diff options
23 files changed, 350 insertions, 58 deletions
diff --git a/src/android/app/src/main/jni/native.cpp b/src/android/app/src/main/jni/native.cpp index 9fa082dd5..9cf71680c 100644 --- a/src/android/app/src/main/jni/native.cpp +++ b/src/android/app/src/main/jni/native.cpp @@ -141,10 +141,6 @@ public: if (nsp->IsExtractedType()) { return InstallError; } - } else if (file_extension == "xci") { - jconst xci = - std::make_shared<FileSys::XCI>(m_vfs->OpenFile(filename, FileSys::Mode::Read)); - nsp = xci->GetSecurePartitionNSP(); } else { return ErrorFilenameExtension; } diff --git a/src/common/settings.h b/src/common/settings.h index ae5e5d2b8..98ab0ec2e 100644 --- a/src/common/settings.h +++ b/src/common/settings.h @@ -351,6 +351,8 @@ struct Values { linkage, false, "disable_shader_loop_safety_checks", Category::RendererDebug}; Setting<bool> enable_renderdoc_hotkey{linkage, false, "renderdoc_hotkey", Category::RendererDebug}; + // TODO: remove this once AMDVLK supports VK_EXT_depth_bias_control + bool renderer_amdvlk_depth_bias_workaround{}; // System SwitchableSetting<Language, true> language_index{linkage, diff --git a/src/core/core.cpp b/src/core/core.cpp index e8300cd05..08cbb8978 100644 --- a/src/core/core.cpp +++ b/src/core/core.cpp @@ -381,6 +381,10 @@ struct System::Impl { room_member->SendGameInfo(game_info); } + // Workarounds: + // Activate this in Super Smash Brothers Ultimate, it only affects AMD cards using AMDVLK + Settings::values.renderer_amdvlk_depth_bias_workaround = program_id == 0x1006A800016E000ULL; + status = SystemResultStatus::Success; return status; } @@ -440,6 +444,9 @@ struct System::Impl { room_member->SendGameInfo(game_info); } + // Workarounds + Settings::values.renderer_amdvlk_depth_bias_workaround = false; + LOG_DEBUG(Core, "Shutdown OK"); } diff --git a/src/core/hle/service/am/am.cpp b/src/core/hle/service/am/am.cpp index e9bd04842..a83622f7c 100644 --- a/src/core/hle/service/am/am.cpp +++ b/src/core/hle/service/am/am.cpp @@ -2094,9 +2094,6 @@ void IApplicationFunctions::PrepareForJit(HLERequestContext& ctx) { void LoopProcess(Nvnflinger::Nvnflinger& nvnflinger, Core::System& system) { auto message_queue = std::make_shared<AppletMessageQueue>(system); - // Needed on game boot - message_queue->PushMessage(AppletMessageQueue::AppletMessage::FocusStateChanged); - auto server_manager = std::make_unique<ServerManager>(system); server_manager->RegisterNamedService( diff --git a/src/core/hle/service/am/applets/applet_mii_edit.cpp b/src/core/hle/service/am/applets/applet_mii_edit.cpp index ff77830d2..50adc7c02 100644 --- a/src/core/hle/service/am/applets/applet_mii_edit.cpp +++ b/src/core/hle/service/am/applets/applet_mii_edit.cpp @@ -145,6 +145,8 @@ void MiiEdit::MiiEditOutput(MiiEditResult result, s32 index) { .index{index}, }; + LOG_INFO(Input, "called, result={}, index={}", result, index); + std::vector<u8> out_data(sizeof(MiiEditAppletOutput)); std::memcpy(out_data.data(), &applet_output, sizeof(MiiEditAppletOutput)); diff --git a/src/core/hle/service/filesystem/fsp_srv.cpp b/src/core/hle/service/filesystem/fsp_srv.cpp index 2492bbc16..c2054e8a0 100644 --- a/src/core/hle/service/filesystem/fsp_srv.cpp +++ b/src/core/hle/service/filesystem/fsp_srv.cpp @@ -329,6 +329,7 @@ public: {13, &IFileSystem::CleanDirectoryRecursively, "CleanDirectoryRecursively"}, {14, &IFileSystem::GetFileTimeStampRaw, "GetFileTimeStampRaw"}, {15, nullptr, "QueryEntry"}, + {16, &IFileSystem::GetFileSystemAttribute, "GetFileSystemAttribute"}, }; RegisterHandlers(functions); } @@ -521,6 +522,46 @@ public: rb.PushRaw(vfs_timestamp); } + void GetFileSystemAttribute(HLERequestContext& ctx) { + LOG_WARNING(Service_FS, "(STUBBED) called"); + + struct FileSystemAttribute { + u8 dir_entry_name_length_max_defined; + u8 file_entry_name_length_max_defined; + u8 dir_path_name_length_max_defined; + u8 file_path_name_length_max_defined; + INSERT_PADDING_BYTES_NOINIT(0x5); + u8 utf16_dir_entry_name_length_max_defined; + u8 utf16_file_entry_name_length_max_defined; + u8 utf16_dir_path_name_length_max_defined; + u8 utf16_file_path_name_length_max_defined; + INSERT_PADDING_BYTES_NOINIT(0x18); + s32 dir_entry_name_length_max; + s32 file_entry_name_length_max; + s32 dir_path_name_length_max; + s32 file_path_name_length_max; + INSERT_PADDING_WORDS_NOINIT(0x5); + s32 utf16_dir_entry_name_length_max; + s32 utf16_file_entry_name_length_max; + s32 utf16_dir_path_name_length_max; + s32 utf16_file_path_name_length_max; + INSERT_PADDING_WORDS_NOINIT(0x18); + INSERT_PADDING_WORDS_NOINIT(0x1); + }; + static_assert(sizeof(FileSystemAttribute) == 0xc0, + "FileSystemAttribute has incorrect size"); + + FileSystemAttribute savedata_attribute{}; + savedata_attribute.dir_entry_name_length_max_defined = true; + savedata_attribute.file_entry_name_length_max_defined = true; + savedata_attribute.dir_entry_name_length_max = 0x40; + savedata_attribute.file_entry_name_length_max = 0x40; + + IPC::ResponseBuilder rb{ctx, 50}; + rb.Push(ResultSuccess); + rb.PushRaw(savedata_attribute); + } + private: VfsDirectoryServiceWrapper backend; SizeGetter size; diff --git a/src/core/hle/service/mii/mii_database_manager.cpp b/src/core/hle/service/mii/mii_database_manager.cpp index c39898594..0080b6705 100644 --- a/src/core/hle/service/mii/mii_database_manager.cpp +++ b/src/core/hle/service/mii/mii_database_manager.cpp @@ -168,7 +168,7 @@ Result DatabaseManager::FindIndex(s32& out_index, const Common::UUID& create_id, return ResultSuccess; } - for (std::size_t i = 0; i <= index; ++i) { + for (std::size_t i = 0; i < index; ++i) { if (database.Get(i).IsSpecial()) { continue; } diff --git a/src/core/hle/service/mii/types/char_info.cpp b/src/core/hle/service/mii/types/char_info.cpp index c82186c73..e90124af4 100644 --- a/src/core/hle/service/mii/types/char_info.cpp +++ b/src/core/hle/service/mii/types/char_info.cpp @@ -37,7 +37,7 @@ void CharInfo::SetFromStoreData(const StoreData& store_data) { eyebrow_aspect = store_data.GetEyebrowAspect(); eyebrow_rotate = store_data.GetEyebrowRotate(); eyebrow_x = store_data.GetEyebrowX(); - eyebrow_y = store_data.GetEyebrowY(); + eyebrow_y = store_data.GetEyebrowY() + 3; nose_type = store_data.GetNoseType(); nose_scale = store_data.GetNoseScale(); nose_y = store_data.GetNoseY(); diff --git a/src/core/hle/service/mii/types/core_data.cpp b/src/core/hle/service/mii/types/core_data.cpp index 1068031e3..970c748ca 100644 --- a/src/core/hle/service/mii/types/core_data.cpp +++ b/src/core/hle/service/mii/types/core_data.cpp @@ -113,7 +113,7 @@ void CoreData::BuildRandom(Age age, Gender gender, Race race) { .values[MiiUtil::GetRandomValue<std::size_t>(eyebrow_type_info.values_count)]); const auto eyebrow_rotate_1{race == Race::Asian ? 6 : 0}; - const auto eyebrow_y{race == Race::Asian ? 9 : 10}; + const auto eyebrow_y{race == Race::Asian ? 6 : 7}; const auto eyebrow_rotate_offset{32 - RawData::EyebrowRotateLookup[eyebrow_rotate_1] + 6}; const auto eyebrow_rotate{ 32 - RawData::EyebrowRotateLookup[static_cast<std::size_t>(data.eyebrow_type.Value())]}; diff --git a/src/core/hle/service/nfc/common/device.cpp b/src/core/hle/service/nfc/common/device.cpp index 05951d8cb..68c407f81 100644 --- a/src/core/hle/service/nfc/common/device.cpp +++ b/src/core/hle/service/nfc/common/device.cpp @@ -1374,7 +1374,7 @@ NFP::AmiiboName NfcDevice::GetAmiiboName(const NFP::AmiiboSettings& settings) co // Convert from utf16 to utf8 const auto amiibo_name_utf8 = Common::UTF16ToUTF8(settings_amiibo_name.data()); - memcpy(amiibo_name.data(), amiibo_name_utf8.data(), amiibo_name_utf8.size() - 1); + memcpy(amiibo_name.data(), amiibo_name_utf8.data(), amiibo_name_utf8.size()); return amiibo_name; } diff --git a/src/video_core/host_shaders/convert_msaa_to_non_msaa.comp b/src/video_core/host_shaders/convert_msaa_to_non_msaa.comp index fc3854d18..66f2ad483 100644 --- a/src/video_core/host_shaders/convert_msaa_to_non_msaa.comp +++ b/src/video_core/host_shaders/convert_msaa_to_non_msaa.comp @@ -15,11 +15,14 @@ void main() { // TODO: Specialization constants for num_samples? const int num_samples = imageSamples(msaa_in); + const ivec3 msaa_size = imageSize(msaa_in); + const ivec3 out_size = imageSize(output_img); + const ivec3 scale = out_size / msaa_size; for (int curr_sample = 0; curr_sample < num_samples; ++curr_sample) { const vec4 pixel = imageLoad(msaa_in, coords, curr_sample); - const int single_sample_x = 2 * coords.x + (curr_sample & 1); - const int single_sample_y = 2 * coords.y + ((curr_sample / 2) & 1); + const int single_sample_x = scale.x * coords.x + (curr_sample & 1); + const int single_sample_y = scale.y * coords.y + ((curr_sample / 2) & 1); const ivec3 dest_coords = ivec3(single_sample_x, single_sample_y, coords.z); if (any(greaterThanEqual(dest_coords, imageSize(output_img)))) { diff --git a/src/video_core/host_shaders/convert_non_msaa_to_msaa.comp b/src/video_core/host_shaders/convert_non_msaa_to_msaa.comp index dedd962f1..c7ce38efa 100644 --- a/src/video_core/host_shaders/convert_non_msaa_to_msaa.comp +++ b/src/video_core/host_shaders/convert_non_msaa_to_msaa.comp @@ -15,9 +15,12 @@ void main() { // TODO: Specialization constants for num_samples? const int num_samples = imageSamples(output_msaa); + const ivec3 msaa_size = imageSize(output_msaa); + const ivec3 out_size = imageSize(img_in); + const ivec3 scale = out_size / msaa_size; for (int curr_sample = 0; curr_sample < num_samples; ++curr_sample) { - const int single_sample_x = 2 * coords.x + (curr_sample & 1); - const int single_sample_y = 2 * coords.y + ((curr_sample / 2) & 1); + const int single_sample_x = scale.x * coords.x + (curr_sample & 1); + const int single_sample_y = scale.y * coords.y + ((curr_sample / 2) & 1); const ivec3 single_coords = ivec3(single_sample_x, single_sample_y, coords.z); if (any(greaterThanEqual(single_coords, imageSize(img_in)))) { diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 289d5b25c..617f92910 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -3,6 +3,7 @@ #include <array> #include <memory> +#include <numeric> #include <optional> #include <utility> @@ -11,7 +12,10 @@ #include "common/assert.h" #include "common/common_types.h" #include "common/div_ceil.h" +#include "common/vector_math.h" #include "video_core/host_shaders/astc_decoder_comp_spv.h" +#include "video_core/host_shaders/convert_msaa_to_non_msaa_comp_spv.h" +#include "video_core/host_shaders/convert_non_msaa_to_msaa_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h" #include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h" #include "video_core/host_shaders/resolve_conditional_render_comp_spv.h" @@ -131,6 +135,33 @@ constexpr DescriptorBankInfo ASTC_BANK_INFO{ .score = 2, }; +constexpr std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> MSAA_DESCRIPTOR_SET_BINDINGS{{ + { + .binding = 0, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, + { + .binding = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = nullptr, + }, +}}; + +constexpr DescriptorBankInfo MSAA_BANK_INFO{ + .uniform_buffers = 0, + .storage_buffers = 0, + .texture_buffers = 0, + .image_buffers = 0, + .textures = 0, + .images = 2, + .score = 2, +}; + constexpr VkDescriptorUpdateTemplateEntry INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE{ .dstBinding = 0, .dstArrayElement = 0, @@ -149,6 +180,15 @@ constexpr VkDescriptorUpdateTemplateEntry QUERIES_SCAN_DESCRIPTOR_UPDATE_TEMPLAT .stride = sizeof(DescriptorUpdateEntry), }; +constexpr VkDescriptorUpdateTemplateEntry MSAA_DESCRIPTOR_UPDATE_TEMPLATE{ + .dstBinding = 0, + .dstArrayElement = 0, + .descriptorCount = 2, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .offset = 0, + .stride = sizeof(DescriptorUpdateEntry), +}; + constexpr std::array<VkDescriptorUpdateTemplateEntry, ASTC_NUM_BINDINGS> ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY{{ { @@ -224,6 +264,9 @@ ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool, }); descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, bank_info); } + if (code.empty()) { + return; + } module = device.GetLogical().CreateShaderModule({ .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO, .pNext = nullptr, @@ -590,4 +633,100 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map, scheduler.Finish(); } +MSAACopyPass::MSAACopyPass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, + StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_) + : ComputePass(device_, descriptor_pool_, MSAA_DESCRIPTOR_SET_BINDINGS, + MSAA_DESCRIPTOR_UPDATE_TEMPLATE, MSAA_BANK_INFO, {}, + CONVERT_NON_MSAA_TO_MSAA_COMP_SPV), + scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, + compute_pass_descriptor_queue{compute_pass_descriptor_queue_} { + const auto make_msaa_pipeline = [this](size_t i, std::span<const u32> code) { + modules[i] = device.GetLogical().CreateShaderModule({ + .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .codeSize = static_cast<u32>(code.size_bytes()), + .pCode = code.data(), + }); + pipelines[i] = device.GetLogical().CreateComputePipeline({ + .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_COMPUTE_BIT, + .module = *modules[i], + .pName = "main", + .pSpecializationInfo = nullptr, + }, + .layout = *layout, + .basePipelineHandle = nullptr, + .basePipelineIndex = 0, + }); + }; + make_msaa_pipeline(0, CONVERT_NON_MSAA_TO_MSAA_COMP_SPV); + make_msaa_pipeline(1, CONVERT_MSAA_TO_NON_MSAA_COMP_SPV); +} + +MSAACopyPass::~MSAACopyPass() = default; + +void MSAACopyPass::CopyImage(Image& dst_image, Image& src_image, + std::span<const VideoCommon::ImageCopy> copies, + bool msaa_to_non_msaa) { + const VkPipeline msaa_pipeline = *pipelines[msaa_to_non_msaa ? 1 : 0]; + scheduler.RequestOutsideRenderPassOperationContext(); + for (const VideoCommon::ImageCopy& copy : copies) { + ASSERT(copy.src_subresource.base_layer == 0); + ASSERT(copy.src_subresource.num_layers == 1); + ASSERT(copy.dst_subresource.base_layer == 0); + ASSERT(copy.dst_subresource.num_layers == 1); + + compute_pass_descriptor_queue.Acquire(); + compute_pass_descriptor_queue.AddImage( + src_image.StorageImageView(copy.src_subresource.base_level)); + compute_pass_descriptor_queue.AddImage( + dst_image.StorageImageView(copy.dst_subresource.base_level)); + const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()}; + + const Common::Vec3<u32> num_dispatches = { + Common::DivCeil(copy.extent.width, 8U), + Common::DivCeil(copy.extent.height, 8U), + copy.extent.depth, + }; + + scheduler.Record([this, dst = dst_image.Handle(), msaa_pipeline, num_dispatches, + descriptor_data](vk::CommandBuffer cmdbuf) { + const VkDescriptorSet set = descriptor_allocator.Commit(); + device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, msaa_pipeline); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {}); + cmdbuf.Dispatch(num_dispatches.x, num_dispatches.y, num_dispatches.z); + const VkImageMemoryBarrier write_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst, + .subresourceRange{ + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier); + }); + } +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index 3ff935639..7b8f938c1 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -11,6 +11,7 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" +#include "video_core/texture_cache/types.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" @@ -130,4 +131,22 @@ private: MemoryAllocator& memory_allocator; }; +class MSAACopyPass final : public ComputePass { +public: + explicit MSAACopyPass(const Device& device_, Scheduler& scheduler_, + DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_, + ComputePassDescriptorQueue& compute_pass_descriptor_queue_); + ~MSAACopyPass(); + + void CopyImage(Image& dst_image, Image& src_image, + std::span<const VideoCommon::ImageCopy> copies, bool msaa_to_non_msaa); + +private: + Scheduler& scheduler; + StagingBufferPool& staging_buffer_pool; + ComputePassDescriptorQueue& compute_pass_descriptor_queue; + std::array<vk::ShaderModule, 2> modules; + std::array<vk::Pipeline, 2> pipelines; +}; + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index c7ce7c312..1628d76d6 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -1014,15 +1014,37 @@ void RasterizerVulkan::UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs) { regs.zeta.format == Tegra::DepthFormat::X8Z24_UNORM || regs.zeta.format == Tegra::DepthFormat::S8Z24_UNORM || regs.zeta.format == Tegra::DepthFormat::V8Z24_UNORM; - if (is_d24 && !device.SupportsD24DepthBuffer()) { + bool force_unorm = ([&] { + if (!is_d24 || device.SupportsD24DepthBuffer()) { + return false; + } + if (device.IsExtDepthBiasControlSupported()) { + return true; + } + if (!Settings::values.renderer_amdvlk_depth_bias_workaround) { + return false; + } // the base formulas can be obtained from here: // https://docs.microsoft.com/en-us/windows/win32/direct3d11/d3d10-graphics-programming-guide-output-merger-stage-depth-bias const double rescale_factor = static_cast<double>(1ULL << (32 - 24)) / (static_cast<double>(0x1.ep+127)); units = static_cast<float>(static_cast<double>(units) * rescale_factor); - } + return false; + })(); scheduler.Record([constant = units, clamp = regs.depth_bias_clamp, - factor = regs.slope_scale_depth_bias](vk::CommandBuffer cmdbuf) { + factor = regs.slope_scale_depth_bias, force_unorm, + precise = device.HasExactDepthBiasControl()](vk::CommandBuffer cmdbuf) { + if (force_unorm) { + VkDepthBiasRepresentationInfoEXT info{ + .sType = VK_STRUCTURE_TYPE_DEPTH_BIAS_REPRESENTATION_INFO_EXT, + .pNext = nullptr, + .depthBiasRepresentation = + VK_DEPTH_BIAS_REPRESENTATION_LEAST_REPRESENTABLE_VALUE_FORCE_UNORM_EXT, + .depthBiasExact = precise ? VK_TRUE : VK_FALSE, + }; + cmdbuf.SetDepthBias(constant, clamp, factor, &info); + return; + } cmdbuf.SetDepthBias(constant, clamp, factor); }); } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 1f9e7acaa..71fdec809 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -176,6 +176,36 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { return allocator.CreateImage(image_ci); } +[[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image, + VkFormat format) { + static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ + .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, + .pNext = nullptr, + .usage = VK_IMAGE_USAGE_STORAGE_BIT, + }; + return device.CreateImageView(VkImageViewCreateInfo{ + .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, + .pNext = &storage_image_view_usage_create_info, + .flags = 0, + .image = image, + .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, + .format = format, + .components{ + .r = VK_COMPONENT_SWIZZLE_IDENTITY, + .g = VK_COMPONENT_SWIZZLE_IDENTITY, + .b = VK_COMPONENT_SWIZZLE_IDENTITY, + .a = VK_COMPONENT_SWIZZLE_IDENTITY, + }, + .subresourceRange{ + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = level, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }); +} + [[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) { switch (VideoCore::Surface::GetFormatType(format)) { case VideoCore::Surface::SurfaceType::ColorTexture: @@ -817,6 +847,10 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue, memory_allocator); } + if (device.IsStorageImageMultisampleSupported()) { + msaa_copy_pass = std::make_unique<MSAACopyPass>( + device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue); + } if (!device.IsKhrImageFormatListSupported()) { return; } @@ -1285,7 +1319,11 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies) { - UNIMPLEMENTED_MSG("Copying images with different samples is not implemented in Vulkan."); + const bool msaa_to_non_msaa = src.info.num_samples > 1 && dst.info.num_samples == 1; + if (msaa_copy_pass) { + return msaa_copy_pass->CopyImage(dst, src, copies, msaa_to_non_msaa); + } + UNIMPLEMENTED_MSG("Copying images with different samples is not supported."); } u64 TextureCacheRuntime::GetDeviceLocalMemory() const { @@ -1333,39 +1371,15 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu if (runtime->device.HasDebuggingToolAttached()) { original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } - static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ - .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, - .pNext = nullptr, - .usage = VK_IMAGE_USAGE_STORAGE_BIT, - }; current_image = *original_image; + storage_image_views.resize(info.resources.levels); if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported() && Settings::values.astc_recompression.GetValue() == Settings::AstcRecompression::Uncompressed) { const auto& device = runtime->device.GetLogical(); - storage_image_views.reserve(info.resources.levels); for (s32 level = 0; level < info.resources.levels; ++level) { - storage_image_views.push_back(device.CreateImageView(VkImageViewCreateInfo{ - .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, - .pNext = &storage_image_view_usage_create_info, - .flags = 0, - .image = *original_image, - .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, - .format = VK_FORMAT_A8B8G8R8_UNORM_PACK32, - .components{ - .r = VK_COMPONENT_SWIZZLE_IDENTITY, - .g = VK_COMPONENT_SWIZZLE_IDENTITY, - .b = VK_COMPONENT_SWIZZLE_IDENTITY, - .a = VK_COMPONENT_SWIZZLE_IDENTITY, - }, - .subresourceRange{ - .aspectMask = aspect_mask, - .baseMipLevel = static_cast<u32>(level), - .levelCount = 1, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, - })); + storage_image_views[level] = + MakeStorageView(device, level, *original_image, VK_FORMAT_A8B8G8R8_UNORM_PACK32); } } } @@ -1496,6 +1510,17 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm DownloadMemory(buffers, offsets, copies); } +VkImageView Image::StorageImageView(s32 level) noexcept { + auto& view = storage_image_views[level]; + if (!view) { + const auto format_info = + MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format); + view = + MakeStorageView(runtime->device.GetLogical(), level, current_image, format_info.format); + } + return *view; +} + bool Image::IsRescaled() const noexcept { return True(flags & ImageFlagBits::Rescaled); } diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 565ce19a9..d6c5a15cc 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -117,6 +117,7 @@ public: BlitImageHelper& blit_image_helper; RenderPassCache& render_pass_cache; std::optional<ASTCDecoderPass> astc_decoder_pass; + std::unique_ptr<MSAACopyPass> msaa_copy_pass; const Settings::ResolutionScalingInfo& resolution; std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats; @@ -161,15 +162,13 @@ public: return aspect_mask; } - [[nodiscard]] VkImageView StorageImageView(s32 level) const noexcept { - return *storage_image_views[level]; - } - /// Returns true when the image is already initialized and mark it as initialized [[nodiscard]] bool ExchangeInitialization() noexcept { return std::exchange(initialized, true); } + VkImageView StorageImageView(s32 level) noexcept; + bool IsRescaled() const noexcept; bool ScaleUp(bool ignore = false); diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 18185610f..3960b135a 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -1059,6 +1059,13 @@ void Device::RemoveUnsuitableExtensions() { RemoveExtensionFeatureIfUnsuitable(extensions.custom_border_color, features.custom_border_color, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME); + // VK_EXT_depth_bias_control + extensions.depth_bias_control = + features.depth_bias_control.depthBiasControl && + features.depth_bias_control.leastRepresentableValueForceUnormRepresentation; + RemoveExtensionFeatureIfUnsuitable(extensions.depth_bias_control, features.depth_bias_control, + VK_EXT_DEPTH_BIAS_CONTROL_EXTENSION_NAME); + // VK_EXT_depth_clip_control extensions.depth_clip_control = features.depth_clip_control.depthClipControl; RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_control, features.depth_clip_control, diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 94f41266d..9be612392 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -41,6 +41,7 @@ VK_DEFINE_HANDLE(VmaAllocator) // Define all features which may be used by the implementation and require an extension here. #define FOR_EACH_VK_FEATURE_EXT(FEATURE) \ FEATURE(EXT, CustomBorderColor, CUSTOM_BORDER_COLOR, custom_border_color) \ + FEATURE(EXT, DepthBiasControl, DEPTH_BIAS_CONTROL, depth_bias_control) \ FEATURE(EXT, DepthClipControl, DEPTH_CLIP_CONTROL, depth_clip_control) \ FEATURE(EXT, ExtendedDynamicState, EXTENDED_DYNAMIC_STATE, extended_dynamic_state) \ FEATURE(EXT, ExtendedDynamicState2, EXTENDED_DYNAMIC_STATE_2, extended_dynamic_state2) \ @@ -96,6 +97,7 @@ VK_DEFINE_HANDLE(VmaAllocator) #define FOR_EACH_VK_RECOMMENDED_EXTENSION(EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_CONDITIONAL_RENDERING_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_CONSERVATIVE_RASTERIZATION_EXTENSION_NAME) \ + EXTENSION_NAME(VK_EXT_DEPTH_BIAS_CONTROL_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME) \ @@ -147,6 +149,9 @@ VK_DEFINE_HANDLE(VmaAllocator) // Define features where the absence of the feature may result in a degraded experience. #define FOR_EACH_VK_RECOMMENDED_FEATURE(FEATURE_NAME) \ FEATURE_NAME(custom_border_color, customBorderColors) \ + FEATURE_NAME(depth_bias_control, depthBiasControl) \ + FEATURE_NAME(depth_bias_control, leastRepresentableValueForceUnormRepresentation) \ + FEATURE_NAME(depth_bias_control, depthBiasExact) \ FEATURE_NAME(extended_dynamic_state, extendedDynamicState) \ FEATURE_NAME(format_a4b4g4r4, formatA4B4G4R4) \ FEATURE_NAME(index_type_uint8, indexTypeUint8) \ @@ -324,6 +329,11 @@ public: return features.shader_float16_int8.shaderInt8; } + /// Returns true if the device supports binding multisample images as storage images. + bool IsStorageImageMultisampleSupported() const { + return features.features.shaderStorageImageMultisample; + } + /// Returns true if the device warp size can potentially be bigger than guest's warp size. bool IsWarpSizePotentiallyBiggerThanGuest() const { return is_warp_potentially_bigger; @@ -459,6 +469,11 @@ public: return extensions.depth_clip_control; } + /// Returns true if the device supports VK_EXT_depth_bias_control. + bool IsExtDepthBiasControlSupported() const { + return extensions.depth_bias_control; + } + /// Returns true if the device supports VK_EXT_shader_viewport_index_layer. bool IsExtShaderViewportIndexLayerSupported() const { return extensions.shader_viewport_index_layer; @@ -619,6 +634,10 @@ public: return features.robustness2.nullDescriptor; } + bool HasExactDepthBiasControl() const { + return features.depth_bias_control.depthBiasExact; + } + u32 GetMaxVertexInputAttributes() const { return properties.properties.limits.maxVertexInputAttributes; } diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 5afba365c..2f3254a97 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -113,6 +113,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdPushDescriptorSetWithTemplateKHR); X(vkCmdSetBlendConstants); X(vkCmdSetDepthBias); + X(vkCmdSetDepthBias2EXT); X(vkCmdSetDepthBounds); X(vkCmdSetEvent); X(vkCmdSetScissor); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 0d4bbe7f7..1e3c0fa64 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -226,6 +226,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants{}; PFN_vkCmdSetCullModeEXT vkCmdSetCullModeEXT{}; PFN_vkCmdSetDepthBias vkCmdSetDepthBias{}; + PFN_vkCmdSetDepthBias2EXT vkCmdSetDepthBias2EXT{}; PFN_vkCmdSetDepthBounds vkCmdSetDepthBounds{}; PFN_vkCmdSetDepthBoundsTestEnableEXT vkCmdSetDepthBoundsTestEnableEXT{}; PFN_vkCmdSetDepthCompareOpEXT vkCmdSetDepthCompareOpEXT{}; @@ -1333,6 +1334,18 @@ public: dld->vkCmdSetDepthBias(handle, constant_factor, clamp, slope_factor); } + void SetDepthBias(float constant_factor, float clamp, float slope_factor, + VkDepthBiasRepresentationInfoEXT* extra) const noexcept { + VkDepthBiasInfoEXT info{ + .sType = VK_STRUCTURE_TYPE_DEPTH_BIAS_INFO_EXT, + .pNext = extra, + .depthBiasConstantFactor = constant_factor, + .depthBiasClamp = clamp, + .depthBiasSlopeFactor = slope_factor, + }; + dld->vkCmdSetDepthBias2EXT(handle, &info); + } + void SetDepthBounds(float min_depth_bounds, float max_depth_bounds) const noexcept { dld->vkCmdSetDepthBounds(handle, min_depth_bounds, max_depth_bounds); } diff --git a/src/yuzu/main.cpp b/src/yuzu/main.cpp index b1b6b9354..adb7b332f 100644 --- a/src/yuzu/main.cpp +++ b/src/yuzu/main.cpp @@ -3113,10 +3113,9 @@ void GMainWindow::OnMenuInstallToNAND() { QFuture<InstallResult> future; InstallResult result; - if (file.endsWith(QStringLiteral("xci"), Qt::CaseInsensitive) || - file.endsWith(QStringLiteral("nsp"), Qt::CaseInsensitive)) { + if (file.endsWith(QStringLiteral("nsp"), Qt::CaseInsensitive)) { - future = QtConcurrent::run([this, &file] { return InstallNSPXCI(file); }); + future = QtConcurrent::run([this, &file] { return InstallNSP(file); }); while (!future.isFinished()) { QCoreApplication::processEvents(); @@ -3175,7 +3174,7 @@ void GMainWindow::OnMenuInstallToNAND() { ui->action_Install_File_NAND->setEnabled(true); } -InstallResult GMainWindow::InstallNSPXCI(const QString& filename) { +InstallResult GMainWindow::InstallNSP(const QString& filename) { const auto qt_raw_copy = [this](const FileSys::VirtualFile& src, const FileSys::VirtualFile& dest, std::size_t block_size) { if (src == nullptr || dest == nullptr) { @@ -3209,9 +3208,7 @@ InstallResult GMainWindow::InstallNSPXCI(const QString& filename) { return InstallResult::Failure; } } else { - const auto xci = std::make_shared<FileSys::XCI>( - vfs->OpenFile(filename.toStdString(), FileSys::Mode::Read)); - nsp = xci->GetSecurePartitionNSP(); + return InstallResult::Failure; } if (nsp->GetStatus() != Loader::ResultStatus::Success) { diff --git a/src/yuzu/main.h b/src/yuzu/main.h index 53bedfab3..ba318eb11 100644 --- a/src/yuzu/main.h +++ b/src/yuzu/main.h @@ -387,7 +387,7 @@ private: void RemoveCacheStorage(u64 program_id); bool SelectRomFSDumpTarget(const FileSys::ContentProvider&, u64 program_id, u64* selected_title_id, u8* selected_content_record_type); - InstallResult InstallNSPXCI(const QString& filename); + InstallResult InstallNSP(const QString& filename); InstallResult InstallNCA(const QString& filename); void MigrateConfigFiles(); void UpdateWindowTitle(std::string_view title_name = {}, std::string_view title_version = {}, |