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 std::string& description,
69 const deUint32 bufferSize,
70 const glu::GLSLVersion glslVersion,
71 const bool useDeviceGroups);
72
73
74 void initPrograms (SourceCollections& sourceCollections) const;
75 TestInstance* createInstance (Context& context) const;
76
77 private:
78 const deUint32 m_bufferSize;
79 const glu::GLSLVersion m_glslVersion;
80 const bool m_useDeviceGroups;
81
82 };
83
BufferSparseResidencyCase(tcu::TestContext & testCtx,const std::string & name,const std::string & description,const deUint32 bufferSize,const glu::GLSLVersion glslVersion,const bool useDeviceGroups)84 BufferSparseResidencyCase::BufferSparseResidencyCase (tcu::TestContext& testCtx,
85 const std::string& name,
86 const std::string& description,
87 const deUint32 bufferSize,
88 const glu::GLSLVersion glslVersion,
89 const bool useDeviceGroups)
90
91 : TestCase (testCtx, name, description)
92 , m_bufferSize (bufferSize)
93 , m_glslVersion (glslVersion)
94 , m_useDeviceGroups (useDeviceGroups)
95 {
96 }
97
initPrograms(SourceCollections & sourceCollections) const98 void BufferSparseResidencyCase::initPrograms (SourceCollections& sourceCollections) const
99 {
100 const char* const versionDecl = glu::getGLSLVersionDeclaration(m_glslVersion);
101 const deUint32 iterationsCount = m_bufferSize / SIZE_OF_UINT_IN_SHADER;
102
103 std::ostringstream src;
104
105 src << versionDecl << "\n"
106 << "layout (local_size_x = 1, local_size_y = 1, local_size_z = 1) in;\n"
107 << "layout(set = 0, binding = 0, std430) readonly buffer Input\n"
108 << "{\n"
109 << " uint data[];\n"
110 << "} sb_in;\n"
111 << "\n"
112 << "layout(set = 0, binding = 1, std430) writeonly buffer Output\n"
113 << "{\n"
114 << " uint result[];\n"
115 << "} sb_out;\n"
116 << "\n"
117 << "void main (void)\n"
118 << "{\n"
119 << " for(int i=0; i<" << iterationsCount << "; ++i) \n"
120 << " {\n"
121 << " sb_out.result[i] = sb_in.data[i];"
122 << " }\n"
123 << "}\n";
124
125 sourceCollections.glslSources.add("comp") << glu::ComputeSource(src.str());
126 }
127
128 class BufferSparseResidencyInstance : public SparseResourcesBaseInstance
129 {
130 public:
131 BufferSparseResidencyInstance (Context& context,
132 const deUint32 bufferSize,
133 const bool useDeviceGroups);
134
135 tcu::TestStatus iterate (void);
136
137 private:
138 const deUint32 m_bufferSize;
139 };
140
BufferSparseResidencyInstance(Context & context,const deUint32 bufferSize,const bool useDeviceGroups)141 BufferSparseResidencyInstance::BufferSparseResidencyInstance (Context& context,
142 const deUint32 bufferSize,
143 const bool useDeviceGroups)
144 : SparseResourcesBaseInstance (context, useDeviceGroups)
145 , m_bufferSize (bufferSize)
146 {
147 }
148
iterate(void)149 tcu::TestStatus BufferSparseResidencyInstance::iterate (void)
150 {
151 const InstanceInterface& instance = m_context.getInstanceInterface();
152 {
153 // Create logical device supporting both sparse and compute operations
154 QueueRequirementsVec queueRequirements;
155 queueRequirements.push_back(QueueRequirements(VK_QUEUE_SPARSE_BINDING_BIT, 1u));
156 queueRequirements.push_back(QueueRequirements(VK_QUEUE_COMPUTE_BIT, 1u));
157
158 createDeviceSupportingQueues(queueRequirements);
159 }
160 const VkPhysicalDevice physicalDevice = getPhysicalDevice();
161 const VkPhysicalDeviceProperties physicalDeviceProperties = getPhysicalDeviceProperties(instance, physicalDevice);
162
163 if (!getPhysicalDeviceFeatures(instance, physicalDevice).sparseResidencyBuffer)
164 TCU_THROW(NotSupportedError, "Sparse partially resident buffers not supported");
165
166 const DeviceInterface& deviceInterface = getDeviceInterface();
167 const Queue& sparseQueue = getQueue(VK_QUEUE_SPARSE_BINDING_BIT, 0);
168 const Queue& computeQueue = getQueue(VK_QUEUE_COMPUTE_BIT, 0);
169
170 // Go through all physical devices
171 for (deUint32 physDevID = 0; physDevID < m_numPhysicalDevices; physDevID++)
172 {
173 const deUint32 firstDeviceID = physDevID;
174 const deUint32 secondDeviceID = (firstDeviceID + 1) % m_numPhysicalDevices;
175
176 VkBufferCreateInfo bufferCreateInfo =
177 {
178 VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType;
179 DE_NULL, // const void* pNext;
180 VK_BUFFER_CREATE_SPARSE_BINDING_BIT |
181 VK_BUFFER_CREATE_SPARSE_RESIDENCY_BIT, // VkBufferCreateFlags flags;
182 m_bufferSize, // VkDeviceSize size;
183 VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
184 VK_BUFFER_USAGE_TRANSFER_SRC_BIT, // VkBufferUsageFlags usage;
185 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
186 0u, // deUint32 queueFamilyIndexCount;
187 DE_NULL // const deUint32* pQueueFamilyIndices;
188 };
189
190 const deUint32 queueFamilyIndices[] = { sparseQueue.queueFamilyIndex, computeQueue.queueFamilyIndex };
191
192 if (sparseQueue.queueFamilyIndex != computeQueue.queueFamilyIndex)
193 {
194 bufferCreateInfo.sharingMode = VK_SHARING_MODE_CONCURRENT;
195 bufferCreateInfo.queueFamilyIndexCount = 2u;
196 bufferCreateInfo.pQueueFamilyIndices = queueFamilyIndices;
197 }
198
199 // Create sparse buffer
200 const Unique<VkBuffer> sparseBuffer(createBuffer(deviceInterface, getDevice(), &bufferCreateInfo));
201
202 // Create sparse buffer memory bind semaphore
203 const Unique<VkSemaphore> bufferMemoryBindSemaphore(createSemaphore(deviceInterface, getDevice()));
204
205 const VkMemoryRequirements bufferMemRequirements = getBufferMemoryRequirements(deviceInterface, getDevice(), *sparseBuffer);
206
207 if (bufferMemRequirements.size > physicalDeviceProperties.limits.sparseAddressSpaceSize)
208 TCU_THROW(NotSupportedError, "Required memory size for sparse resources exceeds device limits");
209
210 DE_ASSERT((bufferMemRequirements.size % bufferMemRequirements.alignment) == 0);
211
212 const deUint32 numSparseSlots = static_cast<deUint32>(bufferMemRequirements.size / bufferMemRequirements.alignment);
213 std::vector<DeviceMemorySp> deviceMemUniquePtrVec;
214
215 {
216 std::vector<VkSparseMemoryBind> sparseMemoryBinds;
217 const deUint32 memoryType = findMatchingMemoryType(instance, getPhysicalDevice(secondDeviceID), bufferMemRequirements, MemoryRequirement::Any);
218
219 if (memoryType == NO_MATCH_FOUND)
220 return tcu::TestStatus::fail("No matching memory type found");
221
222 if (firstDeviceID != secondDeviceID)
223 {
224 VkPeerMemoryFeatureFlags peerMemoryFeatureFlags = (VkPeerMemoryFeatureFlags)0;
225 const deUint32 heapIndex = getHeapIndexForMemoryType(instance, getPhysicalDevice(secondDeviceID), memoryType);
226 deviceInterface.getDeviceGroupPeerMemoryFeatures(getDevice(), heapIndex, firstDeviceID, secondDeviceID, &peerMemoryFeatureFlags);
227
228 if (((peerMemoryFeatureFlags & VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT) == 0) ||
229 ((peerMemoryFeatureFlags & VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT) == 0))
230 {
231 TCU_THROW(NotSupportedError, "Peer memory does not support COPY_SRC and GENERIC_DST");
232 }
233 }
234
235 for (deUint32 sparseBindNdx = 0; sparseBindNdx < numSparseSlots; sparseBindNdx += 2)
236 {
237 const VkSparseMemoryBind sparseMemoryBind = makeSparseMemoryBind(deviceInterface, getDevice(), bufferMemRequirements.alignment, memoryType, bufferMemRequirements.alignment * sparseBindNdx);
238
239 deviceMemUniquePtrVec.push_back(makeVkSharedPtr(Move<VkDeviceMemory>(check<VkDeviceMemory>(sparseMemoryBind.memory), Deleter<VkDeviceMemory>(deviceInterface, getDevice(), DE_NULL))));
240
241 sparseMemoryBinds.push_back(sparseMemoryBind);
242 }
243
244 const VkSparseBufferMemoryBindInfo sparseBufferBindInfo = makeSparseBufferMemoryBindInfo(*sparseBuffer, static_cast<deUint32>(sparseMemoryBinds.size()), &sparseMemoryBinds[0]);
245
246 const VkDeviceGroupBindSparseInfo devGroupBindSparseInfo =
247 {
248 VK_STRUCTURE_TYPE_DEVICE_GROUP_BIND_SPARSE_INFO_KHR, //VkStructureType sType;
249 DE_NULL, //const void* pNext;
250 firstDeviceID, //deUint32 resourceDeviceIndex;
251 secondDeviceID, //deUint32 memoryDeviceIndex;
252 };
253 const VkBindSparseInfo bindSparseInfo =
254 {
255 VK_STRUCTURE_TYPE_BIND_SPARSE_INFO, //VkStructureType sType;
256 usingDeviceGroups() ? &devGroupBindSparseInfo : DE_NULL,//const void* pNext;
257 0u, //deUint32 waitSemaphoreCount;
258 DE_NULL, //const VkSemaphore* pWaitSemaphores;
259 1u, //deUint32 bufferBindCount;
260 &sparseBufferBindInfo, //const VkSparseBufferMemoryBindInfo* pBufferBinds;
261 0u, //deUint32 imageOpaqueBindCount;
262 DE_NULL, //const VkSparseImageOpaqueMemoryBindInfo* pImageOpaqueBinds;
263 0u, //deUint32 imageBindCount;
264 DE_NULL, //const VkSparseImageMemoryBindInfo* pImageBinds;
265 1u, //deUint32 signalSemaphoreCount;
266 &bufferMemoryBindSemaphore.get() //const VkSemaphore* pSignalSemaphores;
267 };
268
269 VK_CHECK(deviceInterface.queueBindSparse(sparseQueue.queueHandle, 1u, &bindSparseInfo, DE_NULL));
270 }
271
272 // Create input buffer
273 const VkBufferCreateInfo inputBufferCreateInfo = makeBufferCreateInfo(m_bufferSize, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT);
274 const Unique<VkBuffer> inputBuffer (createBuffer(deviceInterface, getDevice(), &inputBufferCreateInfo));
275 const de::UniquePtr<Allocation> inputBufferAlloc (bindBuffer(deviceInterface, getDevice(), getAllocator(), *inputBuffer, MemoryRequirement::HostVisible));
276
277
278 std::vector<deUint8> referenceData;
279 referenceData.resize(m_bufferSize);
280
281 for (deUint32 valueNdx = 0; valueNdx < m_bufferSize; ++valueNdx)
282 {
283 referenceData[valueNdx] = static_cast<deUint8>((valueNdx % bufferMemRequirements.alignment) + 1u);
284 }
285
286 deMemcpy(inputBufferAlloc->getHostPtr(), &referenceData[0], m_bufferSize);
287
288 flushAlloc(deviceInterface, getDevice(), *inputBufferAlloc);
289
290 // Create output buffer
291 const VkBufferCreateInfo outputBufferCreateInfo = makeBufferCreateInfo(m_bufferSize, VK_BUFFER_USAGE_TRANSFER_DST_BIT);
292 const Unique<VkBuffer> outputBuffer (createBuffer(deviceInterface, getDevice(), &outputBufferCreateInfo));
293 const de::UniquePtr<Allocation> outputBufferAlloc (bindBuffer(deviceInterface, getDevice(), getAllocator(), *outputBuffer, MemoryRequirement::HostVisible));
294
295 // Create command buffer for compute and data transfer operations
296 const Unique<VkCommandPool> commandPool(makeCommandPool(deviceInterface, getDevice(), computeQueue.queueFamilyIndex));
297 const Unique<VkCommandBuffer> commandBuffer(allocateCommandBuffer(deviceInterface, getDevice(), *commandPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY));
298
299 // Start recording compute and transfer commands
300 beginCommandBuffer(deviceInterface, *commandBuffer);
301
302 // Create descriptor set
303 const Unique<VkDescriptorSetLayout> descriptorSetLayout(
304 DescriptorSetLayoutBuilder()
305 .addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, VK_SHADER_STAGE_COMPUTE_BIT)
306 .addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, VK_SHADER_STAGE_COMPUTE_BIT)
307 .build(deviceInterface, getDevice()));
308
309 // Create compute pipeline
310 const Unique<VkShaderModule> shaderModule(createShaderModule(deviceInterface, getDevice(), m_context.getBinaryCollection().get("comp"), DE_NULL));
311 const Unique<VkPipelineLayout> pipelineLayout(makePipelineLayout(deviceInterface, getDevice(), *descriptorSetLayout));
312 const Unique<VkPipeline> computePipeline(makeComputePipeline(deviceInterface, getDevice(), *pipelineLayout, *shaderModule));
313
314 deviceInterface.cmdBindPipeline(*commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *computePipeline);
315
316 const Unique<VkDescriptorPool> descriptorPool(
317 DescriptorPoolBuilder()
318 .addType(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 2u)
319 .build(deviceInterface, getDevice(), VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, 1u));
320
321 const Unique<VkDescriptorSet> descriptorSet(makeDescriptorSet(deviceInterface, getDevice(), *descriptorPool, *descriptorSetLayout));
322
323 {
324 const VkDescriptorBufferInfo inputBufferInfo = makeDescriptorBufferInfo(*inputBuffer, 0ull, m_bufferSize);
325 const VkDescriptorBufferInfo sparseBufferInfo = makeDescriptorBufferInfo(*sparseBuffer, 0ull, m_bufferSize);
326
327 DescriptorSetUpdateBuilder()
328 .writeSingle(*descriptorSet, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &inputBufferInfo)
329 .writeSingle(*descriptorSet, DescriptorSetUpdateBuilder::Location::binding(1u), VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, &sparseBufferInfo)
330 .update(deviceInterface, getDevice());
331 }
332
333 deviceInterface.cmdBindDescriptorSets(*commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, *pipelineLayout, 0u, 1u, &descriptorSet.get(), 0u, DE_NULL);
334
335 {
336 const VkBufferMemoryBarrier inputBufferBarrier
337 = makeBufferMemoryBarrier( VK_ACCESS_HOST_WRITE_BIT,
338 VK_ACCESS_SHADER_READ_BIT,
339 *inputBuffer,
340 0ull,
341 m_bufferSize);
342
343 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0u, 0u, DE_NULL, 1u, &inputBufferBarrier, 0u, DE_NULL);
344 }
345
346 deviceInterface.cmdDispatch(*commandBuffer, 1u, 1u, 1u);
347
348 {
349 const VkBufferMemoryBarrier sparseBufferBarrier
350 = makeBufferMemoryBarrier( VK_ACCESS_SHADER_WRITE_BIT,
351 VK_ACCESS_TRANSFER_READ_BIT,
352 *sparseBuffer,
353 0ull,
354 m_bufferSize);
355
356 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0u, 0u, DE_NULL, 1u, &sparseBufferBarrier, 0u, DE_NULL);
357 }
358
359 {
360 const VkBufferCopy bufferCopy = makeBufferCopy(0u, 0u, m_bufferSize);
361
362 deviceInterface.cmdCopyBuffer(*commandBuffer, *sparseBuffer, *outputBuffer, 1u, &bufferCopy);
363 }
364
365 {
366 const VkBufferMemoryBarrier outputBufferBarrier
367 = makeBufferMemoryBarrier( VK_ACCESS_TRANSFER_WRITE_BIT,
368 VK_ACCESS_HOST_READ_BIT,
369 *outputBuffer,
370 0ull,
371 m_bufferSize);
372
373 deviceInterface.cmdPipelineBarrier(*commandBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0u, 0u, DE_NULL, 1u, &outputBufferBarrier, 0u, DE_NULL);
374 }
375
376 // End recording compute and transfer commands
377 endCommandBuffer(deviceInterface, *commandBuffer);
378
379 const VkPipelineStageFlags waitStageBits[] = { VK_PIPELINE_STAGE_TRANSFER_BIT };
380
381 // Submit transfer commands for execution and wait for completion
382 submitCommandsAndWait(deviceInterface, getDevice(), computeQueue.queueHandle, *commandBuffer, 1u, &bufferMemoryBindSemaphore.get(),
383 waitStageBits, 0, DE_NULL, usingDeviceGroups(), firstDeviceID);
384
385 // Retrieve data from output buffer to host memory
386 invalidateAlloc(deviceInterface, getDevice(), *outputBufferAlloc);
387
388 const deUint8* outputData = static_cast<const deUint8*>(outputBufferAlloc->getHostPtr());
389
390 // Wait for sparse queue to become idle
391 deviceInterface.queueWaitIdle(sparseQueue.queueHandle);
392
393 // Compare output data with reference data
394 for (deUint32 sparseBindNdx = 0; sparseBindNdx < numSparseSlots; ++sparseBindNdx)
395 {
396 const deUint32 alignment = static_cast<deUint32>(bufferMemRequirements.alignment);
397 const deUint32 offset = alignment * sparseBindNdx;
398 const deUint32 size = sparseBindNdx == (numSparseSlots - 1) ? m_bufferSize % alignment : alignment;
399
400 if (sparseBindNdx % 2u == 0u)
401 {
402 if (deMemCmp(&referenceData[offset], outputData + offset, size) != 0)
403 return tcu::TestStatus::fail("Failed");
404 }
405 else if (physicalDeviceProperties.sparseProperties.residencyNonResidentStrict)
406 {
407 deMemset(&referenceData[offset], 0u, size);
408
409 if (deMemCmp(&referenceData[offset], outputData + offset, size) != 0)
410 return tcu::TestStatus::fail("Failed");
411 }
412 }
413 }
414
415 return tcu::TestStatus::pass("Passed");
416 }
417
createInstance(Context & context) const418 TestInstance* BufferSparseResidencyCase::createInstance (Context& context) const
419 {
420 return new BufferSparseResidencyInstance(context, m_bufferSize, m_useDeviceGroups);
421 }
422
423 } // anonymous ns
424
addBufferSparseResidencyTests(tcu::TestCaseGroup * group,const bool useDeviceGroups)425 void addBufferSparseResidencyTests(tcu::TestCaseGroup* group, const bool useDeviceGroups)
426 {
427 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_10", "", 1 << 10, glu::GLSL_VERSION_440, useDeviceGroups));
428 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_12", "", 1 << 12, glu::GLSL_VERSION_440, useDeviceGroups));
429 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_16", "", 1 << 16, glu::GLSL_VERSION_440, useDeviceGroups));
430 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_17", "", 1 << 17, glu::GLSL_VERSION_440, useDeviceGroups));
431 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_20", "", 1 << 20, glu::GLSL_VERSION_440, useDeviceGroups));
432 group->addChild(new BufferSparseResidencyCase(group->getTestContext(), "buffer_size_2_24", "", 1 << 24, glu::GLSL_VERSION_440, useDeviceGroups));
433 }
434
435 } // sparse
436 } // vkt
437