diff options
25 files changed, 228 insertions, 227 deletions
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index 478246b6f..1cfd3bbc9 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -64,8 +64,10 @@ if (MSVC) else() add_compile_options( -Wall + -Werror=array-bounds -Werror=implicit-fallthrough -Werror=missing-declarations + -Werror=missing-field-initializers -Werror=reorder -Werror=switch -Werror=uninitialized diff --git a/src/core/file_sys/vfs_real.cpp b/src/core/file_sys/vfs_real.cpp index a287eebe3..a44ce6288 100644 --- a/src/core/file_sys/vfs_real.cpp +++ b/src/core/file_sys/vfs_real.cpp @@ -133,8 +133,11 @@ VirtualFile RealVfsFilesystem::MoveFile(std::string_view old_path_, std::string_ } cache.erase(old_path); - file->Open(new_path, "r+b"); - cache.insert_or_assign(new_path, std::move(file)); + if (file->Open(new_path, "r+b")) { + cache.insert_or_assign(new_path, std::move(file)); + } else { + LOG_ERROR(Service_FS, "Failed to open path {} in order to re-cache it", new_path); + } } else { UNREACHABLE(); return nullptr; @@ -214,9 +217,12 @@ VirtualDir RealVfsFilesystem::MoveDirectory(std::string_view old_path_, } auto file = cached.lock(); - file->Open(file_new_path, "r+b"); cache.erase(file_old_path); - cache.insert_or_assign(std::move(file_new_path), std::move(file)); + if (file->Open(file_new_path, "r+b")) { + cache.insert_or_assign(std::move(file_new_path), std::move(file)); + } else { + LOG_ERROR(Service_FS, "Failed to open path {} in order to re-cache it", file_new_path); + } } return OpenDirectory(new_path, Mode::ReadWrite); diff --git a/src/core/hle/service/am/am.cpp b/src/core/hle/service/am/am.cpp index c9808060a..1743bcb2b 100644 --- a/src/core/hle/service/am/am.cpp +++ b/src/core/hle/service/am/am.cpp @@ -856,7 +856,7 @@ public: {25, nullptr, "Terminate"}, {30, &ILibraryAppletAccessor::GetResult, "GetResult"}, {50, nullptr, "SetOutOfFocusApplicationSuspendingEnabled"}, - {60, nullptr, "PresetLibraryAppletGpuTimeSliceZero"}, + {60, &ILibraryAppletAccessor::PresetLibraryAppletGpuTimeSliceZero, "PresetLibraryAppletGpuTimeSliceZero"}, {100, &ILibraryAppletAccessor::PushInData, "PushInData"}, {101, &ILibraryAppletAccessor::PopOutData, "PopOutData"}, {102, nullptr, "PushExtraStorage"}, @@ -900,6 +900,13 @@ private: rb.Push(applet->GetStatus()); } + void PresetLibraryAppletGpuTimeSliceZero(Kernel::HLERequestContext& ctx) { + LOG_WARNING(Service_AM, "(STUBBED) called"); + + IPC::ResponseBuilder rb{ctx, 2}; + rb.Push(RESULT_SUCCESS); + } + void Start(Kernel::HLERequestContext& ctx) { LOG_DEBUG(Service_AM, "called"); diff --git a/src/input_common/sdl/sdl_impl.cpp b/src/input_common/sdl/sdl_impl.cpp index d32eb732a..1b5750937 100644 --- a/src/input_common/sdl/sdl_impl.cpp +++ b/src/input_common/sdl/sdl_impl.cpp @@ -81,10 +81,14 @@ public: } bool RumblePlay(u16 amp_low, u16 amp_high) { + constexpr u32 rumble_max_duration_ms = 1000; + if (sdl_controller) { - return SDL_GameControllerRumble(sdl_controller.get(), amp_low, amp_high, 0) == 0; + return SDL_GameControllerRumble(sdl_controller.get(), amp_low, amp_high, + rumble_max_duration_ms) == 0; } else if (sdl_joystick) { - return SDL_JoystickRumble(sdl_joystick.get(), amp_low, amp_high, 0) == 0; + return SDL_JoystickRumble(sdl_joystick.get(), amp_low, amp_high, + rumble_max_duration_ms) == 0; } return false; diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index e01ea55ab..bb1f8491f 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -288,10 +288,10 @@ target_link_libraries(video_core PRIVATE sirit) if (ENABLE_NSIGHT_AFTERMATH) if (NOT DEFINED ENV{NSIGHT_AFTERMATH_SDK}) - message(ERROR "Environment variable NSIGHT_AFTERMATH_SDK has to be provided") + message(FATAL_ERROR "Environment variable NSIGHT_AFTERMATH_SDK has to be provided") endif() if (NOT WIN32) - message(ERROR "Nsight Aftermath doesn't support non-Windows platforms") + message(FATAL_ERROR "Nsight Aftermath doesn't support non-Windows platforms") endif() target_compile_definitions(video_core PRIVATE HAS_NSIGHT_AFTERMATH) target_include_directories(video_core PRIVATE "$ENV{NSIGHT_AFTERMATH_SDK}/include") diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 9be651e24..116ad1722 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -179,22 +179,22 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume return ProcessMacroBind(argument); case MAXWELL3D_REG_INDEX(firmware[4]): return ProcessFirmwareCall4(); - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 1: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 2: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 3: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 4: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 5: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 6: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 7: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 8: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 9: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 10: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 11: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 12: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15: return StartCBData(method); case MAXWELL3D_REG_INDEX(cb_bind[0]): return ProcessCBBind(0); @@ -287,22 +287,22 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, return; } switch (method) { - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]): - case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data): + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 1: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 2: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 3: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 4: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 5: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 6: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 7: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 8: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 9: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 10: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 11: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 12: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 13: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 14: + case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15: ProcessCBMultiData(method, base_start, amount); break; default: @@ -592,7 +592,7 @@ void Maxwell3D::ProcessCBData(u32 value) { } void Maxwell3D::StartCBData(u32 method) { - constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]); + constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data); cb_data_state.start_pos = regs.const_buffer.cb_pos; cb_data_state.id = method - first_cb_data; cb_data_state.current = method; @@ -605,7 +605,7 @@ void Maxwell3D::ProcessCBMultiData(u32 method, const u32* start_base, u32 amount if (cb_data_state.current != null_cb_data) { FinishCBData(); } - constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]); + constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data); cb_data_state.start_pos = regs.const_buffer.cb_pos; cb_data_state.id = method - first_cb_data; cb_data_state.current = method; diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 326b32228..002d1b3f9 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -1337,7 +1337,7 @@ public: u32 cb_address_high; u32 cb_address_low; u32 cb_pos; - u32 cb_data[NumCBData]; + std::array<u32, NumCBData> cb_data; GPUVAddr BufferAddress() const { return static_cast<GPUVAddr>( diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index 4c7399d5a..73f331d4c 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -55,7 +55,7 @@ foreach(FILENAME IN ITEMS ${SHADER_FILES}) OUTPUT ${SPIRV_HEADER_FILE} COMMAND - ${GLSLANGVALIDATOR} -V ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE} + ${GLSLANGVALIDATOR} -V --quiet ${GLSL_FLAGS} --variable-name ${SPIRV_VARIABLE_NAME} -o ${SPIRV_HEADER_FILE} ${SOURCE_FILE} MAIN_DEPENDENCY ${SOURCE_FILE} ) diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index 65feff588..f5cdf548e 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -314,17 +314,29 @@ void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buf } } +void MemoryManager::FlushRegion(GPUVAddr gpu_addr, size_t size) const { + size_t remaining_size{size}; + size_t page_index{gpu_addr >> page_bits}; + size_t page_offset{gpu_addr & page_mask}; + while (remaining_size > 0) { + const size_t num_bytes{std::min(page_size - page_offset, remaining_size)}; + if (const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; page_addr) { + rasterizer->FlushRegion(*page_addr + page_offset, num_bytes); + } + ++page_index; + page_offset = 0; + remaining_size -= num_bytes; + } +} + void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size) { std::vector<u8> tmp_buffer(size); ReadBlock(gpu_src_addr, tmp_buffer.data(), size); - WriteBlock(gpu_dest_addr, tmp_buffer.data(), size); -} -void MemoryManager::CopyBlockUnsafe(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, - std::size_t size) { - std::vector<u8> tmp_buffer(size); - ReadBlockUnsafe(gpu_src_addr, tmp_buffer.data(), size); - WriteBlockUnsafe(gpu_dest_addr, tmp_buffer.data(), size); + // The output block must be flushed in case it has data modified from the GPU. + // Fixes NPC geometry in Zombie Panic in Wonderland DX + FlushRegion(gpu_dest_addr, size); + WriteBlock(gpu_dest_addr, tmp_buffer.data(), size); } bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const { diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index c35e57689..a52fbbd8c 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -107,7 +107,6 @@ public: */ void ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const; void WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size); - void CopyBlockUnsafe(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size); /** * IsGranularRange checks if a gpu region can be simply read with a pointer. @@ -131,6 +130,8 @@ private: void TryLockPage(PageEntry page_entry, std::size_t size); void TryUnlockPage(PageEntry page_entry, std::size_t size); + void FlushRegion(GPUVAddr gpu_addr, size_t size) const; + [[nodiscard]] static constexpr std::size_t PageEntryIndex(GPUVAddr gpu_addr) { return (gpu_addr >> page_bits) & page_table_mask; } diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index ca7c2c579..85121d9fd 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -110,8 +110,8 @@ VkCompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_ } // namespace Sampler namespace { - -enum : u32 { Attachable = 1, Storage = 2 }; +constexpr u32 Attachable = 1 << 0; +constexpr u32 Storage = 1 << 1; struct FormatTuple { VkFormat format; ///< Vulkan format @@ -222,22 +222,27 @@ constexpr bool IsZetaFormat(PixelFormat pixel_format) { } // Anonymous namespace -FormatInfo SurfaceFormat(const Device& device, FormatType format_type, PixelFormat pixel_format) { - ASSERT(static_cast<std::size_t>(pixel_format) < std::size(tex_format_tuples)); - - auto tuple = tex_format_tuples[static_cast<std::size_t>(pixel_format)]; +FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with_srgb, + PixelFormat pixel_format) { + ASSERT(static_cast<size_t>(pixel_format) < std::size(tex_format_tuples)); + FormatTuple tuple = tex_format_tuples[static_cast<size_t>(pixel_format)]; if (tuple.format == VK_FORMAT_UNDEFINED) { UNIMPLEMENTED_MSG("Unimplemented texture format with pixel format={}", pixel_format); - return {VK_FORMAT_A8B8G8R8_UNORM_PACK32, true, true}; + return FormatInfo{VK_FORMAT_A8B8G8R8_UNORM_PACK32, true, true}; } // Use A8B8G8R8_UNORM on hardware that doesn't support ASTC natively if (!device.IsOptimalAstcSupported() && VideoCore::Surface::IsPixelFormatASTC(pixel_format)) { - const bool is_srgb = VideoCore::Surface::IsPixelFormatSRGB(pixel_format); - tuple.format = is_srgb ? VK_FORMAT_A8B8G8R8_SRGB_PACK32 : VK_FORMAT_A8B8G8R8_UNORM_PACK32; + const bool is_srgb = with_srgb && VideoCore::Surface::IsPixelFormatSRGB(pixel_format); + if (is_srgb) { + tuple.format = VK_FORMAT_A8B8G8R8_SRGB_PACK32; + } else { + tuple.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32; + tuple.usage |= Storage; + } } - const bool attachable = tuple.usage & Attachable; - const bool storage = tuple.usage & Storage; + const bool attachable = (tuple.usage & Attachable) != 0; + const bool storage = (tuple.usage & Storage) != 0; VkFormatFeatureFlags usage{}; switch (format_type) { @@ -671,7 +676,7 @@ VkFrontFace FrontFace(Maxwell::FrontFace front_face) { return {}; } -VkCullModeFlags CullFace(Maxwell::CullFace cull_face) { +VkCullModeFlagBits CullFace(Maxwell::CullFace cull_face) { switch (cull_face) { case Maxwell::CullFace::Front: return VK_CULL_MODE_FRONT_BIT; diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.h b/src/video_core/renderer_vulkan/maxwell_to_vk.h index 537969840..7c34b47dc 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.h +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.h @@ -35,7 +35,15 @@ struct FormatInfo { bool storage; }; -FormatInfo SurfaceFormat(const Device& device, FormatType format_type, PixelFormat pixel_format); +/** + * Returns format properties supported in the host + * @param device Host device + * @param format_type Type of image the buffer will use + * @param with_srgb True when the format can be sRGB when converted to another format (ASTC) + * @param pixel_format Guest pixel format to describe + */ +[[nodiscard]] FormatInfo SurfaceFormat(const Device& device, FormatType format_type, bool with_srgb, + PixelFormat pixel_format); VkShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage); @@ -55,7 +63,7 @@ VkBlendFactor BlendFactor(Maxwell::Blend::Factor factor); VkFrontFace FrontFace(Maxwell::FrontFace front_face); -VkCullModeFlags CullFace(Maxwell::CullFace cull_face); +VkCullModeFlagBits CullFace(Maxwell::CullFace cull_face); VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle); diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index a5214d0bc..d50dca604 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -181,6 +181,7 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules( .pNext = nullptr, .flags = 0, .codeSize = 0, + .pCode = nullptr, }; std::vector<vk::ShaderModule> shader_modules; @@ -326,8 +327,8 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, .rasterizerDiscardEnable = static_cast<VkBool32>(state.rasterize_enable == 0 ? VK_TRUE : VK_FALSE), .polygonMode = VK_POLYGON_MODE_FILL, - .cullMode = - dynamic.cull_enable ? MaxwellToVK::CullFace(dynamic.CullFace()) : VK_CULL_MODE_NONE, + .cullMode = static_cast<VkCullModeFlags>( + dynamic.cull_enable ? MaxwellToVK::CullFace(dynamic.CullFace()) : VK_CULL_MODE_NONE), .frontFace = MaxwellToVK::FrontFace(dynamic.FrontFace()), .depthBiasEnable = state.depth_bias_enable, .depthBiasConstantFactor = 0.0f, diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 02282e36f..8991505ca 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -355,14 +355,12 @@ VKPipelineCache::DecompileShaders(const FixedPipelineState& fixed_state) { SPIRVProgram program; std::vector<VkDescriptorSetLayoutBinding> bindings; - for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { + for (std::size_t index = 1; index < Maxwell::MaxShaderProgram; ++index) { const auto program_enum = static_cast<Maxwell::ShaderProgram>(index); - // Skip stages that are not enabled if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { continue; } - const GPUVAddr gpu_addr = GetShaderAddress(maxwell3d, program_enum); const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); Shader* const shader = cpu_addr ? TryGet(*cpu_addr) : null_shader.get(); @@ -372,12 +370,8 @@ VKPipelineCache::DecompileShaders(const FixedPipelineState& fixed_state) { const auto& entries = shader->GetEntries(); program[stage] = { Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization), - entries}; - - if (program_enum == Maxwell::ShaderProgram::VertexA) { - // VertexB was combined with VertexA, so we skip the VertexB iteration - ++index; - } + entries, + }; const u32 old_binding = specialization.base_binding; specialization.base_binding = diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index ab14922d7..aa7c5d7c6 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -95,20 +95,12 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { } } -[[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info) { - const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, info.format); - VkImageCreateFlags flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; - if (info.type == ImageType::e2D && info.resources.layers >= 6 && - info.size.width == info.size.height) { - flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT; - } - if (info.type == ImageType::e3D) { - flags |= VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT; - } +[[nodiscard]] VkImageUsageFlags ImageUsageFlags(const MaxwellToVK::FormatInfo& info, + PixelFormat format) { VkImageUsageFlags usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT; - if (format_info.attachable) { - switch (VideoCore::Surface::GetFormatType(info.format)) { + if (info.attachable) { + switch (VideoCore::Surface::GetFormatType(format)) { case VideoCore::Surface::SurfaceType::ColorTexture: usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; break; @@ -120,9 +112,33 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { UNREACHABLE_MSG("Invalid surface type"); } } - if (format_info.storage) { + if (info.storage) { usage |= VK_IMAGE_USAGE_STORAGE_BIT; } + return usage; +} + +/// Returns the preferred format for a VkImage +[[nodiscard]] PixelFormat StorageFormat(PixelFormat format) { + switch (format) { + case PixelFormat::A8B8G8R8_SRGB: + return PixelFormat::A8B8G8R8_UNORM; + default: + return format; + } +} + +[[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info) { + const PixelFormat format = StorageFormat(info.format); + const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format); + VkImageCreateFlags flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT; + if (info.type == ImageType::e2D && info.resources.layers >= 6 && + info.size.width == info.size.height) { + flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT; + } + if (info.type == ImageType::e3D) { + flags |= VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT; + } const auto [samples_x, samples_y] = VideoCommon::SamplesLog2(info.num_samples); return VkImageCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, @@ -130,17 +146,16 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { .flags = flags, .imageType = ConvertImageType(info.type), .format = format_info.format, - .extent = - { - .width = info.size.width >> samples_x, - .height = info.size.height >> samples_y, - .depth = info.size.depth, - }, + .extent{ + .width = info.size.width >> samples_x, + .height = info.size.height >> samples_y, + .depth = info.size.depth, + }, .mipLevels = static_cast<u32>(info.resources.levels), .arrayLayers = static_cast<u32>(info.resources.layers), .samples = ConvertSampleCount(info.num_samples), .tiling = VK_IMAGE_TILING_OPTIMAL, - .usage = usage, + .usage = ImageUsageFlags(format_info, format), .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, @@ -209,10 +224,11 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { [[nodiscard]] VkAttachmentDescription AttachmentDescription(const Device& device, const ImageView* image_view) { - const auto pixel_format = image_view->format; + using MaxwellToVK::SurfaceFormat; + const PixelFormat pixel_format = image_view->format; return VkAttachmentDescription{ .flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT, - .format = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, pixel_format).format, + .format = SurfaceFormat(device, FormatType::Optimal, true, pixel_format).format, .samples = image_view->Samples(), .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD, .storeOp = VK_ATTACHMENT_STORE_OP_STORE, @@ -868,11 +884,16 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI std::ranges::transform(swizzle, swizzle.begin(), ConvertGreenRed); } } - const VkFormat vk_format = - MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, format).format; + const auto format_info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format); + const VkFormat vk_format = format_info.format; + const VkImageViewUsageCreateInfo image_view_usage{ + .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, + .pNext = nullptr, + .usage = ImageUsageFlags(format_info, format), + }; const VkImageViewCreateInfo create_info{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, - .pNext = nullptr, + .pNext = &image_view_usage, .flags = 0, .image = image.Handle(), .viewType = VkImageViewType{}, @@ -962,7 +983,7 @@ vk::ImageView ImageView::MakeDepthStencilView(VkImageAspectFlags aspect_mask) { .flags = 0, .image = image_handle, .viewType = ImageViewType(type), - .format = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, format).format, + .format = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format).format, .components{ .r = VK_COMPONENT_SWIZZLE_IDENTITY, .g = VK_COMPONENT_SWIZZLE_IDENTITY, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index a55d405d1..8d29361a1 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -72,7 +72,7 @@ struct TextureCacheRuntime { MemoryAllocator& memory_allocator; StagingBufferPool& staging_buffer_pool; BlitImageHelper& blit_image_helper; - std::unordered_map<RenderPassKey, vk::RenderPass> renderpass_cache; + std::unordered_map<RenderPassKey, vk::RenderPass> renderpass_cache{}; void Finish(); diff --git a/src/video_core/shader/async_shaders.cpp b/src/video_core/shader/async_shaders.cpp index 9707136e9..3b40db9bc 100644 --- a/src/video_core/shader/async_shaders.cpp +++ b/src/video_core/shader/async_shaders.cpp @@ -129,6 +129,15 @@ void AsyncShaders::QueueOpenGLShader(const OpenGL::Device& device, .compiler_settings = compiler_settings, .registry = registry, .cpu_address = cpu_addr, + .pp_cache = nullptr, + .vk_device = nullptr, + .scheduler = nullptr, + .descriptor_pool = nullptr, + .update_descriptor_queue = nullptr, + .bindings{}, + .program{}, + .key{}, + .num_color_buffers = 0, }); cv.notify_one(); } @@ -143,6 +152,15 @@ void AsyncShaders::QueueVulkanShader(Vulkan::VKPipelineCache* pp_cache, std::unique_lock lock(queue_mutex); pending_queue.push({ .backend = Backend::Vulkan, + .device = nullptr, + .shader_type{}, + .uid = 0, + .code{}, + .code_b{}, + .main_offset = 0, + .compiler_settings{}, + .registry{}, + .cpu_address = 0, .pp_cache = pp_cache, .vk_device = &device, .scheduler = &scheduler, diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h index c9840b75e..b54d33763 100644 --- a/src/video_core/shader/node.h +++ b/src/video_core/shader/node.h @@ -465,14 +465,6 @@ public: return operands.size(); } - NodeBlock& GetOperands() { - return operands; - } - - const NodeBlock& GetOperands() const { - return operands; - } - [[nodiscard]] const Node& operator[](std::size_t operand_index) const { return operands.at(operand_index); } diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp index caf5ff362..a4987ffc6 100644 --- a/src/video_core/shader/shader_ir.cpp +++ b/src/video_core/shader/shader_ir.cpp @@ -388,54 +388,9 @@ void ShaderIR::SetInternalFlagsFromInteger(NodeBlock& bb, Node value, bool sets_ if (!sets_cc) { return; } - switch (value->index()) { - case 0: // Operation Node - SearchOperands(bb, value); - break; - case 2: // Genral Purpose Node - if (const auto* gpr = std::get_if<GprNode>(value.get())) { - LOG_DEBUG(HW_GPU, "GprNode: index={}", gpr->GetIndex()); - Node zerop = Operation(OperationCode::LogicalIEqual, std::move(value), - Immediate(gpr->GetIndex())); - SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop)); - } - break; - - default: - Node zerop = Operation(OperationCode::LogicalIEqual, std::move(value), Immediate(0)); - SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop)); - LOG_WARNING(HW_GPU, "Node Type: {}", value->index()); - break; - } -} - -void ShaderIR::SearchOperands(NodeBlock& nb, Node var) { - const auto* op = std::get_if<OperationNode>(var.get()); - if (op == nullptr) { - return; - } - - if (op->GetOperandsCount() == 0) { - return; - } - - for (auto& operand : op->GetOperands()) { - switch (operand->index()) { - case 0: // Operation Node - return SearchOperands(nb, operand); - case 2: // General Purpose Node - if (const auto* gpr = std::get_if<GprNode>(operand.get())) { - LOG_DEBUG(HW_GPU, "Child GprNode: index={}", gpr->GetIndex()); - Node zerop = Operation(OperationCode::LogicalIEqual, std::move(operand), - Immediate(gpr->GetIndex())); - SetInternalFlag(nb, InternalFlag::Zero, std::move(zerop)); - } - break; - default: - LOG_WARNING(HW_GPU, "Child Node Type: {}", operand->index()); - break; - } - } + Node zerop = Operation(OperationCode::LogicalIEqual, std::move(value), Immediate(0)); + SetInternalFlag(bb, InternalFlag::Zero, std::move(zerop)); + LOG_WARNING(HW_GPU, "Condition codes implementation is incomplete"); } Node ShaderIR::BitfieldExtract(Node value, u32 offset, u32 bits) { diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index 0afa39531..0c6ab0f07 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -346,9 +346,6 @@ private: /// Access a bindless image sampler. ImageEntry& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type); - /// Recursive Iteration over the OperationNode operands, searching for GprNodes. - void SearchOperands(NodeBlock& nb, Node var); - /// Extracts a sequence of bits from a node Node BitfieldExtract(Node value, u32 offset, u32 bits); diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index b23424523..bb2cdef81 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -1193,25 +1193,35 @@ u32 MapSizeBytes(const ImageBase& image) { } } -using P = PixelFormat; - -static_assert(CalculateLevelSize(LevelInfo{{1920, 1080}, {0, 2, 0}, {1, 1}, 2, 0}, 0) == 0x7f8000); -static_assert(CalculateLevelSize(LevelInfo{{32, 32}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x4000); - -static_assert(CalculateLevelOffset(P::R8_SINT, {1920, 1080}, {0, 2}, 1, 0, 7) == 0x2afc00); -static_assert(CalculateLevelOffset(P::ASTC_2D_12X12_UNORM, {8192, 4096}, {0, 2}, 1, 0, 12) == - 0x50d200); - -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 0) == 0); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 1) == 0x400000); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 2) == 0x500000); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 3) == 0x540000); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 4) == 0x550000); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 5) == 0x554000); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 6) == 0x555000); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 7) == 0x555400); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 8) == 0x555600); -static_assert(CalculateLevelOffset(P::A8B8G8R8_UNORM, {1024, 1024}, {0, 4}, 1, 0, 9) == 0x555800); +static_assert(CalculateLevelSize(LevelInfo{{1920, 1080, 1}, {0, 2, 0}, {1, 1}, 2, 0}, 0) == + 0x7f8000); +static_assert(CalculateLevelSize(LevelInfo{{32, 32, 1}, {0, 0, 4}, {1, 1}, 4, 0}, 0) == 0x4000); + +static_assert(CalculateLevelOffset(PixelFormat::R8_SINT, {1920, 1080, 1}, {0, 2, 0}, 1, 0, 7) == + 0x2afc00); +static_assert(CalculateLevelOffset(PixelFormat::ASTC_2D_12X12_UNORM, {8192, 4096, 1}, {0, 2, 0}, 1, + 0, 12) == 0x50d200); + +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 0) == 0); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 1) == 0x400000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 2) == 0x500000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 3) == 0x540000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 4) == 0x550000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 5) == 0x554000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 6) == 0x555000); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 7) == 0x555400); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 8) == 0x555600); +static_assert(CalculateLevelOffset(PixelFormat::A8B8G8R8_UNORM, {1024, 1024, 1}, {0, 4, 0}, 1, 0, + 9) == 0x555800); constexpr u32 ValidateLayerSize(PixelFormat format, u32 width, u32 height, u32 block_height, u32 tile_width_spacing, u32 level) { @@ -1221,13 +1231,14 @@ constexpr u32 ValidateLayerSize(PixelFormat format, u32 width, u32 height, u32 b return AlignLayerSize(offset, size, block, DefaultBlockHeight(format), tile_width_spacing); } -static_assert(ValidateLayerSize(P::ASTC_2D_12X12_UNORM, 8192, 4096, 2, 0, 12) == 0x50d800); -static_assert(ValidateLayerSize(P::A8B8G8R8_UNORM, 1024, 1024, 2, 0, 10) == 0x556000); -static_assert(ValidateLayerSize(P::BC3_UNORM, 128, 128, 2, 0, 8) == 0x6000); +static_assert(ValidateLayerSize(PixelFormat::ASTC_2D_12X12_UNORM, 8192, 4096, 2, 0, 12) == + 0x50d800); +static_assert(ValidateLayerSize(PixelFormat::A8B8G8R8_UNORM, 1024, 1024, 2, 0, 10) == 0x556000); +static_assert(ValidateLayerSize(PixelFormat::BC3_UNORM, 128, 128, 2, 0, 8) == 0x6000); -static_assert(ValidateLayerSize(P::A8B8G8R8_UNORM, 518, 572, 4, 3, 1) == 0x190000, +static_assert(ValidateLayerSize(PixelFormat::A8B8G8R8_UNORM, 518, 572, 4, 3, 1) == 0x190000, "Tile width spacing is not working"); -static_assert(ValidateLayerSize(P::BC5_UNORM, 1024, 1024, 3, 4, 11) == 0x160000, +static_assert(ValidateLayerSize(PixelFormat::BC5_UNORM, 1024, 1024, 3, 4, 11) == 0x160000, "Compressed tile width spacing is not working"); } // namespace VideoCommon diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp index 8d10ac29e..7a9d00d4f 100644 --- a/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.cpp @@ -12,21 +12,12 @@ #include <fmt/format.h> -#define VK_NO_PROTOTYPES -#include <vulkan/vulkan.h> - -#include <GFSDK_Aftermath.h> -#include <GFSDK_Aftermath_Defines.h> -#include <GFSDK_Aftermath_GpuCrashDump.h> -#include <GFSDK_Aftermath_GpuCrashDumpDecoding.h> - #include "common/common_paths.h" #include "common/common_types.h" #include "common/file_util.h" #include "common/logging/log.h" #include "common/scope_exit.h" - -#include "video_core/renderer_vulkan/nsight_aftermath_tracker.h" +#include "video_core/vulkan_common/nsight_aftermath_tracker.h" namespace Vulkan { @@ -53,7 +44,7 @@ NsightAftermathTracker::NsightAftermathTracker() { !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GetJSON", &GFSDK_Aftermath_GpuCrashDump_GetJSON)) { LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath function pointers"); - return false; + return; } dump_dir = Common::FS::GetUserPath(Common::FS::UserPath::LogDir) + "gpucrash"; diff --git a/src/video_core/vulkan_common/nsight_aftermath_tracker.h b/src/video_core/vulkan_common/nsight_aftermath_tracker.h index cee3847fb..1ce8d4e8e 100644 --- a/src/video_core/vulkan_common/nsight_aftermath_tracker.h +++ b/src/video_core/vulkan_common/nsight_aftermath_tracker.h @@ -8,8 +8,9 @@ #include <string> #include <vector> -#define VK_NO_PROTOTYPES -#include <vulkan/vulkan.h> +#include "common/common_types.h" +#include "common/dynamic_library.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" #ifdef HAS_NSIGHT_AFTERMATH #include <GFSDK_Aftermath_Defines.h> @@ -17,9 +18,6 @@ #include <GFSDK_Aftermath_GpuCrashDumpDecoding.h> #endif -#include "common/common_types.h" -#include "common/dynamic_library.h" - namespace Vulkan { class NsightAftermathTracker { diff --git a/src/video_core/vulkan_common/vulkan_debug_callback.cpp b/src/video_core/vulkan_common/vulkan_debug_callback.cpp index ea7af8ad4..5c64c9bf7 100644 --- a/src/video_core/vulkan_common/vulkan_debug_callback.cpp +++ b/src/video_core/vulkan_common/vulkan_debug_callback.cpp @@ -39,6 +39,7 @@ vk::DebugUtilsMessenger CreateDebugCallback(const vk::Instance& instance) { VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT | VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT, .pfnUserCallback = Callback, + .pUserData = nullptr, }); } diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 37d7b45a3..5b4209c72 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -84,21 +84,6 @@ VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType } } -[[nodiscard]] bool IsRDNA(std::string_view device_name, VkDriverIdKHR driver_id) { - static constexpr std::array RDNA_DEVICES{ - "5700", - "5600", - "5500", - "5300", - }; - if (driver_id != VK_DRIVER_ID_AMD_PROPRIETARY_KHR) { - return false; - } - return std::any_of(RDNA_DEVICES.begin(), RDNA_DEVICES.end(), [device_name](const char* name) { - return device_name.find(name) != std::string_view::npos; - }); -} - std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(vk::PhysicalDevice physical) { static constexpr std::array formats{ VK_FORMAT_A8B8G8R8_UNORM_PACK32, @@ -436,14 +421,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR "Blacklisting RADV for VK_EXT_extended_dynamic state, likely due to a bug in yuzu"); ext_extended_dynamic_state = false; } - if (ext_extended_dynamic_state && IsRDNA(properties.deviceName, driver_id)) { - // AMD's proprietary driver supports VK_EXT_extended_dynamic_state but on RDNA devices it - // seems to cause stability issues - LOG_WARNING( - Render_Vulkan, - "Blacklisting AMD proprietary on RDNA devices from VK_EXT_extended_dynamic_state"); - ext_extended_dynamic_state = false; - } graphics_queue = logical.GetQueue(graphics_family); present_queue = logical.GetQueue(present_family); |