:
1046 			case VK_FORMAT_ASTC_8x5_UNORM_BLOCK:
1047 			case VK_FORMAT_ASTC_8x6_UNORM_BLOCK:
1048 			case VK_FORMAT_ASTC_8x8_UNORM_BLOCK:
1049 			case VK_FORMAT_ASTC_10x5_UNORM_BLOCK:
1050 			case VK_FORMAT_ASTC_10x6_UNORM_BLOCK:
1051 			case VK_FORMAT_ASTC_10x8_UNORM_BLOCK:
1052 			case VK_FORMAT_ASTC_10x10_UNORM_BLOCK:
1053 			case VK_FORMAT_ASTC_12x10_UNORM_BLOCK:
1054 			case VK_FORMAT_ASTC_12x12_UNORM_BLOCK:
1055 			case VK_FORMAT_ASTC_4x4_SRGB_BLOCK:
1056 			case VK_FORMAT_ASTC_5x4_SRGB_BLOCK:
1057 			case VK_FORMAT_ASTC_5x5_SRGB_BLOCK:
1058 			case VK_FORMAT_ASTC_6x5_SRGB_BLOCK:
1059 			case VK_FORMAT_ASTC_6x6_SRGB_BLOCK:
1060 			case VK_FORMAT_ASTC_8x5_SRGB_BLOCK:
1061 			case VK_FORMAT_ASTC_8x6_SRGB_BLOCK:
1062 			case VK_FORMAT_ASTC_8x8_SRGB_BLOCK:
1063 			case VK_FORMAT_ASTC_10x5_SRGB_BLOCK:
1064 			case VK_FORMAT_ASTC_10x6_SRGB_BLOCK:
1065 			case VK_FORMAT_ASTC_10x8_SRGB_BLOCK:
1066 			case VK_FORMAT_ASTC_10x10_SRGB_BLOCK:
1067 			case VK_FORMAT_ASTC_12x10_SRGB_BLOCK:
1068 			case VK_FORMAT_ASTC_12x12_SRGB_BLOCK:
1069 			case VK_FORMAT_ASTC_4x4_SFLOAT_BLOCK_EXT:
1070 			case VK_FORMAT_ASTC_5x4_SFLOAT_BLOCK_EXT:
1071 			case VK_FORMAT_ASTC_5x5_SFLOAT_BLOCK_EXT:
1072 			case VK_FORMAT_ASTC_6x5_SFLOAT_BLOCK_EXT:
1073 			case VK_FORMAT_ASTC_6x6_SFLOAT_BLOCK_EXT:
1074 			case VK_FORMAT_ASTC_8x5_SFLOAT_BLOCK_EXT:
1075 			case VK_FORMAT_ASTC_8x6_SFLOAT_BLOCK_EXT:
1076 			case VK_FORMAT_ASTC_8x8_SFLOAT_BLOCK_EXT:
1077 			case VK_FORMAT_ASTC_10x5_SFLOAT_BLOCK_EXT:
1078 			case VK_FORMAT_ASTC_10x6_SFLOAT_BLOCK_EXT:
1079 			case VK_FORMAT_ASTC_10x8_SFLOAT_BLOCK_EXT:
1080 			case VK_FORMAT_ASTC_10x10_SFLOAT_BLOCK_EXT:
1081 			case VK_FORMAT_ASTC_12x10_SFLOAT_BLOCK_EXT:
1082 			case VK_FORMAT_ASTC_12x12_SFLOAT_BLOCK_EXT:
1083 				decodeASTC(subresourceRange);
1084 				break;
1085 			default:
1086 				break;
1087 		}
1088 	}
1089 
1090 	if(isCube() && (arrayLayers >= 6))
1091 	{
1092 		VkImageSubresourceLayers subresourceLayers = {
1093 			subresourceRange.aspectMask,
1094 			subresourceRange.baseMipLevel,
1095 			subresourceRange.baseArrayLayer,
1096 			6
1097 		};
1098 
1099 		// Update the borders of all the groups of 6 layers that can be part of a cubemaps but don't
1100 		// touch leftover layers that cannot be part of cubemaps.
1101 		uint32_t lastMipLevel = getLastMipLevel(subresourceRange);
1102 		for(; subresourceLayers.mipLevel <= lastMipLevel; subresourceLayers.mipLevel++)
1103 		{
1104 			for(subresourceLayers.baseArrayLayer = 0;
1105 			    subresourceLayers.baseArrayLayer < arrayLayers - 5;
1106 			    subresourceLayers.baseArrayLayer += 6)
1107 			{
1108 				device->getBlitter()->updateBorders(decompressedImage ? decompressedImage : this, subresourceLayers);
1109 			}
1110 		}
1111 	}
1112 }
1113 
decodeETC2(const VkImageSubresourceRange & subresourceRange) const1114 void Image::decodeETC2(const VkImageSubresourceRange &subresourceRange) const
1115 {
1116 	ASSERT(decompressedImage);
1117 
1118 	ETC_Decoder::InputType inputType = GetInputType(format);
1119 
1120 	uint32_t lastLayer = getLastLayerIndex(subresourceRange);
1121 	uint32_t lastMipLevel = getLastMipLevel(subresourceRange);
1122 
1123 	int bytes = decompressedImage->format.bytes();
1124 	bool fakeAlpha = (format == VK_FORMAT_ETC2_R8G8B8_UNORM_BLOCK) || (format == VK_FORMAT_ETC2_R8G8B8_SRGB_BLOCK);
1125 	size_t sizeToWrite = 0;
1126 
1127 	VkImageSubresourceLayers subresourceLayers = { subresourceRange.aspectMask, subresourceRange.baseMipLevel, subresourceRange.baseArrayLayer, 1 };
1128 	for(; subresourceLayers.baseArrayLayer <= lastLayer; subresourceLayers.baseArrayLayer++)
1129 	{
1130 		for(; subresourceLayers.mipLevel <= lastMipLevel; subresourceLayers.mipLevel++)
1131 		{
1132 			VkExtent3D mipLevelExtent = getMipLevelExtent(static_cast<VkImageAspectFlagBits>(subresourceLayers.aspectMask), subresourceLayers.mipLevel);
1133 
1134 			int pitchB = decompressedImage->rowPitchBytes(VK_IMAGE_ASPECT_COLOR_BIT, subresourceLayers.mipLevel);
1135 
1136 			if(fakeAlpha)
1137 			{
1138 				// To avoid overflow in case of cube textures, which are offset in memory to account for the border,
1139 				// compute the size from the first pixel to the last pixel, excluding any padding or border before
1140 				// the first pixel or after the last pixel.
1141 				sizeToWrite = ((mipLevelExtent.height - 1) * pitchB) + (mipLevelExtent.width * bytes);
1142 			}
1143 
1144 			for(int32_t depth = 0; depth < static_cast<int32_t>(mipLevelExtent.depth); depth++)
1145 			{
1146 				uint8_t *source = static_cast<uint8_t *>(getTexelPointer({ 0, 0, depth }, subresourceLayers));
1147 				uint8_t *dest = static_cast<uint8_t *>(decompressedImage->getTexelPointer({ 0, 0, depth }, subresourceLayers));
1148 
1149 				if(fakeAlpha)
1150 				{
1151 					ASSERT((dest + sizeToWrite) < decompressedImage->end());
1152 					memset(dest, 0xFF, sizeToWrite);
1153 				}
1154 
1155 				ETC_Decoder::Decode(source, dest, mipLevelExtent.width, mipLevelExtent.height,
1156 				                    pitchB, bytes, inputType);
1157 			}
1158 		}
1159 	}
1160 }
1161 
decodeBC(const VkImageSubresourceRange & subresourceRange) const1162 void Image::decodeBC(const VkImageSubresourceRange &subresourceRange) const
1163 {
1164 	ASSERT(decompressedImage);
1165 
1166 	int n = GetBCn(format);
1167 	int noAlphaU = GetNoAlphaOrUnsigned(format);
1168 
1169 	uint32_t lastLayer = getLastLayerIndex(subresourceRange);
1170 	uint32_t lastMipLevel = getLastMipLevel(subresourceRange);
1171 
1172 	int bytes = decompressedImage->format.bytes();
1173 
1174 	VkImageSubresourceLayers subresourceLayers = { subresourceRange.aspectMask, subresourceRange.baseMipLevel, subresourceRange.baseArrayLayer, 1 };
1175 	for(; subresourceLayers.baseArrayLayer <= lastLayer; subresourceLayers.baseArrayLayer++)
1176 	{
1177 		for(; subresourceLayers.mipLevel <= lastMipLevel; subresourceLayers.mipLevel++)
1178 		{
1179 			VkExtent3D mipLevelExtent = getMipLevelExtent(static_cast<VkImageAspectFlagBits>(subresourceLayers.aspectMask), subresourceLayers.mipLevel);
1180 
1181 			int pitchB = decompressedImage->rowPitchBytes(VK_IMAGE_ASPECT_COLOR_BIT, subresourceLayers.mipLevel);
1182 
1183 			for(int32_t depth = 0; depth < static_cast<int32_t>(mipLevelExtent.depth); depth++)
1184 			{
1185 				uint8_t *source = static_cast<uint8_t *>(getTexelPointer({ 0, 0, depth }, subresourceLayers));
1186 				uint8_t *dest = static_cast<uint8_t *>(decompressedImage->getTexelPointer({ 0, 0, depth }, subresourceLayers));
1187 
1188 				BC_Decoder::Decode(source, dest, mipLevelExtent.width, mipLevelExtent.height,
1189 				                   pitchB, bytes, n, noAlphaU);
1190 			}
1191 		}
1192 	}
1193 }
1194 
decodeASTC(const VkImageSubresourceRange & subresourceRange) const1195 void Image::decodeASTC(const VkImageSubresourceRange &subresourceRange) const
1196 {
1197 	ASSERT(decompressedImage);
1198 
1199 	int xBlockSize = format.blockWidth();
1200 	int yBlockSize = format.blockHeight();
1201 	int zBlockSize = 1;
1202 	bool isUnsigned = format.isUnsignedComponent(0);
1203 
1204 	uint32_t lastLayer = getLastLayerIndex(subresourceRange);
1205 	uint32_t lastMipLevel = getLastMipLevel(subresourceRange);
1206 
1207 	int bytes = decompressedImage->format.bytes();
1208 
1209 	VkImageSubresourceLayers subresourceLayers = { subresourceRange.aspectMask, subresourceRange.baseMipLevel, subresourceRange.baseArrayLayer, 1 };
1210 	for(; subresourceLayers.baseArrayLayer <= lastLayer; subresourceLayers.baseArrayLayer++)
1211 	{
1212 		for(; subresourceLayers.mipLevel <= lastMipLevel; subresourceLayers.mipLevel++)
1213 		{
1214 			VkExtent3D mipLevelExtent = getMipLevelExtent(static_cast<VkImageAspectFlagBits>(subresourceLayers.aspectMask), subresourceLayers.mipLevel);
1215 
1216 			int xblocks = (mipLevelExtent.width + xBlockSize - 1) / xBlockSize;
1217 			int yblocks = (mipLevelExtent.height + yBlockSize - 1) / yBlockSize;
1218 			int zblocks = (zBlockSize > 1) ? (mipLevelExtent.depth + zBlockSize - 1) / zBlockSize : 1;
1219 
1220 			if(xblocks <= 0 || yblocks <= 0 || zblocks <= 0)
1221 			{
1222 				continue;
1223 			}
1224 
1225 			int pitchB = decompressedImage->rowPitchBytes(VK_IMAGE_ASPECT_COLOR_BIT, subresourceLayers.mipLevel);
1226 			int sliceB = decompressedImage->slicePitchBytes(VK_IMAGE_ASPECT_COLOR_BIT, subresourceLayers.mipLevel);
1227 
1228 			for(int32_t depth = 0; depth < static_cast<int32_t>(mipLevelExtent.depth); depth++)
1229 			{
1230 				uint8_t *source = static_cast<uint8_t *>(getTexelPointer({ 0, 0, depth }, subresourceLayers));
1231 				uint8_t *dest = static_cast<uint8_t *>(decompressedImage->getTexelPointer({ 0, 0, depth }, subresourceLayers));
1232 
1233 				ASTC_Decoder::Decode(source, dest, mipLevelExtent.width, mipLevelExtent.height, mipLevelExtent.depth, bytes, pitchB, sliceB,
1234 				                     xBlockSize, yBlockSize, zBlockSize, xblocks, yblocks, zblocks, isUnsigned);
1235 			}
1236 		}
1237 	}
1238 }
1239 
1240 }  // namespace vk
1241