1 /*------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
4 *
5 * Copyright (c) 2016 The Khronos Group Inc.
6 *
7 * Licensed under the Apache License, Version 2.0 (the "License");
8 * you may not use this file except in compliance with the License.
9 * You may obtain a copy of the License at
10 *
11 * http://www.apache.org/licenses/LICENSE-2.0
12 *
13 * Unless required by applicable law or agreed to in writing, software
14 * distributed under the License is distributed on an "AS IS" BASIS,
15 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 * See the License for the specific language governing permissions and
17 * limitations under the License.
18 *
19 *//*!
20 * \file vktSparseResourcesBufferSparseResidency.cpp
21 * \brief Sparse partially resident buffers tests
22 *//*--------------------------------------------------------------------*/
23
24 #include "vktSparseResourcesBufferSparseResidency.hpp"
25 #include "vktSparseResourcesTestsUtil.hpp"
26 #include "vktSparseResourcesBase.hpp"
27 #include "vktTestCaseUtil.hpp"
28
29 #include "vkDefs.hpp"
30 #include "vkRef.hpp"
31 #include "vkRefUtil.hpp"
32 #include "vkPlatform.hpp"
33 #include "vkPrograms.hpp"
34 #include "vkRefUtil.hpp"
35 #include "vkMemUtil.hpp"
36 #include "vkBarrierUtil.hpp"
37 #include "vkQueryUtil.hpp"
38 #include "vkBuilderUtil.hpp"
39 #include "vkTypeUtil.hpp"
40 #include "vkCmdUtil.hpp"
41 #include "vkObjUtil.hpp"
42
43 #include "deStringUtil.hpp"
44 #include "deUniquePtr.hpp"
45
46 #include <string>
47 #include <vector>
48
49 using namespace vk;
50
51 namespace vkt
52 {
53 namespace sparse
54 {
55 namespace
56 {
57
58 enum ShaderParameters
59 {
60 SIZE_OF_UINT_IN_SHADER = 4u,
61 };
62
63 class BufferSparseResidencyCase : public TestCase
64 {
65 public:
66 BufferSparseResidencyCase (tcu::TestContext& testCtx,
67 const std::string& name,
68 const deUint32 bufferSize,
69 const glu::GLSLVersion glslVersion,
70 const bool useDeviceGroups);
71
72
73 void initPrograms (SourceCollections& sourceCollections) const;
74 TestInstance* createInstance (Context& context) const;
75
76 private:
77 const deUint32 m_bufferSize;
78 const glu::GLSLVersion m_glslVersion;
79 const bool m_useDeviceGroups;
80
81 };
82
BufferSparseResidencyCase(tcu::TestContext & testCtx,const std::string & name,const deUint32 bufferSize,const glu::GLSLVersion glslVersion,const bool useDeviceGroups)83 BufferSparseResidencyCase::BufferSparseResidencyCase (tcu::TestContext& testCtx,
84 const std::string& name,
85 const deUint32 bufferSize,
86 const glu::GLSLVersion glslVersion,
87 const bool useDeviceGroups)
88
89 : TestCase (testCtx, name)
90 , m_bufferSize (bufferSize)
91 , m_glslVersion (glslVersion)
92 , m_useDeviceGroups (useDeviceGroups)
93 {
94 }
95
initPrograms(SourceCollections & sourceCollections) const96 void BufferSparseResidencyCase::initPrograms (SourceCollections& sourceCollections) const
97 {
98 const char* const versionDecl = glu::getGLSLVersionDeclaration(m_glslVersion);
99 const deUint32 iterationsCount = m_bufferSize / SIZE_OF_UINT_IN_SHADER;
100
101 std::ostringstream src;
102
103 src << versionDecl << "\n"
104 << "layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;\n"
105 << "layout(set = 0, binding = 0, std430) readonly buffer Input\n"
106 << "{\n"
107 << " uint data[];\n"
108 << "} sb_in;\n"
109 << "\n"
110 << "layout(set = 0, binding = 1, std430) writeonly buffer Output\n"
111 << "{\n"
112 << " uint result[];\n"
113 << "} sb_out;\n"
114 << "\n"
115 << "void main (void)\n"
116 << "{\n"
117 << " for(int i=0; i<" << iterationsCount << "; ++i) \n"
118 << " {\n"
119 << " sb_out.result[i] = sb_in.data[i];"
120 << " }\n"
121 << "}\n";
122
123 sourceCollections.glslSources.add("comp") << glu::ComputeSource(src.str());
124 }
125
126 class BufferSparseResidencyInstance : public SparseResourcesBaseInstance
127 {
128 public:
129 BufferSparseResidencyInstance (Context& context,
130 const deUint32 bufferSize,
131 const bool useDeviceGroups);
132
133 tcu::TestStatus iterate (void);
134
135 private:
136 const deUint32 m_bufferSize;
137 };
138
BufferSparseResidencyInstance(Context & context,const deUint32 bufferSize,const bool useDeviceGroups)139 BufferSparseResidencyInstance::BufferSparseResidencyInstance (Context& context,
140 const deUint32 bufferSize,
141 const bool useDeviceGroups)
142 : SparseResourcesBaseInstance (context, useDeviceGroups)
143 , m_bufferSize (bufferSize)
144 {
145 }
146
iterate(void)147 tcu::TestStatus BufferSparseResidencyInstance::iterate (void)
148 {
149 const InstanceInterface& instance = m_context.getInstanceInterface();
150 {
151 // Create logical device supporting both sparse and compute operations
152 QueueRequirementsVec queueRequirements;
153 queueRequirements.push_back(QueueRequirements(VK_QUEUE_SPARSE_BINDING_BIT, 1u));
154 queueRequirements.push_back(QueueRequirements(VK_QUEUE_COMPUTE_BIT, 1u));
155
156 createDeviceSupportingQueues(queueRequirements);
157 }
158 const VkPhysicalDevice physicalDevice = getPhysicalDevice();
159 const VkPhysicalDeviceProperties physicalDeviceProperties = getPhysicalDeviceProperties(instance, physicalDevice);
160
161 if (!getPhysicalDeviceFeatures(instance, physicalDevice).sparseResidencyBuffer)
162 TCU_THROW(NotSupportedError, "Sparse partially resident buffers not supported");
163
164 const DeviceInterface& deviceInterface = getDeviceInterface();
165 const Queue& sparseQueue = getQueue(VK_QUEUE_SPARSE_BINDING_BIT, 0);
166 const Queue& computeQueue = getQueue(VK_QUEUE_COMPUTE_BIT, 0);
167
168 // Go through all physical devices
169 for (deUint32 physDevID = 0; physDevID < m_numPhysicalDevices; physDevID++)
170 {
171 const deUint32 firstDeviceID = physDevID;
172 const deUint32 secondDeviceID = (firstDeviceID + 1) % m_numPhysicalDevices;
173
174 VkBufferCreateInfo bufferCreateInfo =
175 {
176 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
177 DE_NULL, // const void* pNext;
178 VK_BUFFER_CREATE_SPARSE_BINDING_BIT |
179 VK_BUFFER_CREATE_SPARSE_RESIDENCY_BIT, // VkBufferCreateFlags flags;
180 m_bufferSize, // VkDeviceSize size;
181 VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
182 VK_BUFFER_USAGE_TRANSFER_SRC_BIT, // VkBufferUsageFlags usage;
183 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
184 0u, // deUint32 queueFamilyIndexCount;
185 DE_NULL // const deUint32* pQueueFamilyIndices;
186 };
187
188 const deUint32 queueFamilyIndices[] = { sparseQueue.queueFamilyIndex, computeQueue.queueFamilyIndex };
189
190 if (sparseQueue.queueFamilyIndex != computeQueue.queueFamilyIndex)
191 {
192 bufferCreateInfo.sharingMode = VK_SHARING_MODE_CONCURRENT;
193 bufferCreateInfo.queueFamilyIndexCount = 2u;
194 bufferCreateInfo.pQueueFamilyIndices = queueFamilyIndices;
195 }
196
197 // Create sparse buffer
198 const Unique<VkBuffer> sparseBuffer(createBuffer(deviceInterface, getDevice(), &bufferCreateInfo));
199
200 // Create sparse buffer memory bind semaphore
201 const Unique<VkSemaphore> bufferMemoryBindSemaphore(createSemaphore(deviceInterface, getDevice()));
202
203 const VkMemoryRequirements bufferMemRequirements = getBufferMemoryRequirements(deviceInterface, getDevice(), *sparseBuffer);
204
205 if (bufferMemRequirements.size > physicalDeviceProperties.limits.sparseAddressSpaceSize)
206 TCU_THROW(NotSupportedError, "Required memory size for sparse resources exceeds device limits");
207
208 DE_ASSERT((bufferMemRequirements.size % bufferMemRequirements.alignment) == 0);
209
210 const deUint32 numSparseSlots = static_cast<deUint32>(bufferMemRequirements.size / bufferMemRequirements.alignment);
211 std::vector<DeviceMemorySp> deviceMemUniquePtrVec;
212
213 {
214 std::vector<VkSparseMemoryBind> sparseMemoryBinds;
215 const deUint32 memoryType = findMatchingMemoryType(instance, getPhysicalDevice(secondDeviceID), bufferMemRequirements, MemoryRequirement::Any);
216
217 if (memoryType == NO_MATCH_FOUND)
218 return tcu::TestStatus::fail("No matching memory type found");
219
220 if (firstDeviceID != secondDeviceID)
221 {
222 VkPeerMemoryFeatureFlags peerMemoryFeatureFlags = (VkPeerMemoryFeatureFlags)0;
223 const deUint32 heapIndex = getHeapIndexForMemoryType(instance, getPhysicalDevice(secondDeviceID), memoryType);
224 deviceInterface.getDeviceGroupPeerMemoryFeatures(getDevice(), heapIndex, firstDeviceID, secondDeviceID, &peerMemoryFeatureFlags);
225
226 if (((peerMemoryFeatureFlags & VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT) == 0) ||
227 ((peerMemoryFeatureFlags & VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT) == 0))
228 {
229 TCU_THROW(NotSupportedError, "Peer memory does not support COPY_SRC and GENERIC_DST");
230 }
231 }
232
233 for (deUint32 sparseBindNdx = 0; sparseBindNdx < numSparseSlots; sparseBindNdx += 2)
234 {
235 const VkSparseMemoryBind sparseMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(), bufferMemRequirements.alignment, memoryType, bufferMemRequirements.alignment * sparseBindNdx);
236
237 deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(sparseMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));
238
239 sparseMemoryBinds.push_back(sparseMemoryBind);
240 }
241
242 const VkSparseBufferMemoryBindInfo sparseBufferBindInfo = makeSparseBufferMemoryBindInfo(*sparseBuffer, static_cast<deUint32>(sparseMemoryBinds.size()), &sparseMemoryBinds[0]);
243
244 const VkDeviceGroupBindSparseInfo devGroupBindSparseInfo =
245 {
246 VK_STRUCTURE_TYPE_DEVICE_GROUP_BIND_SPARSE_INFO, //VkStructureType sType;
247 DE_NULL, //const void* pNext;
248 firstDeviceID, //deUint32 resourceDeviceIndex;
249 secondDeviceID, //deUint32 memoryDeviceIndex;
250 };
251 const VkBindSparseInfo bindSparseInfo =
252 {
253 VK_STRUCTURE_TYPE_BIND_SPARSE_INFO, //VkStructureType sType;
254 usingDeviceGroups() ? &devGroupBindSparseInfo : DE_NULL,//const void* pNext;
255 0u, //deUint32 waitSemaphoreCount;
256 DE_NULL, //const VkSemaphore* pWaitSemaphores;
257 1u, //deUint32 bufferBindCount;
258 &sparseBufferBindInfo, //const VkSparseBufferMemoryBindInfo* pBufferBinds;
259 0u, //deUint32 imageOpaqueBindCount;
260 DE_NULL, //const VkSparseImageOpaqueMemoryBindInfo* pImageOpaqueBinds;
261 0u, //deUint32 imageBindCount;
262 DE_NULL, //const VkSparseImageMemoryBindInfo* pImageBinds;
263 1u, //deUint32 signalSemaphoreCount;
264 &bufferMemoryBindSemaphore.get() //const VkSemaphore* pSignalSemaphores;
265 };
266
267 VK_CHECK(deviceInterface.queueBindSparse(sparseQueue.queueHandle, 1u, &bindSparseInfo, DE_NULL));
268 }
269
270 // Create input buffer
271 const VkBufferCreateInfo inputBufferCreateInfo = makeBufferCreateInfo(m_bufferSize, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT);
272 const Unique<VkBuffer> inputBuffer (createBuffer(deviceInterface, getDevice(), &inputBufferCreateInfo));
273 const de::UniquePtr<Allocation> inputBufferAlloc (bindBuffer(deviceInterface, getDevice(), getAllocator(), *inputBuffer, MemoryRequirement::HostVisible));
274
275
276 std::vector<deUint8> referenceData;
277 referenceData.resize(m_bufferSize);
278
279 for (deUint32 valueNdx = 0; valueNdx < m_bufferSize; ++valueNdx)
280 {
281 referenceData[valueNdx] = static_cast<deUint8>((valueNdx % bufferMemRequirements.alignment) + 1u);
282 }
283
284 deMemcpy(inputBufferAlloc->getHostPtr(), &referenceData[0], m_bufferSize);
285
286 flushAlloc(deviceInterface, getDevice(), *inputBufferAlloc);
287
288 // Create output buffer
289 const VkBufferCreateInfo outputBufferCreateInfo = makeBufferCreateInfo(m_bufferSize, VK_BUFFER_USAGE_TRANSFER_DST_BIT);
290 const Unique<VkBuffer> outputBuffer (createBuffer(deviceInterface, getDevice(), &outputBufferCreateInfo));
291 const de::UniquePtr<Allocation> outputBufferAlloc (bindBuffer(deviceInterface, getDevice(), getAllocator(), *outputBuffer, MemoryRequirement::HostVisible));
292
293 // Create command buffer for compute and data transfer operations
294 const Unique<VkCommandPool> commandPool(makeCommandPool(deviceInterface, getDevice(), computeQueue.queueFamilyIndex));
295 const Unique<VkCommandBuffer> commandBuffer(allocateCommandBuffer(deviceInterface, getDevice(), *commandPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY));
296
297 // Start recording compute and transfer commands
298 beginCommandBuffer(deviceInterface, *commandBuffer);
299
300 // Create descriptor set
301 const Unique<VkDescriptorSetLayout> descriptorSetLayout(
302 DescriptorSetLayoutBuilder()
303 .addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, VK_SHADER_STAGE_COMPUTE_BIT)
304 .addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, VK_SHADER_STAGE_COMPUTE_BIT)
305 .build(deviceInterface, getDevice()));
306
307 // Create compute pipeline
308 const Unique<VkShaderModule> shaderModule(createShaderModule(deviceInterface, getDevice(), m_context.getBinaryCollection().get("comp"), DE_NULL));
309 const Unique<VkPipelineLayout> pipelineLayout(makePipelineLayout(deviceInterface, getDevice(), *descriptorSetLayout));
310 const Unique<VkPipeline> computePipeline(makeComputePipeline(deviceInterface, getDevice(), *pipelineLayout, *shaderModule));
311
312 deviceInterface.cmdBindPipeline(*commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *computePipeline);
313
314 const Unique<VkDescriptorPool> descriptorPool(
315 DescriptorPoolBuilder()
316 .addType(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 2u)
317 .build(deviceInterface, getDevice(), VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, 1u));
318
319 const Unique<VkDescriptorSet> descriptorSet(makeDescriptorSet(deviceInterface, getDevice(), *descriptorPool, *descriptorSetLayout));
320
321 {
322 const VkDescriptorBufferInfo inputBufferInfo = makeDescriptorBufferInfo(*inputBuffer, 0ull, m_bufferSize);
323 const VkDescriptorBufferInfo sparseBufferInfo = makeDescriptorBufferInfo(*sparseBuffer, 0ull, m_bufferSize);
324
325 DescriptorSetUpdateBuilder()
326 .writeSingle(*descriptorSet, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &inputBufferInfo)
327 .writeSingle(*descriptorSet, DescriptorSetUpdateBuilder::Location::binding(1u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &sparseBufferInfo)
328 .update(deviceInterface, getDevice());
329 }
330
331 deviceInterface.cmdBindDescriptorSets(*commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *pipelineLayout, 0u, 1u, &descriptorSet.get(), 0u, DE_NULL);
332
333 {
334 const VkBufferMemoryBarrier inputBufferBarrier
335 = makeBufferMemoryBarrier( VK_ACCESS_HOST_WRITE_BIT,
336 VK_ACCESS_SHADER_READ_BIT,
337 *inputBuffer,
338 0ull,
339 m_bufferSize);
340
341 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0u, 0u, DE_NULL, 1u, &inputBufferBarrier, 0u, DE_NULL);
342 }
343
344 deviceInterface.cmdDispatch(*commandBuffer, 1u, 1u, 1u);
345
346 {
347 const VkBufferMemoryBarrier sparseBufferBarrier
348 = makeBufferMemoryBarrier( VK_ACCESS_SHADER_WRITE_BIT,
349 VK_ACCESS_TRANSFER_READ_BIT,
350 *sparseBuffer,
351 0ull,
352 m_bufferSize);
353
354 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, DE_NULL, 1u, &sparseBufferBarrier, 0u, DE_NULL);
355 }
356
357 {
358 const VkBufferCopy bufferCopy = makeBufferCopy(0u, 0u, m_bufferSize);
359
360 deviceInterface.cmdCopyBuffer(*commandBuffer, *sparseBuffer, *outputBuffer, 1u, &bufferCopy);
361 }
362
363 {
364 const VkBufferMemoryBarrier outputBufferBarrier
365 = makeBufferMemoryBarrier( VK_ACCESS_TRANSFER_WRITE_BIT,
366 VK_ACCESS_HOST_READ_BIT,
367 *outputBuffer,
368 0ull,
369 m_bufferSize);
370
371 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0u, 0u, DE_NULL, 1u, &outputBufferBarrier, 0u, DE_NULL);
372 }
373
374 // End recording compute and transfer commands
375 endCommandBuffer(deviceInterface, *commandBuffer);
376
377 const VkPipelineStageFlags waitStageBits[] = { VK_PIPELINE_STAGE_TRANSFER_BIT };
378
379 // Submit transfer commands for execution and wait for completion
380 submitCommandsAndWait(deviceInterface, getDevice(), computeQueue.queueHandle, *commandBuffer, 1u, &bufferMemoryBindSemaphore.get(),
381 waitStageBits, 0, DE_NULL, usingDeviceGroups(), firstDeviceID);
382
383 // Retrieve data from output buffer to host memory
384 invalidateAlloc(deviceInterface, getDevice(), *outputBufferAlloc);
385
386 const deUint8* outputData = static_cast<const deUint8*>(outputBufferAlloc->getHostPtr());
387
388 // Wait for sparse queue to become idle
389 deviceInterface.queueWaitIdle(sparseQueue.queueHandle);
390
391 // Compare output data with reference data
392 for (deUint32 sparseBindNdx = 0; sparseBindNdx < numSparseSlots; ++sparseBindNdx)
393 {
394 const deUint32 alignment = static_cast<deUint32>(bufferMemRequirements.alignment);
395 const deUint32 offset = alignment * sparseBindNdx;
396 const deUint32 size = sparseBindNdx == (numSparseSlots - 1) ? m_bufferSize % alignment : alignment;
397
398 if (sparseBindNdx % 2u == 0u)
399 {
400 if (deMemCmp(&referenceData[offset], outputData + offset, size) != 0)
401 return tcu::TestStatus::fail("Failed");
402 }
403 else if (physicalDeviceProperties.sparseProperties.residencyNonResidentStrict)
404 {
405 deMemset(&referenceData[offset], 0u, size);
406
407 if (deMemCmp(&referenceData[offset], outputData + offset, size) != 0)
408 return tcu::TestStatus::fail("Failed");
409 }
410 }
411 }
412
413 return tcu::TestStatus::pass("Passed");
414 }
415
createInstance(Context & context) const416 TestInstance* BufferSparseResidencyCase::createInstance (Context& context) const
417 {
418 return new BufferSparseResidencyInstance(context, m_bufferSize, m_useDeviceGroups);
419 }
420
421 } // anonymous ns
422
addBufferSparseResidencyTests(tcu::TestCaseGroup * group,const bool useDeviceGroups)423 void addBufferSparseResidencyTests(tcu::TestCaseGroup* group, const bool useDeviceGroups)
424 {
425 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_10", 1 << 10, glu::GLSL_VERSION_440, useDeviceGroups));
426 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_12", 1 << 12, glu::GLSL_VERSION_440, useDeviceGroups));
427 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_16", 1 << 16, glu::GLSL_VERSION_440, useDeviceGroups));
428 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_17", 1 << 17, glu::GLSL_VERSION_440, useDeviceGroups));
429 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_20", 1 << 20, glu::GLSL_VERSION_440, useDeviceGroups));
430 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_24", 1 << 24, glu::GLSL_VERSION_440, useDeviceGroups));
431 }
432
433 } // sparse
434 } // vkt
435