diff options
| author | 2021-01-05 17:44:56 -0800 | |
|---|---|---|
| committer | 2021-01-05 17:44:56 -0800 | |
| commit | 275b96a0e22e40a0b1d9f49fa8ba3ec07e8dcc65 (patch) | |
| tree | 75f9aa4727d41015f841ba83f4af36335f83771f /src/video_core/vulkan_common | |
| parent | Merge pull request #5292 from ReinUsesLisp/empty-set (diff) | |
| parent | vulkan_device: Allow creating a device without surface (diff) | |
| download | yuzu-275b96a0e22e40a0b1d9f49fa8ba3ec07e8dcc65.tar.gz yuzu-275b96a0e22e40a0b1d9f49fa8ba3ec07e8dcc65.tar.xz yuzu-275b96a0e22e40a0b1d9f49fa8ba3ec07e8dcc65.zip | |
Merge pull request #5289 from ReinUsesLisp/vulkan-device
vulkan_common: Move device abstraction to the common directory and allow surfaceless devices
Diffstat (limited to 'src/video_core/vulkan_common')
| -rw-r--r-- | src/video_core/vulkan_common/nsight_aftermath_tracker.cpp | 212 | ||||
| -rw-r--r-- | src/video_core/vulkan_common/nsight_aftermath_tracker.h | 82 | ||||
| -rw-r--r-- | src/video_core/vulkan_common/vulkan_device.cpp | 886 | ||||
| -rw-r--r-- | src/video_core/vulkan_common/vulkan_device.h | 307 |
4 files changed, 1487 insertions, 0 deletions
diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp new file mode 100644 index 000000000..8d10ac29e --- /dev/null +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp | |||
| @@ -0,0 +1,212 @@ | |||
| 1 | // Copyright 2020 yuzu Emulator Project | ||
| 2 | // Licensed under GPLv2 or any later version | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #ifdef HAS_NSIGHT_AFTERMATH | ||
| 6 | |||
| 7 | #include <mutex> | ||
| 8 | #include <string> | ||
| 9 | #include <string_view> | ||
| 10 | #include <utility> | ||
| 11 | #include <vector> | ||
| 12 | |||
| 13 | #include <fmt/format.h> | ||
| 14 | |||
| 15 | #define VK_NO_PROTOTYPES | ||
| 16 | #include <vulkan/vulkan.h> | ||
| 17 | |||
| 18 | #include <GFSDK_Aftermath.h> | ||
| 19 | #include <GFSDK_Aftermath_Defines.h> | ||
| 20 | #include <GFSDK_Aftermath_GpuCrashDump.h> | ||
| 21 | #include <GFSDK_Aftermath_GpuCrashDumpDecoding.h> | ||
| 22 | |||
| 23 | #include "common/common_paths.h" | ||
| 24 | #include "common/common_types.h" | ||
| 25 | #include "common/file_util.h" | ||
| 26 | #include "common/logging/log.h" | ||
| 27 | #include "common/scope_exit.h" | ||
| 28 | |||
| 29 | #include "video_core/renderer_vulkan/nsight_aftermath_tracker.h" | ||
| 30 | |||
| 31 | namespace Vulkan { | ||
| 32 | |||
| 33 | static constexpr char AFTERMATH_LIB_NAME[] = "GFSDK_Aftermath_Lib.x64.dll"; | ||
| 34 | |||
| 35 | NsightAftermathTracker::NsightAftermathTracker() { | ||
| 36 | if (!dl.Open(AFTERMATH_LIB_NAME)) { | ||
| 37 | LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath DLL"); | ||
| 38 | return; | ||
| 39 | } | ||
| 40 | if (!dl.GetSymbol("GFSDK_Aftermath_DisableGpuCrashDumps", | ||
| 41 | &GFSDK_Aftermath_DisableGpuCrashDumps) || | ||
| 42 | !dl.GetSymbol("GFSDK_Aftermath_EnableGpuCrashDumps", | ||
| 43 | &GFSDK_Aftermath_EnableGpuCrashDumps) || | ||
| 44 | !dl.GetSymbol("GFSDK_Aftermath_GetShaderDebugInfoIdentifier", | ||
| 45 | &GFSDK_Aftermath_GetShaderDebugInfoIdentifier) || | ||
| 46 | !dl.GetSymbol("GFSDK_Aftermath_GetShaderHashSpirv", &GFSDK_Aftermath_GetShaderHashSpirv) || | ||
| 47 | !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_CreateDecoder", | ||
| 48 | &GFSDK_Aftermath_GpuCrashDump_CreateDecoder) || | ||
| 49 | !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_DestroyDecoder", | ||
| 50 | &GFSDK_Aftermath_GpuCrashDump_DestroyDecoder) || | ||
| 51 | !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GenerateJSON", | ||
| 52 | &GFSDK_Aftermath_GpuCrashDump_GenerateJSON) || | ||
| 53 | !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GetJSON", | ||
| 54 | &GFSDK_Aftermath_GpuCrashDump_GetJSON)) { | ||
| 55 | LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath function pointers"); | ||
| 56 | return false; | ||
| 57 | } | ||
| 58 | dump_dir = Common::FS::GetUserPath(Common::FS::UserPath::LogDir) + "gpucrash"; | ||
| 59 | |||
| 60 | void(Common::FS::DeleteDirRecursively(dump_dir)); | ||
| 61 | if (!Common::FS::CreateDir(dump_dir)) { | ||
| 62 | LOG_ERROR(Render_Vulkan, "Failed to create Nsight Aftermath dump directory"); | ||
| 63 | return; | ||
| 64 | } | ||
| 65 | if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_EnableGpuCrashDumps( | ||
| 66 | GFSDK_Aftermath_Version_API, GFSDK_Aftermath_GpuCrashDumpWatchedApiFlags_Vulkan, | ||
| 67 | GFSDK_Aftermath_GpuCrashDumpFeatureFlags_Default, GpuCrashDumpCallback, | ||
| 68 | ShaderDebugInfoCallback, CrashDumpDescriptionCallback, this))) { | ||
| 69 | LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_EnableGpuCrashDumps failed"); | ||
| 70 | return; | ||
| 71 | } | ||
| 72 | LOG_INFO(Render_Vulkan, "Nsight Aftermath dump directory is \"{}\"", dump_dir); | ||
| 73 | initialized = true; | ||
| 74 | } | ||
| 75 | |||
| 76 | NsightAftermathTracker::~NsightAftermathTracker() { | ||
| 77 | if (initialized) { | ||
| 78 | (void)GFSDK_Aftermath_DisableGpuCrashDumps(); | ||
| 79 | } | ||
| 80 | } | ||
| 81 | |||
| 82 | void NsightAftermathTracker::SaveShader(const std::vector<u32>& spirv) const { | ||
| 83 | if (!initialized) { | ||
| 84 | return; | ||
| 85 | } | ||
| 86 | |||
| 87 | std::vector<u32> spirv_copy = spirv; | ||
| 88 | GFSDK_Aftermath_SpirvCode shader; | ||
| 89 | shader.pData = spirv_copy.data(); | ||
| 90 | shader.size = static_cast<u32>(spirv_copy.size() * 4); | ||
| 91 | |||
| 92 | std::scoped_lock lock{mutex}; | ||
| 93 | |||
| 94 | GFSDK_Aftermath_ShaderHash hash; | ||
| 95 | if (!GFSDK_Aftermath_SUCCEED( | ||
| 96 | GFSDK_Aftermath_GetShaderHashSpirv(GFSDK_Aftermath_Version_API, &shader, &hash))) { | ||
| 97 | LOG_ERROR(Render_Vulkan, "Failed to hash SPIR-V module"); | ||
| 98 | return; | ||
| 99 | } | ||
| 100 | |||
| 101 | Common::FS::IOFile file(fmt::format("{}/source_{:016x}.spv", dump_dir, hash.hash), "wb"); | ||
| 102 | if (!file.IsOpen()) { | ||
| 103 | LOG_ERROR(Render_Vulkan, "Failed to dump SPIR-V module with hash={:016x}", hash.hash); | ||
| 104 | return; | ||
| 105 | } | ||
| 106 | if (file.WriteArray(spirv.data(), spirv.size()) != spirv.size()) { | ||
| 107 | LOG_ERROR(Render_Vulkan, "Failed to write SPIR-V module with hash={:016x}", hash.hash); | ||
| 108 | return; | ||
| 109 | } | ||
| 110 | } | ||
| 111 | |||
| 112 | void NsightAftermathTracker::OnGpuCrashDumpCallback(const void* gpu_crash_dump, | ||
| 113 | u32 gpu_crash_dump_size) { | ||
| 114 | std::scoped_lock lock{mutex}; | ||
| 115 | |||
| 116 | LOG_CRITICAL(Render_Vulkan, "called"); | ||
| 117 | |||
| 118 | GFSDK_Aftermath_GpuCrashDump_Decoder decoder; | ||
| 119 | if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_CreateDecoder( | ||
| 120 | GFSDK_Aftermath_Version_API, gpu_crash_dump, gpu_crash_dump_size, &decoder))) { | ||
| 121 | LOG_ERROR(Render_Vulkan, "Failed to create decoder"); | ||
| 122 | return; | ||
| 123 | } | ||
| 124 | SCOPE_EXIT({ GFSDK_Aftermath_GpuCrashDump_DestroyDecoder(decoder); }); | ||
| 125 | |||
| 126 | u32 json_size = 0; | ||
| 127 | if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_GenerateJSON( | ||
| 128 | decoder, GFSDK_Aftermath_GpuCrashDumpDecoderFlags_ALL_INFO, | ||
| 129 | GFSDK_Aftermath_GpuCrashDumpFormatterFlags_NONE, nullptr, nullptr, nullptr, nullptr, | ||
| 130 | this, &json_size))) { | ||
| 131 | LOG_ERROR(Render_Vulkan, "Failed to generate JSON"); | ||
| 132 | return; | ||
| 133 | } | ||
| 134 | std::vector<char> json(json_size); | ||
| 135 | if (!GFSDK_Aftermath_SUCCEED( | ||
| 136 | GFSDK_Aftermath_GpuCrashDump_GetJSON(decoder, json_size, json.data()))) { | ||
| 137 | LOG_ERROR(Render_Vulkan, "Failed to query JSON"); | ||
| 138 | return; | ||
| 139 | } | ||
| 140 | |||
| 141 | const std::string base_name = [this] { | ||
| 142 | const int id = dump_id++; | ||
| 143 | if (id == 0) { | ||
| 144 | return fmt::format("{}/crash.nv-gpudmp", dump_dir); | ||
| 145 | } else { | ||
| 146 | return fmt::format("{}/crash_{}.nv-gpudmp", dump_dir, id); | ||
| 147 | } | ||
| 148 | }(); | ||
| 149 | |||
| 150 | std::string_view dump_view(static_cast<const char*>(gpu_crash_dump), gpu_crash_dump_size); | ||
| 151 | if (Common::FS::WriteStringToFile(false, base_name, dump_view) != gpu_crash_dump_size) { | ||
| 152 | LOG_ERROR(Render_Vulkan, "Failed to write dump file"); | ||
| 153 | return; | ||
| 154 | } | ||
| 155 | const std::string_view json_view(json.data(), json.size()); | ||
| 156 | if (Common::FS::WriteStringToFile(true, base_name + ".json", json_view) != json.size()) { | ||
| 157 | LOG_ERROR(Render_Vulkan, "Failed to write JSON"); | ||
| 158 | return; | ||
| 159 | } | ||
| 160 | } | ||
| 161 | |||
| 162 | void NsightAftermathTracker::OnShaderDebugInfoCallback(const void* shader_debug_info, | ||
| 163 | u32 shader_debug_info_size) { | ||
| 164 | std::scoped_lock lock{mutex}; | ||
| 165 | |||
| 166 | GFSDK_Aftermath_ShaderDebugInfoIdentifier identifier; | ||
| 167 | if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GetShaderDebugInfoIdentifier( | ||
| 168 | GFSDK_Aftermath_Version_API, shader_debug_info, shader_debug_info_size, &identifier))) { | ||
| 169 | LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_GetShaderDebugInfoIdentifier failed"); | ||
| 170 | return; | ||
| 171 | } | ||
| 172 | |||
| 173 | const std::string path = | ||
| 174 | fmt::format("{}/shader_{:016x}{:016x}.nvdbg", dump_dir, identifier.id[0], identifier.id[1]); | ||
| 175 | Common::FS::IOFile file(path, "wb"); | ||
| 176 | if (!file.IsOpen()) { | ||
| 177 | LOG_ERROR(Render_Vulkan, "Failed to create file {}", path); | ||
| 178 | return; | ||
| 179 | } | ||
| 180 | if (file.WriteBytes(static_cast<const u8*>(shader_debug_info), shader_debug_info_size) != | ||
| 181 | shader_debug_info_size) { | ||
| 182 | LOG_ERROR(Render_Vulkan, "Failed to write file {}", path); | ||
| 183 | return; | ||
| 184 | } | ||
| 185 | } | ||
| 186 | |||
| 187 | void NsightAftermathTracker::OnCrashDumpDescriptionCallback( | ||
| 188 | PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description) { | ||
| 189 | add_description(GFSDK_Aftermath_GpuCrashDumpDescriptionKey_ApplicationName, "yuzu"); | ||
| 190 | } | ||
| 191 | |||
| 192 | void NsightAftermathTracker::GpuCrashDumpCallback(const void* gpu_crash_dump, | ||
| 193 | u32 gpu_crash_dump_size, void* user_data) { | ||
| 194 | static_cast<NsightAftermathTracker*>(user_data)->OnGpuCrashDumpCallback(gpu_crash_dump, | ||
| 195 | gpu_crash_dump_size); | ||
| 196 | } | ||
| 197 | |||
| 198 | void NsightAftermathTracker::ShaderDebugInfoCallback(const void* shader_debug_info, | ||
| 199 | u32 shader_debug_info_size, void* user_data) { | ||
| 200 | static_cast<NsightAftermathTracker*>(user_data)->OnShaderDebugInfoCallback( | ||
| 201 | shader_debug_info, shader_debug_info_size); | ||
| 202 | } | ||
| 203 | |||
| 204 | void NsightAftermathTracker::CrashDumpDescriptionCallback( | ||
| 205 | PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data) { | ||
| 206 | static_cast<NsightAftermathTracker*>(user_data)->OnCrashDumpDescriptionCallback( | ||
| 207 | add_description); | ||
| 208 | } | ||
| 209 | |||
| 210 | } // namespace Vulkan | ||
| 211 | |||
| 212 | #endif // HAS_NSIGHT_AFTERMATH | ||
diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.h b/src/video_core/vulkan_common/nsight_aftermath_tracker.h new file mode 100644 index 000000000..cee3847fb --- /dev/null +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.h | |||
| @@ -0,0 +1,82 @@ | |||
| 1 | // Copyright 2020 yuzu Emulator Project | ||
| 2 | // Licensed under GPLv2 or any later version | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #pragma once | ||
| 6 | |||
| 7 | #include <mutex> | ||
| 8 | #include <string> | ||
| 9 | #include <vector> | ||
| 10 | |||
| 11 | #define VK_NO_PROTOTYPES | ||
| 12 | #include <vulkan/vulkan.h> | ||
| 13 | |||
| 14 | #ifdef HAS_NSIGHT_AFTERMATH | ||
| 15 | #include <GFSDK_Aftermath_Defines.h> | ||
| 16 | #include <GFSDK_Aftermath_GpuCrashDump.h> | ||
| 17 | #include <GFSDK_Aftermath_GpuCrashDumpDecoding.h> | ||
| 18 | #endif | ||
| 19 | |||
| 20 | #include "common/common_types.h" | ||
| 21 | #include "common/dynamic_library.h" | ||
| 22 | |||
| 23 | namespace Vulkan { | ||
| 24 | |||
| 25 | class NsightAftermathTracker { | ||
| 26 | public: | ||
| 27 | NsightAftermathTracker(); | ||
| 28 | ~NsightAftermathTracker(); | ||
| 29 | |||
| 30 | NsightAftermathTracker(const NsightAftermathTracker&) = delete; | ||
| 31 | NsightAftermathTracker& operator=(const NsightAftermathTracker&) = delete; | ||
| 32 | |||
| 33 | // Delete move semantics because Aftermath initialization uses a pointer to this. | ||
| 34 | NsightAftermathTracker(NsightAftermathTracker&&) = delete; | ||
| 35 | NsightAftermathTracker& operator=(NsightAftermathTracker&&) = delete; | ||
| 36 | |||
| 37 | void SaveShader(const std::vector<u32>& spirv) const; | ||
| 38 | |||
| 39 | private: | ||
| 40 | #ifdef HAS_NSIGHT_AFTERMATH | ||
| 41 | static void GpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size, | ||
| 42 | void* user_data); | ||
| 43 | |||
| 44 | static void ShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size, | ||
| 45 | void* user_data); | ||
| 46 | |||
| 47 | static void CrashDumpDescriptionCallback( | ||
| 48 | PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data); | ||
| 49 | |||
| 50 | void OnGpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size); | ||
| 51 | |||
| 52 | void OnShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size); | ||
| 53 | |||
| 54 | void OnCrashDumpDescriptionCallback( | ||
| 55 | PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description); | ||
| 56 | |||
| 57 | mutable std::mutex mutex; | ||
| 58 | |||
| 59 | std::string dump_dir; | ||
| 60 | int dump_id = 0; | ||
| 61 | |||
| 62 | bool initialized = false; | ||
| 63 | |||
| 64 | Common::DynamicLibrary dl; | ||
| 65 | PFN_GFSDK_Aftermath_DisableGpuCrashDumps GFSDK_Aftermath_DisableGpuCrashDumps; | ||
| 66 | PFN_GFSDK_Aftermath_EnableGpuCrashDumps GFSDK_Aftermath_EnableGpuCrashDumps; | ||
| 67 | PFN_GFSDK_Aftermath_GetShaderDebugInfoIdentifier GFSDK_Aftermath_GetShaderDebugInfoIdentifier; | ||
| 68 | PFN_GFSDK_Aftermath_GetShaderHashSpirv GFSDK_Aftermath_GetShaderHashSpirv; | ||
| 69 | PFN_GFSDK_Aftermath_GpuCrashDump_CreateDecoder GFSDK_Aftermath_GpuCrashDump_CreateDecoder; | ||
| 70 | PFN_GFSDK_Aftermath_GpuCrashDump_DestroyDecoder GFSDK_Aftermath_GpuCrashDump_DestroyDecoder; | ||
| 71 | PFN_GFSDK_Aftermath_GpuCrashDump_GenerateJSON GFSDK_Aftermath_GpuCrashDump_GenerateJSON; | ||
| 72 | PFN_GFSDK_Aftermath_GpuCrashDump_GetJSON GFSDK_Aftermath_GpuCrashDump_GetJSON; | ||
| 73 | #endif | ||
| 74 | }; | ||
| 75 | |||
| 76 | #ifndef HAS_NSIGHT_AFTERMATH | ||
| 77 | inline NsightAftermathTracker::NsightAftermathTracker() = default; | ||
| 78 | inline NsightAftermathTracker::~NsightAftermathTracker() = default; | ||
| 79 | inline void NsightAftermathTracker::SaveShader(const std::vector<u32>&) const {} | ||
| 80 | #endif | ||
| 81 | |||
| 82 | } // namespace Vulkan | ||
diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp new file mode 100644 index 000000000..75173324e --- /dev/null +++ b/src/video_core/vulkan_common/vulkan_device.cpp | |||
| @@ -0,0 +1,886 @@ | |||
| 1 | // Copyright 2018 yuzu Emulator Project | ||
| 2 | // Licensed under GPLv2 or any later version | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #include <bitset> | ||
| 6 | #include <chrono> | ||
| 7 | #include <optional> | ||
| 8 | #include <string_view> | ||
| 9 | #include <thread> | ||
| 10 | #include <unordered_set> | ||
| 11 | #include <utility> | ||
| 12 | #include <vector> | ||
| 13 | |||
| 14 | #include "common/assert.h" | ||
| 15 | #include "core/settings.h" | ||
| 16 | #include "video_core/vulkan_common/nsight_aftermath_tracker.h" | ||
| 17 | #include "video_core/vulkan_common/vulkan_device.h" | ||
| 18 | #include "video_core/vulkan_common/vulkan_wrapper.h" | ||
| 19 | |||
| 20 | namespace Vulkan { | ||
| 21 | |||
| 22 | namespace { | ||
| 23 | |||
| 24 | namespace Alternatives { | ||
| 25 | |||
| 26 | constexpr std::array Depth24UnormS8_UINT{ | ||
| 27 | VK_FORMAT_D32_SFLOAT_S8_UINT, | ||
| 28 | VK_FORMAT_D16_UNORM_S8_UINT, | ||
| 29 | VkFormat{}, | ||
| 30 | }; | ||
| 31 | |||
| 32 | constexpr std::array Depth16UnormS8_UINT{ | ||
| 33 | VK_FORMAT_D24_UNORM_S8_UINT, | ||
| 34 | VK_FORMAT_D32_SFLOAT_S8_UINT, | ||
| 35 | VkFormat{}, | ||
| 36 | }; | ||
| 37 | |||
| 38 | } // namespace Alternatives | ||
| 39 | |||
| 40 | constexpr std::array REQUIRED_EXTENSIONS{ | ||
| 41 | VK_KHR_SWAPCHAIN_EXTENSION_NAME, | ||
| 42 | VK_KHR_MAINTENANCE1_EXTENSION_NAME, | ||
| 43 | VK_KHR_STORAGE_BUFFER_STORAGE_CLASS_EXTENSION_NAME, | ||
| 44 | VK_KHR_SHADER_DRAW_PARAMETERS_EXTENSION_NAME, | ||
| 45 | VK_KHR_16BIT_STORAGE_EXTENSION_NAME, | ||
| 46 | VK_KHR_8BIT_STORAGE_EXTENSION_NAME, | ||
| 47 | VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME, | ||
| 48 | VK_KHR_DESCRIPTOR_UPDATE_TEMPLATE_EXTENSION_NAME, | ||
| 49 | VK_KHR_TIMELINE_SEMAPHORE_EXTENSION_NAME, | ||
| 50 | VK_KHR_SAMPLER_MIRROR_CLAMP_TO_EDGE_EXTENSION_NAME, | ||
| 51 | VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME, | ||
| 52 | VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME, | ||
| 53 | VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME, | ||
| 54 | VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME, | ||
| 55 | }; | ||
| 56 | |||
| 57 | template <typename T> | ||
| 58 | void SetNext(void**& next, T& data) { | ||
| 59 | *next = &data; | ||
| 60 | next = &data.pNext; | ||
| 61 | } | ||
| 62 | |||
| 63 | constexpr const VkFormat* GetFormatAlternatives(VkFormat format) { | ||
| 64 | switch (format) { | ||
| 65 | case VK_FORMAT_D24_UNORM_S8_UINT: | ||
| 66 | return Alternatives::Depth24UnormS8_UINT.data(); | ||
| 67 | case VK_FORMAT_D16_UNORM_S8_UINT: | ||
| 68 | return Alternatives::Depth16UnormS8_UINT.data(); | ||
| 69 | default: | ||
| 70 | return nullptr; | ||
| 71 | } | ||
| 72 | } | ||
| 73 | |||
| 74 | VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType format_type) { | ||
| 75 | switch (format_type) { | ||
| 76 | case FormatType::Linear: | ||
| 77 | return properties.linearTilingFeatures; | ||
| 78 | case FormatType::Optimal: | ||
| 79 | return properties.optimalTilingFeatures; | ||
| 80 | case FormatType::Buffer: | ||
| 81 | return properties.bufferFeatures; | ||
| 82 | default: | ||
| 83 | return {}; | ||
| 84 | } | ||
| 85 | } | ||
| 86 | |||
| 87 | [[nodiscard]] bool IsRDNA(std::string_view device_name, VkDriverIdKHR driver_id) { | ||
| 88 | static constexpr std::array RDNA_DEVICES{ | ||
| 89 | "5700", | ||
| 90 | "5600", | ||
| 91 | "5500", | ||
| 92 | "5300", | ||
| 93 | }; | ||
| 94 | if (driver_id != VK_DRIVER_ID_AMD_PROPRIETARY_KHR) { | ||
| 95 | return false; | ||
| 96 | } | ||
| 97 | return std::any_of(RDNA_DEVICES.begin(), RDNA_DEVICES.end(), [device_name](const char* name) { | ||
| 98 | return device_name.find(name) != std::string_view::npos; | ||
| 99 | }); | ||
| 100 | } | ||
| 101 | |||
| 102 | std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties( | ||
| 103 | vk::PhysicalDevice physical, const vk::InstanceDispatch& dld) { | ||
| 104 | static constexpr std::array formats{ | ||
| 105 | VK_FORMAT_A8B8G8R8_UNORM_PACK32, | ||
| 106 | VK_FORMAT_A8B8G8R8_UINT_PACK32, | ||
| 107 | VK_FORMAT_A8B8G8R8_SNORM_PACK32, | ||
| 108 | VK_FORMAT_A8B8G8R8_SINT_PACK32, | ||
| 109 | VK_FORMAT_A8B8G8R8_SRGB_PACK32, | ||
| 110 | VK_FORMAT_B5G6R5_UNORM_PACK16, | ||
| 111 | VK_FORMAT_A2B10G10R10_UNORM_PACK32, | ||
| 112 | VK_FORMAT_A2B10G10R10_UINT_PACK32, | ||
| 113 | VK_FORMAT_A1R5G5B5_UNORM_PACK16, | ||
| 114 | VK_FORMAT_R32G32B32A32_SFLOAT, | ||
| 115 | VK_FORMAT_R32G32B32A32_SINT, | ||
| 116 | VK_FORMAT_R32G32B32A32_UINT, | ||
| 117 | VK_FORMAT_R32G32_SFLOAT, | ||
| 118 | VK_FORMAT_R32G32_SINT, | ||
| 119 | VK_FORMAT_R32G32_UINT, | ||
| 120 | VK_FORMAT_R16G16B16A16_SINT, | ||
| 121 | VK_FORMAT_R16G16B16A16_UINT, | ||
| 122 | VK_FORMAT_R16G16B16A16_SNORM, | ||
| 123 | VK_FORMAT_R16G16B16A16_UNORM, | ||
| 124 | VK_FORMAT_R16G16_UNORM, | ||
| 125 | VK_FORMAT_R16G16_SNORM, | ||
| 126 | VK_FORMAT_R16G16_SFLOAT, | ||
| 127 | VK_FORMAT_R16G16_SINT, | ||
| 128 | VK_FORMAT_R16_UNORM, | ||
| 129 | VK_FORMAT_R16_UINT, | ||
| 130 | VK_FORMAT_R8G8B8A8_SRGB, | ||
| 131 | VK_FORMAT_R8G8_UNORM, | ||
| 132 | VK_FORMAT_R8G8_SNORM, | ||
| 133 | VK_FORMAT_R8G8_SINT, | ||
| 134 | VK_FORMAT_R8G8_UINT, | ||
| 135 | VK_FORMAT_R8_UNORM, | ||
| 136 | VK_FORMAT_R8_SNORM, | ||
| 137 | VK_FORMAT_R8_SINT, | ||
| 138 | VK_FORMAT_R8_UINT, | ||
| 139 | VK_FORMAT_B10G11R11_UFLOAT_PACK32, | ||
| 140 | VK_FORMAT_R32_SFLOAT, | ||
| 141 | VK_FORMAT_R32_UINT, | ||
| 142 | VK_FORMAT_R32_SINT, | ||
| 143 | VK_FORMAT_R16_SFLOAT, | ||
| 144 | VK_FORMAT_R16G16B16A16_SFLOAT, | ||
| 145 | VK_FORMAT_B8G8R8A8_UNORM, | ||
| 146 | VK_FORMAT_B8G8R8A8_SRGB, | ||
| 147 | VK_FORMAT_R4G4B4A4_UNORM_PACK16, | ||
| 148 | VK_FORMAT_D32_SFLOAT, | ||
| 149 | VK_FORMAT_D16_UNORM, | ||
| 150 | VK_FORMAT_D16_UNORM_S8_UINT, | ||
| 151 | VK_FORMAT_D24_UNORM_S8_UINT, | ||
| 152 | VK_FORMAT_D32_SFLOAT_S8_UINT, | ||
| 153 | VK_FORMAT_BC1_RGBA_UNORM_BLOCK, | ||
| 154 | VK_FORMAT_BC2_UNORM_BLOCK, | ||
| 155 | VK_FORMAT_BC3_UNORM_BLOCK, | ||
| 156 | VK_FORMAT_BC4_UNORM_BLOCK, | ||
| 157 | VK_FORMAT_BC4_SNORM_BLOCK, | ||
| 158 | VK_FORMAT_BC5_UNORM_BLOCK, | ||
| 159 | VK_FORMAT_BC5_SNORM_BLOCK, | ||
| 160 | VK_FORMAT_BC7_UNORM_BLOCK, | ||
| 161 | VK_FORMAT_BC6H_UFLOAT_BLOCK, | ||
| 162 | VK_FORMAT_BC6H_SFLOAT_BLOCK, | ||
| 163 | VK_FORMAT_BC1_RGBA_SRGB_BLOCK, | ||
| 164 | VK_FORMAT_BC2_SRGB_BLOCK, | ||
| 165 | VK_FORMAT_BC3_SRGB_BLOCK, | ||
| 166 | VK_FORMAT_BC7_SRGB_BLOCK, | ||
| 167 | VK_FORMAT_ASTC_4x4_UNORM_BLOCK, | ||
| 168 | VK_FORMAT_ASTC_4x4_SRGB_BLOCK, | ||
| 169 | VK_FORMAT_ASTC_5x4_UNORM_BLOCK, | ||
| 170 | VK_FORMAT_ASTC_5x4_SRGB_BLOCK, | ||
| 171 | VK_FORMAT_ASTC_5x5_UNORM_BLOCK, | ||
| 172 | VK_FORMAT_ASTC_5x5_SRGB_BLOCK, | ||
| 173 | VK_FORMAT_ASTC_6x5_UNORM_BLOCK, | ||
| 174 | VK_FORMAT_ASTC_6x5_SRGB_BLOCK, | ||
| 175 | VK_FORMAT_ASTC_6x6_UNORM_BLOCK, | ||
| 176 | VK_FORMAT_ASTC_6x6_SRGB_BLOCK, | ||
| 177 | VK_FORMAT_ASTC_8x5_UNORM_BLOCK, | ||
| 178 | VK_FORMAT_ASTC_8x5_SRGB_BLOCK, | ||
| 179 | VK_FORMAT_ASTC_8x6_UNORM_BLOCK, | ||
| 180 | VK_FORMAT_ASTC_8x6_SRGB_BLOCK, | ||
| 181 | VK_FORMAT_ASTC_8x8_UNORM_BLOCK, | ||
| 182 | VK_FORMAT_ASTC_8x8_SRGB_BLOCK, | ||
| 183 | VK_FORMAT_ASTC_10x5_UNORM_BLOCK, | ||
| 184 | VK_FORMAT_ASTC_10x5_SRGB_BLOCK, | ||
| 185 | VK_FORMAT_ASTC_10x6_UNORM_BLOCK, | ||
| 186 | VK_FORMAT_ASTC_10x6_SRGB_BLOCK, | ||
| 187 | VK_FORMAT_ASTC_10x8_UNORM_BLOCK, | ||
| 188 | VK_FORMAT_ASTC_10x8_SRGB_BLOCK, | ||
| 189 | VK_FORMAT_ASTC_10x10_UNORM_BLOCK, | ||
| 190 | VK_FORMAT_ASTC_10x10_SRGB_BLOCK, | ||
| 191 | VK_FORMAT_ASTC_12x10_UNORM_BLOCK, | ||
| 192 | VK_FORMAT_ASTC_12x10_SRGB_BLOCK, | ||
| 193 | VK_FORMAT_ASTC_12x12_UNORM_BLOCK, | ||
| 194 | VK_FORMAT_ASTC_12x12_SRGB_BLOCK, | ||
| 195 | VK_FORMAT_ASTC_8x6_UNORM_BLOCK, | ||
| 196 | VK_FORMAT_ASTC_8x6_SRGB_BLOCK, | ||
| 197 | VK_FORMAT_ASTC_6x5_UNORM_BLOCK, | ||
| 198 | VK_FORMAT_ASTC_6x5_SRGB_BLOCK, | ||
| 199 | VK_FORMAT_E5B9G9R9_UFLOAT_PACK32, | ||
| 200 | }; | ||
| 201 | std::unordered_map<VkFormat, VkFormatProperties> format_properties; | ||
| 202 | for (const auto format : formats) { | ||
| 203 | format_properties.emplace(format, physical.GetFormatProperties(format)); | ||
| 204 | } | ||
| 205 | return format_properties; | ||
| 206 | } | ||
| 207 | |||
| 208 | } // Anonymous namespace | ||
| 209 | |||
| 210 | Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR surface, | ||
| 211 | const vk::InstanceDispatch& dld_) | ||
| 212 | : instance{instance_}, dld{dld_}, physical{physical_}, properties{physical.GetProperties()}, | ||
| 213 | format_properties{GetFormatProperties(physical, dld)} { | ||
| 214 | CheckSuitability(); | ||
| 215 | SetupFamilies(surface); | ||
| 216 | SetupFeatures(); | ||
| 217 | |||
| 218 | const auto queue_cis = GetDeviceQueueCreateInfos(); | ||
| 219 | const std::vector extensions = LoadExtensions(); | ||
| 220 | |||
| 221 | VkPhysicalDeviceFeatures2 features2{ | ||
| 222 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2, | ||
| 223 | .pNext = nullptr, | ||
| 224 | }; | ||
| 225 | const void* first_next = &features2; | ||
| 226 | void** next = &features2.pNext; | ||
| 227 | |||
| 228 | features2.features = { | ||
| 229 | .robustBufferAccess = false, | ||
| 230 | .fullDrawIndexUint32 = false, | ||
| 231 | .imageCubeArray = true, | ||
| 232 | .independentBlend = true, | ||
| 233 | .geometryShader = true, | ||
| 234 | .tessellationShader = true, | ||
| 235 | .sampleRateShading = false, | ||
| 236 | .dualSrcBlend = false, | ||
| 237 | .logicOp = false, | ||
| 238 | .multiDrawIndirect = false, | ||
| 239 | .drawIndirectFirstInstance = false, | ||
| 240 | .depthClamp = true, | ||
| 241 | .depthBiasClamp = true, | ||
| 242 | .fillModeNonSolid = false, | ||
| 243 | .depthBounds = false, | ||
| 244 | .wideLines = false, | ||
| 245 | .largePoints = true, | ||
| 246 | .alphaToOne = false, | ||
| 247 | .multiViewport = true, | ||
| 248 | .samplerAnisotropy = true, | ||
| 249 | .textureCompressionETC2 = false, | ||
| 250 | .textureCompressionASTC_LDR = is_optimal_astc_supported, | ||
| 251 | .textureCompressionBC = false, | ||
| 252 | .occlusionQueryPrecise = true, | ||
| 253 | .pipelineStatisticsQuery = false, | ||
| 254 | .vertexPipelineStoresAndAtomics = true, | ||
| 255 | .fragmentStoresAndAtomics = true, | ||
| 256 | .shaderTessellationAndGeometryPointSize = false, | ||
| 257 | .shaderImageGatherExtended = true, | ||
| 258 | .shaderStorageImageExtendedFormats = false, | ||
| 259 | .shaderStorageImageMultisample = true, | ||
| 260 | .shaderStorageImageReadWithoutFormat = is_formatless_image_load_supported, | ||
| 261 | .shaderStorageImageWriteWithoutFormat = true, | ||
| 262 | .shaderUniformBufferArrayDynamicIndexing = false, | ||
| 263 | .shaderSampledImageArrayDynamicIndexing = false, | ||
| 264 | .shaderStorageBufferArrayDynamicIndexing = false, | ||
| 265 | .shaderStorageImageArrayDynamicIndexing = false, | ||
| 266 | .shaderClipDistance = false, | ||
| 267 | .shaderCullDistance = false, | ||
| 268 | .shaderFloat64 = false, | ||
| 269 | .shaderInt64 = false, | ||
| 270 | .shaderInt16 = false, | ||
| 271 | .shaderResourceResidency = false, | ||
| 272 | .shaderResourceMinLod = false, | ||
| 273 | .sparseBinding = false, | ||
| 274 | .sparseResidencyBuffer = false, | ||
| 275 | .sparseResidencyImage2D = false, | ||
| 276 | .sparseResidencyImage3D = false, | ||
| 277 | .sparseResidency2Samples = false, | ||
| 278 | .sparseResidency4Samples = false, | ||
| 279 | .sparseResidency8Samples = false, | ||
| 280 | .sparseResidency16Samples = false, | ||
| 281 | .sparseResidencyAliased = false, | ||
| 282 | .variableMultisampleRate = false, | ||
| 283 | .inheritedQueries = false, | ||
| 284 | }; | ||
| 285 | VkPhysicalDeviceTimelineSemaphoreFeaturesKHR timeline_semaphore{ | ||
| 286 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR, | ||
| 287 | .pNext = nullptr, | ||
| 288 | .timelineSemaphore = true, | ||
| 289 | }; | ||
| 290 | SetNext(next, timeline_semaphore); | ||
| 291 | |||
| 292 | VkPhysicalDevice16BitStorageFeaturesKHR bit16_storage{ | ||
| 293 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES_KHR, | ||
| 294 | .pNext = nullptr, | ||
| 295 | .storageBuffer16BitAccess = false, | ||
| 296 | .uniformAndStorageBuffer16BitAccess = true, | ||
| 297 | .storagePushConstant16 = false, | ||
| 298 | .storageInputOutput16 = false, | ||
| 299 | }; | ||
| 300 | SetNext(next, bit16_storage); | ||
| 301 | |||
| 302 | VkPhysicalDevice8BitStorageFeaturesKHR bit8_storage{ | ||
| 303 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR, | ||
| 304 | .pNext = nullptr, | ||
| 305 | .storageBuffer8BitAccess = false, | ||
| 306 | .uniformAndStorageBuffer8BitAccess = true, | ||
| 307 | .storagePushConstant8 = false, | ||
| 308 | }; | ||
| 309 | SetNext(next, bit8_storage); | ||
| 310 | |||
| 311 | VkPhysicalDeviceHostQueryResetFeaturesEXT host_query_reset{ | ||
| 312 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES_EXT, | ||
| 313 | .hostQueryReset = true, | ||
| 314 | }; | ||
| 315 | SetNext(next, host_query_reset); | ||
| 316 | |||
| 317 | VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8; | ||
| 318 | if (is_float16_supported) { | ||
| 319 | float16_int8 = { | ||
| 320 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR, | ||
| 321 | .pNext = nullptr, | ||
| 322 | .shaderFloat16 = true, | ||
| 323 | .shaderInt8 = false, | ||
| 324 | }; | ||
| 325 | SetNext(next, float16_int8); | ||
| 326 | } else { | ||
| 327 | LOG_INFO(Render_Vulkan, "Device doesn't support float16 natively"); | ||
| 328 | } | ||
| 329 | |||
| 330 | if (!nv_viewport_swizzle) { | ||
| 331 | LOG_INFO(Render_Vulkan, "Device doesn't support viewport swizzles"); | ||
| 332 | } | ||
| 333 | |||
| 334 | VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR std430_layout; | ||
| 335 | if (khr_uniform_buffer_standard_layout) { | ||
| 336 | std430_layout = { | ||
| 337 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR, | ||
| 338 | .pNext = nullptr, | ||
| 339 | .uniformBufferStandardLayout = true, | ||
| 340 | }; | ||
| 341 | SetNext(next, std430_layout); | ||
| 342 | } else { | ||
| 343 | LOG_INFO(Render_Vulkan, "Device doesn't support packed UBOs"); | ||
| 344 | } | ||
| 345 | |||
| 346 | VkPhysicalDeviceIndexTypeUint8FeaturesEXT index_type_uint8; | ||
| 347 | if (ext_index_type_uint8) { | ||
| 348 | index_type_uint8 = { | ||
| 349 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT, | ||
| 350 | .pNext = nullptr, | ||
| 351 | .indexTypeUint8 = true, | ||
| 352 | }; | ||
| 353 | SetNext(next, index_type_uint8); | ||
| 354 | } else { | ||
| 355 | LOG_INFO(Render_Vulkan, "Device doesn't support uint8 indexes"); | ||
| 356 | } | ||
| 357 | |||
| 358 | VkPhysicalDeviceTransformFeedbackFeaturesEXT transform_feedback; | ||
| 359 | if (ext_transform_feedback) { | ||
| 360 | transform_feedback = { | ||
| 361 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT, | ||
| 362 | .pNext = nullptr, | ||
| 363 | .transformFeedback = true, | ||
| 364 | .geometryStreams = true, | ||
| 365 | }; | ||
| 366 | SetNext(next, transform_feedback); | ||
| 367 | } else { | ||
| 368 | LOG_INFO(Render_Vulkan, "Device doesn't support transform feedbacks"); | ||
| 369 | } | ||
| 370 | |||
| 371 | VkPhysicalDeviceCustomBorderColorFeaturesEXT custom_border; | ||
| 372 | if (ext_custom_border_color) { | ||
| 373 | custom_border = { | ||
| 374 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT, | ||
| 375 | .pNext = nullptr, | ||
| 376 | .customBorderColors = VK_TRUE, | ||
| 377 | .customBorderColorWithoutFormat = VK_TRUE, | ||
| 378 | }; | ||
| 379 | SetNext(next, custom_border); | ||
| 380 | } else { | ||
| 381 | LOG_INFO(Render_Vulkan, "Device doesn't support custom border colors"); | ||
| 382 | } | ||
| 383 | |||
| 384 | VkPhysicalDeviceExtendedDynamicStateFeaturesEXT dynamic_state; | ||
| 385 | if (ext_extended_dynamic_state) { | ||
| 386 | dynamic_state = { | ||
| 387 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT, | ||
| 388 | .pNext = nullptr, | ||
| 389 | .extendedDynamicState = VK_TRUE, | ||
| 390 | }; | ||
| 391 | SetNext(next, dynamic_state); | ||
| 392 | } else { | ||
| 393 | LOG_INFO(Render_Vulkan, "Device doesn't support extended dynamic state"); | ||
| 394 | } | ||
| 395 | |||
| 396 | VkPhysicalDeviceRobustness2FeaturesEXT robustness2; | ||
| 397 | if (ext_robustness2) { | ||
| 398 | robustness2 = { | ||
| 399 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT, | ||
| 400 | .pNext = nullptr, | ||
| 401 | .robustBufferAccess2 = false, | ||
| 402 | .robustImageAccess2 = true, | ||
| 403 | .nullDescriptor = true, | ||
| 404 | }; | ||
| 405 | SetNext(next, robustness2); | ||
| 406 | } else { | ||
| 407 | LOG_INFO(Render_Vulkan, "Device doesn't support robustness2"); | ||
| 408 | } | ||
| 409 | |||
| 410 | if (!ext_depth_range_unrestricted) { | ||
| 411 | LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted"); | ||
| 412 | } | ||
| 413 | |||
| 414 | VkDeviceDiagnosticsConfigCreateInfoNV diagnostics_nv; | ||
| 415 | if (nv_device_diagnostics_config) { | ||
| 416 | nsight_aftermath_tracker = std::make_unique<NsightAftermathTracker>(); | ||
| 417 | |||
| 418 | diagnostics_nv = { | ||
| 419 | .sType = VK_STRUCTURE_TYPE_DEVICE_DIAGNOSTICS_CONFIG_CREATE_INFO_NV, | ||
| 420 | .pNext = &features2, | ||
| 421 | .flags = VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_SHADER_DEBUG_INFO_BIT_NV | | ||
| 422 | VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_RESOURCE_TRACKING_BIT_NV | | ||
| 423 | VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_AUTOMATIC_CHECKPOINTS_BIT_NV, | ||
| 424 | }; | ||
| 425 | first_next = &diagnostics_nv; | ||
| 426 | } | ||
| 427 | logical = vk::Device::Create(physical, queue_cis, extensions, first_next, dld); | ||
| 428 | |||
| 429 | CollectTelemetryParameters(); | ||
| 430 | CollectToolingInfo(); | ||
| 431 | |||
| 432 | if (ext_extended_dynamic_state && driver_id == VK_DRIVER_ID_MESA_RADV) { | ||
| 433 | LOG_WARNING( | ||
| 434 | Render_Vulkan, | ||
| 435 | "Blacklisting RADV for VK_EXT_extended_dynamic state, likely due to a bug in yuzu"); | ||
| 436 | ext_extended_dynamic_state = false; | ||
| 437 | } | ||
| 438 | if (ext_extended_dynamic_state && IsRDNA(properties.deviceName, driver_id)) { | ||
| 439 | // AMD's proprietary driver supports VK_EXT_extended_dynamic_state but on RDNA devices it | ||
| 440 | // seems to cause stability issues | ||
| 441 | LOG_WARNING( | ||
| 442 | Render_Vulkan, | ||
| 443 | "Blacklisting AMD proprietary on RDNA devices from VK_EXT_extended_dynamic_state"); | ||
| 444 | ext_extended_dynamic_state = false; | ||
| 445 | } | ||
| 446 | |||
| 447 | graphics_queue = logical.GetQueue(graphics_family); | ||
| 448 | present_queue = logical.GetQueue(present_family); | ||
| 449 | |||
| 450 | use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue(); | ||
| 451 | } | ||
| 452 | |||
| 453 | Device::~Device() = default; | ||
| 454 | |||
| 455 | VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, | ||
| 456 | FormatType format_type) const { | ||
| 457 | if (IsFormatSupported(wanted_format, wanted_usage, format_type)) { | ||
| 458 | return wanted_format; | ||
| 459 | } | ||
| 460 | // The wanted format is not supported by hardware, search for alternatives | ||
| 461 | const VkFormat* alternatives = GetFormatAlternatives(wanted_format); | ||
| 462 | if (alternatives == nullptr) { | ||
| 463 | UNREACHABLE_MSG("Format={} with usage={} and type={} has no defined alternatives and host " | ||
| 464 | "hardware does not support it", | ||
| 465 | wanted_format, wanted_usage, format_type); | ||
| 466 | return wanted_format; | ||
| 467 | } | ||
| 468 | |||
| 469 | std::size_t i = 0; | ||
| 470 | for (VkFormat alternative = *alternatives; alternative; alternative = alternatives[++i]) { | ||
| 471 | if (!IsFormatSupported(alternative, wanted_usage, format_type)) { | ||
| 472 | continue; | ||
| 473 | } | ||
| 474 | LOG_WARNING(Render_Vulkan, | ||
| 475 | "Emulating format={} with alternative format={} with usage={} and type={}", | ||
| 476 | wanted_format, alternative, wanted_usage, format_type); | ||
| 477 | return alternative; | ||
| 478 | } | ||
| 479 | |||
| 480 | // No alternatives found, panic | ||
| 481 | UNREACHABLE_MSG("Format={} with usage={} and type={} is not supported by the host hardware and " | ||
| 482 | "doesn't support any of the alternatives", | ||
| 483 | wanted_format, wanted_usage, format_type); | ||
| 484 | return wanted_format; | ||
| 485 | } | ||
| 486 | |||
| 487 | void Device::ReportLoss() const { | ||
| 488 | LOG_CRITICAL(Render_Vulkan, "Device loss occured!"); | ||
| 489 | |||
| 490 | // Wait for the log to flush and for Nsight Aftermath to dump the results | ||
| 491 | std::this_thread::sleep_for(std::chrono::seconds{15}); | ||
| 492 | } | ||
| 493 | |||
| 494 | void Device::SaveShader(const std::vector<u32>& spirv) const { | ||
| 495 | if (nsight_aftermath_tracker) { | ||
| 496 | nsight_aftermath_tracker->SaveShader(spirv); | ||
| 497 | } | ||
| 498 | } | ||
| 499 | |||
| 500 | bool Device::IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const { | ||
| 501 | // Disable for now to avoid converting ASTC twice. | ||
| 502 | static constexpr std::array astc_formats = { | ||
| 503 | VK_FORMAT_ASTC_4x4_UNORM_BLOCK, VK_FORMAT_ASTC_4x4_SRGB_BLOCK, | ||
| 504 | VK_FORMAT_ASTC_5x4_UNORM_BLOCK, VK_FORMAT_ASTC_5x4_SRGB_BLOCK, | ||
| 505 | VK_FORMAT_ASTC_5x5_UNORM_BLOCK, VK_FORMAT_ASTC_5x5_SRGB_BLOCK, | ||
| 506 | VK_FORMAT_ASTC_6x5_UNORM_BLOCK, VK_FORMAT_ASTC_6x5_SRGB_BLOCK, | ||
| 507 | VK_FORMAT_ASTC_6x6_UNORM_BLOCK, VK_FORMAT_ASTC_6x6_SRGB_BLOCK, | ||
| 508 | VK_FORMAT_ASTC_8x5_UNORM_BLOCK, VK_FORMAT_ASTC_8x5_SRGB_BLOCK, | ||
| 509 | VK_FORMAT_ASTC_8x6_UNORM_BLOCK, VK_FORMAT_ASTC_8x6_SRGB_BLOCK, | ||
| 510 | VK_FORMAT_ASTC_8x8_UNORM_BLOCK, VK_FORMAT_ASTC_8x8_SRGB_BLOCK, | ||
| 511 | VK_FORMAT_ASTC_10x5_UNORM_BLOCK, VK_FORMAT_ASTC_10x5_SRGB_BLOCK, | ||
| 512 | VK_FORMAT_ASTC_10x6_UNORM_BLOCK, VK_FORMAT_ASTC_10x6_SRGB_BLOCK, | ||
| 513 | VK_FORMAT_ASTC_10x8_UNORM_BLOCK, VK_FORMAT_ASTC_10x8_SRGB_BLOCK, | ||
| 514 | VK_FORMAT_ASTC_10x10_UNORM_BLOCK, VK_FORMAT_ASTC_10x10_SRGB_BLOCK, | ||
| 515 | VK_FORMAT_ASTC_12x10_UNORM_BLOCK, VK_FORMAT_ASTC_12x10_SRGB_BLOCK, | ||
| 516 | VK_FORMAT_ASTC_12x12_UNORM_BLOCK, VK_FORMAT_ASTC_12x12_SRGB_BLOCK, | ||
| 517 | }; | ||
| 518 | if (!features.textureCompressionASTC_LDR) { | ||
| 519 | return false; | ||
| 520 | } | ||
| 521 | const auto format_feature_usage{ | ||
| 522 | VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT | VK_FORMAT_FEATURE_BLIT_SRC_BIT | | ||
| 523 | VK_FORMAT_FEATURE_BLIT_DST_BIT | VK_FORMAT_FEATURE_TRANSFER_SRC_BIT | | ||
| 524 | VK_FORMAT_FEATURE_TRANSFER_DST_BIT}; | ||
| 525 | for (const auto format : astc_formats) { | ||
| 526 | const auto physical_format_properties{physical.GetFormatProperties(format)}; | ||
| 527 | if ((physical_format_properties.optimalTilingFeatures & format_feature_usage) == 0) { | ||
| 528 | return false; | ||
| 529 | } | ||
| 530 | } | ||
| 531 | return true; | ||
| 532 | } | ||
| 533 | |||
| 534 | bool Device::TestDepthStencilBlits() const { | ||
| 535 | static constexpr VkFormatFeatureFlags required_features = | ||
| 536 | VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT; | ||
| 537 | const auto test_features = [](VkFormatProperties props) { | ||
| 538 | return (props.optimalTilingFeatures & required_features) == required_features; | ||
| 539 | }; | ||
| 540 | return test_features(format_properties.at(VK_FORMAT_D32_SFLOAT_S8_UINT)) && | ||
| 541 | test_features(format_properties.at(VK_FORMAT_D24_UNORM_S8_UINT)); | ||
| 542 | } | ||
| 543 | |||
| 544 | bool Device::IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, | ||
| 545 | FormatType format_type) const { | ||
| 546 | const auto it = format_properties.find(wanted_format); | ||
| 547 | if (it == format_properties.end()) { | ||
| 548 | UNIMPLEMENTED_MSG("Unimplemented format query={}", wanted_format); | ||
| 549 | return true; | ||
| 550 | } | ||
| 551 | const auto supported_usage = GetFormatFeatures(it->second, format_type); | ||
| 552 | return (supported_usage & wanted_usage) == wanted_usage; | ||
| 553 | } | ||
| 554 | |||
| 555 | void Device::CheckSuitability() const { | ||
| 556 | std::bitset<REQUIRED_EXTENSIONS.size()> available_extensions; | ||
| 557 | for (const VkExtensionProperties& property : physical.EnumerateDeviceExtensionProperties()) { | ||
| 558 | for (std::size_t i = 0; i < REQUIRED_EXTENSIONS.size(); ++i) { | ||
| 559 | if (available_extensions[i]) { | ||
| 560 | continue; | ||
| 561 | } | ||
| 562 | const std::string_view name{property.extensionName}; | ||
| 563 | available_extensions[i] = name == REQUIRED_EXTENSIONS[i]; | ||
| 564 | } | ||
| 565 | } | ||
| 566 | for (size_t i = 0; i < REQUIRED_EXTENSIONS.size(); ++i) { | ||
| 567 | if (available_extensions[i]) { | ||
| 568 | continue; | ||
| 569 | } | ||
| 570 | LOG_ERROR(Render_Vulkan, "Missing required extension: {}", REQUIRED_EXTENSIONS[i]); | ||
| 571 | throw vk::Exception(VK_ERROR_EXTENSION_NOT_PRESENT); | ||
| 572 | } | ||
| 573 | struct LimitTuple { | ||
| 574 | u32 minimum; | ||
| 575 | u32 value; | ||
| 576 | const char* name; | ||
| 577 | }; | ||
| 578 | const VkPhysicalDeviceLimits& limits{properties.limits}; | ||
| 579 | const std::array limits_report{ | ||
| 580 | LimitTuple{65536, limits.maxUniformBufferRange, "maxUniformBufferRange"}, | ||
| 581 | LimitTuple{16, limits.maxViewports, "maxViewports"}, | ||
| 582 | LimitTuple{8, limits.maxColorAttachments, "maxColorAttachments"}, | ||
| 583 | LimitTuple{8, limits.maxClipDistances, "maxClipDistances"}, | ||
| 584 | }; | ||
| 585 | for (const auto& tuple : limits_report) { | ||
| 586 | if (tuple.value < tuple.minimum) { | ||
| 587 | LOG_ERROR(Render_Vulkan, "{} has to be {} or greater but it is {}", tuple.name, | ||
| 588 | tuple.minimum, tuple.value); | ||
| 589 | throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); | ||
| 590 | } | ||
| 591 | } | ||
| 592 | const VkPhysicalDeviceFeatures features{physical.GetFeatures()}; | ||
| 593 | const std::array feature_report{ | ||
| 594 | std::make_pair(features.vertexPipelineStoresAndAtomics, "vertexPipelineStoresAndAtomics"), | ||
| 595 | std::make_pair(features.imageCubeArray, "imageCubeArray"), | ||
| 596 | std::make_pair(features.independentBlend, "independentBlend"), | ||
| 597 | std::make_pair(features.depthClamp, "depthClamp"), | ||
| 598 | std::make_pair(features.samplerAnisotropy, "samplerAnisotropy"), | ||
| 599 | std::make_pair(features.largePoints, "largePoints"), | ||
| 600 | std::make_pair(features.multiViewport, "multiViewport"), | ||
| 601 | std::make_pair(features.depthBiasClamp, "depthBiasClamp"), | ||
| 602 | std::make_pair(features.geometryShader, "geometryShader"), | ||
| 603 | std::make_pair(features.tessellationShader, "tessellationShader"), | ||
| 604 | std::make_pair(features.occlusionQueryPrecise, "occlusionQueryPrecise"), | ||
| 605 | std::make_pair(features.fragmentStoresAndAtomics, "fragmentStoresAndAtomics"), | ||
| 606 | std::make_pair(features.shaderImageGatherExtended, "shaderImageGatherExtended"), | ||
| 607 | std::make_pair(features.shaderStorageImageMultisample, "shaderStorageImageMultisample"), | ||
| 608 | std::make_pair(features.shaderStorageImageWriteWithoutFormat, | ||
| 609 | "shaderStorageImageWriteWithoutFormat"), | ||
| 610 | }; | ||
| 611 | for (const auto& [is_supported, name] : feature_report) { | ||
| 612 | if (is_supported) { | ||
| 613 | continue; | ||
| 614 | } | ||
| 615 | LOG_ERROR(Render_Vulkan, "Missing required feature: {}", name); | ||
| 616 | throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); | ||
| 617 | } | ||
| 618 | } | ||
| 619 | |||
| 620 | std::vector<const char*> Device::LoadExtensions() { | ||
| 621 | std::vector<const char*> extensions; | ||
| 622 | extensions.reserve(7 + REQUIRED_EXTENSIONS.size()); | ||
| 623 | extensions.insert(extensions.begin(), REQUIRED_EXTENSIONS.begin(), REQUIRED_EXTENSIONS.end()); | ||
| 624 | |||
| 625 | bool has_khr_shader_float16_int8{}; | ||
| 626 | bool has_ext_subgroup_size_control{}; | ||
| 627 | bool has_ext_transform_feedback{}; | ||
| 628 | bool has_ext_custom_border_color{}; | ||
| 629 | bool has_ext_extended_dynamic_state{}; | ||
| 630 | bool has_ext_robustness2{}; | ||
| 631 | for (const VkExtensionProperties& extension : physical.EnumerateDeviceExtensionProperties()) { | ||
| 632 | const auto test = [&](std::optional<std::reference_wrapper<bool>> status, const char* name, | ||
| 633 | bool push) { | ||
| 634 | if (extension.extensionName != std::string_view(name)) { | ||
| 635 | return; | ||
| 636 | } | ||
| 637 | if (push) { | ||
| 638 | extensions.push_back(name); | ||
| 639 | } | ||
| 640 | if (status) { | ||
| 641 | status->get() = true; | ||
| 642 | } | ||
| 643 | }; | ||
| 644 | test(nv_viewport_swizzle, VK_NV_VIEWPORT_SWIZZLE_EXTENSION_NAME, true); | ||
| 645 | test(khr_uniform_buffer_standard_layout, | ||
| 646 | VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true); | ||
| 647 | test(has_khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false); | ||
| 648 | test(ext_depth_range_unrestricted, VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME, true); | ||
| 649 | test(ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true); | ||
| 650 | test(ext_sampler_filter_minmax, VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME, true); | ||
| 651 | test(ext_shader_viewport_index_layer, VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME, | ||
| 652 | true); | ||
| 653 | test(ext_tooling_info, VK_EXT_TOOLING_INFO_EXTENSION_NAME, true); | ||
| 654 | test(ext_shader_stencil_export, VK_EXT_SHADER_STENCIL_EXPORT_EXTENSION_NAME, true); | ||
| 655 | test(has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME, false); | ||
| 656 | test(has_ext_custom_border_color, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME, false); | ||
| 657 | test(has_ext_extended_dynamic_state, VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME, false); | ||
| 658 | test(has_ext_robustness2, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME, false); | ||
| 659 | test(has_ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false); | ||
| 660 | if (Settings::values.renderer_debug) { | ||
| 661 | test(nv_device_diagnostics_config, VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, | ||
| 662 | true); | ||
| 663 | } | ||
| 664 | } | ||
| 665 | |||
| 666 | VkPhysicalDeviceFeatures2KHR features; | ||
| 667 | features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR; | ||
| 668 | |||
| 669 | VkPhysicalDeviceProperties2KHR physical_properties; | ||
| 670 | physical_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR; | ||
| 671 | |||
| 672 | if (has_khr_shader_float16_int8) { | ||
| 673 | VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8_features; | ||
| 674 | float16_int8_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR; | ||
| 675 | float16_int8_features.pNext = nullptr; | ||
| 676 | features.pNext = &float16_int8_features; | ||
| 677 | |||
| 678 | physical.GetFeatures2KHR(features); | ||
| 679 | is_float16_supported = float16_int8_features.shaderFloat16; | ||
| 680 | extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME); | ||
| 681 | } | ||
| 682 | |||
| 683 | if (has_ext_subgroup_size_control) { | ||
| 684 | VkPhysicalDeviceSubgroupSizeControlFeaturesEXT subgroup_features; | ||
| 685 | subgroup_features.sType = | ||
| 686 | VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_FEATURES_EXT; | ||
| 687 | subgroup_features.pNext = nullptr; | ||
| 688 | features.pNext = &subgroup_features; | ||
| 689 | physical.GetFeatures2KHR(features); | ||
| 690 | |||
| 691 | VkPhysicalDeviceSubgroupSizeControlPropertiesEXT subgroup_properties; | ||
| 692 | subgroup_properties.sType = | ||
| 693 | VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES_EXT; | ||
| 694 | subgroup_properties.pNext = nullptr; | ||
| 695 | physical_properties.pNext = &subgroup_properties; | ||
| 696 | physical.GetProperties2KHR(physical_properties); | ||
| 697 | |||
| 698 | is_warp_potentially_bigger = subgroup_properties.maxSubgroupSize > GuestWarpSize; | ||
| 699 | |||
| 700 | if (subgroup_features.subgroupSizeControl && | ||
| 701 | subgroup_properties.minSubgroupSize <= GuestWarpSize && | ||
| 702 | subgroup_properties.maxSubgroupSize >= GuestWarpSize) { | ||
| 703 | extensions.push_back(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME); | ||
| 704 | guest_warp_stages = subgroup_properties.requiredSubgroupSizeStages; | ||
| 705 | } | ||
| 706 | } else { | ||
| 707 | is_warp_potentially_bigger = true; | ||
| 708 | } | ||
| 709 | |||
| 710 | if (has_ext_transform_feedback) { | ||
| 711 | VkPhysicalDeviceTransformFeedbackFeaturesEXT tfb_features; | ||
| 712 | tfb_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT; | ||
| 713 | tfb_features.pNext = nullptr; | ||
| 714 | features.pNext = &tfb_features; | ||
| 715 | physical.GetFeatures2KHR(features); | ||
| 716 | |||
| 717 | VkPhysicalDeviceTransformFeedbackPropertiesEXT tfb_properties; | ||
| 718 | tfb_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT; | ||
| 719 | tfb_properties.pNext = nullptr; | ||
| 720 | physical_properties.pNext = &tfb_properties; | ||
| 721 | physical.GetProperties2KHR(physical_properties); | ||
| 722 | |||
| 723 | if (tfb_features.transformFeedback && tfb_features.geometryStreams && | ||
| 724 | tfb_properties.maxTransformFeedbackStreams >= 4 && | ||
| 725 | tfb_properties.maxTransformFeedbackBuffers && tfb_properties.transformFeedbackQueries && | ||
| 726 | tfb_properties.transformFeedbackDraw) { | ||
| 727 | extensions.push_back(VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME); | ||
| 728 | ext_transform_feedback = true; | ||
| 729 | } | ||
| 730 | } | ||
| 731 | |||
| 732 | if (has_ext_custom_border_color) { | ||
| 733 | VkPhysicalDeviceCustomBorderColorFeaturesEXT border_features; | ||
| 734 | border_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT; | ||
| 735 | border_features.pNext = nullptr; | ||
| 736 | features.pNext = &border_features; | ||
| 737 | physical.GetFeatures2KHR(features); | ||
| 738 | |||
| 739 | if (border_features.customBorderColors && border_features.customBorderColorWithoutFormat) { | ||
| 740 | extensions.push_back(VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME); | ||
| 741 | ext_custom_border_color = true; | ||
| 742 | } | ||
| 743 | } | ||
| 744 | |||
| 745 | if (has_ext_extended_dynamic_state) { | ||
| 746 | VkPhysicalDeviceExtendedDynamicStateFeaturesEXT dynamic_state; | ||
| 747 | dynamic_state.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT; | ||
| 748 | dynamic_state.pNext = nullptr; | ||
| 749 | features.pNext = &dynamic_state; | ||
| 750 | physical.GetFeatures2KHR(features); | ||
| 751 | |||
| 752 | if (dynamic_state.extendedDynamicState) { | ||
| 753 | extensions.push_back(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); | ||
| 754 | ext_extended_dynamic_state = true; | ||
| 755 | } | ||
| 756 | } | ||
| 757 | |||
| 758 | if (has_ext_robustness2) { | ||
| 759 | VkPhysicalDeviceRobustness2FeaturesEXT robustness2; | ||
| 760 | robustness2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT; | ||
| 761 | robustness2.pNext = nullptr; | ||
| 762 | features.pNext = &robustness2; | ||
| 763 | physical.GetFeatures2KHR(features); | ||
| 764 | if (robustness2.nullDescriptor && robustness2.robustImageAccess2) { | ||
| 765 | extensions.push_back(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME); | ||
| 766 | ext_robustness2 = true; | ||
| 767 | } | ||
| 768 | } | ||
| 769 | |||
| 770 | return extensions; | ||
| 771 | } | ||
| 772 | |||
| 773 | void Device::SetupFamilies(VkSurfaceKHR surface) { | ||
| 774 | const std::vector queue_family_properties = physical.GetQueueFamilyProperties(); | ||
| 775 | std::optional<u32> graphics; | ||
| 776 | std::optional<u32> present; | ||
| 777 | for (u32 index = 0; index < static_cast<u32>(queue_family_properties.size()); ++index) { | ||
| 778 | if (graphics && (present || !surface)) { | ||
| 779 | break; | ||
| 780 | } | ||
| 781 | const VkQueueFamilyProperties& queue_family = queue_family_properties[index]; | ||
| 782 | if (queue_family.queueCount == 0) { | ||
| 783 | continue; | ||
| 784 | } | ||
| 785 | if (queue_family.queueFlags & VK_QUEUE_GRAPHICS_BIT) { | ||
| 786 | graphics = index; | ||
| 787 | } | ||
| 788 | if (surface && physical.GetSurfaceSupportKHR(index, surface)) { | ||
| 789 | present = index; | ||
| 790 | } | ||
| 791 | } | ||
| 792 | if (!graphics) { | ||
| 793 | LOG_ERROR(Render_Vulkan, "Device lacks a graphics queue"); | ||
| 794 | throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); | ||
| 795 | } | ||
| 796 | if (surface && !present) { | ||
| 797 | LOG_ERROR(Render_Vulkan, "Device lacks a present queue"); | ||
| 798 | throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); | ||
| 799 | } | ||
| 800 | graphics_family = *graphics; | ||
| 801 | present_family = *present; | ||
| 802 | } | ||
| 803 | |||
| 804 | void Device::SetupFeatures() { | ||
| 805 | const auto supported_features{physical.GetFeatures()}; | ||
| 806 | is_formatless_image_load_supported = supported_features.shaderStorageImageReadWithoutFormat; | ||
| 807 | is_blit_depth_stencil_supported = TestDepthStencilBlits(); | ||
| 808 | is_optimal_astc_supported = IsOptimalAstcSupported(supported_features); | ||
| 809 | } | ||
| 810 | |||
| 811 | void Device::CollectTelemetryParameters() { | ||
| 812 | VkPhysicalDeviceDriverPropertiesKHR driver{ | ||
| 813 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR, | ||
| 814 | .pNext = nullptr, | ||
| 815 | .driverID = {}, | ||
| 816 | .driverName = {}, | ||
| 817 | .driverInfo = {}, | ||
| 818 | .conformanceVersion = {}, | ||
| 819 | }; | ||
| 820 | |||
| 821 | VkPhysicalDeviceProperties2KHR device_properties{ | ||
| 822 | .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR, | ||
| 823 | .pNext = &driver, | ||
| 824 | .properties = {}, | ||
| 825 | }; | ||
| 826 | physical.GetProperties2KHR(device_properties); | ||
| 827 | |||
| 828 | driver_id = driver.driverID; | ||
| 829 | vendor_name = driver.driverName; | ||
| 830 | |||
| 831 | const std::vector extensions = physical.EnumerateDeviceExtensionProperties(); | ||
| 832 | reported_extensions.reserve(std::size(extensions)); | ||
| 833 | for (const auto& extension : extensions) { | ||
| 834 | reported_extensions.emplace_back(extension.extensionName); | ||
| 835 | } | ||
| 836 | } | ||
| 837 | |||
| 838 | void Device::CollectToolingInfo() { | ||
| 839 | if (!ext_tooling_info) { | ||
| 840 | return; | ||
| 841 | } | ||
| 842 | const auto vkGetPhysicalDeviceToolPropertiesEXT = | ||
| 843 | reinterpret_cast<PFN_vkGetPhysicalDeviceToolPropertiesEXT>( | ||
| 844 | dld.vkGetInstanceProcAddr(instance, "vkGetPhysicalDeviceToolPropertiesEXT")); | ||
| 845 | if (!vkGetPhysicalDeviceToolPropertiesEXT) { | ||
| 846 | return; | ||
| 847 | } | ||
| 848 | u32 tool_count = 0; | ||
| 849 | if (vkGetPhysicalDeviceToolPropertiesEXT(physical, &tool_count, nullptr) != VK_SUCCESS) { | ||
| 850 | return; | ||
| 851 | } | ||
| 852 | std::vector<VkPhysicalDeviceToolPropertiesEXT> tools(tool_count); | ||
| 853 | if (vkGetPhysicalDeviceToolPropertiesEXT(physical, &tool_count, tools.data()) != VK_SUCCESS) { | ||
| 854 | return; | ||
| 855 | } | ||
| 856 | for (const VkPhysicalDeviceToolPropertiesEXT& tool : tools) { | ||
| 857 | const std::string_view name = tool.name; | ||
| 858 | LOG_INFO(Render_Vulkan, "{}", name); | ||
| 859 | has_renderdoc = has_renderdoc || name == "RenderDoc"; | ||
| 860 | has_nsight_graphics = has_nsight_graphics || name == "NVIDIA Nsight Graphics"; | ||
| 861 | } | ||
| 862 | } | ||
| 863 | |||
| 864 | std::vector<VkDeviceQueueCreateInfo> Device::GetDeviceQueueCreateInfos() const { | ||
| 865 | static constexpr float QUEUE_PRIORITY = 1.0f; | ||
| 866 | |||
| 867 | std::unordered_set<u32> unique_queue_families{graphics_family, present_family}; | ||
| 868 | std::vector<VkDeviceQueueCreateInfo> queue_cis; | ||
| 869 | queue_cis.reserve(unique_queue_families.size()); | ||
| 870 | |||
| 871 | for (const u32 queue_family : unique_queue_families) { | ||
| 872 | auto& ci = queue_cis.emplace_back(VkDeviceQueueCreateInfo{ | ||
| 873 | .sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO, | ||
| 874 | .pNext = nullptr, | ||
| 875 | .flags = 0, | ||
| 876 | .queueFamilyIndex = queue_family, | ||
| 877 | .queueCount = 1, | ||
| 878 | .pQueuePriorities = nullptr, | ||
| 879 | }); | ||
| 880 | ci.pQueuePriorities = &QUEUE_PRIORITY; | ||
| 881 | } | ||
| 882 | |||
| 883 | return queue_cis; | ||
| 884 | } | ||
| 885 | |||
| 886 | } // namespace Vulkan | ||
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h new file mode 100644 index 000000000..a973c3ce4 --- /dev/null +++ b/src/video_core/vulkan_common/vulkan_device.h | |||
| @@ -0,0 +1,307 @@ | |||
| 1 | // Copyright 2018 yuzu Emulator Project | ||
| 2 | // Licensed under GPLv2 or any later version | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #pragma once | ||
| 6 | |||
| 7 | #include <string> | ||
| 8 | #include <string_view> | ||
| 9 | #include <unordered_map> | ||
| 10 | #include <vector> | ||
| 11 | |||
| 12 | #include "common/common_types.h" | ||
| 13 | #include "video_core/vulkan_common/vulkan_wrapper.h" | ||
| 14 | |||
| 15 | namespace Vulkan { | ||
| 16 | |||
| 17 | class NsightAftermathTracker; | ||
| 18 | |||
| 19 | /// Format usage descriptor. | ||
| 20 | enum class FormatType { Linear, Optimal, Buffer }; | ||
| 21 | |||
| 22 | /// Subgroup size of the guest emulated hardware (Nvidia has 32 threads per subgroup). | ||
| 23 | const u32 GuestWarpSize = 32; | ||
| 24 | |||
| 25 | /// Handles data specific to a physical device. | ||
| 26 | class Device final { | ||
| 27 | public: | ||
| 28 | explicit Device(VkInstance instance, vk::PhysicalDevice physical, VkSurfaceKHR surface, | ||
| 29 | const vk::InstanceDispatch& dld); | ||
| 30 | ~Device(); | ||
| 31 | |||
| 32 | /** | ||
| 33 | * Returns a format supported by the device for the passed requeriments. | ||
| 34 | * @param wanted_format The ideal format to be returned. It may not be the returned format. | ||
| 35 | * @param wanted_usage The usage that must be fulfilled even if the format is not supported. | ||
| 36 | * @param format_type Format type usage. | ||
| 37 | * @returns A format supported by the device. | ||
| 38 | */ | ||
| 39 | VkFormat GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, | ||
| 40 | FormatType format_type) const; | ||
| 41 | |||
| 42 | /// Reports a device loss. | ||
| 43 | void ReportLoss() const; | ||
| 44 | |||
| 45 | /// Reports a shader to Nsight Aftermath. | ||
| 46 | void SaveShader(const std::vector<u32>& spirv) const; | ||
| 47 | |||
| 48 | /// Returns the dispatch loader with direct function pointers of the device. | ||
| 49 | const vk::DeviceDispatch& GetDispatchLoader() const { | ||
| 50 | return dld; | ||
| 51 | } | ||
| 52 | |||
| 53 | /// Returns the logical device. | ||
| 54 | const vk::Device& GetLogical() const { | ||
| 55 | return logical; | ||
| 56 | } | ||
| 57 | |||
| 58 | /// Returns the physical device. | ||
| 59 | vk::PhysicalDevice GetPhysical() const { | ||
| 60 | return physical; | ||
| 61 | } | ||
| 62 | |||
| 63 | /// Returns the main graphics queue. | ||
| 64 | vk::Queue GetGraphicsQueue() const { | ||
| 65 | return graphics_queue; | ||
| 66 | } | ||
| 67 | |||
| 68 | /// Returns the main present queue. | ||
| 69 | vk::Queue GetPresentQueue() const { | ||
| 70 | return present_queue; | ||
| 71 | } | ||
| 72 | |||
| 73 | /// Returns main graphics queue family index. | ||
| 74 | u32 GetGraphicsFamily() const { | ||
| 75 | return graphics_family; | ||
| 76 | } | ||
| 77 | |||
| 78 | /// Returns main present queue family index. | ||
| 79 | u32 GetPresentFamily() const { | ||
| 80 | return present_family; | ||
| 81 | } | ||
| 82 | |||
| 83 | /// Returns the current Vulkan API version provided in Vulkan-formatted version numbers. | ||
| 84 | u32 ApiVersion() const { | ||
| 85 | return properties.apiVersion; | ||
| 86 | } | ||
| 87 | |||
| 88 | /// Returns the current driver version provided in Vulkan-formatted version numbers. | ||
| 89 | u32 GetDriverVersion() const { | ||
| 90 | return properties.driverVersion; | ||
| 91 | } | ||
| 92 | |||
| 93 | /// Returns the device name. | ||
| 94 | std::string_view GetModelName() const { | ||
| 95 | return properties.deviceName; | ||
| 96 | } | ||
| 97 | |||
| 98 | /// Returns the driver ID. | ||
| 99 | VkDriverIdKHR GetDriverID() const { | ||
| 100 | return driver_id; | ||
| 101 | } | ||
| 102 | |||
| 103 | /// Returns uniform buffer alignment requeriment. | ||
| 104 | VkDeviceSize GetUniformBufferAlignment() const { | ||
| 105 | return properties.limits.minUniformBufferOffsetAlignment; | ||
| 106 | } | ||
| 107 | |||
| 108 | /// Returns storage alignment requeriment. | ||
| 109 | VkDeviceSize GetStorageBufferAlignment() const { | ||
| 110 | return properties.limits.minStorageBufferOffsetAlignment; | ||
| 111 | } | ||
| 112 | |||
| 113 | /// Returns the maximum range for storage buffers. | ||
| 114 | VkDeviceSize GetMaxStorageBufferRange() const { | ||
| 115 | return properties.limits.maxStorageBufferRange; | ||
| 116 | } | ||
| 117 | |||
| 118 | /// Returns the maximum size for push constants. | ||
| 119 | VkDeviceSize GetMaxPushConstantsSize() const { | ||
| 120 | return properties.limits.maxPushConstantsSize; | ||
| 121 | } | ||
| 122 | |||
| 123 | /// Returns the maximum size for shared memory. | ||
| 124 | u32 GetMaxComputeSharedMemorySize() const { | ||
| 125 | return properties.limits.maxComputeSharedMemorySize; | ||
| 126 | } | ||
| 127 | |||
| 128 | /// Returns true if ASTC is natively supported. | ||
| 129 | bool IsOptimalAstcSupported() const { | ||
| 130 | return is_optimal_astc_supported; | ||
| 131 | } | ||
| 132 | |||
| 133 | /// Returns true if the device supports float16 natively | ||
| 134 | bool IsFloat16Supported() const { | ||
| 135 | return is_float16_supported; | ||
| 136 | } | ||
| 137 | |||
| 138 | /// Returns true if the device warp size can potentially be bigger than guest's warp size. | ||
| 139 | bool IsWarpSizePotentiallyBiggerThanGuest() const { | ||
| 140 | return is_warp_potentially_bigger; | ||
| 141 | } | ||
| 142 | |||
| 143 | /// Returns true if the device can be forced to use the guest warp size. | ||
| 144 | bool IsGuestWarpSizeSupported(VkShaderStageFlagBits stage) const { | ||
| 145 | return guest_warp_stages & stage; | ||
| 146 | } | ||
| 147 | |||
| 148 | /// Returns true if formatless image load is supported. | ||
| 149 | bool IsFormatlessImageLoadSupported() const { | ||
| 150 | return is_formatless_image_load_supported; | ||
| 151 | } | ||
| 152 | |||
| 153 | /// Returns true when blitting from and to depth stencil images is supported. | ||
| 154 | bool IsBlitDepthStencilSupported() const { | ||
| 155 | return is_blit_depth_stencil_supported; | ||
| 156 | } | ||
| 157 | |||
| 158 | /// Returns true if the device supports VK_NV_viewport_swizzle. | ||
| 159 | bool IsNvViewportSwizzleSupported() const { | ||
| 160 | return nv_viewport_swizzle; | ||
| 161 | } | ||
| 162 | |||
| 163 | /// Returns true if the device supports VK_EXT_scalar_block_layout. | ||
| 164 | bool IsKhrUniformBufferStandardLayoutSupported() const { | ||
| 165 | return khr_uniform_buffer_standard_layout; | ||
| 166 | } | ||
| 167 | |||
| 168 | /// Returns true if the device supports VK_EXT_index_type_uint8. | ||
| 169 | bool IsExtIndexTypeUint8Supported() const { | ||
| 170 | return ext_index_type_uint8; | ||
| 171 | } | ||
| 172 | |||
| 173 | /// Returns true if the device supports VK_EXT_sampler_filter_minmax. | ||
| 174 | bool IsExtSamplerFilterMinmaxSupported() const { | ||
| 175 | return ext_sampler_filter_minmax; | ||
| 176 | } | ||
| 177 | |||
| 178 | /// Returns true if the device supports VK_EXT_depth_range_unrestricted. | ||
| 179 | bool IsExtDepthRangeUnrestrictedSupported() const { | ||
| 180 | return ext_depth_range_unrestricted; | ||
| 181 | } | ||
| 182 | |||
| 183 | /// Returns true if the device supports VK_EXT_shader_viewport_index_layer. | ||
| 184 | bool IsExtShaderViewportIndexLayerSupported() const { | ||
| 185 | return ext_shader_viewport_index_layer; | ||
| 186 | } | ||
| 187 | |||
| 188 | /// Returns true if the device supports VK_EXT_transform_feedback. | ||
| 189 | bool IsExtTransformFeedbackSupported() const { | ||
| 190 | return ext_transform_feedback; | ||
| 191 | } | ||
| 192 | |||
| 193 | /// Returns true if the device supports VK_EXT_custom_border_color. | ||
| 194 | bool IsExtCustomBorderColorSupported() const { | ||
| 195 | return ext_custom_border_color; | ||
| 196 | } | ||
| 197 | |||
| 198 | /// Returns true if the device supports VK_EXT_extended_dynamic_state. | ||
| 199 | bool IsExtExtendedDynamicStateSupported() const { | ||
| 200 | return ext_extended_dynamic_state; | ||
| 201 | } | ||
| 202 | |||
| 203 | /// Returns true if the device supports VK_EXT_shader_stencil_export. | ||
| 204 | bool IsExtShaderStencilExportSupported() const { | ||
| 205 | return ext_shader_stencil_export; | ||
| 206 | } | ||
| 207 | |||
| 208 | /// Returns true when a known debugging tool is attached. | ||
| 209 | bool HasDebuggingToolAttached() const { | ||
| 210 | return has_renderdoc || has_nsight_graphics; | ||
| 211 | } | ||
| 212 | |||
| 213 | /// Returns the vendor name reported from Vulkan. | ||
| 214 | std::string_view GetVendorName() const { | ||
| 215 | return vendor_name; | ||
| 216 | } | ||
| 217 | |||
| 218 | /// Returns the list of available extensions. | ||
| 219 | const std::vector<std::string>& GetAvailableExtensions() const { | ||
| 220 | return reported_extensions; | ||
| 221 | } | ||
| 222 | |||
| 223 | /// Returns true if the setting for async shader compilation is enabled. | ||
| 224 | bool UseAsynchronousShaders() const { | ||
| 225 | return use_asynchronous_shaders; | ||
| 226 | } | ||
| 227 | |||
| 228 | private: | ||
| 229 | /// Checks if the physical device is suitable. | ||
| 230 | void CheckSuitability() const; | ||
| 231 | |||
| 232 | /// Loads extensions into a vector and stores available ones in this object. | ||
| 233 | std::vector<const char*> LoadExtensions(); | ||
| 234 | |||
| 235 | /// Sets up queue families. | ||
| 236 | void SetupFamilies(VkSurfaceKHR surface); | ||
| 237 | |||
| 238 | /// Sets up device features. | ||
| 239 | void SetupFeatures(); | ||
| 240 | |||
| 241 | /// Collects telemetry information from the device. | ||
| 242 | void CollectTelemetryParameters(); | ||
| 243 | |||
| 244 | /// Collects information about attached tools. | ||
| 245 | void CollectToolingInfo(); | ||
| 246 | |||
| 247 | /// Returns a list of queue initialization descriptors. | ||
| 248 | std::vector<VkDeviceQueueCreateInfo> GetDeviceQueueCreateInfos() const; | ||
| 249 | |||
| 250 | /// Returns true if ASTC textures are natively supported. | ||
| 251 | bool IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const; | ||
| 252 | |||
| 253 | /// Returns true if the device natively supports blitting depth stencil images. | ||
| 254 | bool TestDepthStencilBlits() const; | ||
| 255 | |||
| 256 | /// Returns true if a format is supported. | ||
| 257 | bool IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, | ||
| 258 | FormatType format_type) const; | ||
| 259 | |||
| 260 | VkInstance instance; ///< Vulkan instance. | ||
| 261 | vk::DeviceDispatch dld; ///< Device function pointers. | ||
| 262 | vk::PhysicalDevice physical; ///< Physical device. | ||
| 263 | VkPhysicalDeviceProperties properties; ///< Device properties. | ||
| 264 | vk::Device logical; ///< Logical device. | ||
| 265 | vk::Queue graphics_queue; ///< Main graphics queue. | ||
| 266 | vk::Queue present_queue; ///< Main present queue. | ||
| 267 | u32 instance_version{}; ///< Vulkan onstance version. | ||
| 268 | u32 graphics_family{}; ///< Main graphics queue family index. | ||
| 269 | u32 present_family{}; ///< Main present queue family index. | ||
| 270 | VkDriverIdKHR driver_id{}; ///< Driver ID. | ||
| 271 | VkShaderStageFlags guest_warp_stages{}; ///< Stages where the guest warp size can be forced.ed | ||
| 272 | bool is_optimal_astc_supported{}; ///< Support for native ASTC. | ||
| 273 | bool is_float16_supported{}; ///< Support for float16 arithmetics. | ||
| 274 | bool is_warp_potentially_bigger{}; ///< Host warp size can be bigger than guest. | ||
| 275 | bool is_formatless_image_load_supported{}; ///< Support for shader image read without format. | ||
| 276 | bool is_blit_depth_stencil_supported{}; ///< Support for blitting from and to depth stencil. | ||
| 277 | bool nv_viewport_swizzle{}; ///< Support for VK_NV_viewport_swizzle. | ||
| 278 | bool khr_uniform_buffer_standard_layout{}; ///< Support for std430 on UBOs. | ||
| 279 | bool ext_index_type_uint8{}; ///< Support for VK_EXT_index_type_uint8. | ||
| 280 | bool ext_sampler_filter_minmax{}; ///< Support for VK_EXT_sampler_filter_minmax. | ||
| 281 | bool ext_depth_range_unrestricted{}; ///< Support for VK_EXT_depth_range_unrestricted. | ||
| 282 | bool ext_shader_viewport_index_layer{}; ///< Support for VK_EXT_shader_viewport_index_layer. | ||
| 283 | bool ext_tooling_info{}; ///< Support for VK_EXT_tooling_info. | ||
| 284 | bool ext_transform_feedback{}; ///< Support for VK_EXT_transform_feedback. | ||
| 285 | bool ext_custom_border_color{}; ///< Support for VK_EXT_custom_border_color. | ||
| 286 | bool ext_extended_dynamic_state{}; ///< Support for VK_EXT_extended_dynamic_state. | ||
| 287 | bool ext_robustness2{}; ///< Support for VK_EXT_robustness2. | ||
| 288 | bool ext_shader_stencil_export{}; ///< Support for VK_EXT_shader_stencil_export. | ||
| 289 | bool nv_device_diagnostics_config{}; ///< Support for VK_NV_device_diagnostics_config. | ||
| 290 | bool has_renderdoc{}; ///< Has RenderDoc attached | ||
| 291 | bool has_nsight_graphics{}; ///< Has Nsight Graphics attached | ||
| 292 | |||
| 293 | // Asynchronous Graphics Pipeline setting | ||
| 294 | bool use_asynchronous_shaders{}; ///< Setting to use asynchronous shaders/graphics pipeline | ||
| 295 | |||
| 296 | // Telemetry parameters | ||
| 297 | std::string vendor_name; ///< Device's driver name. | ||
| 298 | std::vector<std::string> reported_extensions; ///< Reported Vulkan extensions. | ||
| 299 | |||
| 300 | /// Format properties dictionary. | ||
| 301 | std::unordered_map<VkFormat, VkFormatProperties> format_properties; | ||
| 302 | |||
| 303 | /// Nsight Aftermath GPU crash tracker | ||
| 304 | std::unique_ptr<NsightAftermathTracker> nsight_aftermath_tracker; | ||
| 305 | }; | ||
| 306 | |||
| 307 | } // namespace Vulkan | ||