1 /*------------------------------------------------------------------------
2 * Vulkan Conformance Tests
3 * ------------------------
4 *
5 * Copyright (c) 2019 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
21 * \brief Ray Tracing Build tests
22 *//*--------------------------------------------------------------------*/
23
24 #include "vktRayTracingBuildTests.hpp"
25
26 #include "vkDefs.hpp"
27
28 #include "vktTestCase.hpp"
29 #include "vkCmdUtil.hpp"
30 #include "vkObjUtil.hpp"
31 #include "vkBuilderUtil.hpp"
32 #include "vkBarrierUtil.hpp"
33 #include "vkBufferWithMemory.hpp"
34 #include "vkImageWithMemory.hpp"
35 #include "vkImageUtil.hpp"
36 #include "vkTypeUtil.hpp"
37
38 #include "tcuTextureUtil.hpp"
39
40 #include "vkRayTracingUtil.hpp"
41
42 #include "deClock.h"
43
44 #include <cmath>
45 #include <limits>
46 #include <iostream>
47
48 namespace vkt
49 {
50 namespace RayTracing
51 {
52 namespace
53 {
54 using namespace vk;
55 using namespace std;
56
57 static const VkFlags ALL_RAY_TRACING_STAGES = VK_SHADER_STAGE_RAYGEN_BIT_KHR
58 | VK_SHADER_STAGE_ANY_HIT_BIT_KHR
59 | VK_SHADER_STAGE_CLOSEST_HIT_BIT_KHR
60 | VK_SHADER_STAGE_MISS_BIT_KHR
61 | VK_SHADER_STAGE_INTERSECTION_BIT_KHR
62 | VK_SHADER_STAGE_CALLABLE_BIT_KHR;
63
64 enum TestType
65 {
66 TEST_TYPE_TRIANGLES,
67 TEST_TYPE_AABBS,
68 TEST_TYPE_MIXED,
69 };
70
71 struct CaseDef
72 {
73 TestType testType;
74 deUint32 width;
75 deUint32 height;
76 deUint32 squaresGroupCount;
77 deUint32 geometriesGroupCount;
78 deUint32 instancesGroupCount;
79 bool deferredOperation;
80 deUint32 workerThreadsCount;
81 bool deviceBuild;
82 };
83
getShaderGroupSize(const InstanceInterface& vki, const VkPhysicalDevice physicalDevice)84 deUint32 getShaderGroupSize (const InstanceInterface& vki,
85 const VkPhysicalDevice physicalDevice)
86 {
87 de::MovePtr<RayTracingProperties> rayTracingPropertiesKHR;
88
89 rayTracingPropertiesKHR = makeRayTracingProperties(vki, physicalDevice);
90 return rayTracingPropertiesKHR->getShaderGroupHandleSize();
91 }
92
getShaderGroupBaseAlignment(const InstanceInterface& vki, const VkPhysicalDevice physicalDevice)93 deUint32 getShaderGroupBaseAlignment (const InstanceInterface& vki,
94 const VkPhysicalDevice physicalDevice)
95 {
96 de::MovePtr<RayTracingProperties> rayTracingPropertiesKHR;
97
98 rayTracingPropertiesKHR = makeRayTracingProperties(vki, physicalDevice);
99 return rayTracingPropertiesKHR->getShaderGroupBaseAlignment();
100 }
101
makeImageCreateInfo(deUint32 width, deUint32 height, VkFormat format)102 VkImageCreateInfo makeImageCreateInfo (deUint32 width, deUint32 height, VkFormat format)
103 {
104 const VkImageUsageFlags usage = VK_IMAGE_USAGE_STORAGE_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;
105 const VkImageCreateInfo imageCreateInfo =
106 {
107 VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType;
108 DE_NULL, // const void* pNext;
109 (VkImageCreateFlags)0u, // VkImageCreateFlags flags;
110 VK_IMAGE_TYPE_2D, // VkImageType imageType;
111 format, // VkFormat format;
112 makeExtent3D(width, height, 1u), // VkExtent3D extent;
113 1u, // deUint32 mipLevels;
114 1u, // deUint32 arrayLayers;
115 VK_SAMPLE_COUNT_1_BIT, // VkSampleCountFlagBits samples;
116 VK_IMAGE_TILING_OPTIMAL, // VkImageTiling tiling;
117 usage, // VkImageUsageFlags usage;
118 VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode;
119 0u, // deUint32 queueFamilyIndexCount;
120 DE_NULL, // const deUint32* pQueueFamilyIndices;
121 VK_IMAGE_LAYOUT_UNDEFINED // VkImageLayout initialLayout;
122 };
123
124 return imageCreateInfo;
125 }
126
127 class RayTracingBuildTestInstance : public TestInstance
128 {
129 public:
130 typedef de::SharedPtr<BottomLevelAccelerationStructure> BlasPtr;
131 typedef de::SharedPtr<TopLevelAccelerationStructure> TlasPtr;
132 typedef BottomLevelAccelerationStructurePool BlasPool;
133
134 RayTracingBuildTestInstance (Context& context, const CaseDef& data);
135 ~RayTracingBuildTestInstance (void);
136 tcu::TestStatus iterate (void);
137
138 protected:
139 bool verifyAllocationCount () const;
140 void checkSupportInInstance (void) const;
141 deUint32 validateBuffer (de::MovePtr<BufferWithMemory> buffer);
142 de::MovePtr<BufferWithMemory> runTest (bool useGpuBuild,
143 deUint32 workerThreadsCount);
144 TlasPtr initTopAccelerationStructure (bool useGpuBuild,
145 deUint32 workerThreadsCount,
146 const BlasPool& pool);
147 void createTopAccelerationStructure (VkCommandBuffer cmdBuffer,
148 TopLevelAccelerationStructure* tlas);
149 void initBottomAccelerationStructures (BlasPool& pool,
150 bool useGpuBuild,
151 deUint32 workerThreadsCount) const;
152 void initBottomAccelerationStructure (BlasPtr blas,
153 bool useGpuBuild,
154 deUint32 workerThreadsCount,
155 tcu::UVec2& startPos,
156 bool triangles) const;
157
158 private:
159 CaseDef m_data;
160 const VkFormat m_format;
161 };
162
RayTracingBuildTestInstance(Context& context, const CaseDef& data)163 RayTracingBuildTestInstance::RayTracingBuildTestInstance (Context& context, const CaseDef& data)
164 : vkt::TestInstance (context)
165 , m_data (data)
166 , m_format (VK_FORMAT_R32_UINT)
167 {
168 }
169
~RayTracingBuildTestInstance(void)170 RayTracingBuildTestInstance::~RayTracingBuildTestInstance (void)
171 {
172 }
173
174 class RayTracingTestCase : public TestCase
175 {
176 public:
177 RayTracingTestCase (tcu::TestContext& context, const char* name, const char* desc, const CaseDef data);
178 ~RayTracingTestCase (void);
179
180 virtual void initPrograms (SourceCollections& programCollection) const;
181 virtual TestInstance* createInstance (Context& context) const;
182 virtual void checkSupport (Context& context) const;
183
184 private:
185 CaseDef m_data;
186 };
187
RayTracingTestCase(tcu::TestContext& context, const char* name, const char* desc, const CaseDef data)188 RayTracingTestCase::RayTracingTestCase (tcu::TestContext& context, const char* name, const char* desc, const CaseDef data)
189 : vkt::TestCase (context, name, desc)
190 , m_data (data)
191 {
192 DE_ASSERT((m_data.width * m_data.height) == (m_data.squaresGroupCount * m_data.geometriesGroupCount * m_data.instancesGroupCount));
193 }
194
~RayTracingTestCase(void)195 RayTracingTestCase::~RayTracingTestCase (void)
196 {
197 }
198
checkSupport(Context& context) const199 void RayTracingTestCase::checkSupport (Context& context) const
200 {
201 context.requireDeviceFunctionality("VK_KHR_acceleration_structure");
202 context.requireDeviceFunctionality("VK_KHR_ray_tracing_pipeline");
203
204 const VkPhysicalDeviceRayTracingPipelineFeaturesKHR& rayTracingPipelineFeaturesKHR = context.getRayTracingPipelineFeatures();
205 if (rayTracingPipelineFeaturesKHR.rayTracingPipeline == DE_FALSE )
206 TCU_THROW(NotSupportedError, "Requires VkPhysicalDeviceRayTracingPipelineFeaturesKHR.rayTracingPipeline");
207
208 const VkPhysicalDeviceAccelerationStructureFeaturesKHR& accelerationStructureFeaturesKHR = context.getAccelerationStructureFeatures();
209 if (accelerationStructureFeaturesKHR.accelerationStructure == DE_FALSE)
210 TCU_THROW(TestError, "VK_KHR_ray_tracing_pipeline requires VkPhysicalDeviceAccelerationStructureFeaturesKHR.accelerationStructure");
211
212 if (!m_data.deviceBuild)
213 {
214 context.requireDeviceFunctionality("VK_KHR_deferred_host_operations");
215 if (accelerationStructureFeaturesKHR.accelerationStructureHostCommands == DE_FALSE)
216 TCU_THROW(NotSupportedError, "Requires VkPhysicalDeviceAccelerationStructureFeaturesKHR.accelerationStructureHostCommands");
217 }
218 }
219
initPrograms(SourceCollections& programCollection) const220 void RayTracingTestCase::initPrograms (SourceCollections& programCollection) const
221 {
222 const vk::ShaderBuildOptions buildOptions(programCollection.usedVulkanVersion, vk::SPIRV_VERSION_1_4, 0u, true);
223 {
224 std::stringstream css;
225 css <<
226 "#version 460 core\n"
227 "#extension GL_EXT_ray_tracing : require\n"
228 "layout(location = 0) rayPayloadInEXT vec3 hitValue;\n"
229 "hitAttributeEXT vec3 attribs;\n"
230 "layout(r32ui, set = 0, binding = 0) uniform uimage2D result;\n"
231 "void main()\n"
232 "{\n"
233 " uvec4 color = uvec4(1,0,0,1);\n"
234 " imageStore(result, ivec2(gl_LaunchIDEXT.xy), color);\n"
235 "}\n";
236
237 programCollection.glslSources.add("ahit") << glu::AnyHitSource(updateRayTracingGLSL(css.str())) << buildOptions;
238 }
239
240 {
241 std::stringstream css;
242 css <<
243 "#version 460 core\n"
244 "#extension GL_EXT_ray_tracing : require\n"
245 "layout(location = 0) rayPayloadInEXT dummyPayload { vec4 dummy; };\n"
246 "layout(r32ui, set = 0, binding = 0) uniform uimage2D result;\n"
247 "void main()\n"
248 "{\n"
249 " uvec4 color = uvec4(2,0,0,1);\n"
250 " imageStore(result, ivec2(gl_LaunchIDEXT.xy), color);\n"
251 "}\n";
252
253 programCollection.glslSources.add("miss") << glu::MissSource(updateRayTracingGLSL(css.str())) << buildOptions;
254 }
255
256 {
257 std::stringstream css;
258 css <<
259 "#version 460 core\n"
260 "#extension GL_EXT_ray_tracing : require\n"
261 "hitAttributeEXT vec3 hitAttribute;\n"
262 "void main()\n"
263 "{\n"
264 " reportIntersectionEXT(1.0f, 0);\n"
265 "}\n";
266
267 programCollection.glslSources.add("sect") << glu::IntersectionSource(updateRayTracingGLSL(css.str())) << buildOptions;
268 }
269
270 programCollection.glslSources.add("rgen") << glu::RaygenSource(updateRayTracingGLSL(getCommonRayGenerationShader())) << buildOptions;
271 }
272
createInstance(Context& context) const273 TestInstance* RayTracingTestCase::createInstance (Context& context) const
274 {
275 return new RayTracingBuildTestInstance(context, m_data);
276 }
277
278 auto RayTracingBuildTestInstance::initTopAccelerationStructure (bool useGpuBuild,
279 deUint32 workerThreadsCount,
280 const BlasPool& pool) -> TlasPtr
281 {
282 de::MovePtr<TopLevelAccelerationStructure> result = makeTopLevelAccelerationStructure();
283 const std::vector<BlasPtr>& blases = pool.structures();
284
285 result->setInstanceCount(blases.size());
286 result->setBuildType(useGpuBuild ? VK_ACCELERATION_STRUCTURE_BUILD_TYPE_DEVICE_KHR : VK_ACCELERATION_STRUCTURE_BUILD_TYPE_HOST_KHR);
287 result->setDeferredOperation(m_data.deferredOperation, workerThreadsCount);
288
289 for (size_t instanceNdx = 0; instanceNdx < blases.size(); ++instanceNdx)
290 {
291 const bool triangles = (m_data.testType == TEST_TYPE_TRIANGLES) || (m_data.testType == TEST_TYPE_MIXED && (instanceNdx & 1) == 0);
292 deUint32 instanceShaderBindingTableRecordOffset = triangles ? 0 : 1;
293
294 result->addInstance(blases[instanceNdx], vk::identityMatrix3x4, 0, 0xFF, instanceShaderBindingTableRecordOffset);
295 }
296
297 return TlasPtr(result.release());
298 }
299
createTopAccelerationStructure(VkCommandBuffer cmdBuffer, TopLevelAccelerationStructure* tlas)300 void RayTracingBuildTestInstance::createTopAccelerationStructure (VkCommandBuffer cmdBuffer,
301 TopLevelAccelerationStructure* tlas)
302 {
303 const DeviceInterface& vkd = m_context.getDeviceInterface();
304 const VkDevice device = m_context.getDevice();
305 Allocator& allocator = m_context.getDefaultAllocator();
306
307 tlas->createAndBuild(vkd, device, cmdBuffer, allocator);
308 }
309
initBottomAccelerationStructure(BlasPtr blas, bool useGpuBuild, deUint32 workerThreadsCount, tcu::UVec2& startPos, bool triangles) const310 void RayTracingBuildTestInstance::initBottomAccelerationStructure (BlasPtr blas,
311 bool useGpuBuild,
312 deUint32 workerThreadsCount,
313 tcu::UVec2& startPos,
314 bool triangles) const
315 {
316 blas->setBuildType(useGpuBuild ? VK_ACCELERATION_STRUCTURE_BUILD_TYPE_DEVICE_KHR : VK_ACCELERATION_STRUCTURE_BUILD_TYPE_HOST_KHR);
317 blas->setDeferredOperation(m_data.deferredOperation, workerThreadsCount);
318 blas->setGeometryCount(m_data.geometriesGroupCount);
319
320 for (size_t geometryNdx = 0; geometryNdx < m_data.geometriesGroupCount; ++geometryNdx)
321 {
322 std::vector<tcu::Vec3> geometryData;
323
324 geometryData.reserve(m_data.squaresGroupCount * (triangles ? 3u : 2u));
325
326 for (size_t squareNdx = 0; squareNdx < m_data.squaresGroupCount; ++squareNdx)
327 {
328 const deUint32 n = m_data.width * startPos.y() + startPos.x();
329 const float x0 = float(startPos.x() + 0) / float(m_data.width);
330 const float y0 = float(startPos.y() + 0) / float(m_data.height);
331 const float x1 = float(startPos.x() + 1) / float(m_data.width);
332 const float y1 = float(startPos.y() + 1) / float(m_data.height);
333 const float z = (n % 7 == 0) ? +1.0f : -1.0f;
334 const deUint32 m = (n + 13) % (m_data.width * m_data.height);
335
336 if (triangles)
337 {
338 const float xm = (x0 + x1) / 2.0f;
339 const float ym = (y0 + y1) / 2.0f;
340
341 geometryData.push_back(tcu::Vec3(x0, y0, z));
342 geometryData.push_back(tcu::Vec3(x1, ym, z));
343 geometryData.push_back(tcu::Vec3(xm, y1, z));
344 }
345 else
346 {
347 geometryData.push_back(tcu::Vec3(x0, y0, z));
348 geometryData.push_back(tcu::Vec3(x1, y1, z));
349 }
350
351 startPos.y() = m / m_data.width;
352 startPos.x() = m % m_data.width;
353 }
354
355 blas->addGeometry(geometryData, triangles);
356 }
357 }
358
initBottomAccelerationStructures(BlasPool& pool, bool useGpuBuild, deUint32 workerThreadsCount) const359 void RayTracingBuildTestInstance::initBottomAccelerationStructures (BlasPool& pool,
360 bool useGpuBuild,
361 deUint32 workerThreadsCount) const
362 {
363 tcu::UVec2 startPos {};
364 const DeviceInterface& vkd = m_context.getDeviceInterface();
365 const VkDevice device = m_context.getDevice();
366 Allocator& allocator = m_context.getDefaultAllocator();
367 const VkDeviceSize maxBuffSize = 3 * (VkDeviceSize(1) << 30); // 3GB
368
369 for (size_t instanceNdx = 0; instanceNdx < m_data.instancesGroupCount; ++instanceNdx) pool.add();
370
371 const std::vector<BlasPtr>& blases = pool.structures();
372
373 for (size_t instanceNdx = 0; instanceNdx < m_data.instancesGroupCount; ++instanceNdx)
374 {
375 const bool triangles = (m_data.testType == TEST_TYPE_TRIANGLES) || (m_data.testType == TEST_TYPE_MIXED && (instanceNdx & 1) == 0);
376 initBottomAccelerationStructure(blases[instanceNdx], useGpuBuild, workerThreadsCount, startPos, triangles);
377 }
378
379 pool.batchCreateAdjust(vkd, device, allocator, maxBuffSize);
380 }
381
verifyAllocationCount() const382 bool RayTracingBuildTestInstance::verifyAllocationCount () const
383 {
384 BlasPool pool {};
385 tcu::UVec2 startPos {};
386 const DeviceInterface& vkd = m_context.getDeviceInterface();
387 const VkDevice device = m_context.getDevice();
388 auto& log = m_context.getTestContext().getLog();
389 const size_t avvailableAllocCount = m_context.getDeviceProperties().limits.maxMemoryAllocationCount;
390 const VkDeviceSize maxBufferSize = 3 * (VkDeviceSize(1) << 30); // 3GB
391
392
393 for (size_t instanceNdx = 0; instanceNdx < m_data.instancesGroupCount; ++instanceNdx) pool.add();
394
395 const std::vector<BlasPtr>& blases = pool.structures();
396
397 for (size_t instanceNdx = 0; instanceNdx < m_data.instancesGroupCount; ++instanceNdx)
398 {
399 const bool triangles = (m_data.testType == TEST_TYPE_TRIANGLES) || (m_data.testType == TEST_TYPE_MIXED && (instanceNdx & 1) == 0);
400 initBottomAccelerationStructure(blases[instanceNdx], true, 0, startPos, triangles);
401 }
402
403 const size_t poolAllocationCount = pool.getAllocationCount(vkd, device, maxBufferSize);
404 const size_t requiredAllocationCount = poolAllocationCount + 120;
405
406 log << tcu::TestLog::Message
407 << "The test consumes " << poolAllocationCount
408 << " allocations out of " << avvailableAllocCount << " available"
409 << tcu::TestLog::EndMessage;
410
411 return (requiredAllocationCount < avvailableAllocCount);
412 }
413
runTest(bool useGpuBuild, deUint32 workerThreadsCount)414 de::MovePtr<BufferWithMemory> RayTracingBuildTestInstance::runTest (bool useGpuBuild, deUint32 workerThreadsCount)
415 {
416 const InstanceInterface& vki = m_context.getInstanceInterface();
417 const DeviceInterface& vkd = m_context.getDeviceInterface();
418 const VkDevice device = m_context.getDevice();
419 const VkPhysicalDevice physicalDevice = m_context.getPhysicalDevice();
420 const deUint32 queueFamilyIndex = m_context.getUniversalQueueFamilyIndex();
421 const VkQueue queue = m_context.getUniversalQueue();
422 Allocator& allocator = m_context.getDefaultAllocator();
423 const deUint32 pixelCount = m_data.width * m_data.height;
424 const deUint32 shaderGroupHandleSize = getShaderGroupSize(vki, physicalDevice);
425 const deUint32 shaderGroupBaseAlignment = getShaderGroupBaseAlignment(vki, physicalDevice);
426
427 const Move<VkDescriptorSetLayout> descriptorSetLayout = DescriptorSetLayoutBuilder()
428 .addSingleBinding(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, ALL_RAY_TRACING_STAGES)
429 .addSingleBinding(VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR, ALL_RAY_TRACING_STAGES)
430 .build(vkd, device);
431 const Move<VkDescriptorPool> descriptorPool = DescriptorPoolBuilder()
432 .addType(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE)
433 .addType(VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR)
434 .build(vkd, device, VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, 1u);
435 const Move<VkDescriptorSet> descriptorSet = makeDescriptorSet(vkd, device, *descriptorPool, *descriptorSetLayout);
436 const Move<VkPipelineLayout> pipelineLayout = makePipelineLayout(vkd, device, descriptorSetLayout.get());
437 const Move<VkCommandPool> cmdPool = createCommandPool(vkd, device, 0, queueFamilyIndex);
438 const Move<VkCommandBuffer> cmdBuffer = allocateCommandBuffer(vkd, device, *cmdPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY);
439
440 de::MovePtr<RayTracingPipeline> rayTracingPipeline = de::newMovePtr<RayTracingPipeline>();
441 Move<VkShaderModule> raygenShader = createShaderModule(vkd, device, m_context.getBinaryCollection().get("rgen"), 0);
442 Move<VkShaderModule> hitShader = createShaderModule(vkd, device, m_context.getBinaryCollection().get("ahit"), 0);
443 Move<VkShaderModule> missShader = createShaderModule(vkd, device, m_context.getBinaryCollection().get("miss"), 0);
444 Move<VkShaderModule> intersectionShader = createShaderModule(vkd, device, m_context.getBinaryCollection().get("sect"), 0);
445 rayTracingPipeline->addShader(VK_SHADER_STAGE_RAYGEN_BIT_KHR, *raygenShader, 0u);
446 rayTracingPipeline->addShader(VK_SHADER_STAGE_ANY_HIT_BIT_KHR, *hitShader, 1u);
447 rayTracingPipeline->addShader(VK_SHADER_STAGE_ANY_HIT_BIT_KHR, *hitShader, 2u);
448 rayTracingPipeline->addShader(VK_SHADER_STAGE_INTERSECTION_BIT_KHR, *intersectionShader, 2u);
449 rayTracingPipeline->addShader(VK_SHADER_STAGE_MISS_BIT_KHR, *missShader, 3u);
450 Move<VkPipeline> pipeline = rayTracingPipeline->createPipeline(vkd, device, *pipelineLayout);
451 const de::MovePtr<BufferWithMemory> raygenShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, *pipeline, allocator, shaderGroupHandleSize, shaderGroupBaseAlignment, 0u, 1u);
452 const de::MovePtr<BufferWithMemory> hitShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, *pipeline, allocator, shaderGroupHandleSize, shaderGroupBaseAlignment, 1u, 2u);
453 const de::MovePtr<BufferWithMemory> missShaderBindingTable = rayTracingPipeline->createShaderBindingTable(vkd, device, *pipeline, allocator, shaderGroupHandleSize, shaderGroupBaseAlignment, 3u, 1u);
454 const VkStridedDeviceAddressRegionKHR raygenShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, raygenShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
455 const VkStridedDeviceAddressRegionKHR hitShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, hitShaderBindingTable->get(), 0), shaderGroupHandleSize, 2u * shaderGroupHandleSize);
456 const VkStridedDeviceAddressRegionKHR missShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(getBufferDeviceAddress(vkd, device, missShaderBindingTable->get(), 0), shaderGroupHandleSize, shaderGroupHandleSize);
457 const VkStridedDeviceAddressRegionKHR callableShaderBindingTableRegion = makeStridedDeviceAddressRegionKHR(DE_NULL, 0, 0);
458
459 const VkImageCreateInfo imageCreateInfo = makeImageCreateInfo(m_data.width, m_data.height, m_format);
460 const VkImageSubresourceRange imageSubresourceRange = makeImageSubresourceRange(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 1u, 0, 1u);
461 const de::MovePtr<ImageWithMemory> image = de::MovePtr<ImageWithMemory>(new ImageWithMemory(vkd, device, allocator, imageCreateInfo, MemoryRequirement::Any));
462 const Move<VkImageView> imageView = makeImageView(vkd, device, **image, VK_IMAGE_VIEW_TYPE_2D, m_format, imageSubresourceRange);
463
464 const VkBufferCreateInfo bufferCreateInfo = makeBufferCreateInfo(pixelCount*sizeof(deUint32), VK_BUFFER_USAGE_TRANSFER_DST_BIT);
465 const VkImageSubresourceLayers bufferImageSubresourceLayers = makeImageSubresourceLayers(VK_IMAGE_ASPECT_COLOR_BIT, 0u, 0u, 1u);
466 const VkBufferImageCopy bufferImageRegion = makeBufferImageCopy(makeExtent3D(m_data.width, m_data.height, 1u), bufferImageSubresourceLayers);
467 de::MovePtr<BufferWithMemory> buffer = de::MovePtr<BufferWithMemory>(new BufferWithMemory(vkd, device, allocator, bufferCreateInfo, MemoryRequirement::HostVisible));
468
469 const VkDescriptorImageInfo descriptorImageInfo = makeDescriptorImageInfo(DE_NULL, *imageView, VK_IMAGE_LAYOUT_GENERAL);
470
471 const VkImageMemoryBarrier preImageBarrier = makeImageMemoryBarrier(0u, VK_ACCESS_TRANSFER_WRITE_BIT,
472 VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
473 **image, imageSubresourceRange);
474 const VkImageMemoryBarrier postImageBarrier = makeImageMemoryBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
475 VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_IMAGE_LAYOUT_GENERAL,
476 **image, imageSubresourceRange);
477 const VkMemoryBarrier postTraceMemoryBarrier = makeMemoryBarrier(VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT);
478 const VkMemoryBarrier postCopyMemoryBarrier = makeMemoryBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_HOST_READ_BIT);
479 const VkClearValue clearValue = makeClearValueColorU32(5u, 5u, 5u, 255u);
480
481 TlasPtr topLevelAccelerationStructure;
482 BottomLevelAccelerationStructurePool blasPool;
483
484 initBottomAccelerationStructures(blasPool, useGpuBuild, workerThreadsCount);
485 blasPool.batchBuild(vkd, device, *cmdPool, queue);
486
487 beginCommandBuffer(vkd, *cmdBuffer, 0u);
488 {
489 cmdPipelineImageMemoryBarrier(vkd, *cmdBuffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, &preImageBarrier);
490 vkd.cmdClearColorImage(*cmdBuffer, **image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, &clearValue.color, 1, &imageSubresourceRange);
491 cmdPipelineImageMemoryBarrier(vkd, *cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, &postImageBarrier);
492
493 topLevelAccelerationStructure = initTopAccelerationStructure(useGpuBuild, workerThreadsCount, blasPool);
494 createTopAccelerationStructure(*cmdBuffer, topLevelAccelerationStructure.get());
495
496 VkWriteDescriptorSetAccelerationStructureKHR accelerationStructureWriteDescriptorSet =
497 {
498 VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET_ACCELERATION_STRUCTURE_KHR, // VkStructureType sType;
499 DE_NULL, // const void* pNext;
500 1u, // deUint32 accelerationStructureCount;
501 topLevelAccelerationStructure->getPtr(), // const VkAccelerationStructureKHR* pAccelerationStructures;
502 };
503
504 DescriptorSetUpdateBuilder()
505 .writeSingle(*descriptorSet, DescriptorSetUpdateBuilder::Location::binding(0u), VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, &descriptorImageInfo)
506 .writeSingle(*descriptorSet, DescriptorSetUpdateBuilder::Location::binding(1u), VK_DESCRIPTOR_TYPE_ACCELERATION_STRUCTURE_KHR, &accelerationStructureWriteDescriptorSet)
507 .update(vkd, device);
508
509 vkd.cmdBindDescriptorSets(*cmdBuffer, VK_PIPELINE_BIND_POINT_RAY_TRACING_KHR, *pipelineLayout, 0, 1, &descriptorSet.get(), 0, DE_NULL);
510
511 vkd.cmdBindPipeline(*cmdBuffer, VK_PIPELINE_BIND_POINT_RAY_TRACING_KHR, *pipeline);
512
513 cmdTraceRays(vkd,
514 *cmdBuffer,
515 &raygenShaderBindingTableRegion,
516 &missShaderBindingTableRegion,
517 &hitShaderBindingTableRegion,
518 &callableShaderBindingTableRegion,
519 m_data.width, m_data.height, 1);
520
521 cmdPipelineMemoryBarrier(vkd, *cmdBuffer, VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_KHR, VK_PIPELINE_STAGE_TRANSFER_BIT, &postTraceMemoryBarrier);
522
523 vkd.cmdCopyImageToBuffer(*cmdBuffer, **image, VK_IMAGE_LAYOUT_GENERAL, **buffer, 1u, &bufferImageRegion);
524
525 cmdPipelineMemoryBarrier(vkd, *cmdBuffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, &postCopyMemoryBarrier);
526 }
527 endCommandBuffer(vkd, *cmdBuffer);
528
529 submitCommandsAndWait(vkd, device, queue, cmdBuffer.get());
530
531 invalidateMappedMemoryRange(vkd, device, buffer->getAllocation().getMemory(), buffer->getAllocation().getOffset(), pixelCount * sizeof(deUint32));
532
533 return buffer;
534 }
535
checkSupportInInstance(void) const536 void RayTracingBuildTestInstance::checkSupportInInstance (void) const
537 {
538 const InstanceInterface& vki = m_context.getInstanceInterface();
539 const VkPhysicalDevice physicalDevice = m_context.getPhysicalDevice();
540 de::MovePtr<RayTracingProperties> rayTracingProperties = makeRayTracingProperties(vki, physicalDevice);
541
542 if (rayTracingProperties->getMaxPrimitiveCount() < m_data.squaresGroupCount)
543 TCU_THROW(NotSupportedError, "Triangles required more than supported");
544
545 if (rayTracingProperties->getMaxGeometryCount() < m_data.geometriesGroupCount)
546 TCU_THROW(NotSupportedError, "Geometries required more than supported");
547
548 if (rayTracingProperties->getMaxInstanceCount() < m_data.instancesGroupCount)
549 TCU_THROW(NotSupportedError, "Instances required more than supported");
550
551 if (!verifyAllocationCount())
552 TCU_THROW(NotSupportedError, "Memory allocations required more than supported");
553 }
554
validateBuffer(de::MovePtr<BufferWithMemory> buffer)555 deUint32 RayTracingBuildTestInstance::validateBuffer (de::MovePtr<BufferWithMemory> buffer)
556 {
557 const deUint32* bufferPtr = (deUint32*)buffer->getAllocation().getHostPtr();
558 deUint32 failures = 0;
559 deUint32 pos = 0;
560
561 for (deUint32 y = 0; y < m_data.height; ++y)
562 for (deUint32 x = 0; x < m_data.width; ++x)
563 {
564 const deUint32 anyHitValue = 1;
565 const deUint32 missValue = 2;
566
567 const deUint32 n = m_data.width * y + x;
568 const deUint32 expectedValue = (n % 7 == 0) ? missValue : anyHitValue;
569
570 if (bufferPtr[pos] != expectedValue)
571 {
572 if (m_data.testType == TEST_TYPE_AABBS || m_data.testType == TEST_TYPE_MIXED)
573 {
574 // In the case of AABB geometries, implementations may increase their size in
575 // an acceleration structure in order to mitigate precision issues. This may
576 // result in false positives being reported to the application."
577
578 if (bufferPtr[pos] != anyHitValue)
579 {
580 failures++;
581 }
582 }
583 else
584 {
585 failures++;
586 }
587 }
588
589 ++pos;
590 }
591
592 return failures;
593 }
594
iterate(void)595 tcu::TestStatus RayTracingBuildTestInstance::iterate (void)
596 {
597 checkSupportInInstance();
598
599 const deUint32 failures = validateBuffer(runTest(m_data.deviceBuild, m_data.workerThreadsCount));
600
601 return (failures == 0) ? tcu::TestStatus::pass("Pass") : tcu::TestStatus::fail("failures=" + de::toString(failures));
602 }
603
604 } // anonymous
605
buildTest(tcu::TestCaseGroup* testParentGroup, deUint32 threadsCount, bool deviceBuild)606 static void buildTest (tcu::TestCaseGroup* testParentGroup, deUint32 threadsCount, bool deviceBuild)
607 {
608 const char* tests[] =
609 {
610 "level_primitives",
611 "level_geometries",
612 "level_instances"
613 };
614 const deUint32 sizes[] = { 4, 16, 64, 256, 1024 };
615 const deUint32 factors[] = { 1, 4 };
616 const bool deferredOperation = threadsCount != 0;
617 tcu::TestContext& testCtx = testParentGroup->getTestContext();
618
619 for (size_t testsNdx = 0; testsNdx < DE_LENGTH_OF_ARRAY(tests); ++testsNdx)
620 {
621 de::MovePtr<tcu::TestCaseGroup> group(new tcu::TestCaseGroup(testCtx, tests[testsNdx], ""));
622
623 for (size_t factorNdx = 0; factorNdx < DE_LENGTH_OF_ARRAY(factors); ++factorNdx)
624 for (size_t sizesNdx = 0; sizesNdx < DE_LENGTH_OF_ARRAY(sizes); ++sizesNdx)
625 {
626 const deUint32 factor = factors[factorNdx];
627 const deUint32 largestGroup = sizes[sizesNdx] * sizes[sizesNdx] / factor / factor;
628 const deUint32 squaresGroupCount = testsNdx == 0 ? largestGroup : factor;
629 const deUint32 geometriesGroupCount = testsNdx == 1 ? largestGroup : factor;
630 const deUint32 instancesGroupCount = testsNdx == 2 ? largestGroup : factor;
631 const CaseDef caseDef =
632 {
633 TEST_TYPE_TRIANGLES, // TestType testType;
634 sizes[sizesNdx], // deUint32 width;
635 sizes[sizesNdx], // deUint32 height;
636 squaresGroupCount, // deUint32 squaresGroupCount;
637 geometriesGroupCount, // deUint32 geometriesGroupCount;
638 instancesGroupCount, // deUint32 instancesGroupCount;
639 deferredOperation, // bool deferredOperation;
640 threadsCount, // deUint32 workerThreadsCount;
641 deviceBuild // bool deviceBuild;
642 };
643 const std::string suffix = de::toString(caseDef.instancesGroupCount) + '_' + de::toString(caseDef.geometriesGroupCount) + '_' + de::toString(caseDef.squaresGroupCount);
644 const std::string testName = "triangles_" + suffix;
645
646 if (squaresGroupCount == 0 || geometriesGroupCount == 0 || instancesGroupCount == 0)
647 continue;
648
649 group->addChild(new RayTracingTestCase(testCtx, testName.c_str(), "", caseDef));
650 }
651
652 for (size_t factorNdx = 0; factorNdx < DE_LENGTH_OF_ARRAY(factors); ++factorNdx)
653 for (size_t sizesNdx = 0; sizesNdx < DE_LENGTH_OF_ARRAY(sizes); ++sizesNdx)
654 {
655 const deUint32 factor = factors[factorNdx];
656 const deUint32 largestGroup = sizes[sizesNdx] * sizes[sizesNdx] / factor / factor;
657 const deUint32 squaresGroupCount = testsNdx == 0 ? largestGroup : factor;
658 const deUint32 geometriesGroupCount = testsNdx == 1 ? largestGroup : factor;
659 const deUint32 instancesGroupCount = testsNdx == 2 ? largestGroup : factor;
660 const CaseDef caseDef =
661 {
662 TEST_TYPE_AABBS, // TestType testType;
663 sizes[sizesNdx], // deUint32 width;
664 sizes[sizesNdx], // deUint32 height;
665 squaresGroupCount, // deUint32 squaresGroupCount;
666 geometriesGroupCount, // deUint32 geometriesGroupCount;
667 instancesGroupCount, // deUint32 instancesGroupCount;
668 deferredOperation, // bool deferredOperation;
669 threadsCount, // deUint32 workerThreadsCount;
670 deviceBuild // bool deviceBuild;
671 };
672 const std::string suffix = de::toString(caseDef.instancesGroupCount) + '_' + de::toString(caseDef.geometriesGroupCount) + '_' + de::toString(caseDef.squaresGroupCount);
673 const std::string testName = "aabbs_" + suffix;
674
675 if (squaresGroupCount == 0 || geometriesGroupCount == 0 || instancesGroupCount == 0)
676 continue;
677
678 group->addChild(new RayTracingTestCase(testCtx, testName.c_str(), "", caseDef));
679 }
680
681 for (size_t factorNdx = 0; factorNdx < DE_LENGTH_OF_ARRAY(factors); ++factorNdx)
682 for (size_t sizesNdx = 0; sizesNdx < DE_LENGTH_OF_ARRAY(sizes); ++sizesNdx)
683 {
684 const deUint32 factor = factors[factorNdx];
685 const deUint32 largestGroup = sizes[sizesNdx] * sizes[sizesNdx] / factor / factor;
686 const deUint32 squaresGroupCount = testsNdx == 0 ? largestGroup : factor;
687 const deUint32 geometriesGroupCount = testsNdx == 1 ? largestGroup : factor;
688 const deUint32 instancesGroupCount = testsNdx == 2 ? largestGroup : factor;
689 const CaseDef caseDef =
690 {
691 TEST_TYPE_MIXED, // TestType testType;
692 sizes[sizesNdx], // deUint32 width;
693 sizes[sizesNdx], // deUint32 height;
694 squaresGroupCount, // deUint32 squaresGroupCount;
695 geometriesGroupCount, // deUint32 geometriesGroupCount;
696 instancesGroupCount, // deUint32 instancesGroupCount;
697 deferredOperation, // bool deferredOperation;
698 threadsCount, // deUint32 workerThreadsCount;
699 deviceBuild // bool deviceBuild;
700 };
701 const std::string suffix = de::toString(caseDef.instancesGroupCount) + '_' + de::toString(caseDef.geometriesGroupCount) + '_' + de::toString(caseDef.squaresGroupCount);
702 const std::string testName = "mixed_" + suffix;
703
704 if (squaresGroupCount < 2 || geometriesGroupCount < 2 || instancesGroupCount < 2)
705 continue;
706
707 group->addChild(new RayTracingTestCase(testCtx, testName.c_str(), "", caseDef));
708 }
709
710 testParentGroup->addChild(group.release());
711 }
712 }
713
createBuildTests(tcu::TestContext& testCtx)714 tcu::TestCaseGroup* createBuildTests (tcu::TestContext& testCtx)
715 {
716 de::MovePtr<tcu::TestCaseGroup> buildGroup(new tcu::TestCaseGroup(testCtx, "build", "Ray tracing build tests"));
717
718 const deUint32 threads[] = { 0, 1, 2, 3, 4, 8, std::numeric_limits<deUint32>::max() };
719
720 for (const auto threadCount : threads)
721 {
722 auto buildTargeGroup = [&](bool deviceBuild) -> void
723 {
724 DE_ASSERT(!(threadCount != 0 && deviceBuild));
725
726 string groupName, groupDesc;
727 if (deviceBuild)
728 {
729 groupName = "gpu";
730 groupDesc = "Compare results of run with acceleration structures build on GPU";
731 }
732 else
733 {
734 groupName = "cpu";
735 groupDesc = "Compare results of run with acceleration structures build on CPU";
736 }
737
738 if (threadCount != 0)
739 {
740 groupName += threadCount == std::numeric_limits<deUint32>::max()
741 ? "ht_max" : "ht_" + de::toString(threadCount);
742 groupDesc = "Compare results of run with acceleration structures build on CPU and using host threading";
743 }
744
745 de::MovePtr<tcu::TestCaseGroup> groupGpuCpuHt(new tcu::TestCaseGroup(testCtx, groupName.c_str(), groupDesc.c_str()));
746 buildTest(groupGpuCpuHt.get(), threadCount, deviceBuild);
747 buildGroup->addChild(groupGpuCpuHt.release());
748 };
749
750 if (threadCount == 0)
751 {
752 buildTargeGroup(true);
753 }
754 buildTargeGroup(false);
755 }
756
757 return buildGroup.release();
758 }
759
760 } // RayTracing
761 } // vkt
762