| 1 | // Copyright 2018 The SwiftShader Authors. All Rights Reserved. |
| 2 | // |
| 3 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | // you may not use this file except in compliance with the License. |
| 5 | // You may obtain a copy of the License at |
| 6 | // |
| 7 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | // |
| 9 | // Unless required by applicable law or agreed to in writing, software |
| 10 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | // See the License for the specific language governing permissions and |
| 13 | // limitations under the License. |
| 14 | |
| 15 | #include "VkDevice.hpp" |
| 16 | |
| 17 | #include "VkConfig.h" |
| 18 | #include "VkDebug.hpp" |
| 19 | #include "VkDescriptorSetLayout.hpp" |
| 20 | #include "VkFence.hpp" |
| 21 | #include "VkQueue.hpp" |
| 22 | #include "Device/Blitter.hpp" |
| 23 | |
| 24 | #include <chrono> |
| 25 | #include <climits> |
| 26 | #include <new> // Must #include this to use "placement new" |
| 27 | |
| 28 | namespace |
| 29 | { |
| 30 | std::chrono::time_point<std::chrono::system_clock, std::chrono::nanoseconds> now() |
| 31 | { |
| 32 | return std::chrono::time_point_cast<std::chrono::nanoseconds>(std::chrono::system_clock::now()); |
| 33 | } |
| 34 | } |
| 35 | |
| 36 | namespace vk |
| 37 | { |
| 38 | |
| 39 | std::shared_ptr<rr::Routine> Device::SamplingRoutineCache::query(const vk::Device::SamplingRoutineCache::Key& key) const |
| 40 | { |
| 41 | return cache.query(key); |
| 42 | } |
| 43 | |
| 44 | void Device::SamplingRoutineCache::add(const vk::Device::SamplingRoutineCache::Key& key, const std::shared_ptr<rr::Routine>& routine) |
| 45 | { |
| 46 | ASSERT(routine); |
| 47 | cache.add(key, routine); |
| 48 | } |
| 49 | |
| 50 | rr::Routine* Device::SamplingRoutineCache::queryConst(const vk::Device::SamplingRoutineCache::Key& key) const |
| 51 | { |
| 52 | return cache.queryConstCache(key).get(); |
| 53 | } |
| 54 | |
| 55 | void Device::SamplingRoutineCache::updateConstCache() |
| 56 | { |
| 57 | cache.updateConstCache(); |
| 58 | } |
| 59 | |
| 60 | Device::Device(const VkDeviceCreateInfo* pCreateInfo, void* mem, PhysicalDevice *physicalDevice, const VkPhysicalDeviceFeatures *enabledFeatures, const std::shared_ptr<marl::Scheduler>& scheduler) |
| 61 | : physicalDevice(physicalDevice), |
| 62 | queues(reinterpret_cast<Queue*>(mem)), |
| 63 | enabledExtensionCount(pCreateInfo->enabledExtensionCount), |
| 64 | enabledFeatures(enabledFeatures ? *enabledFeatures : VkPhysicalDeviceFeatures{}), // "Setting pEnabledFeatures to NULL and not including a VkPhysicalDeviceFeatures2 in the pNext member of VkDeviceCreateInfo is equivalent to setting all members of the structure to VK_FALSE." |
| 65 | scheduler(scheduler) |
| 66 | { |
| 67 | for(uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++) |
| 68 | { |
| 69 | const VkDeviceQueueCreateInfo& queueCreateInfo = pCreateInfo->pQueueCreateInfos[i]; |
| 70 | queueCount += queueCreateInfo.queueCount; |
| 71 | } |
| 72 | |
| 73 | uint32_t queueID = 0; |
| 74 | for(uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++) |
| 75 | { |
| 76 | const VkDeviceQueueCreateInfo& queueCreateInfo = pCreateInfo->pQueueCreateInfos[i]; |
| 77 | |
| 78 | for(uint32_t j = 0; j < queueCreateInfo.queueCount; j++, queueID++) |
| 79 | { |
| 80 | new (&queues[queueID]) Queue(this, scheduler.get()); |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | extensions = reinterpret_cast<ExtensionName*>(static_cast<uint8_t*>(mem) + (sizeof(Queue) * queueCount)); |
| 85 | for(uint32_t i = 0; i < enabledExtensionCount; i++) |
| 86 | { |
| 87 | strncpy(extensions[i], pCreateInfo->ppEnabledExtensionNames[i], VK_MAX_EXTENSION_NAME_SIZE); |
| 88 | } |
| 89 | |
| 90 | if(pCreateInfo->enabledLayerCount) |
| 91 | { |
| 92 | // "The ppEnabledLayerNames and enabledLayerCount members of VkDeviceCreateInfo are deprecated and their values must be ignored by implementations." |
| 93 | UNIMPLEMENTED("enabledLayerCount" ); // TODO(b/119321052): UNIMPLEMENTED() should be used only for features that must still be implemented. Use a more informational macro here. |
| 94 | } |
| 95 | |
| 96 | // FIXME (b/119409619): use an allocator here so we can control all memory allocations |
| 97 | blitter.reset(new sw::Blitter()); |
| 98 | samplingRoutineCache.reset(new SamplingRoutineCache()); |
| 99 | } |
| 100 | |
| 101 | void Device::destroy(const VkAllocationCallbacks* pAllocator) |
| 102 | { |
| 103 | for(uint32_t i = 0; i < queueCount; i++) |
| 104 | { |
| 105 | queues[i].~Queue(); |
| 106 | } |
| 107 | |
| 108 | vk::deallocate(queues, pAllocator); |
| 109 | } |
| 110 | |
| 111 | size_t Device::ComputeRequiredAllocationSize(const VkDeviceCreateInfo* pCreateInfo) |
| 112 | { |
| 113 | uint32_t queueCount = 0; |
| 114 | for(uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++) |
| 115 | { |
| 116 | queueCount += pCreateInfo->pQueueCreateInfos[i].queueCount; |
| 117 | } |
| 118 | |
| 119 | return (sizeof(Queue) * queueCount) + (pCreateInfo->enabledExtensionCount * sizeof(ExtensionName)); |
| 120 | } |
| 121 | |
| 122 | bool Device::hasExtension(const char* extensionName) const |
| 123 | { |
| 124 | for(uint32_t i = 0; i < enabledExtensionCount; i++) |
| 125 | { |
| 126 | if(strncmp(extensions[i], extensionName, VK_MAX_EXTENSION_NAME_SIZE) == 0) |
| 127 | { |
| 128 | return true; |
| 129 | } |
| 130 | } |
| 131 | return false; |
| 132 | } |
| 133 | |
| 134 | VkQueue Device::getQueue(uint32_t queueFamilyIndex, uint32_t queueIndex) const |
| 135 | { |
| 136 | ASSERT(queueFamilyIndex == 0); |
| 137 | |
| 138 | return queues[queueIndex]; |
| 139 | } |
| 140 | |
| 141 | VkResult Device::waitForFences(uint32_t fenceCount, const VkFence* pFences, VkBool32 waitAll, uint64_t timeout) |
| 142 | { |
| 143 | using time_point = std::chrono::time_point<std::chrono::system_clock, std::chrono::nanoseconds>; |
| 144 | const time_point start = now(); |
| 145 | const uint64_t max_timeout = (LLONG_MAX - start.time_since_epoch().count()); |
| 146 | bool infiniteTimeout = (timeout > max_timeout); |
| 147 | const time_point end_ns = start + std::chrono::nanoseconds(std::min(max_timeout, timeout)); |
| 148 | |
| 149 | if(waitAll != VK_FALSE) // All fences must be signaled |
| 150 | { |
| 151 | for(uint32_t i = 0; i < fenceCount; i++) |
| 152 | { |
| 153 | if(timeout == 0) |
| 154 | { |
| 155 | if(Cast(pFences[i])->getStatus() != VK_SUCCESS) // At least one fence is not signaled |
| 156 | { |
| 157 | return VK_TIMEOUT; |
| 158 | } |
| 159 | } |
| 160 | else if(infiniteTimeout) |
| 161 | { |
| 162 | if(Cast(pFences[i])->wait() != VK_SUCCESS) // At least one fence is not signaled |
| 163 | { |
| 164 | return VK_TIMEOUT; |
| 165 | } |
| 166 | } |
| 167 | else |
| 168 | { |
| 169 | if(Cast(pFences[i])->wait(end_ns) != VK_SUCCESS) // At least one fence is not signaled |
| 170 | { |
| 171 | return VK_TIMEOUT; |
| 172 | } |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | return VK_SUCCESS; |
| 177 | } |
| 178 | else // At least one fence must be signaled |
| 179 | { |
| 180 | marl::containers::vector<marl::Event, 8> events; |
| 181 | for(uint32_t i = 0; i < fenceCount; i++) |
| 182 | { |
| 183 | events.push_back(Cast(pFences[i])->getEvent()); |
| 184 | } |
| 185 | |
| 186 | auto any = marl::Event::any(events.begin(), events.end()); |
| 187 | |
| 188 | if(timeout == 0) |
| 189 | { |
| 190 | return any.isSignalled() ? VK_SUCCESS : VK_TIMEOUT; |
| 191 | } |
| 192 | else if (infiniteTimeout) |
| 193 | { |
| 194 | any.wait(); |
| 195 | return VK_SUCCESS; |
| 196 | } |
| 197 | else |
| 198 | { |
| 199 | return any.wait_until(end_ns) ? VK_SUCCESS : VK_TIMEOUT; |
| 200 | } |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | VkResult Device::waitIdle() |
| 205 | { |
| 206 | for(uint32_t i = 0; i < queueCount; i++) |
| 207 | { |
| 208 | queues[i].waitIdle(); |
| 209 | } |
| 210 | |
| 211 | return VK_SUCCESS; |
| 212 | } |
| 213 | |
| 214 | void Device::getDescriptorSetLayoutSupport(const VkDescriptorSetLayoutCreateInfo* pCreateInfo, |
| 215 | VkDescriptorSetLayoutSupport* pSupport) const |
| 216 | { |
| 217 | // Mark everything as unsupported |
| 218 | pSupport->supported = VK_FALSE; |
| 219 | } |
| 220 | |
| 221 | void Device::updateDescriptorSets(uint32_t descriptorWriteCount, const VkWriteDescriptorSet* pDescriptorWrites, |
| 222 | uint32_t descriptorCopyCount, const VkCopyDescriptorSet* pDescriptorCopies) |
| 223 | { |
| 224 | for(uint32_t i = 0; i < descriptorWriteCount; i++) |
| 225 | { |
| 226 | DescriptorSetLayout::WriteDescriptorSet(this, pDescriptorWrites[i]); |
| 227 | } |
| 228 | |
| 229 | for(uint32_t i = 0; i < descriptorCopyCount; i++) |
| 230 | { |
| 231 | DescriptorSetLayout::CopyDescriptorSet(pDescriptorCopies[i]); |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | Device::SamplingRoutineCache* Device::getSamplingRoutineCache() const |
| 236 | { |
| 237 | return samplingRoutineCache.get(); |
| 238 | } |
| 239 | |
| 240 | rr::Routine* Device::findInConstCache(const SamplingRoutineCache::Key& key) const |
| 241 | { |
| 242 | return samplingRoutineCache->queryConst(key); |
| 243 | } |
| 244 | |
| 245 | void Device::updateSamplingRoutineConstCache() |
| 246 | { |
| 247 | std::unique_lock<std::mutex> lock(samplingRoutineCacheMutex); |
| 248 | samplingRoutineCache->updateConstCache(); |
| 249 | } |
| 250 | |
| 251 | std::mutex& Device::getSamplingRoutineCacheMutex() |
| 252 | { |
| 253 | return samplingRoutineCacheMutex; |
| 254 | } |
| 255 | |
| 256 | } // namespace vk |
| 257 | |