• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*------------------------------------------------------------------------
2  * Vulkan Conformance Tests
3  * ------------------------
4  *
5  * Copyright (c) 2016 The Khronos Group Inc.
6  * Copyright (c) 2016 The Android Open Source Project
7  *
8  * Licensed under the Apache License, Version 2.0 (the "License");
9  * you may not use this file except in compliance with the License.
10  * You may obtain a copy of the License at
11  *
12  *      http://www.apache.org/licenses/LICENSE-2.0
13  *
14  * Unless required by applicable law or agreed to in writing, software
15  * distributed under the License is distributed on an "AS IS" BASIS,
16  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17  * See the License for the specific language governing permissions and
18  * limitations under the License.
19  *
20  *//*!
21  * \file
22  * \brief Multisampled image load/store Tests
23  *//*--------------------------------------------------------------------*/
24 
25 #include "vktImageMultisampleLoadStoreTests.hpp"
26 #include "vktTestCaseUtil.hpp"
27 #include "vktImageTestsUtil.hpp"
28 #include "vktImageLoadStoreUtil.hpp"
29 #include "vktImageTexture.hpp"
30 
31 #include "vkDefs.hpp"
32 #include "vkRef.hpp"
33 #include "vkRefUtil.hpp"
34 #include "vkPlatform.hpp"
35 #include "vkPrograms.hpp"
36 #include "vkMemUtil.hpp"
37 #include "vkBarrierUtil.hpp"
38 #include "vkBuilderUtil.hpp"
39 #include "vkQueryUtil.hpp"
40 #include "vkImageUtil.hpp"
41 #include "vkCmdUtil.hpp"
42 #include "vkObjUtil.hpp"
43 
44 #include "deUniquePtr.hpp"
45 
46 #include "tcuTextureUtil.hpp"
47 #include "tcuTestLog.hpp"
48 
49 #include <string>
50 #include <vector>
51 
52 namespace vkt
53 {
54 namespace image
55 {
56 namespace
57 {
58 using namespace vk;
59 using de::MovePtr;
60 using de::UniquePtr;
61 using tcu::IVec3;
62 
63 static const VkFormat CHECKSUM_IMAGE_FORMAT = VK_FORMAT_R32_SINT;
64 
65 struct CaseDef
66 {
67 	Texture					texture;
68 	VkFormat				format;
69 	VkSampleCountFlagBits	numSamples;
70 	bool					singleLayerBind;
71 };
72 
73 //  Multisampled storage image test.
74 //
75 //  Pass 1: Write a slightly different color pattern per-sample to the whole image.
76 //  Pass 2: Read samples of the same image and check if color values are in the expected range.
77 //          Write back results as a checksum image and verify them on the host.
78 //  Each checksum image pixel should contain an integer equal to the number of samples.
79 
initPrograms(SourceCollections & programCollection,const CaseDef caseDef)80 void initPrograms (SourceCollections& programCollection, const CaseDef caseDef)
81 {
82 	const int			dimension			= (caseDef.singleLayerBind ? caseDef.texture.layerDimension() : caseDef.texture.dimension());
83 	const std::string	texelCoordStr		= (dimension == 1 ? "gx" : dimension == 2 ? "ivec2(gx, gy)" : dimension == 3 ? "ivec3(gx, gy, gz)" : "");
84 
85 	const ImageType		usedImageType		= (caseDef.singleLayerBind ? getImageTypeForSingleLayer(caseDef.texture.type()) : caseDef.texture.type());
86 	const std::string	formatQualifierStr	= getShaderImageFormatQualifier(mapVkFormat(caseDef.format));
87 	const std::string	msImageTypeStr		= getShaderImageType(mapVkFormat(caseDef.format), usedImageType, (caseDef.texture.numSamples() > 1));
88 
89 	const std::string	xMax				= de::toString(caseDef.texture.size().x() - 1);
90 	const std::string	yMax				= de::toString(caseDef.texture.size().y() - 1);
91 	const std::string	signednessPrefix	= isUintFormat(caseDef.format) ? "u" : isIntFormat(caseDef.format) ? "i" : "";
92 	const std::string	gvec4Expr			= signednessPrefix + "vec4";
93 	const int			numColorComponents	= tcu::getNumUsedChannels(mapVkFormat(caseDef.format).order);
94 
95 	const float			storeColorScale		= computeStoreColorScale(caseDef.format, caseDef.texture.size());
96 	const float			storeColorBias		= computeStoreColorBias(caseDef.format);
97 	DE_ASSERT(colorScaleAndBiasAreValid(caseDef.format, storeColorScale, storeColorBias));
98 
99 	const std::string	colorScaleExpr		= (storeColorScale == 1.0f ? "" : "*" + de::toString(storeColorScale))
100 											+ (storeColorBias == 0.0f ? "" : " + float(" + de::toString(storeColorBias) + ")");
101 	const std::string	colorExpr			=
102 		gvec4Expr + "("
103 		+                           "gx^gy^gz^(sampleNdx >> 5)^(sampleNdx & 31), "		// we "split" sampleNdx to keep this value in [0, 31] range for numSamples = 64 case
104 		+ (numColorComponents > 1 ? "(" + xMax + "-gx)^gy^gz, "              : "0, ")
105 		+ (numColorComponents > 2 ? "gx^(" + yMax + "-gy)^gz, "              : "0, ")
106 		+ (numColorComponents > 3 ? "(" + xMax + "-gx)^(" + yMax + "-gy)^gz" : "1")
107 		+ ")" + colorScaleExpr;
108 
109 	// Store shader
110 	{
111 		std::ostringstream src;
112 		src << glu::getGLSLVersionDeclaration(glu::GLSL_VERSION_450) << "\n"
113 			<< "\n"
114 			<< "layout(local_size_x = 1) in;\n"
115 			<< "layout(set = 0, binding = 1, " << formatQualifierStr << ") writeonly uniform " << msImageTypeStr << " u_msImage;\n";
116 
117 		if (caseDef.singleLayerBind)
118 			src << "layout(set = 0, binding = 0) readonly uniform Constants {\n"
119 				<< "    int u_layerNdx;\n"
120 				<< "};\n";
121 
122 		src << "\n"
123 			<< "void main (void)\n"
124 			<< "{\n"
125 			<< "    int gx = int(gl_GlobalInvocationID.x);\n"
126 			<< "    int gy = int(gl_GlobalInvocationID.y);\n"
127 			<< "    int gz = " << (caseDef.singleLayerBind ? "u_layerNdx" : "int(gl_GlobalInvocationID.z)") << ";\n"
128 			<< "\n"
129 			<< "    for (int sampleNdx = 0; sampleNdx < " << caseDef.texture.numSamples() <<"; ++sampleNdx) {\n"
130 			<< "        imageStore(u_msImage, " << texelCoordStr << ", sampleNdx, " << colorExpr << ");\n"
131 			<< "    }\n"
132 			<< "}\n";
133 
134 		programCollection.glslSources.add("comp_store") << glu::ComputeSource(src.str());
135 	}
136 
137 	// Load shader
138 	{
139 		const tcu::TextureFormat	checksumFormat			= mapVkFormat(CHECKSUM_IMAGE_FORMAT);
140 		const std::string			checksumImageTypeStr	= getShaderImageType(checksumFormat, usedImageType);
141 		const bool					useExactCompare			= isIntegerFormat(caseDef.format);
142 
143 		std::ostringstream src;
144 		src << glu::getGLSLVersionDeclaration(glu::GLSL_VERSION_450) << "\n"
145 			<< "\n"
146 			<< "layout(local_size_x = 1) in;\n"
147 			<< "layout(set = 0, binding = 1, " << formatQualifierStr << ") readonly  uniform " << msImageTypeStr << " u_msImage;\n"
148 			<< "layout(set = 0, binding = 2, " << getShaderImageFormatQualifier(checksumFormat) << ") writeonly uniform " << checksumImageTypeStr << " u_checksumImage;\n";
149 
150 		if (caseDef.singleLayerBind)
151 			src << "layout(set = 0, binding = 0) readonly uniform Constants {\n"
152 				<< "    int u_layerNdx;\n"
153 				<< "};\n";
154 
155 		src << "\n"
156 			<< "void main (void)\n"
157 			<< "{\n"
158 			<< "    int gx = int(gl_GlobalInvocationID.x);\n"
159 			<< "    int gy = int(gl_GlobalInvocationID.y);\n"
160 			<< "    int gz = " << (caseDef.singleLayerBind ? "u_layerNdx" : "int(gl_GlobalInvocationID.z)") << ";\n"
161 			<< "\n"
162 			<< "    int checksum = 0;\n"
163 			<< "    for (int sampleNdx = 0; sampleNdx < " << caseDef.texture.numSamples() <<"; ++sampleNdx) {\n"
164 			<< "        " << gvec4Expr << " color = imageLoad(u_msImage, " << texelCoordStr << ", sampleNdx);\n";
165 
166 		if (useExactCompare)
167 			src << "        if (color == " << colorExpr << ")\n"
168 				<< "            ++checksum;\n";
169 		else
170 			src << "        " << gvec4Expr << " diff  = abs(abs(color) - abs(" << colorExpr << "));\n"
171 				<< "        if (all(lessThan(diff, " << gvec4Expr << "(0.02))))\n"
172 				<< "            ++checksum;\n";
173 
174 		src << "    }\n"
175 			<< "\n"
176 			<< "    imageStore(u_checksumImage, " << texelCoordStr << ", ivec4(checksum));\n"
177 			<< "}\n";
178 
179 		programCollection.glslSources.add("comp_load") << glu::ComputeSource(src.str());
180 	}
181 }
182 
checkSupport(Context & context,const CaseDef caseDef)183 void checkSupport (Context& context, const CaseDef caseDef)
184 {
185 	context.requireDeviceCoreFeature(DEVICE_CORE_FEATURE_SHADER_STORAGE_IMAGE_MULTISAMPLE);
186 
187 	VkImageFormatProperties		imageFormatProperties;
188 	const VkResult				imageFormatResult		= context.getInstanceInterface().getPhysicalDeviceImageFormatProperties(
189 		context.getPhysicalDevice(), caseDef.format, VK_IMAGE_TYPE_2D, VK_IMAGE_TILING_OPTIMAL, VK_IMAGE_USAGE_STORAGE_BIT, (VkImageCreateFlags)0, &imageFormatProperties);
190 
191 	if (imageFormatResult == VK_ERROR_FORMAT_NOT_SUPPORTED)
192 		TCU_THROW(NotSupportedError, "Format is not supported");
193 
194 	if ((imageFormatProperties.sampleCounts & caseDef.numSamples) != caseDef.numSamples)
195 		TCU_THROW(NotSupportedError, "Requested sample count is not supported");
196 }
197 
198 //! Helper function to deal with per-layer resources.
insertImageViews(const DeviceInterface & vk,const VkDevice device,const CaseDef & caseDef,const VkFormat format,const VkImage image,std::vector<SharedVkImageView> * const pOutImageViews)199 void insertImageViews (const DeviceInterface& vk, const VkDevice device, const CaseDef& caseDef, const VkFormat format, const VkImage image, std::vector<SharedVkImageView>* const pOutImageViews)
200 {
201 	if (caseDef.singleLayerBind)
202 	{
203 		pOutImageViews->clear();
204 		pOutImageViews->resize(caseDef.texture.numLayers());
205 		for (int layerNdx = 0; layerNdx < caseDef.texture.numLayers(); ++layerNdx)
206 		{
207 			(*pOutImageViews)[layerNdx] = makeVkSharedPtr(makeImageView(
208 				vk, device, image, mapImageViewType(getImageTypeForSingleLayer(caseDef.texture.type())), format,
209 				makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 1u, layerNdx, 1u)));
210 		}
211 	}
212 	else // bind all layers at once
213 	{
214 		pOutImageViews->clear();
215 		pOutImageViews->resize(1);
216 		(*pOutImageViews)[0] = makeVkSharedPtr(makeImageView(
217 			vk, device, image, mapImageViewType(caseDef.texture.type()), format,
218 			makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 1u, 0u, caseDef.texture.numLayers())));
219 	}
220 }
221 
222 //! Helper function to deal with per-layer resources.
insertDescriptorSets(const DeviceInterface & vk,const VkDevice device,const CaseDef & caseDef,const VkDescriptorPool descriptorPool,const VkDescriptorSetLayout descriptorSetLayout,std::vector<SharedVkDescriptorSet> * const pOutDescriptorSets)223 void insertDescriptorSets (const DeviceInterface& vk, const VkDevice device, const CaseDef& caseDef, const VkDescriptorPool descriptorPool, const VkDescriptorSetLayout descriptorSetLayout, std::vector<SharedVkDescriptorSet>* const pOutDescriptorSets)
224 {
225 	if (caseDef.singleLayerBind)
226 	{
227 		pOutDescriptorSets->clear();
228 		pOutDescriptorSets->resize(caseDef.texture.numLayers());
229 		for (int layerNdx = 0; layerNdx < caseDef.texture.numLayers(); ++layerNdx)
230 			(*pOutDescriptorSets)[layerNdx] = makeVkSharedPtr(makeDescriptorSet(vk, device, descriptorPool, descriptorSetLayout));
231 	}
232 	else // bind all layers at once
233 	{
234 		pOutDescriptorSets->clear();
235 		pOutDescriptorSets->resize(1);
236 		(*pOutDescriptorSets)[0] = makeVkSharedPtr(makeDescriptorSet(vk, device, descriptorPool, descriptorSetLayout));
237 	}
238 }
239 
test(Context & context,const CaseDef caseDef)240 tcu::TestStatus test (Context& context, const CaseDef caseDef)
241 {
242 	const InstanceInterface&	vki					= context.getInstanceInterface();
243 	const VkPhysicalDevice		physDevice			= context.getPhysicalDevice();
244 	const DeviceInterface&		vk					= context.getDeviceInterface();
245 	const VkDevice				device				= context.getDevice();
246 	const VkQueue				queue				= context.getUniversalQueue();
247 	const deUint32				queueFamilyIndex	= context.getUniversalQueueFamilyIndex();
248 	Allocator&					allocator			= context.getDefaultAllocator();
249 
250 	// Images
251 
252 	const UniquePtr<Image> msImage(new Image(
253 		vk, device, allocator, makeImageCreateInfo(caseDef.texture, caseDef.format, VK_IMAGE_USAGE_STORAGE_BIT, 0u), MemoryRequirement::Any));
254 
255 	const UniquePtr<Image> checksumImage(new Image(
256 		vk, device, allocator,
257 		makeImageCreateInfo(Texture(caseDef.texture, 1), CHECKSUM_IMAGE_FORMAT, VK_IMAGE_USAGE_STORAGE_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT, 0u),
258 		MemoryRequirement::Any));
259 
260 	// Buffer used to pass constants to the shader.
261 
262 	const int			numLayers					= caseDef.texture.numLayers();
263 	const VkDeviceSize	bufferChunkSize				= getOptimalUniformBufferChunkSize(vki, physDevice, sizeof(deInt32));
264 	const VkDeviceSize	constantsBufferSizeBytes	= numLayers * bufferChunkSize;
265 	UniquePtr<Buffer>	constantsBuffer				(new Buffer(vk, device, allocator, makeBufferCreateInfo(constantsBufferSizeBytes, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT),
266 													 MemoryRequirement::HostVisible));
267 
268 	{
269 		const Allocation&	alloc	= constantsBuffer->getAllocation();
270 		deUint8* const		basePtr = static_cast<deUint8*>(alloc.getHostPtr());
271 
272 		deMemset(alloc.getHostPtr(), 0, static_cast<size_t>(constantsBufferSizeBytes));
273 
274 		for (int layerNdx = 0; layerNdx < numLayers; ++layerNdx)
275 		{
276 			deInt32* const valuePtr = reinterpret_cast<deInt32*>(basePtr + layerNdx * bufferChunkSize);
277 			*valuePtr = layerNdx;
278 		}
279 
280 		flushAlloc(vk, device, alloc);
281 	}
282 
283 	const VkDeviceSize	resultBufferSizeBytes	= getImageSizeBytes(caseDef.texture.size(), CHECKSUM_IMAGE_FORMAT);
284 	UniquePtr<Buffer>	resultBuffer			(new Buffer(vk, device, allocator, makeBufferCreateInfo(resultBufferSizeBytes, VK_BUFFER_USAGE_TRANSFER_DST_BIT),
285 												 MemoryRequirement::HostVisible));
286 
287 	{
288 		const Allocation& alloc = resultBuffer->getAllocation();
289 		deMemset(alloc.getHostPtr(), 0, static_cast<size_t>(resultBufferSizeBytes));
290 		flushAlloc(vk, device, alloc);
291 	}
292 
293 	// Descriptors
294 
295 	Unique<VkDescriptorSetLayout> descriptorSetLayout(DescriptorSetLayoutBuilder()
296 		.addSingleBinding(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, VK_SHADER_STAGE_COMPUTE_BIT)
297 		.addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, VK_SHADER_STAGE_COMPUTE_BIT)
298 		.addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, VK_SHADER_STAGE_COMPUTE_BIT)
299 		.build(vk, device));
300 
301 	Unique<VkDescriptorPool> descriptorPool(DescriptorPoolBuilder()
302 		.addType(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, numLayers)
303 		.addType(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, numLayers)
304 		.addType(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, numLayers)
305 		.build(vk, device, VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, numLayers));
306 
307 	std::vector<SharedVkDescriptorSet>	allDescriptorSets;
308 	std::vector<SharedVkImageView>		allMultisampledImageViews;
309 	std::vector<SharedVkImageView>		allChecksumImageViews;
310 
311 	insertDescriptorSets(vk, device, caseDef, *descriptorPool, *descriptorSetLayout, &allDescriptorSets);
312 	insertImageViews	(vk, device, caseDef, caseDef.format, **msImage, &allMultisampledImageViews);
313 	insertImageViews	(vk, device, caseDef, CHECKSUM_IMAGE_FORMAT, **checksumImage, &allChecksumImageViews);
314 
315 	// Prepare commands
316 
317 	const Unique<VkPipelineLayout>	pipelineLayout	(makePipelineLayout(vk, device, *descriptorSetLayout));
318 	const Unique<VkCommandPool>		cmdPool			(createCommandPool(vk, device, VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT, queueFamilyIndex));
319 	const Unique<VkCommandBuffer>	cmdBuffer		(allocateCommandBuffer(vk, device, *cmdPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY));
320 
321 	const tcu::IVec3				workSize				= (caseDef.singleLayerBind ? caseDef.texture.layerSize() : caseDef.texture.size());
322 	const int						loopNumLayers			= (caseDef.singleLayerBind ? numLayers : 1);
323 	const VkImageSubresourceRange	subresourceAllLayers	= makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 1u, 0u, caseDef.texture.numLayers());
324 
325 	// Pass 1: Write MS image
326 	{
327 		const Unique<VkShaderModule>	shaderModule	(createShaderModule	(vk, device, context.getBinaryCollection().get("comp_store"), 0));
328 		const Unique<VkPipeline>		pipeline		(makeComputePipeline(vk, device, *pipelineLayout, *shaderModule));
329 
330 		beginCommandBuffer(vk, *cmdBuffer);
331 		vk.cmdBindPipeline(*cmdBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
332 
333 		{
334 			const VkImageMemoryBarrier barriers[] =
335 			{
336 				makeImageMemoryBarrier((VkAccessFlags)0, VK_ACCESS_SHADER_WRITE_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_GENERAL, **msImage, subresourceAllLayers),
337 				makeImageMemoryBarrier((VkAccessFlags)0, VK_ACCESS_SHADER_WRITE_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_GENERAL, **checksumImage, subresourceAllLayers),
338 			};
339 
340 			vk.cmdPipelineBarrier(*cmdBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, (VkDependencyFlags)0,
341 				0u, DE_NULL, 0u, DE_NULL, DE_LENGTH_OF_ARRAY(barriers), barriers);
342 		}
343 
344 		for (int layerNdx = 0; layerNdx < loopNumLayers; ++layerNdx)
345 		{
346 			const VkDescriptorSet			descriptorSet					= **allDescriptorSets[layerNdx];
347 			const VkDescriptorImageInfo		descriptorMultiImageInfo		= makeDescriptorImageInfo(DE_NULL, **allMultisampledImageViews[layerNdx], VK_IMAGE_LAYOUT_GENERAL);
348 			const VkDescriptorBufferInfo	descriptorConstantsBufferInfo	= makeDescriptorBufferInfo(constantsBuffer->get(), layerNdx*bufferChunkSize, bufferChunkSize);
349 
350 			DescriptorSetUpdateBuilder()
351 				.writeSingle(descriptorSet, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, &descriptorConstantsBufferInfo)
352 				.writeSingle(descriptorSet, DescriptorSetUpdateBuilder::Location::binding(1u), VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, &descriptorMultiImageInfo)
353 				.update(vk, device);
354 
355 			vk.cmdBindDescriptorSets(*cmdBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *pipelineLayout, 0u, 1u, &descriptorSet, 0u, DE_NULL);
356 			vk.cmdDispatch(*cmdBuffer, workSize.x(), workSize.y(), workSize.z());
357 		}
358 
359 		endCommandBuffer(vk, *cmdBuffer);
360 		submitCommandsAndWait(vk, device, queue, *cmdBuffer);
361 	}
362 
363 	// Pass 2: "Resolve" MS image in compute shader
364 	{
365 		const Unique<VkShaderModule>	shaderModule	(createShaderModule	(vk, device, context.getBinaryCollection().get("comp_load"), 0));
366 		const Unique<VkPipeline>		pipeline		(makeComputePipeline(vk, device, *pipelineLayout, *shaderModule));
367 
368 		beginCommandBuffer(vk, *cmdBuffer);
369 		vk.cmdBindPipeline(*cmdBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
370 
371 		{
372 			const VkImageMemoryBarrier barriers[] =
373 			{
374 				makeImageMemoryBarrier(VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT, VK_IMAGE_LAYOUT_GENERAL, VK_IMAGE_LAYOUT_GENERAL, **msImage, subresourceAllLayers),
375 			};
376 
377 			vk.cmdPipelineBarrier(*cmdBuffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, (VkDependencyFlags)0,
378 				0u, DE_NULL, 0u, DE_NULL, DE_LENGTH_OF_ARRAY(barriers), barriers);
379 		}
380 
381 		for (int layerNdx = 0; layerNdx < loopNumLayers; ++layerNdx)
382 		{
383 			const VkDescriptorSet			descriptorSet					= **allDescriptorSets[layerNdx];
384 			const VkDescriptorImageInfo		descriptorMultiImageInfo		= makeDescriptorImageInfo(DE_NULL, **allMultisampledImageViews[layerNdx], VK_IMAGE_LAYOUT_GENERAL);
385 			const VkDescriptorImageInfo		descriptorChecksumImageInfo		= makeDescriptorImageInfo(DE_NULL, **allChecksumImageViews[layerNdx], VK_IMAGE_LAYOUT_GENERAL);
386 			const VkDescriptorBufferInfo	descriptorConstantsBufferInfo	= makeDescriptorBufferInfo(constantsBuffer->get(), layerNdx*bufferChunkSize, bufferChunkSize);
387 
388 			DescriptorSetUpdateBuilder()
389 				.writeSingle(descriptorSet, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, &descriptorConstantsBufferInfo)
390 				.writeSingle(descriptorSet, DescriptorSetUpdateBuilder::Location::binding(1u), VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, &descriptorMultiImageInfo)
391 				.writeSingle(descriptorSet, DescriptorSetUpdateBuilder::Location::binding(2u), VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, &descriptorChecksumImageInfo)
392 				.update(vk, device);
393 
394 			vk.cmdBindDescriptorSets(*cmdBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *pipelineLayout, 0u, 1u, &descriptorSet, 0u, DE_NULL);
395 			vk.cmdDispatch(*cmdBuffer, workSize.x(), workSize.y(), workSize.z());
396 		}
397 
398 		endCommandBuffer(vk, *cmdBuffer);
399 		submitCommandsAndWait(vk, device, queue, *cmdBuffer);
400 	}
401 
402 	// Retrieve result
403 	{
404 		beginCommandBuffer(vk, *cmdBuffer);
405 
406 		{
407 			const VkImageMemoryBarrier barriers[] =
408 			{
409 				makeImageMemoryBarrier(VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT, VK_IMAGE_LAYOUT_GENERAL, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, **checksumImage, subresourceAllLayers),
410 			};
411 			vk.cmdPipelineBarrier(*cmdBuffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, (VkDependencyFlags)0,
412 				0u, DE_NULL, 0u, DE_NULL, DE_LENGTH_OF_ARRAY(barriers), barriers);
413 		}
414 		{
415 			const VkBufferImageCopy copyRegion = makeBufferImageCopy(makeExtent3D(caseDef.texture.layerSize()), caseDef.texture.numLayers());
416 			vk.cmdCopyImageToBuffer(*cmdBuffer, **checksumImage, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, **resultBuffer, 1u, &copyRegion);
417 		}
418 		{
419 			const VkBufferMemoryBarrier barriers[] =
420 			{
421 				makeBufferMemoryBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_HOST_READ_BIT, **resultBuffer, 0ull, resultBufferSizeBytes),
422 			};
423 			vk.cmdPipelineBarrier(*cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, (VkDependencyFlags)0,
424 				0u, DE_NULL, DE_LENGTH_OF_ARRAY(barriers), barriers, 0u, DE_NULL);
425 		}
426 
427 		endCommandBuffer(vk, *cmdBuffer);
428 		submitCommandsAndWait(vk, device, queue, *cmdBuffer);
429 	}
430 
431 	// Verify
432 	{
433 		const Allocation& alloc = resultBuffer->getAllocation();
434 		invalidateAlloc(vk, device, alloc);
435 
436 		const IVec3		imageSize			= caseDef.texture.size();
437 		const deInt32*	pDataPtr			= static_cast<deInt32*>(alloc.getHostPtr());
438 		const deInt32	expectedChecksum	= caseDef.texture.numSamples();
439 
440 		for (int layer = 0; layer < imageSize.z(); ++layer)
441 		for (int y = 0; y < imageSize.y(); ++y)
442 		for (int x = 0; x < imageSize.x(); ++x)
443 		{
444 			if (*pDataPtr != expectedChecksum)
445 			{
446 				context.getTestContext().getLog()
447 					<< tcu::TestLog::Message << "Some sample colors were incorrect at (x, y, layer) = (" << x << ", " << y << ", " << layer << ")"	<< tcu::TestLog::EndMessage
448 					<< tcu::TestLog::Message << "Checksum value is " << *pDataPtr << " but expected " << expectedChecksum << tcu::TestLog::EndMessage;
449 
450 				return tcu::TestStatus::fail("Some sample colors were incorrect");
451 			}
452 			++pDataPtr;
453 		}
454 
455 		return tcu::TestStatus::pass("OK");
456 	}
457 }
458 
459 } // anonymous ns
460 
createImageMultisampleLoadStoreTests(tcu::TestContext & testCtx)461 tcu::TestCaseGroup* createImageMultisampleLoadStoreTests (tcu::TestContext& testCtx)
462 {
463 	const Texture textures[] =
464 	{
465 		// \note Shader code is tweaked to work with image size of 32, take a look if this needs to be modified.
466 		Texture(IMAGE_TYPE_2D,			tcu::IVec3(32,	32,	1),		1),
467 		Texture(IMAGE_TYPE_2D_ARRAY,	tcu::IVec3(32,	32,	1),		4),
468 	};
469 
470 	static const VkFormat formats[] =
471 	{
472 		VK_FORMAT_R32G32B32A32_SFLOAT,
473 		VK_FORMAT_R16G16B16A16_SFLOAT,
474 		VK_FORMAT_R32_SFLOAT,
475 
476 		VK_FORMAT_R32G32B32A32_UINT,
477 		VK_FORMAT_R16G16B16A16_UINT,
478 		VK_FORMAT_R8G8B8A8_UINT,
479 		VK_FORMAT_R32_UINT,
480 
481 		VK_FORMAT_R32G32B32A32_SINT,
482 		VK_FORMAT_R16G16B16A16_SINT,
483 		VK_FORMAT_R8G8B8A8_SINT,
484 		VK_FORMAT_R32_SINT,
485 
486 		VK_FORMAT_R8G8B8A8_UNORM,
487 
488 		VK_FORMAT_R8G8B8A8_SNORM,
489 	};
490 
491 	static const VkSampleCountFlagBits samples[] =
492 	{
493 		VK_SAMPLE_COUNT_2_BIT,
494 		VK_SAMPLE_COUNT_4_BIT,
495 		VK_SAMPLE_COUNT_8_BIT,
496 		VK_SAMPLE_COUNT_16_BIT,
497 		VK_SAMPLE_COUNT_32_BIT,
498 		VK_SAMPLE_COUNT_64_BIT,
499 	};
500 
501 	MovePtr<tcu::TestCaseGroup> testGroup(new tcu::TestCaseGroup(testCtx, "load_store_multisample", "Multisampled image store and load"));
502 
503 	for (int baseTextureNdx = 0; baseTextureNdx < DE_LENGTH_OF_ARRAY(textures); ++baseTextureNdx)
504 	{
505 		const Texture&				baseTexture			= textures[baseTextureNdx];
506 		MovePtr<tcu::TestCaseGroup>	imageViewGroup		(new tcu::TestCaseGroup(testCtx, getImageTypeName(baseTexture.type()).c_str(), ""));
507 		const int					numLayerBindModes	= (baseTexture.numLayers() == 1 ? 1 : 2);
508 
509 		for (int formatNdx = 0; formatNdx < DE_LENGTH_OF_ARRAY(formats); ++formatNdx)
510 		for (int layerBindMode = 0; layerBindMode < numLayerBindModes; ++layerBindMode)
511 		{
512 			const bool					singleLayerBind	= (layerBindMode != 0);
513 			const std::string			formatGroupName	= getFormatShortString(formats[formatNdx]) + (singleLayerBind ? "_single_layer" : "");
514 			MovePtr<tcu::TestCaseGroup>	formatGroup		(new tcu::TestCaseGroup(testCtx, formatGroupName.c_str(), ""));
515 
516 			for (int samplesNdx = 0; samplesNdx < DE_LENGTH_OF_ARRAY(samples); ++samplesNdx)
517 			{
518 				const std::string	samplesCaseName = "samples_" + de::toString(samples[samplesNdx]);
519 
520 				const CaseDef		caseDef =
521 				{
522 					Texture(baseTexture, samples[samplesNdx]),
523 					formats[formatNdx],
524 					samples[samplesNdx],
525 					singleLayerBind,
526 				};
527 
528 				addFunctionCaseWithPrograms(formatGroup.get(), samplesCaseName, "", checkSupport, initPrograms, test, caseDef);
529 			}
530 			imageViewGroup->addChild(formatGroup.release());
531 		}
532 		testGroup->addChild(imageViewGroup.release());
533 	}
534 
535 	return testGroup.release();
536 }
537 
538 } // image
539 } // vkt
540