From 1d4ccdbb0cf01add647247a4ab64b5060b1bb8eb Mon Sep 17 00:00:00 2001 From: Duck912 <350528965+Duck912@users.noreply.github.com> Date: Tue, 3 Nov 2026 12:10:30 -0500 Subject: video_core: read the framebuffer before recording the upscaled-output blit BlitColor(VkImageView, VkImage, ...), which draws FSR 3.1's output into the game's DLSS output, captured the Framebuffer pointer and read its render pass, handle and size on the worker thread. The pointer points into the texture cache's framebuffer slots, which can be reallocated before the worker records the command, so the render pass could begin with a stale or null render pass and framebuffer. FSR 3.1 records quickly enough that this rarely shows; with a slower first evaluation (NVIDIA DLSS creating its feature) it crashed in vkCmdBeginRenderPass at the first evaluation in every run. Read the three values on the calling thread and capture those instead. Nothing else changes. Assisted-by: Claude Opus 5.5 (debugging, implementation) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com> --- src/video_core/renderer_vulkan/blit_image.cpp | 18 +++++++++++------- 1 file changed, 11 insertions(+), 7 deletions(-) diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index 0628da1d0f..4a274b2915 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -644,10 +644,8 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) }); } -void BeginRenderPass(vk::CommandBuffer& cmdbuf, const Framebuffer* framebuffer) { - const VkRenderPass render_pass = framebuffer->RenderPass(); - const VkFramebuffer framebuffer_handle = framebuffer->Handle(); - const VkExtent2D render_area = framebuffer->RenderArea(); +void BeginRenderPass(vk::CommandBuffer& cmdbuf, VkRenderPass render_pass, + VkFramebuffer framebuffer_handle, VkExtent2D render_area) { const VkRenderPassBeginInfo renderpass_bi{ .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, .pNext = nullptr, @@ -784,11 +782,17 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView }; const VkPipelineLayout layout = *one_texture_pipeline_layout; const VkPipeline pipeline = FindOrEmplaceColorPipeline(key); + // Read the framebuffer now: the texture cache's framebuffer slots can move before the + // worker records this. + const VkRenderPass render_pass = dst_framebuffer->RenderPass(); + const VkFramebuffer framebuffer = dst_framebuffer->Handle(); + const VkExtent2D render_area = dst_framebuffer->RenderArea(); scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([this, dst_framebuffer, src_image_view, src_image, src_sampler, dst_region, - src_region, src_size, pipeline, layout](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, render_pass, framebuffer, render_area, src_image_view, src_image, + src_sampler, dst_region, src_region, src_size, pipeline, + layout](vk::CommandBuffer cmdbuf) { TransitionImageLayout(cmdbuf, src_image, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL); - BeginRenderPass(cmdbuf, dst_framebuffer); + BeginRenderPass(cmdbuf, render_pass, framebuffer, render_area); const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit(); // The source was just transitioned: describe it in that layout. UpdateOneTextureDescriptorSet(device, descriptor_set, src_sampler, src_image_view, -- cgit v1.2.3 From 63e87a22c60394bc4f427ee5f8d6ddbce7b0ac7f Mon Sep 17 00:00:00 2001 From: Duck912 <350528965+Duck912@users.noreply.github.com> Date: Tue, 3 Nov 2026 12:13:35 -0500 Subject: video_core: NVIDIA DLSS through NVIDIA Streamline in place of a game's DLSS Adds a Graphics setting, "Use NVIDIA DLSS in place of a game's DLSS" (nvidia_dlss, off by default). When it is on, in Windows builds configured with EGLANTINE_STREAMLINE_SDK, the evaluation Eglantine already replaces with FSR 3.1 goes to NVIDIA DLSS on GeForce RTX cards, with the same inputs (colour, depth, motion vectors, jitter). When DLSS can't run, FSR 3.1 runs as before. Only Streamline's MIT-licensed headers are compiled in; nothing of NVIDIA's is linked or shipped. At run time Eglantine looks for sl.interposer.dll in the user folder's "streamline" folder (or EDEN_STREAMLINE_DIR, or "streamline" next to the executable), checks its Streamline signature and starts Streamline with manual hooking. Streamline's plugins come from that folder or the NVIDIA app's downloads. DLSS itself is nvngx_dlss.dll in the same folder, else the newest DLSS 310 or later that the NVIDIA app has downloaded, checked for NVIDIA's Authenticode signature and copied as streamline/nvidia-app/nvngx_dlss.dll. Like other tools that call DLSS outside a game, it uses its own fixed project id with the custom engine type. - Preset K for all modes (EDEN_NX2_DLSS_PRESET overrides it): the default Performance model (M) and L draw thin dark horizontal lines in some frames at 2x resolution scale; K doesn't. - HDR input with auto exposure; the mode follows the render/output ratio. - Buffer device address stays enabled when DLSS needs it, for DLSS only. - The upscaler waits for the device instead of submitting when it is destroyed: by then the game's channels are released, and a submit read the state tracker's stale flags (a crash when stopping a game). With the setting off, or in builds without the SDK path (the setting is hidden there), nothing changes. Testing: Windows 11, RTX 5080, Fire Emblem: Fortune's Weave docked at 1x (720p -> 1440p) and 2x (1440p -> 2880p), CLI runs and GUI play, stopping and restarting games with DLSS on. Not tested on other GPUs or games. Assisted-by: Claude Opus 5.5 (implementation, debugging) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com> --- src/common/fs/fs_paths.h | 1 + src/common/fs/path_util.cpp | 1 + src/common/fs/path_util.h | 1 + src/common/settings.h | 2 + src/qt_common/config/shared_translation.cpp | 5 + src/video_core/CMakeLists.txt | 18 + src/video_core/renderer_vulkan/renderer_vulkan.cpp | 6 + src/video_core/renderer_vulkan/renderer_vulkan.h | 3 + .../renderer_vulkan/vk_streamline_dlss.cpp | 386 +++++++++++++++++ .../renderer_vulkan/vk_streamline_dlss.h | 80 ++++ .../renderer_vulkan/vk_texture_cache.cpp | 122 +++++- src/video_core/renderer_vulkan/vk_texture_cache.h | 17 + src/video_core/texture_cache/texture_cache.h | 4 +- src/video_core/vulkan_common/streamline.cpp | 463 +++++++++++++++++++++ src/video_core/vulkan_common/streamline.h | 63 +++ src/video_core/vulkan_common/streamline_api.h | 36 ++ src/video_core/vulkan_common/vulkan_device.cpp | 47 +++ src/video_core/vulkan_common/vulkan_device.h | 7 + src/video_core/vulkan_common/vulkan_instance.cpp | 20 + src/yuzu/configuration/configure_graphics.cpp | 6 + 20 files changed, 1271 insertions(+), 17 deletions(-) create mode 100644 src/video_core/renderer_vulkan/vk_streamline_dlss.cpp create mode 100644 src/video_core/renderer_vulkan/vk_streamline_dlss.h create mode 100644 src/video_core/vulkan_common/streamline.cpp create mode 100644 src/video_core/vulkan_common/streamline.h create mode 100644 src/video_core/vulkan_common/streamline_api.h diff --git a/src/common/fs/fs_paths.h b/src/common/fs/fs_paths.h index ae25846fc2..763a8759dd 100644 --- a/src/common/fs/fs_paths.h +++ b/src/common/fs/fs_paths.h @@ -28,6 +28,7 @@ #define SCREENSHOTS_DIR "screenshots" #define SDMC_DIR "sdmc" #define SHADER_DIR "shader" +#define STREAMLINE_DIR "streamline" #define TAS_DIR "tas" #define ICONS_DIR "icons" diff --git a/src/common/fs/path_util.cpp b/src/common/fs/path_util.cpp index 82ea8f6cb2..f77da2e301 100644 --- a/src/common/fs/path_util.cpp +++ b/src/common/fs/path_util.cpp @@ -168,6 +168,7 @@ public: GenerateEdenPath(EdenPath::SDMCDir, eden_path / SDMC_DIR); GenerateEdenPath(EdenPath::TASDir, eden_path / TAS_DIR); GenerateEdenPath(EdenPath::IconsDir, eden_path / ICONS_DIR); + GenerateEdenPath(EdenPath::StreamlineDir, eden_path / STREAMLINE_DIR); // config GenerateEdenPath(EdenPath::ConfigDir, eden_path_config); // cache diff --git a/src/common/fs/path_util.h b/src/common/fs/path_util.h index e8fa5614ae..940dbb9186 100644 --- a/src/common/fs/path_util.h +++ b/src/common/fs/path_util.h @@ -34,6 +34,7 @@ enum class EdenPath { ShaderDir, // Where shaders are stored. TASDir, // Where TAS scripts are stored. IconsDir, // Where Icons for Windows shortcuts are stored. + StreamlineDir, // Where the user-supplied NVIDIA Streamline loader is stored. }; // migration/compat dirs diff --git a/src/common/settings.h b/src/common/settings.h index 6669f98f4c..dc442a7180 100644 --- a/src/common/settings.h +++ b/src/common/settings.h @@ -378,6 +378,8 @@ struct Values { Specialization::Percentage, true, true}; + // NVIDIA DLSS (through NVIDIA Streamline) in place of a game's DLSS. + SwitchableSetting nvidia_dlss{linkage, false, "nvidia_dlss", Category::Renderer}; SwitchableSetting aspect_ratio{linkage, AspectRatio::R16_9, "aspect_ratio", diff --git a/src/qt_common/config/shared_translation.cpp b/src/qt_common/config/shared_translation.cpp index c5907e1b65..b395b2959c 100644 --- a/src/qt_common/config/shared_translation.cpp +++ b/src/qt_common/config/shared_translation.cpp @@ -149,6 +149,11 @@ std::unique_ptr InitializeTranslations(QObject* parent) { tr("Strength of RCAS (AMD's contrast-adaptive sharpening) on the upscaler used in place " "of a game's DLSS (AMD FSR 3.1); 0% turns it off.\nAlso sets the sharpening " "of the AMD FSR 1 and SGSR window adapting filters.")); + INSERT(Settings, nvidia_dlss, tr("Use NVIDIA DLSS in place of a game's DLSS"), + tr("Uses NVIDIA DLSS on GeForce RTX cards in place of a Switch 2 game's own DLSS, " + "instead of AMD FSR 3.1.\nNeeds NVIDIA's Streamline loader, and NVIDIA DLSS unless " + "the NVIDIA app has downloaded it; Eglantine does not ship NVIDIA's files.\nTakes " + "effect when a game starts.")); INSERT(Settings, anti_aliasing, tr("Anti-Aliasing Method:"), tr("The anti-aliasing method to use.\nSMAA offers the best quality.\nFXAA " "can produce a more stable picture in lower resolutions.")); diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index d0e84c324c..e474d08fba 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -499,4 +499,22 @@ if (ANDROID AND ARCHITECTURE_arm64) target_link_libraries(video_core PRIVATE adrenotools) endif() +# NVIDIA DLSS through NVIDIA Streamline, in place of a game's DLSS (before FSR 3.1) when the +# setting is on. Only Streamline's MIT-licensed headers are compiled in and nothing of NVIDIA's is +# linked or copied: sl.interposer.dll and the DLSS plugins are loaded at run time from the user's PC. +set(EGLANTINE_STREAMLINE_SDK "" CACHE PATH "NVIDIA Streamline SDK directory (headers only)") +if (EGLANTINE_STREAMLINE_SDK AND WIN32) + target_sources(video_core PRIVATE + renderer_vulkan/vk_streamline_dlss.cpp + renderer_vulkan/vk_streamline_dlss.h + vulkan_common/streamline.cpp + vulkan_common/streamline.h + vulkan_common/streamline_api.h + ) + target_include_directories(video_core PRIVATE ${EGLANTINE_STREAMLINE_SDK}/include) + target_link_libraries(video_core PRIVATE version wintrust crypt32) + target_compile_definitions(video_core PUBLIC EGLANTINE_STREAMLINE) + message(STATUS "NVIDIA DLSS through Streamline: ${EGLANTINE_STREAMLINE_SDK}") +endif() + create_target_directory_groups(video_core) diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 27ca531186..fc0eea84ab 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -41,6 +41,9 @@ #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_surface.h" #include "video_core/vulkan_common/vulkan_wrapper.h" +#ifdef EGLANTINE_STREAMLINE +#include "video_core/vulkan_common/streamline.h" +#endif #ifdef __ANDROID__ #include #endif @@ -130,6 +133,9 @@ try , gpu(gpu_) , library(OpenLibrary(context.get())) , dld() +#ifdef EGLANTINE_STREAMLINE + , streamline_started(Streamline::Start()) +#endif // Create raw Vulkan instance first , instance(CreateInstance(*library, dld, diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h index 87422dd935..67587cde37 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.h +++ b/src/video_core/renderer_vulkan/renderer_vulkan.h @@ -83,6 +83,9 @@ private: std::shared_ptr library; vk::InstanceDispatch dld; +#ifdef EGLANTINE_STREAMLINE + bool streamline_started; ///< NVIDIA Streamline starts before the instance. +#endif // Keep original handles for compatibility with existing code vk::Instance instance; diff --git a/src/video_core/renderer_vulkan/vk_streamline_dlss.cpp b/src/video_core/renderer_vulkan/vk_streamline_dlss.cpp new file mode 100644 index 0000000000..aef32735f9 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_streamline_dlss.cpp @@ -0,0 +1,386 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// NVIDIA DLSS (through NVIDIA Streamline) in place of a game's DLSS evaluation. + +#include +#include +#include + +#include "common/logging.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_streamline_dlss.h" +#include "video_core/vulkan_common/streamline.h" +#include "video_core/vulkan_common/streamline_api.h" +#include "video_core/vulkan_common/vulkan_device.h" +#include "video_core/vulkan_common/vulkan_memory_allocator.h" + +namespace Vulkan { +namespace { + +constexpr VkFormat OUTPUT_FORMAT = VK_FORMAT_R16G16B16A16_SFLOAT; +// Streamline clears the output before DLSS writes it. +constexpr VkImageUsageFlags OUTPUT_USAGE = VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT | + VK_IMAGE_USAGE_TRANSFER_SRC_BIT | + VK_IMAGE_USAGE_TRANSFER_DST_BIT; + +int EnvInt(const char* name, int fallback) { + const char* const value = std::getenv(name); + return value ? std::atoi(value) : fallback; +} + +// DLSS validates the render size against the mode's range: pick the mode for this ratio. +sl::DLSSMode ModeFor(VkExtent2D render, VkExtent2D upscale) { + const f32 ratio = static_cast(render.width) / static_cast(upscale.width); + if (ratio >= 0.99f) { + return sl::DLSSMode::eDLAA; + } + if (ratio >= 0.65f) { + return sl::DLSSMode::eMaxQuality; + } + if (ratio >= 0.57f) { + return sl::DLSSMode::eBalanced; + } + if (ratio >= 0.49f) { + return sl::DLSSMode::eMaxPerformance; + } + return sl::DLSSMode::eUltraPerformance; +} + +// Preset K for every mode: NVIDIA's defaults for Performance/Ultra Performance (M, L) draw thin +// horizontal lines in Fire Emblem at 2x resolution scale. EDEN_NX2_DLSS_PRESET=J|K|L|M picks +// another model, EDEN_NX2_DLSS_PRESET=default leaves the choice to NVIDIA. +sl::DLSSPreset Preset() { + const char* const value = std::getenv("EDEN_NX2_DLSS_PRESET"); + if (value && value[0] >= 'J' && value[0] <= 'M') { + return static_cast(static_cast(sl::DLSSPreset::ePresetJ) + + static_cast(value[0] - 'J')); + } + if (value && std::string_view{value} == "default") { + return sl::DLSSPreset::eDefault; + } + return sl::DLSSPreset::ePresetK; +} + +sl::float4x4 Identity() { + sl::float4x4 matrix; + matrix.setRow(0, {1.0f, 0.0f, 0.0f, 0.0f}); + matrix.setRow(1, {0.0f, 1.0f, 0.0f, 0.0f}); + matrix.setRow(2, {0.0f, 0.0f, 1.0f, 0.0f}); + matrix.setRow(3, {0.0f, 0.0f, 0.0f, 1.0f}); + return matrix; +} + +VkImageMemoryBarrier Barrier(const StreamlineDlssUpscaler::Image& image, VkAccessFlags src_access, + VkAccessFlags dst_access, VkImageLayout old_layout, + VkImageLayout new_layout) { + VkImageSubresourceRange range = image.range; + range.aspectMask = image.barrier_aspect; + return VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = src_access, + .dstAccessMask = dst_access, + .oldLayout = old_layout, + .newLayout = new_layout, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = image.image, + .subresourceRange = range, + }; +} + +sl::Resource Resource(VkImage image, VkImageView view, VkFormat format, VkImageLayout layout, + VkExtent2D size, VkImageUsageFlags usage) { + sl::Resource resource{sl::ResourceType::eTex2d, image, nullptr, view, + static_cast(layout)}; + resource.width = size.width; + resource.height = size.height; + resource.nativeFormat = static_cast(format); + resource.mipLevels = 1; + resource.arrayLayers = 1; + resource.flags = 0; + resource.usage = usage; + return resource; +} + +} // Anonymous namespace + +StreamlineDlssUpscaler::StreamlineDlssUpscaler(const Device& device_, Scheduler& scheduler_, + MemoryAllocator& memory_allocator_) + : device{device_}, scheduler{scheduler_}, memory_allocator{memory_allocator_} { + if (!Streamline::IsDlssAvailable()) { + throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); + } + const sl::Result result = Streamline::GetApi().get_feature_function( + sl::kFeatureDLSS, "slDLSSSetOptions", set_options); + if (result != sl::Result::eOk || !set_options) { + LOG_ERROR(Render_Vulkan, "Streamline's DLSS has no slDLSSSetOptions (result {})", + static_cast(result)); + throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); + } + point_sampler = device.GetLogical().CreateSampler(VkSamplerCreateInfo{ + .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .magFilter = VK_FILTER_NEAREST, + .minFilter = VK_FILTER_NEAREST, + .mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST, + .addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, + .addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, + .addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE, + .mipLodBias = 0.0f, + .anisotropyEnable = VK_FALSE, + .maxAnisotropy = 1.0f, + .compareEnable = VK_FALSE, + .compareOp = VK_COMPARE_OP_NEVER, + .minLod = 0.0f, + .maxLod = VK_LOD_CLAMP_NONE, + .borderColor = VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK, + .unnormalizedCoordinates = VK_FALSE, + }); +} + +StreamlineDlssUpscaler::~StreamlineDlssUpscaler() { + // Not scheduler.Finish(): this runs when the renderer goes away, after the game's channels are + // released, and a submit would touch their state. Nothing more gets recorded by then. + void(device.GetLogical().WaitIdle()); + if (Streamline::IsDlssAvailable()) { + Streamline::GetApi().free_resources(sl::kFeatureDLSS, sl::ViewportHandle{0u}); + } +} + +void StreamlineDlssUpscaler::CreateOutput(VkExtent2D upscale_size) { + if (output_image) { + scheduler.Finish(); // the old output may still be in use + } + output_image = memory_allocator.CreateImage(VkImageCreateInfo{ + .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .imageType = VK_IMAGE_TYPE_2D, + .format = OUTPUT_FORMAT, + .extent = {upscale_size.width, upscale_size.height, 1}, + .mipLevels = 1, + .arrayLayers = 1, + .samples = VK_SAMPLE_COUNT_1_BIT, + .tiling = VK_IMAGE_TILING_OPTIMAL, + .usage = OUTPUT_USAGE, + .sharingMode = VK_SHARING_MODE_EXCLUSIVE, + .queueFamilyIndexCount = 0, + .pQueueFamilyIndices = nullptr, + .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED, + }); + output_view = device.GetLogical().CreateImageView(VkImageViewCreateInfo{ + .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .image = *output_image, + .viewType = VK_IMAGE_VIEW_TYPE_2D, + .format = OUTPUT_FORMAT, + .components{}, + .subresourceRange{ + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }); + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([image = *output_image](vk::CommandBuffer cmdbuf) { + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = image, + .subresourceRange{ + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }); + }); +} + +bool StreamlineDlssUpscaler::Upscale(const Inputs& in) { + if (failed) { + return false; + } + bool reset = in.reset; + bool options_changed = false; + if (!output_image || in.render_size.width != render_extent.width || + in.render_size.height != render_extent.height || + in.upscale_size.width != upscale_extent.width || + in.upscale_size.height != upscale_extent.height) { + CreateOutput(in.upscale_size); + render_extent = in.render_size; + upscale_extent = in.upscale_size; + options_changed = true; + reset = true; + } + + // The game's own setup (Fortune's Weave: render-size motion vectors, reversed depth). Its + // colour is linear and goes above 1 with no exposure passed, so by default DLSS gets the HDR + // path with its own auto exposure. EDEN_NX2_DLSS_HDR=0 / EDEN_NX2_DLSS_AUTO_EXPOSURE=0 turn + // those off. + sl::DLSSOptions options{}; + options.mode = ModeFor(in.render_size, in.upscale_size); + options.outputWidth = in.upscale_size.width; + options.outputHeight = in.upscale_size.height; + options.colorBuffersHDR = EnvInt("EDEN_NX2_DLSS_HDR", 1) != 0 ? sl::eTrue : sl::eFalse; + options.useAutoExposure = + EnvInt("EDEN_NX2_DLSS_AUTO_EXPOSURE", 1) != 0 ? sl::eTrue : sl::eFalse; + const sl::DLSSPreset preset = Preset(); + options.dlaaPreset = preset; + options.qualityPreset = preset; + options.balancedPreset = preset; + options.performancePreset = preset; + options.ultraPerformancePreset = preset; + options.ultraQualityPreset = preset; + if (options_changed) { + LOG_INFO(Render_Vulkan, "NVIDIA DLSS (Streamline): {}x{} -> {}x{}, mode {}, preset {}", + in.render_size.width, in.render_size.height, in.upscale_size.width, + in.upscale_size.height, static_cast(options.mode), + static_cast(preset)); + } + + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([this, in, reset, options, options_changed, index = frame_index++, + output_image_handle = *output_image, + output_view_handle = *output_view](vk::CommandBuffer cmdbuf) { + const Streamline::Api& api = Streamline::GetApi(); + const sl::ViewportHandle viewport{0u}; + if (options_changed) { + const auto set = reinterpret_cast(set_options); + if (const sl::Result result = set(viewport, options); result != sl::Result::eOk) { + if (!failed.exchange(true)) { + LOG_ERROR(Render_Vulkan, + "Streamline rejected the DLSS options (result {}); using FSR 3.1", + static_cast(result)); + } + return; + } + } + sl::FrameToken* frame{}; + if (const sl::Result result = api.get_new_frame_token(frame, &index); + result != sl::Result::eOk || !frame) { + if (!failed.exchange(true)) { + LOG_ERROR(Render_Vulkan, "Streamline gave no frame token (result {}); using FSR 3.1", + static_cast(result)); + } + return; + } + + constexpr VkAccessFlags WRITES = + VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT; + // The inputs come from the game's draws and dispatches and stay in GENERAL otherwise; + // DLSS reads them as sampled images. + const std::array to_read{ + Barrier(in.color, WRITES, VK_ACCESS_SHADER_READ_BIT, VK_IMAGE_LAYOUT_GENERAL, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL), + Barrier(in.depth, WRITES, VK_ACCESS_SHADER_READ_BIT, VK_IMAGE_LAYOUT_GENERAL, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL), + Barrier(in.motion_vectors, WRITES, VK_ACCESS_SHADER_READ_BIT, + VK_IMAGE_LAYOUT_GENERAL, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL), + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, {}, {}, to_read); + + // DLSS super resolution only reads the jitter, motion and depth conventions; the camera + // fields are for frame generation, which this doesn't use. + const f32 rw = static_cast(in.render_size.width); + const f32 rh = static_cast(in.render_size.height); + sl::Constants constants{}; + constants.cameraViewToClip = Identity(); + constants.clipToCameraView = Identity(); + constants.clipToLensClip = Identity(); + constants.clipToPrevClip = Identity(); + constants.prevClipToClip = Identity(); + constants.jitterOffset = {in.jitter[0], in.jitter[1]}; + // Streamline wants motion in [-1,1] screen units; ours turns into render pixels. + constants.mvecScale = {in.motion_vector_scale[0] / rw, in.motion_vector_scale[1] / rh}; + constants.cameraPinholeOffset = {0.0f, 0.0f}; + constants.cameraPos = {0.0f, 0.0f, 0.0f}; + constants.cameraUp = {0.0f, 1.0f, 0.0f}; + constants.cameraRight = {1.0f, 0.0f, 0.0f}; + constants.cameraFwd = {0.0f, 0.0f, 1.0f}; + constants.cameraNear = 0.1f; + constants.cameraFar = 1000.0f; + constants.cameraFOV = 1.0f; + constants.cameraAspectRatio = rw / rh; + constants.depthInverted = sl::eTrue; + constants.cameraMotionIncluded = sl::eTrue; + constants.motionVectors3D = sl::eFalse; + constants.reset = reset ? sl::eTrue : sl::eFalse; + constants.motionVectorsJittered = sl::eFalse; + api.set_constants(constants, *frame, viewport); + + sl::Resource color = Resource(in.color.image, in.color.view, in.color.format, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, in.render_size, + VK_IMAGE_USAGE_SAMPLED_BIT); + sl::Resource depth = Resource(in.depth.image, in.depth.view, in.depth.format, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, in.render_size, + VK_IMAGE_USAGE_SAMPLED_BIT); + sl::Resource motion = Resource(in.motion_vectors.image, in.motion_vectors.view, + in.motion_vectors.format, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, in.render_size, + VK_IMAGE_USAGE_SAMPLED_BIT); + sl::Resource output = Resource(output_image_handle, output_view_handle, OUTPUT_FORMAT, + VK_IMAGE_LAYOUT_GENERAL, in.upscale_size, OUTPUT_USAGE); + const sl::Extent render_area{0, 0, in.render_size.width, in.render_size.height}; + const sl::Extent output_area{0, 0, in.upscale_size.width, in.upscale_size.height}; + // Local tags: everything stays put until the evaluation is recorded. + sl::ResourceTag color_tag{&color, sl::kBufferTypeScalingInputColor, + sl::ResourceLifecycle::eValidUntilEvaluate, &render_area}; + sl::ResourceTag depth_tag{&depth, sl::kBufferTypeDepth, + sl::ResourceLifecycle::eValidUntilEvaluate, &render_area}; + sl::ResourceTag motion_tag{&motion, sl::kBufferTypeMotionVectors, + sl::ResourceLifecycle::eValidUntilEvaluate, &render_area}; + sl::ResourceTag output_tag{&output, sl::kBufferTypeScalingOutputColor, + sl::ResourceLifecycle::eValidUntilEvaluate, &output_area}; + const sl::BaseStructure* inputs[] = {&viewport, &color_tag, &depth_tag, &motion_tag, + &output_tag}; + const sl::Result result = + api.evaluate_feature(sl::kFeatureDLSS, *frame, inputs, std::size(inputs), + reinterpret_cast(*cmdbuf)); + if (result != sl::Result::eOk && !failed.exchange(true)) { + LOG_ERROR(Render_Vulkan, "NVIDIA DLSS evaluation failed (Streamline result {}); " + "using FSR 3.1 from now on", + static_cast(result)); + } + + const std::array to_general{ + Barrier(in.color, VK_ACCESS_SHADER_READ_BIT, 0, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_IMAGE_LAYOUT_GENERAL), + Barrier(in.depth, VK_ACCESS_SHADER_READ_BIT, 0, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_IMAGE_LAYOUT_GENERAL), + Barrier(in.motion_vectors, VK_ACCESS_SHADER_READ_BIT, 0, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_IMAGE_LAYOUT_GENERAL), + }; + const VkMemoryBarrier output_written{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, output_written, {}, + to_general); + }); + // DLSS binds its own pipelines and descriptors on the command buffer. + scheduler.InvalidateState(); + return true; +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_streamline_dlss.h b/src/video_core/renderer_vulkan/vk_streamline_dlss.h new file mode 100644 index 0000000000..9f8704fe15 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_streamline_dlss.h @@ -0,0 +1,80 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// NVIDIA DLSS (through NVIDIA Streamline) in place of a game's DLSS +// evaluation. NVIDIA's DLLs come from the user's PC; see vulkan_common/streamline.h. + +#pragma once + +#include +#include + +#include "common/common_types.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" + +namespace Vulkan { + +class Device; +class MemoryAllocator; +class Scheduler; + +class StreamlineDlssUpscaler { +public: + struct Image { + VkImageView view; + VkImage image; + VkFormat format; + VkImageSubresourceRange range; ///< The view's level and layer, with the view's aspect. + VkImageAspectFlags barrier_aspect; ///< Every aspect of the image's format. + }; + + struct Inputs { + Image color; ///< Jittered linear colour at the render size. + Image depth; ///< Device depth (reversed) at the render size. + Image motion_vectors; ///< Motion to the previous frame at the render size. + VkExtent2D render_size; + VkExtent2D upscale_size; + /// This frame's sample offset in render pixels, as the game hands its own DLSS. + std::array jitter; + /// Turns a motion vector into render pixels (the render size for UV offsets). + std::array motion_vector_scale; + bool reset; + }; + + /// Throws vk::Exception when Streamline has no DLSS for this device. + explicit StreamlineDlssUpscaler(const Device& device, Scheduler& scheduler, + MemoryAllocator& memory_allocator); + ~StreamlineDlssUpscaler(); + + /// Records one evaluation into the output image (RGBA16F at the upscale size, GENERAL + /// layout). Returns false when DLSS can't run, so the caller can fall back to FSR 3.1. + bool Upscale(const Inputs& inputs); + + [[nodiscard]] VkImage OutputImage() const noexcept { + return *output_image; + } + [[nodiscard]] VkImageView OutputView() const noexcept { + return *output_view; + } + /// Nearest-neighbour clamp sampler (normalized coordinates), for reading the output. + [[nodiscard]] VkSampler PointSampler() const noexcept { + return *point_sampler; + } + +private: + void CreateOutput(VkExtent2D upscale_size); + + const Device& device; + Scheduler& scheduler; + MemoryAllocator& memory_allocator; + void* set_options{}; ///< slDLSSSetOptions + VkExtent2D render_extent{}; + VkExtent2D upscale_extent{}; + u32 frame_index{}; + vk::Image output_image; + vk::ImageView output_view; + vk::Sampler point_sampler; + std::atomic failed{}; +}; + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index e3f427dead..d7093997c4 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -39,6 +39,9 @@ #include "video_core/texture_cache/util.h" #include "video_core/vulkan_common/vulkan_device.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" +#ifdef EGLANTINE_STREAMLINE +#include "video_core/vulkan_common/streamline.h" +#endif #include "video_core/vulkan_common/vulkan_wrapper.h" #include "video_core/textures/decoders.h" @@ -1556,6 +1559,21 @@ bool TextureCacheRuntime::UpscaleWithFsr3(Framebuffer* dst_framebuffer, ImageVie if (!WantsTemporalUpscaleDepth()) { return false; } + const VkExtent2D extent = dst_framebuffer->RenderArea(); + // A gap in the evaluations (a menu, a movie, a load) starts the history again. + const auto now = std::chrono::steady_clock::now(); + const f32 frame_time_ms = + fsr3_last_evaluation + ? std::chrono::duration(now - *fsr3_last_evaluation).count() + : 1000.0f / 30.0f; + const bool reset = !fsr3_last_evaluation || frame_time_ms > 250.0f; + fsr3_last_evaluation = now; +#ifdef EGLANTINE_STREAMLINE + if (UpscaleWithDlss(dst_framebuffer, input_color, motion_vectors, depth, render_size, jitter, + motion_scale, reset)) { + return true; + } +#endif if (!fsr3_upscaler) { if (!Fsr3Upscaler::IsSupported(device)) { LOG_WARNING(Render_Vulkan, "FSR 3.1 needs formats or features this device lacks; " @@ -1574,15 +1592,6 @@ bool TextureCacheRuntime::UpscaleWithFsr3(Framebuffer* dst_framebuffer, ImageVie return false; } } - const VkExtent2D extent = dst_framebuffer->RenderArea(); - // A gap in the evaluations (a menu, a movie, a load) starts the history again. - const auto now = std::chrono::steady_clock::now(); - const f32 frame_time_ms = - fsr3_last_evaluation - ? std::chrono::duration(now - *fsr3_last_evaluation).count() - : 1000.0f / 30.0f; - const bool reset = !fsr3_last_evaluation || frame_time_ms > 250.0f; - fsr3_last_evaluation = now; // FSR's own RCAS, set by the sharpness slider: FSR takes 0-1 for 2 to 0 stops, and the slider // stores stops x 100, so this is the percentage the slider shows; 0% (stored 200) is off. // EDEN_NX2_TEMPORAL_SHARPEN=|off overrides the slider. @@ -1608,15 +1617,28 @@ bool TextureCacheRuntime::UpscaleWithFsr3(Framebuffer* dst_framebuffer, ImageVie .sharpen = sharpen, .sharpness = 1.0f - stops / 2.0f, }); - // Draw the result into the DLSS output, which keeps its own format; the blit leaves FSR's - // output image in SHADER_READ_ONLY_OPTIMAL, so return it to GENERAL for the next frame. + DrawUpscaledOutput(dst_framebuffer, fsr3_upscaler->OutputView(), fsr3_upscaler->OutputImage(), + fsr3_upscaler->PointSampler()); + static bool logged{}; + if (!logged) { + logged = true; + LOG_INFO(Render_Vulkan, "DLSS replaced by FSR 3.1: {}x{} -> {}x{}", render_size[0], + render_size[1], extent.width, extent.height); + } + return true; +} + +void TextureCacheRuntime::DrawUpscaledOutput(Framebuffer* dst_framebuffer, VkImageView view, + VkImage output, VkSampler sampler) { + // Draw the result into the DLSS output, which keeps its own format; the blit leaves the + // upscaler's output image in SHADER_READ_ONLY_OPTIMAL, so return it to GENERAL for the next + // frame. + const VkExtent2D extent = dst_framebuffer->RenderArea(); const Region2D region{ .start = {0, 0}, .end = {static_cast(extent.width), static_cast(extent.height)}, }; - const VkImage output = fsr3_upscaler->OutputImage(); - blit_image_helper.BlitColor(dst_framebuffer, fsr3_upscaler->OutputView(), output, - fsr3_upscaler->PointSampler(), region, region, + blit_image_helper.BlitColor(dst_framebuffer, view, output, sampler, region, region, {extent.width, extent.height, 1}); scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([output](vk::CommandBuffer cmdbuf) { @@ -1641,14 +1663,84 @@ bool TextureCacheRuntime::UpscaleWithFsr3(Framebuffer* dst_framebuffer, ImageVie cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, barrier); }); +} + +#ifdef EGLANTINE_STREAMLINE +bool TextureCacheRuntime::UpscaleWithDlss(Framebuffer* dst_framebuffer, ImageView& input_color, + ImageView& motion_vectors, ImageView& depth, + std::array render_size, + std::array jitter, + std::array motion_scale, bool reset) { + // EDEN_NX2_DLSS_REPLACEMENT=fsr keeps FSR 3.1 for A/B runs. + static const bool force_fsr = [] { + const char* const value = std::getenv("EDEN_NX2_DLSS_REPLACEMENT"); + return value && std::string_view{value} == "fsr"; + }(); + if (force_fsr || dlss_unsupported) { + return false; + } + if (!dlss_upscaler) { + // Off in the settings, or Streamline didn't start: it has said why. + if (!Streamline::IsDlssAvailable()) { + dlss_unsupported = true; + return false; + } + try { + dlss_upscaler = + std::make_unique(device, scheduler, memory_allocator); + } catch (const vk::Exception& exception) { + LOG_WARNING(Render_Vulkan, "NVIDIA DLSS unavailable ({}); using FSR 3.1", + exception.what()); + dlss_unsupported = true; + return false; + } + } + const auto image = [&](ImageView& view, VkImageView handle, VkImageAspectFlags aspect) { + return StreamlineDlssUpscaler::Image{ + .view = handle, + .image = view.ImageHandle(), + .format = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, view.format) + .format, + .range{ + .aspectMask = aspect, + .baseMipLevel = static_cast(view.range.base.level), + .levelCount = 1, + .baseArrayLayer = static_cast(view.range.base.layer), + .layerCount = 1, + }, + .barrier_aspect = ImageAspectMask(view.format), + }; + }; + const VkExtent2D extent = dst_framebuffer->RenderArea(); + const bool upscaled = dlss_upscaler->Upscale({ + .color = image(input_color, input_color.Handle(Shader::TextureType::Color2D), + VK_IMAGE_ASPECT_COLOR_BIT), + .depth = image(depth, depth.DepthView(), VK_IMAGE_ASPECT_DEPTH_BIT), + .motion_vectors = image(motion_vectors, + motion_vectors.Handle(Shader::TextureType::Color2D), + VK_IMAGE_ASPECT_COLOR_BIT), + .render_size = {render_size[0], render_size[1]}, + .upscale_size = extent, + .jitter = jitter, + .motion_vector_scale = {motion_scale[0] * static_cast(render_size[0]), + motion_scale[1] * static_cast(render_size[1])}, + .reset = reset, + }); + if (!upscaled) { + dlss_unsupported = true; + return false; + } + DrawUpscaledOutput(dst_framebuffer, dlss_upscaler->OutputView(), dlss_upscaler->OutputImage(), + dlss_upscaler->PointSampler()); static bool logged{}; if (!logged) { logged = true; - LOG_INFO(Render_Vulkan, "DLSS replaced by FSR 3.1: {}x{} -> {}x{}", render_size[0], + LOG_INFO(Render_Vulkan, "DLSS replaced by NVIDIA DLSS: {}x{} -> {}x{}", render_size[0], render_size[1], extent.width, extent.height); } return true; } +#endif bool TextureCacheRuntime::TemporalUpscale(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& input_color, ImageView& motion_vectors, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index e7dbb97628..44bd21615b 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -17,6 +17,9 @@ #include "shader_recompiler/shader_info.h" #include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_fsr3_upscaler.h" +#ifdef EGLANTINE_STREAMLINE +#include "video_core/renderer_vulkan/vk_streamline_dlss.h" +#endif #include "video_core/renderer_vulkan/vk_render_pass_cache.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/texture_cache/image_view_base.h" @@ -208,6 +211,20 @@ public: bool fsr3_unsupported{}; std::optional fsr3_last_evaluation; + /// Draws an upscaler's RGBA16F output into the DLSS output and returns it to GENERAL. + void DrawUpscaledOutput(Framebuffer* dst_framebuffer, VkImageView view, VkImage image, + VkSampler sampler); +#ifdef EGLANTINE_STREAMLINE + /// NVIDIA's DLSS (through NVIDIA Streamline) in place of the game's, + /// before FSR 3.1, when the setting is on. Returns false to fall back to FSR 3.1. + bool UpscaleWithDlss(Framebuffer* dst_framebuffer, ImageView& input_color, + ImageView& motion_vectors, ImageView& depth, + std::array render_size, std::array jitter, + std::array motion_scale, bool reset); + std::unique_ptr dlss_upscaler; + bool dlss_unsupported{}; +#endif + /// TemporalUpscale's history: its previous result (images[current]) and the image the next /// result is stored to. struct TemporalHistory { diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index ea82ab02e7..a34c591aca 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -1124,8 +1124,8 @@ bool TextureCache

::TemporalUpscale(const Tegra::Engines::Fermi2D::Surface& ds has_deleted_images = false; const ImageId motion_id = FindOrInsertImage(motion_info, motion_vectors.Address(), RelaxedOptions{}); - // The depth only feeds FSR 3.1 (looked up only when FSR can run); without it the stand-in - // runs. + // The depth only feeds NVIDIA DLSS and FSR 3.1 (looked up only when they can run); without + // it the stand-in runs. ImageId depth_id{}; if (depth && motion_id && runtime.WantsTemporalUpscaleDepth()) { depth_id = FindOrInsertImage(ImageInfo(*depth), depth->Address(), RelaxedOptions{}); diff --git a/src/video_core/vulkan_common/streamline.cpp b/src/video_core/vulkan_common/streamline.cpp new file mode 100644 index 0000000000..966de82cb5 --- /dev/null +++ b/src/video_core/vulkan_common/streamline.cpp @@ -0,0 +1,463 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// NVIDIA DLSS through NVIDIA Streamline, with Streamline's manual hooking. + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#include "common/fs/path_util.h" +#include "common/logging.h" +#include "common/settings.h" +#include "video_core/vulkan_common/streamline.h" +#include "video_core/vulkan_common/streamline_api.h" + +#include + +#include +#include + +namespace Vulkan::Streamline { +namespace { + +// A custom engine names itself with its own project id; an application id from NVIDIA is only +// needed without an engine type. Keep it fixed: the driver may key behaviour off it. +constexpr const char* PROJECT_ID = "8d3e2f9a-6b41-4c7e-9a52-1f0c3b7d5e64"; +constexpr const wchar_t* INTERPOSER = L"sl.interposer.dll"; +constexpr const wchar_t* MODEL = L"nvngx_dlss.dll"; +// The first DLSS with preset K. +constexpr u32 MIN_DLSS_MAJOR = 310; + +using Version = std::array; + +struct State { + HMODULE module{}; + Api api{}; + std::wstring plugin_dir; + std::wstring model_dir; + std::vector instance_extensions; + std::vector device_extensions; + bool started{}; + bool attached{}; +}; + +State& GetState() { + static State state; + return state; +} + +int LogLevel() { + const char* const value = std::getenv("EDEN_NX2_DLSS_LOG"); + return value ? std::atoi(value) : 0; +} + +void LogFromStreamline(sl::LogType type, const char* message) { + std::string_view text{message ? message : ""}; + while (!text.empty() && (text.back() == '\n' || text.back() == '\r')) { + text.remove_suffix(1); + } + switch (type) { + case sl::LogType::eError: + LOG_ERROR(Render_Vulkan, "Streamline: {}", text); + break; + case sl::LogType::eWarn: + LOG_WARNING(Render_Vulkan, "Streamline: {}", text); + break; + default: + if (LogLevel() > 0) { + LOG_INFO(Render_Vulkan, "Streamline: {}", text); + } + break; + } +} + +/// Where the user may put sl.interposer.dll: EDEN_STREAMLINE_DIR, the user folder's "streamline" +/// folder, then a "streamline" folder next to the executable. +std::vector SearchDirs() { + std::vector dirs; + if (const char* const dir = std::getenv("EDEN_STREAMLINE_DIR"); dir && *dir) { + dirs.emplace_back(dir); + } + dirs.push_back(UserFolder()); + dirs.push_back(Common::FS::GetExeDirectory() / "streamline"); + return dirs; +} + +template +bool Resolve(HMODULE module, const char* name, T*& function) { + function = reinterpret_cast(GetProcAddress(module, name)); + if (!function) { + LOG_ERROR(Render_Vulkan, "sl.interposer.dll has no {}", name); + } + return function != nullptr; +} + +std::optional ReadVersion(const std::filesystem::path& path) { + DWORD handle = 0; + const DWORD size = GetFileVersionInfoSizeW(path.c_str(), &handle); + if (size == 0) { + return std::nullopt; + } + std::vector data(size); + VS_FIXEDFILEINFO* info{}; + UINT length = 0; + if (!GetFileVersionInfoW(path.c_str(), 0, size, data.data()) || + !VerQueryValueW(data.data(), L"\\", reinterpret_cast(&info), &length) || !info) { + return std::nullopt; + } + return Version{HIWORD(info->dwFileVersionMS), LOWORD(info->dwFileVersionMS), + HIWORD(info->dwFileVersionLS), LOWORD(info->dwFileVersionLS)}; +} + +std::string VersionText(const std::filesystem::path& path) { + const std::optional version = path.empty() ? std::nullopt : ReadVersion(path); + if (!version) { + return {}; + } + const Version& v = *version; + return v[3] == 0 ? fmt::format("{}.{}.{}", v[0], v[1], v[2]) + : fmt::format("{}.{}.{}.{}", v[0], v[1], v[2], v[3]); +} + +/// The NVIDIA app keeps every DLSS it downloads under ProgramData; pick the newest. +std::filesystem::path FindAppModel() { + const wchar_t* const program_data = _wgetenv(L"ProgramData"); + if (!program_data) { + return {}; + } + const std::filesystem::path root = + std::filesystem::path{program_data} / L"NVIDIA" / L"NGX" / L"models" / L"dlss" / L"versions"; + std::filesystem::path best_path; + Version best{}; + std::error_code ec; + for (const auto& version_dir : std::filesystem::directory_iterator(root, ec)) { + std::error_code files_ec; + for (const auto& file : + std::filesystem::directory_iterator(version_dir.path() / L"files", files_ec)) { + if (file.path().extension() != L".bin") { + continue; + } + const std::optional version = ReadVersion(file.path()); + if (version && (*version)[0] >= MIN_DLSS_MAJOR && *version > best) { + best = *version; + best_path = file.path(); + } + } + } + return best_path; +} + +/// Copies the NVIDIA app's DLSS into our folder as nvngx_dlss.dll, the name Streamline loads. +std::filesystem::path UseAppModel(const DlssFiles& files) { + const std::filesystem::path dir = UserFolder() / L"nvidia-app"; + const std::filesystem::path target = dir / MODEL; + std::error_code ec; + const bool current = std::filesystem::is_regular_file(target, ec) && + std::filesystem::file_size(target, ec) == + std::filesystem::file_size(files.app_model, ec) && + ReadVersion(target) == ReadVersion(files.app_model); + if (!current) { + if (!IsSignedByNvidia(files.app_model)) { + LOG_WARNING(Render_Vulkan, "The NVIDIA app's DLSS {} is not signed by NVIDIA", + files.app_model.string()); + return {}; + } + std::filesystem::create_directories(dir, ec); + std::filesystem::copy_file(files.app_model, target, + std::filesystem::copy_options::overwrite_existing, ec); + if (ec) { + LOG_WARNING(Render_Vulkan, "Could not copy the NVIDIA app's DLSS to {}: {}", + target.string(), ec.message()); + return {}; + } + } + return dir; +} + +bool Load(State& state, const std::filesystem::path& path) { + if (!IsStreamlineModule(path)) { + LOG_ERROR(Render_Vulkan, "{} is not signed by NVIDIA; not loading it", + path.string()); + return false; + } + state.module = LoadLibraryW(path.c_str()); + if (!state.module) { + LOG_ERROR(Render_Vulkan, "Could not load {} (error {})", path.string(), GetLastError()); + return false; + } + Api& api = state.api; + const HMODULE m = state.module; + if (Resolve(m, "slInit", api.init) && Resolve(m, "slShutdown", api.shutdown) && + Resolve(m, "slIsFeatureSupported", api.is_feature_supported) && + Resolve(m, "slGetFeatureRequirements", api.get_feature_requirements) && + Resolve(m, "slGetFeatureVersion", api.get_feature_version) && + Resolve(m, "slGetFeatureFunction", api.get_feature_function) && + Resolve(m, "slGetNewFrameToken", api.get_new_frame_token) && + Resolve(m, "slSetConstants", api.set_constants) && + Resolve(m, "slEvaluateFeature", api.evaluate_feature) && + Resolve(m, "slFreeResources", api.free_resources) && + Resolve(m, "slSetVulkanInfo", api.set_vulkan_info)) { + return true; + } + FreeLibrary(state.module); + state.module = nullptr; + state.api = {}; + return false; +} + +std::vector Names(uint32_t count, const char** names) { + std::vector result; + for (uint32_t i = 0; i < count; ++i) { + result.emplace_back(names[i]); + } + return result; +} + +} // Anonymous namespace + +std::filesystem::path UserFolder() { + return Common::FS::GetEdenPath(Common::FS::EdenPath::StreamlineDir); +} + +bool IsStreamlineModule(const std::filesystem::path& path) { + return sl::security::verifyEmbeddedSignature(path.c_str()); +} + +bool IsSignedByNvidia(const std::filesystem::path& path) { + const std::wstring file_path = std::filesystem::absolute(path).wstring(); + WINTRUST_FILE_INFO file{}; + file.cbStruct = sizeof(file); + file.pcwszFilePath = file_path.c_str(); + WINTRUST_DATA data{}; + data.cbStruct = sizeof(data); + data.dwUIChoice = WTD_UI_NONE; + data.fdwRevocationChecks = WTD_REVOKE_NONE; + data.dwUnionChoice = WTD_CHOICE_FILE; + data.pFile = &file; + data.dwStateAction = WTD_STATEACTION_VERIFY; + data.dwProvFlags = WTD_CACHE_ONLY_URL_RETRIEVAL; + GUID action = WINTRUST_ACTION_GENERIC_VERIFY_V2; + bool valid = WinVerifyTrust(nullptr, &action, &data) == ERROR_SUCCESS; + if (valid) { + CRYPT_PROVIDER_DATA* const provider = WTHelperProvDataFromStateData(data.hWVTStateData); + CRYPT_PROVIDER_SGNR* const signer = + provider ? WTHelperGetProvSignerFromChain(provider, 0, FALSE, 0) : nullptr; + CRYPT_PROVIDER_CERT* const cert = signer ? WTHelperGetProvCertFromChain(signer, 0) : nullptr; + wchar_t name[128]{}; + valid = cert && + CertGetNameStringW(cert->pCert, CERT_NAME_SIMPLE_DISPLAY_TYPE, 0, nullptr, name, + static_cast(std::size(name))) > 1 && + std::wstring_view{name} == L"NVIDIA Corporation"; + } + data.dwStateAction = WTD_STATEACTION_CLOSE; + WinVerifyTrust(nullptr, &action, &data); + return valid; +} + +DlssFiles FindDlssFiles() { + DlssFiles files; + files.folder = UserFolder(); + std::error_code ec; + for (const std::filesystem::path& dir : SearchDirs()) { + if (std::filesystem::is_regular_file(dir / INTERPOSER, ec)) { + files.folder = dir; + files.interposer = dir / INTERPOSER; + break; + } + } + if (std::filesystem::is_regular_file(files.folder / MODEL, ec)) { + files.model = files.folder / MODEL; + } + files.app_model = FindAppModel(); + files.interposer_version = VersionText(files.interposer); + files.model_version = VersionText(files.model); + files.app_model_version = VersionText(files.app_model); + return files; +} + +bool Start() { + State& state = GetState(); + if (state.started) { + return true; + } + if (!Settings::values.nvidia_dlss.GetValue()) { + return false; + } + if (!state.module) { + const DlssFiles files = FindDlssFiles(); + if (files.interposer.empty()) { + LOG_WARNING(Render_Vulkan, + "NVIDIA DLSS is on, but sl.interposer.dll (NVIDIA Streamline) is not in " + "{} (the Graphics settings can download it); using FSR 3.1", + files.folder.string()); + return false; + } + if (!Load(state, files.interposer)) { + return false; + } + state.plugin_dir = files.folder.wstring(); + state.model_dir.clear(); + if (files.model.empty() && !files.app_model.empty()) { + if (const std::filesystem::path dir = UseAppModel(files); !dir.empty()) { + state.model_dir = dir.wstring(); + LOG_INFO(Render_Vulkan, "Using the NVIDIA app's DLSS {}", files.app_model_version); + } + } else if (files.model.empty()) { + LOG_WARNING(Render_Vulkan, "No nvngx_dlss.dll in {} and no DLSS from the NVIDIA app", + files.folder.string()); + } + } + + // Streamline's plugins come from the loader's folder or the NVIDIA app's downloads (OTA); the + // DLSS model from the loader's folder, else the NVIDIA app's copy. + const wchar_t* plugin_paths[] = {state.plugin_dir.c_str(), state.model_dir.c_str()}; + const sl::Feature features[] = {sl::kFeatureDLSS}; + sl::Preferences preferences{}; + preferences.logLevel = LogLevel() >= 2 ? sl::LogLevel::eVerbose : sl::LogLevel::eDefault; + preferences.logMessageCallback = LogFromStreamline; + preferences.pathsToPlugins = plugin_paths; + preferences.numPathsToPlugins = state.model_dir.empty() ? 1 : 2; + preferences.flags = sl::PreferenceFlags::eDisableCLStateTracking | + sl::PreferenceFlags::eUseManualHooking | sl::PreferenceFlags::eAllowOTA | + sl::PreferenceFlags::eLoadDownloadedPlugins; + preferences.featuresToLoad = features; + preferences.numFeaturesToLoad = 1; + preferences.engine = sl::EngineType::eCustom; + preferences.engineVersion = "1.0"; + preferences.projectId = PROJECT_ID; + preferences.renderAPI = sl::RenderAPI::eVulkan; + const Api& api = state.api; + if (const sl::Result result = api.init(preferences, sl::kSDKVersion); + result != sl::Result::eOk) { + LOG_ERROR(Render_Vulkan, "NVIDIA Streamline failed to start (result {}); using FSR 3.1", + static_cast(result)); + return false; + } + sl::FeatureRequirements requirements{}; + if (const sl::Result result = api.get_feature_requirements(sl::kFeatureDLSS, requirements); + result != sl::Result::eOk) { + LOG_WARNING(Render_Vulkan, + "Streamline found no DLSS plugin (result {}); put sl.dlss.dll next to " + "sl.interposer.dll or install the NVIDIA app. Using FSR 3.1", + static_cast(result)); + api.shutdown(); + return false; + } + if (requirements.vkNumComputeQueuesRequired != 0 || + requirements.vkNumGraphicsQueuesRequired != 0) { + LOG_WARNING(Render_Vulkan, "This DLSS wants extra Vulkan queues; using FSR 3.1"); + api.shutdown(); + return false; + } + state.instance_extensions = + Names(requirements.vkNumInstanceExtensions, requirements.vkInstanceExtensions); + state.device_extensions = + Names(requirements.vkNumDeviceExtensions, requirements.vkDeviceExtensions); + // The KHR and EXT buffer device address extensions can't both be enabled. + const auto& device_extensions = state.device_extensions; + if (std::ranges::find(device_extensions, VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME) != + device_extensions.end()) { + std::erase(state.device_extensions, VK_EXT_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME); + } + sl::FeatureVersion version{}; + api.get_feature_version(sl::kFeatureDLSS, version); + LOG_INFO(Render_Vulkan, "NVIDIA Streamline {} started from {} (DLSS {})", + version.versionSL.toStr(), std::filesystem::path{state.plugin_dir}.string(), + version.versionNGX.toStr()); + state.started = true; + return true; +} + +bool WantsDevice() { + const State& state = GetState(); + return state.started && !state.attached; +} + +std::vector InstanceExtensions() { + const State& state = GetState(); + return state.started ? state.instance_extensions : std::vector{}; +} + +std::vector DeviceExtensions() { + const State& state = GetState(); + return state.started ? state.device_extensions : std::vector{}; +} + +bool AttachDevice(VkInstance instance, VkPhysicalDevice physical, VkDevice device, + u32 graphics_family, vk::DeviceDispatch& dld) { + State& state = GetState(); + if (!WantsDevice()) { + return false; + } + sl::VulkanInfo info{}; + info.device = device; + info.instance = instance; + info.physicalDevice = physical; + info.computeQueueFamily = graphics_family; + info.graphicsQueueFamily = graphics_family; + if (const sl::Result result = state.api.set_vulkan_info(info); result != sl::Result::eOk) { + LOG_ERROR(Render_Vulkan, "Streamline did not take the Vulkan device (result {})", + static_cast(result)); + Stop(); + return false; + } + sl::AdapterInfo adapter{}; + adapter.vkPhysicalDevice = physical; + if (const sl::Result result = state.api.is_feature_supported(sl::kFeatureDLSS, adapter); + result != sl::Result::eOk) { + LOG_WARNING(Render_Vulkan, + "NVIDIA DLSS is not supported on this GPU or driver (Streamline result {}); " + "using FSR 3.1", + static_cast(result)); + Stop(); + return false; + } + // Manual hooking: Streamline does its per-frame work in these. + const auto hook = [&](auto& function, const char* name) { + if (const FARPROC proxy = GetProcAddress(state.module, name)) { + function = reinterpret_cast>(proxy); + } else { + LOG_WARNING(Render_Vulkan, "sl.interposer.dll has no {}", name); + } + }; + hook(dld.vkCreateSwapchainKHR, "vkCreateSwapchainKHR"); + hook(dld.vkDestroySwapchainKHR, "vkDestroySwapchainKHR"); + hook(dld.vkGetSwapchainImagesKHR, "vkGetSwapchainImagesKHR"); + hook(dld.vkAcquireNextImageKHR, "vkAcquireNextImageKHR"); + hook(dld.vkQueuePresentKHR, "vkQueuePresentKHR"); + hook(dld.vkDeviceWaitIdle, "vkDeviceWaitIdle"); + state.attached = true; + LOG_INFO(Render_Vulkan, "NVIDIA DLSS is available through Streamline"); + return true; +} + +bool IsDlssAvailable() { + const State& state = GetState(); + return state.started && state.attached; +} + +void Stop() { + State& state = GetState(); + if (!state.started) { + return; + } + const sl::Result result = state.api.shutdown(); + state.started = false; + state.attached = false; + LOG_INFO(Render_Vulkan, "NVIDIA Streamline stopped (result {})", static_cast(result)); +} + +const Api& GetApi() { + return GetState().api; +} + +} // namespace Vulkan::Streamline diff --git a/src/video_core/vulkan_common/streamline.h b/src/video_core/vulkan_common/streamline.h new file mode 100644 index 0000000000..47e55f7150 --- /dev/null +++ b/src/video_core/vulkan_common/streamline.h @@ -0,0 +1,63 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// NVIDIA DLSS through NVIDIA Streamline. Only Streamline's MIT-licensed +// headers are built in: NVIDIA's DLLs are loaded at run time from the user's PC, and none ship +// with the emulator. + +#pragma once + +#include +#include +#include + +#include "common/common_types.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" + +namespace Vulkan::Streamline { + +/// NVIDIA's files DLSS needs, as found on this PC. Paths are empty when missing. +struct DlssFiles { + std::filesystem::path folder; ///< Where the loader is, or where it should go + std::filesystem::path interposer; ///< sl.interposer.dll (Streamline's loader) + std::filesystem::path model; ///< nvngx_dlss.dll next to the loader + std::filesystem::path app_model; ///< Newest DLSS the NVIDIA app downloaded + std::string interposer_version; + std::string model_version; + std::string app_model_version; +}; + +[[nodiscard]] DlssFiles FindDlssFiles(); + +/// The user folder's "streamline" folder. +[[nodiscard]] std::filesystem::path UserFolder(); + +/// Streamline's own modules carry a second, Streamline-only signature. +[[nodiscard]] bool IsStreamlineModule(const std::filesystem::path& path); + +/// A valid Authenticode signature from NVIDIA Corporation (DLSS models). +[[nodiscard]] bool IsSignedByNvidia(const std::filesystem::path& path); + +/// When the NVIDIA DLSS setting is on, loads the user's sl.interposer.dll (checking NVIDIA's +/// signature) and starts Streamline for DLSS. Call before creating the renderer's instance. +bool Start(); + +/// Whether Streamline is running and waiting for the renderer's device. +[[nodiscard]] bool WantsDevice(); + +/// Instance and device extensions DLSS asks for (empty when Streamline isn't running). +[[nodiscard]] std::vector InstanceExtensions(); +[[nodiscard]] std::vector DeviceExtensions(); + +/// Hands the renderer's device to Streamline and routes the swapchain calls through it, as +/// Streamline's manual hooking needs. Stops Streamline and returns false when DLSS can't run. +bool AttachDevice(VkInstance instance, VkPhysicalDevice physical, VkDevice device, + u32 graphics_family, vk::DeviceDispatch& dld); + +/// Whether DLSS can run on the attached device. +[[nodiscard]] bool IsDlssAvailable(); + +/// Shuts Streamline down; call before the device is destroyed. +void Stop(); + +} // namespace Vulkan::Streamline diff --git a/src/video_core/vulkan_common/streamline_api.h b/src/video_core/vulkan_common/streamline_api.h new file mode 100644 index 0000000000..0a82c1dd7d --- /dev/null +++ b/src/video_core/vulkan_common/streamline_api.h @@ -0,0 +1,36 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// Streamline's API as loaded from the user's sl.interposer.dll. Only for the +// Streamline sources: it pulls in and Streamline's (MIT-licensed) headers. + +#pragma once + +#include + +#include "video_core/vulkan_common/vulkan_wrapper.h" + +#include +#include +#include + +namespace Vulkan::Streamline { + +struct Api { + PFun_slInit* init{}; + PFun_slShutdown* shutdown{}; + PFun_slIsFeatureSupported* is_feature_supported{}; + PFun_slGetFeatureRequirements* get_feature_requirements{}; + PFun_slGetFeatureVersion* get_feature_version{}; + PFun_slGetFeatureFunction* get_feature_function{}; + PFun_slGetNewFrameToken* get_new_frame_token{}; + PFun_slSetConstants* set_constants{}; + PFun_slEvaluateFeature* evaluate_feature{}; + PFun_slFreeResources* free_resources{}; + PFun_slSetVulkanInfo* set_vulkan_info{}; +}; + +/// The loaded functions; only valid while IsDlssAvailable(). +[[nodiscard]] const Api& GetApi(); + +} // namespace Vulkan::Streamline diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 01e13cb384..6309599877 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -32,6 +32,9 @@ #include "video_core/vulkan_common/vulkan_device.h" #include "video_core/vulkan_common/vulkan_wrapper.h" #include "video_core/gpu_logging/gpu_logging.h" +#ifdef EGLANTINE_STREAMLINE +#include "video_core/vulkan_common/streamline.h" +#endif #if defined(__ANDROID__) && defined(ARCHITECTURE_arm64) #include @@ -743,6 +746,15 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR RemoveExtensionFeature(extensions.descriptor_buffer, features.descriptor_buffer, VK_EXT_DESCRIPTOR_BUFFER_EXTENSION_NAME); } +#ifdef EGLANTINE_STREAMLINE + // NVIDIA's DLSS needs buffer device addresses: keep the feature enabled for + // it, but leave Eglantine's own buffers as they are without descriptor buffers. + const bool dlss_wants_device_address = + streamline_extensions && features.buffer_device_address.bufferDeviceAddress; + if (!extensions.descriptor_buffer && dlss_wants_device_address) { + extensions.buffer_device_address = false; + } else +#endif if (!extensions.descriptor_buffer) { RemoveExtensionFeature(extensions.buffer_device_address, features.buffer_device_address, VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME); @@ -753,6 +765,15 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR graphics_queue = logical.GetQueue(graphics_family); present_queue = logical.GetQueue(present_family); +#ifdef EGLANTINE_STREAMLINE + if (streamline_extensions) { + streamline_attached = + Streamline::AttachDevice(instance, physical, *logical, graphics_family, dld); + } else if (Streamline::WantsDevice()) { + Streamline::Stop(); + } +#endif + VmaVulkanFunctions functions{}; functions.vkGetInstanceProcAddr = dld.vkGetInstanceProcAddr; functions.vkGetDeviceProcAddr = dld.vkGetDeviceProcAddr; @@ -806,6 +827,11 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR } Device::~Device() { +#ifdef EGLANTINE_STREAMLINE + if (streamline_attached) { + Streamline::Stop(); + } +#endif SaveStaticPipelineCache(); ShutdownGPULogging(); for (auto& pools : separate_pools) { @@ -1123,6 +1149,27 @@ bool Device::GetSuitability(bool requires_swapchain) { extensions.robustness_2 = false; } +#ifdef EGLANTINE_STREAMLINE + // The extensions NVIDIA's DLSS asks for, where the device has them all. + if (Streamline::WantsDevice()) { + const std::vector wanted = Streamline::DeviceExtensions(); + const bool all = + !wanted.empty() && std::ranges::all_of(wanted, [&](const std::string& name) { + return supported_extensions.contains(name); + }); + std::string names; + for (const std::string& name : wanted) { + if (all) { + loaded_extensions.insert(name); + } + names += (names.empty() ? "" : ", ") + name; + } + streamline_extensions = all; + LOG_INFO(Render_Vulkan, "NVIDIA DLSS device extensions: {} ({})", + all ? "enabled" : "unavailable", names); + } +#endif + #undef FEATURE_EXTENSION #undef EXTENSION diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index edc091318c..3115f4547c 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -289,6 +289,11 @@ public: return physical; } + /// Returns the instance the device was created from. + VkInstance GetInstance() const { + return instance; + } + VkPipelineCache StaticPipelineCache() const noexcept { return *static_pipeline_cache; } @@ -1326,6 +1331,8 @@ private: // Telemetry parameters std::set> supported_extensions; ///< Reported Vulkan extensions. std::set> loaded_extensions; ///< Loaded Vulkan extensions. + bool streamline_extensions{}; ///< NVIDIA DLSS's extensions are loaded. + bool streamline_attached{}; ///< Streamline has this device. std::vector valid_heap_memory; ///< Heaps used. size_t main_heap{}; ///< Largest device-local heap, where images live. u64 heap_size_limit{}; ///< EDEN_NX2_VRAM_LIMIT_MIB for device-local heaps, 0 if unset. diff --git a/src/video_core/vulkan_common/vulkan_instance.cpp b/src/video_core/vulkan_common/vulkan_instance.cpp index 7f3d8ed41f..3ab924e37f 100644 --- a/src/video_core/vulkan_common/vulkan_instance.cpp +++ b/src/video_core/vulkan_common/vulkan_instance.cpp @@ -18,6 +18,11 @@ #include "core/frontend/emu_window.h" #include "video_core/vulkan_common/vulkan_instance.h" #include "video_core/vulkan_common/vulkan_wrapper.h" +#ifdef EGLANTINE_STREAMLINE +#include +#include +#include "video_core/vulkan_common/streamline.h" +#endif namespace Vulkan { namespace { @@ -82,6 +87,21 @@ namespace { #endif if (enable_validation && AreExtensionsSupported(dld, properties, std::array{VK_EXT_DEBUG_UTILS_EXTENSION_NAME})) extensions.push_back(VK_EXT_DEBUG_UTILS_EXTENSION_NAME); +#ifdef EGLANTINE_STREAMLINE + // What NVIDIA's DLSS asks for through Streamline, where available. Static: the returned + // pointers must outlive this call. + static std::vector streamline; + streamline = Streamline::InstanceExtensions(); + for (const std::string& name : streamline) { + const bool available = std::ranges::any_of(properties, [&](const VkExtensionProperties& p) { + return name == p.extensionName; + }); + const bool listed = std::ranges::any_of(extensions, [&](const char* e) { return name == e; }); + if (available && !listed) { + extensions.push_back(name.c_str()); + } + } +#endif return extensions; } diff --git a/src/yuzu/configuration/configure_graphics.cpp b/src/yuzu/configuration/configure_graphics.cpp index 074b6f405b..3a1c3bf2c2 100644 --- a/src/yuzu/configuration/configure_graphics.cpp +++ b/src/yuzu/configuration/configure_graphics.cpp @@ -226,6 +226,12 @@ void ConfigureGraphics::Setup(const ConfigurationShared::Builder& builder) { continue; } #endif +#ifndef EGLANTINE_STREAMLINE + // Only builds with NVIDIA Streamline (Windows) can use it. + if (setting->Id() == Settings::values.nvidia_dlss.Id()) { + continue; + } +#endif ConfigurationShared::Widget* widget = [&]() { if (setting->Id() == Settings::values.fsr_sharpening_slider.Id()) { -- cgit v1.2.3 From 98cc44dadd7ab4b63cf0685360ac50738e8e5c3e Mon Sep 17 00:00:00 2001 From: Duck912 <350528965+Duck912@users.noreply.github.com> Date: Tue, 3 Nov 2026 12:13:48 -0500 Subject: yuzu: show NVIDIA's DLSS files in the Graphics tab and fetch them from NVIDIA Under the NVIDIA DLSS checkbox, a panel says whether the Streamline loader and DLSS were found and which versions, with "Open Folder" and a link to NVIDIA's Streamline releases. When files are missing, "Download from NVIDIA" takes only those files out of NVIDIA's Streamline 2.14.16 release zip on GitHub with HTTP range requests (about 1 MB for the loader, 32 MB with DLSS, of the 276 MB zip), after showing the Streamline and DLSS licence texts from the release for acceptance. Each file's CRC and signature are checked before it is saved to the user folder's "streamline" folder. Assisted-by: Claude Opus 5.5 (implementation) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com> --- src/qt_common/config/shared_translation.cpp | 4 +- src/yuzu/CMakeLists.txt | 4 + src/yuzu/configuration/configure_graphics.cpp | 15 + src/yuzu/configuration/nvidia_dlss_panel.cpp | 549 ++++++++++++++++++++++++++ src/yuzu/configuration/nvidia_dlss_panel.h | 34 ++ 5 files changed, 604 insertions(+), 2 deletions(-) create mode 100644 src/yuzu/configuration/nvidia_dlss_panel.cpp create mode 100644 src/yuzu/configuration/nvidia_dlss_panel.h diff --git a/src/qt_common/config/shared_translation.cpp b/src/qt_common/config/shared_translation.cpp index b395b2959c..342cc725c0 100644 --- a/src/qt_common/config/shared_translation.cpp +++ b/src/qt_common/config/shared_translation.cpp @@ -152,8 +152,8 @@ std::unique_ptr InitializeTranslations(QObject* parent) { INSERT(Settings, nvidia_dlss, tr("Use NVIDIA DLSS in place of a game's DLSS"), tr("Uses NVIDIA DLSS on GeForce RTX cards in place of a Switch 2 game's own DLSS, " "instead of AMD FSR 3.1.\nNeeds NVIDIA's Streamline loader, and NVIDIA DLSS unless " - "the NVIDIA app has downloaded it; Eglantine does not ship NVIDIA's files.\nTakes " - "effect when a game starts.")); + "the NVIDIA app has downloaded it; Eglantine does not ship NVIDIA's files. Use the " + "download button below.\nTakes effect when a game starts.")); INSERT(Settings, anti_aliasing, tr("Anti-Aliasing Method:"), tr("The anti-aliasing method to use.\nSMAA offers the best quality.\nFXAA " "can produce a more stable picture in lower resolutions.")); diff --git a/src/yuzu/CMakeLists.txt b/src/yuzu/CMakeLists.txt index f5895edd10..f05da88c40 100644 --- a/src/yuzu/CMakeLists.txt +++ b/src/yuzu/CMakeLists.txt @@ -74,6 +74,8 @@ add_executable(yuzu configuration/configure_general.ui configuration/configure_graphics.cpp configuration/configure_graphics.h + configuration/nvidia_dlss_panel.cpp + configuration/nvidia_dlss_panel.h configuration/configure_graphics.ui configuration/configure_graphics_advanced.cpp configuration/configure_graphics_advanced.h @@ -407,6 +409,8 @@ endif() target_link_libraries(yuzu PRIVATE nlohmann_json::nlohmann_json) target_link_libraries(yuzu PRIVATE common core input_common frontend_common network video_core qt_common) target_link_libraries(yuzu PRIVATE Boost::headers Qt6::Widgets Qt6::Charts Qt6::Concurrent) +# Unpacks NVIDIA's files from the Streamline release (nvidia_dlss_panel.cpp). +target_link_libraries(yuzu PRIVATE ZLIB::ZLIB) target_link_libraries(yuzu PRIVATE ${PLATFORM_LIBRARIES} Threads::Threads) if (ENABLE_OPENGL) target_link_libraries(yuzu PRIVATE glad) diff --git a/src/yuzu/configuration/configure_graphics.cpp b/src/yuzu/configuration/configure_graphics.cpp index 3a1c3bf2c2..09ae6adf6e 100644 --- a/src/yuzu/configuration/configure_graphics.cpp +++ b/src/yuzu/configuration/configure_graphics.cpp @@ -35,6 +35,7 @@ #include "ui_configure_graphics.h" #include "yuzu/configuration/configuration_shared.h" #include "yuzu/configuration/configure_graphics.h" +#include "yuzu/configuration/nvidia_dlss_panel.h" #include "yuzu/configuration/shared_widget.h" #ifdef HAS_RESHADE @@ -328,6 +329,20 @@ void ConfigureGraphics::Setup(const ConfigurationShared::Builder& builder) { row->insertWidget(1, post_shader_button); hold_graphics.emplace(setting->Id(), widget); +#endif +#ifdef EGLANTINE_STREAMLINE + } else if (setting->Id() == Settings::values.nvidia_dlss.Id()) { + // NVIDIA's files and a download button under the checkbox. + QWidget* group = new QWidget(this); + QVBoxLayout* group_layout = new QVBoxLayout(group); + group_layout->setContentsMargins(0, 0, 0, 0); + group_layout->setSpacing(2); + group_layout->addWidget(widget); + NvidiaDlssPanel* panel = new NvidiaDlssPanel(group); + group_layout->addWidget(panel); + panel->setVisible(widget->checkbox->isChecked()); + connect(widget->checkbox, &QCheckBox::toggled, panel, &QWidget::setVisible); + hold_graphics.emplace(setting->Id(), group); #endif } else if (setting->Id() == Settings::values.frame_gen.Id()) { frame_gen_checkbox = widget->checkbox; diff --git a/src/yuzu/configuration/nvidia_dlss_panel.cpp b/src/yuzu/configuration/nvidia_dlss_panel.cpp new file mode 100644 index 0000000000..fa1af23c4a --- /dev/null +++ b/src/yuzu/configuration/nvidia_dlss_panel.cpp @@ -0,0 +1,549 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#ifdef EGLANTINE_STREAMLINE + +#include "common/httplib.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "common/common_types.h" +#include "common/logging.h" +#include "video_core/vulkan_common/streamline.h" +#include "yuzu/configuration/nvidia_dlss_panel.h" + +#ifdef YUZU_BUNDLED_OPENSSL +#include +#endif + +namespace { + +namespace Streamline = Vulkan::Streamline; + +constexpr const char* RELEASE_PAGE = "https://github.com/NVIDIA-RTX/Streamline/releases"; +constexpr const char* RELEASE_VERSION = "2.14.16"; +constexpr const char* ZIP_HOST = "https://github.com"; +constexpr const char* ZIP_PATH = + "/NVIDIA-RTX/Streamline/releases/download/v2.14.16/streamline-sdk-v2.14.16.zip"; +constexpr const char* INTERPOSER_ENTRY = "bin/x64/sl.interposer.dll"; +constexpr const char* MODEL_ENTRY = "bin/x64/nvngx_dlss.dll"; +constexpr const char* STREAMLINE_LICENSE_ENTRY = "license.txt"; +constexpr const char* DLSS_LICENSE_ENTRY = "bin/x64/nvngx_dlss.license.txt"; + +QString Tr(const char* text) { + return QCoreApplication::translate("NvidiaDlssPanel", text); +} + +u16 U16(const std::string& data, size_t at) { + return static_cast(static_cast(data[at]) | static_cast(data[at + 1]) << 8); +} + +u32 U32(const std::string& data, size_t at) { + return static_cast(U16(data, at)) | static_cast(U16(data, at + 2)) << 16; +} + +u64 U64(const std::string& data, size_t at) { + return static_cast(U32(data, at)) | static_cast(U32(data, at + 4)) << 32; +} + +struct ZipEntry { + u16 method{}; + u32 crc{}; + u64 compressed{}; + u64 size{}; + u64 offset{}; +}; + +/// Reads single files out of a zip on a web server with range requests, so only the files +/// needed are downloaded (about 32 MB of the 276 MB release). +class RemoteZip { +public: + /// Follows GitHub's redirect to the file's (short-lived) download address. + bool Resolve() { + host = ZIP_HOST; + path = ZIP_PATH; + for (int redirect = 0; redirect < 5; ++redirect) { + const auto result = MakeClient()->Get(path, httplib::Headers{{"Range", "bytes=0-0"}}); + if (!result) { + error = httplib::to_string(result.error()); + return false; + } + if (result->status == 206) { + // "bytes 0-0/"; the server doesn't take "last N bytes" ranges. + const std::string range = result->get_header_value("Content-Range"); + const size_t slash = range.rfind('/'); + zip_size = slash == std::string::npos + ? 0 + : std::strtoull(range.c_str() + slash + 1, nullptr, 10); + if (zip_size == 0) { + error = "no file size"; + return false; + } + return true; + } + const std::string location = result->get_header_value("Location"); + const size_t scheme = location.find("://"); + const size_t slash = + scheme == std::string::npos ? std::string::npos : location.find('/', scheme + 3); + if (result->status < 300 || result->status >= 400 || slash == std::string::npos) { + error = fmt::format("HTTP {}", result->status); + return false; + } + host = location.substr(0, slash); + path = location.substr(slash); + } + error = "too many redirects"; + return false; + } + + bool Open() { + if (!Resolve()) { + return false; + } + const u64 tail_start = zip_size > 65558 ? zip_size - 65558 : 0; + const std::optional tail = + Fetch(fmt::format("bytes={}-{}", tail_start, zip_size - 1)); + if (!tail) { + return false; + } + const size_t end = tail->rfind(std::string{"PK\x05\x06", 4}); + if (end == std::string::npos || end + 22 > tail->size()) { + error = "no zip directory"; + return false; + } + u64 directory_size = U32(*tail, end + 12); + u64 directory_offset = U32(*tail, end + 16); + if (directory_offset == 0xFFFFFFFF && end >= 20 && + tail->compare(end - 20, 4, std::string{"PK\x06\x07", 4}) == 0) { + const u64 record = U64(*tail, end - 20 + 8); + const std::optional zip64 = + Fetch(fmt::format("bytes={}-{}", record, record + 55)); + if (!zip64 || zip64->size() < 56) { + return false; + } + directory_size = U64(*zip64, 40); + directory_offset = U64(*zip64, 48); + } + const std::optional directory = Fetch( + fmt::format("bytes={}-{}", directory_offset, directory_offset + directory_size - 1)); + if (!directory) { + return false; + } + for (size_t at = 0; at + 46 <= directory->size() && U32(*directory, at) == 0x02014b50;) { + const u16 name_length = U16(*directory, at + 28); + const u16 extra_length = U16(*directory, at + 30); + const u16 comment_length = U16(*directory, at + 32); + ZipEntry entry{ + .method = U16(*directory, at + 10), + .crc = U32(*directory, at + 16), + .compressed = U32(*directory, at + 20), + .size = U32(*directory, at + 24), + .offset = U32(*directory, at + 42), + }; + const std::string name = directory->substr(at + 46, name_length); + // Zip64 sizes and offset, in this order, for the fields that overflowed. + const size_t extra_end = at + 46 + name_length + extra_length; + for (size_t extra = at + 46 + name_length; extra + 4 <= extra_end;) { + const u16 id = U16(*directory, extra); + const u16 length = U16(*directory, extra + 2); + if (id == 0x0001) { + size_t field = extra + 4; + for (u64* value : {&entry.size, &entry.compressed, &entry.offset}) { + if (*value == 0xFFFFFFFF) { + *value = U64(*directory, field); + field += 8; + } + } + } + extra += 4 + length; + } + entries.emplace(name, entry); + at += 46 + name_length + extra_length + comment_length; + } + return true; + } + + const ZipEntry* Find(const std::string& name) const { + const auto it = entries.find(name); + return it == entries.end() ? nullptr : &it->second; + } + + std::optional Read(const ZipEntry& entry, + const std::function& progress = {}) { + const std::optional header = + Fetch(fmt::format("bytes={}-{}", entry.offset, entry.offset + 29)); + if (!header) { + return std::nullopt; + } + if (header->size() < 30 || U32(*header, 0) != 0x04034b50) { + error = "bad zip entry"; + return std::nullopt; + } + const u64 start = entry.offset + 30 + U16(*header, 26) + U16(*header, 28); + std::string data; + if (entry.compressed != 0) { + std::optional fetched = + Fetch(fmt::format("bytes={}-{}", start, start + entry.compressed - 1), progress); + if (!fetched) { + return std::nullopt; + } + data = std::move(*fetched); + } + std::string out; + if (entry.method == 0) { + out = std::move(data); + } else if (entry.method == 8) { + out.resize(entry.size); + z_stream stream{}; + if (inflateInit2(&stream, -MAX_WBITS) != Z_OK) { + error = "zlib"; + return std::nullopt; + } + stream.next_in = reinterpret_cast(data.data()); + stream.avail_in = static_cast(data.size()); + stream.next_out = reinterpret_cast(out.data()); + stream.avail_out = static_cast(out.size()); + const int result = inflate(&stream, Z_FINISH); + inflateEnd(&stream); + if (result != Z_STREAM_END || stream.total_out != entry.size) { + error = "could not unpack"; + return std::nullopt; + } + } else { + error = "unknown zip compression"; + return std::nullopt; + } + const uLong crc = crc32(0L, reinterpret_cast(out.data()), + static_cast(out.size())); + if (crc != entry.crc) { + error = "damaged download (checksum)"; + return std::nullopt; + } + return out; + } + + bool canceled{}; + std::string error; + +private: + std::unique_ptr MakeClient() const { + auto client = std::make_unique(host); + client->set_connection_timeout(30); + client->set_read_timeout(30); + client->set_write_timeout(30); +#ifdef YUZU_BUNDLED_OPENSSL + client->load_ca_cert_store(kCert, sizeof(kCert)); +#endif + return client; + } + + std::optional Fetch(const std::string& range, + const std::function& progress = {}) { + std::string body; + int status = 0; + const auto result = MakeClient()->Get( + path, httplib::Headers{{"Range", range}}, + [&status](const httplib::Response& response) { + status = response.status; + return response.status == 206; + }, + [&](const char* data, size_t length) { + body.append(data, length); + if (progress && !progress(body.size())) { + canceled = true; + return false; + } + return true; + }); + if (!result || status != 206) { + error = result ? fmt::format("HTTP {}", status) : httplib::to_string(result.error()); + return std::nullopt; + } + return body; + } + + std::string host; + std::string path; + u64 zip_size{}; + std::map entries; +}; + +QString Named(const QString& name, const std::string& version) { + return version.empty() ? name : QStringLiteral("%1 %2").arg(name, QString::fromStdString(version)); +} + +QString PathText(const std::filesystem::path& path) { + return QString::fromStdWString(path.wstring()); +} + +} // Anonymous namespace + +NvidiaDlssPanel::NvidiaDlssPanel(QWidget* parent) : QWidget(parent) { + QVBoxLayout* layout = new QVBoxLayout(this); + layout->setContentsMargins(24, 0, 0, 8); + layout->setSpacing(6); + + QHBoxLayout* status_row = new QHBoxLayout; + status_row->setSpacing(8); + icon_label = new QLabel(this); + icon_label->setTextFormat(Qt::RichText); + icon_label->setAlignment(Qt::AlignTop | Qt::AlignHCenter); + icon_label->setFixedWidth(16); + status_label = new QLabel(this); + status_label->setTextFormat(Qt::RichText); + status_label->setWordWrap(true); + status_row->addWidget(icon_label); + status_row->addWidget(status_label, 1); + + QHBoxLayout* button_row = new QHBoxLayout; + button_row->setContentsMargins(24, 0, 0, 0); + download_button = new QPushButton(this); + folder_button = new QPushButton(Tr("Open Folder"), this); + QLabel* about = new QLabel( + QStringLiteral("%2").arg(QString::fromLatin1(RELEASE_PAGE), + Tr("Where do these files come from?")), + this); + about->setOpenExternalLinks(true); + about->setToolTip(Tr("NVIDIA's Streamline releases on GitHub. The files are in bin\\x64 of " + "the release zip.")); + button_row->addWidget(download_button); + button_row->addWidget(folder_button); + button_row->addStretch(1); + button_row->addWidget(about); + + layout->addLayout(status_row); + layout->addLayout(button_row); + + connect(download_button, &QPushButton::clicked, this, [this] { Download(); }); + connect(folder_button, &QPushButton::clicked, this, [] { + const std::filesystem::path folder = Streamline::FindDlssFiles().folder; + std::error_code ec; + std::filesystem::create_directories(folder, ec); + QDesktopServices::openUrl(QUrl::fromLocalFile(PathText(folder))); + }); + Refresh(); +} + +void NvidiaDlssPanel::showEvent(QShowEvent* event) { + Refresh(); + QWidget::showEvent(event); +} + +void NvidiaDlssPanel::Refresh() { + const Streamline::DlssFiles files = Streamline::FindDlssFiles(); + const bool has_loader = !files.interposer.empty(); + const bool has_model = !files.model.empty() || !files.app_model.empty(); + const QString dlss = !files.model.empty() + ? Named(QStringLiteral("DLSS"), files.model_version) + : Tr("%1 from the NVIDIA app") + .arg(Named(QStringLiteral("DLSS"), files.app_model_version)); + + QString status; + QString button; + if (has_loader && has_model) { + status = Tr("Ready. Found %1 and %2.") + .arg(Named(QStringLiteral("NVIDIA Streamline"), files.interposer_version), dlss); + } else if (has_model) { + status = Tr("One file needed: NVIDIA's Streamline loader (sl.interposer.dll). " + "%1 will be used. Until then, games use AMD FSR 3.1.") + .arg(dlss); + button = Tr("Download from NVIDIA (1 MB)"); + } else if (has_loader) { + status = Tr("One file needed: NVIDIA DLSS (nvngx_dlss.dll). Until then, games use " + "AMD FSR 3.1."); + button = Tr("Download from NVIDIA (31 MB)"); + } else { + status = Tr("NVIDIA's files needed: the Streamline loader and DLSS " + "(sl.interposer.dll and nvngx_dlss.dll). Until then, games use AMD FSR 3.1."); + button = Tr("Download from NVIDIA (32 MB)"); + } + const bool ready = has_loader && has_model; + icon_label->setText(ready ? QStringLiteral("✔") + : QStringLiteral("⚠")); + status_label->setText(status); + status_label->setToolTip(Tr("Folder: %1").arg(PathText(files.folder))); + download_button->setText(button); + download_button->setVisible(!ready); +} + +void NvidiaDlssPanel::Download() { + const Streamline::DlssFiles files = Streamline::FindDlssFiles(); + std::vector wanted; + if (files.interposer.empty()) { + wanted.emplace_back(INTERPOSER_ENTRY); + } + if (files.model.empty() && files.app_model.empty()) { + wanted.emplace_back(MODEL_ENTRY); + } + if (wanted.empty()) { + return; + } + const std::filesystem::path folder = Streamline::UserFolder(); + const auto fail = [this](const std::string& error) { + LOG_ERROR(Frontend, "NVIDIA DLSS download failed: {}", error); + QMessageBox::warning(this, Tr("NVIDIA DLSS"), + Tr("Could not download NVIDIA's files: %1") + .arg(QString::fromStdString(error))); + Refresh(); + }; + + RemoteZip zip; + std::string licences; + { + QProgressDialog busy(Tr("Contacting NVIDIA's download..."), QString{}, 0, 0, this); + busy.setWindowTitle(Tr("NVIDIA DLSS")); + busy.setWindowModality(Qt::WindowModal); + busy.setMinimumDuration(0); + busy.show(); + QApplication::processEvents(); + if (!zip.Open()) { + busy.close(); + fail(zip.error); + return; + } + std::vector> texts{ + {Tr("NVIDIA Streamline"), STREAMLINE_LICENSE_ENTRY}}; + if (std::ranges::find(wanted, MODEL_ENTRY) != wanted.end()) { + texts.emplace_back(Tr("NVIDIA DLSS"), DLSS_LICENSE_ENTRY); + } + for (const auto& [title, entry_name] : texts) { + const ZipEntry* const entry = zip.Find(entry_name); + const std::optional text = entry ? zip.Read(*entry) : std::nullopt; + licences += fmt::format("==== {} ====\n\n{}\n\n", title.toStdString(), + text ? *text : std::string{"(see the release on GitHub)"}); + } + } + + QStringList names; + for (const std::string& entry : wanted) { + names << QString::fromStdString(std::filesystem::path{entry}.filename().string()); + } + QDialog dialog(this); + dialog.setWindowTitle(Tr("Download NVIDIA DLSS")); + QVBoxLayout* layout = new QVBoxLayout(&dialog); + QLabel* intro = new QLabel( + Tr("Eglantine will download %1 from NVIDIA's Streamline %2 release on GitHub into %3.
" + "These files are NVIDIA's, under NVIDIA's licence terms below.") + .arg(names.join(QStringLiteral(" and ")), QString::fromLatin1(RELEASE_VERSION), + PathText(folder)), + &dialog); + intro->setWordWrap(true); + QPlainTextEdit* licence_text = new QPlainTextEdit(QString::fromStdString(licences), &dialog); + licence_text->setReadOnly(true); + licence_text->setMinimumSize(640, 380); + QDialogButtonBox* buttons = new QDialogButtonBox(&dialog); + buttons->addButton(Tr("Accept and Download"), QDialogButtonBox::AcceptRole); + buttons->addButton(QDialogButtonBox::Cancel); + connect(buttons, &QDialogButtonBox::accepted, &dialog, &QDialog::accept); + connect(buttons, &QDialogButtonBox::rejected, &dialog, &QDialog::reject); + layout->addWidget(intro); + layout->addWidget(licence_text, 1); + layout->addWidget(buttons); + if (dialog.exec() != QDialog::Accepted) { + return; + } + + u64 total = 0; + std::vector entries; + for (const std::string& name : wanted) { + const ZipEntry* const entry = zip.Find(name); + if (!entry) { + fail(name + " is not in the release"); + return; + } + entries.push_back(entry); + total += entry->compressed; + } + QProgressDialog progress(Tr("Downloading NVIDIA's DLSS files..."), Tr("Cancel"), 0, 1000, + this); + progress.setWindowTitle(Tr("NVIDIA DLSS")); + progress.setWindowModality(Qt::WindowModal); + progress.setMinimumDuration(0); + progress.setValue(0); + QApplication::processEvents(); + // The download address from the first step only lasts a few minutes. + if (!zip.Resolve()) { + progress.close(); + fail(zip.error); + return; + } + std::error_code ec; + std::filesystem::create_directories(folder, ec); + u64 done = 0; + for (size_t i = 0; i < wanted.size(); ++i) { + const std::optional contents = zip.Read(*entries[i], [&](u64 received) { + progress.setValue(static_cast((done + received) * 1000 / std::max(total, 1))); + QApplication::processEvents(); + return !progress.wasCanceled(); + }); + if (!contents) { + progress.close(); + if (!zip.canceled) { + fail(zip.error); + } + return; + } + done += entries[i]->compressed; + const std::filesystem::path target = folder / std::filesystem::path{wanted[i]}.filename(); + std::filesystem::path partial = target; + partial += L".download"; + { + std::ofstream file(partial, std::ios::binary | std::ios::trunc); + file.write(contents->data(), static_cast(contents->size())); + if (!file) { + progress.close(); + fail("could not write " + partial.string()); + return; + } + } + const bool signed_ok = wanted[i] == INTERPOSER_ENTRY ? Streamline::IsStreamlineModule(partial) + : Streamline::IsSignedByNvidia(partial); + if (!signed_ok) { + std::filesystem::remove(partial, ec); + progress.close(); + fail(target.filename().string() + " is not signed by NVIDIA"); + return; + } + std::filesystem::rename(partial, target, ec); + if (ec) { + std::filesystem::remove(partial, ec); + progress.close(); + fail("could not save " + target.string()); + return; + } + LOG_INFO(Frontend, "Downloaded {} from NVIDIA's Streamline {} release", target.string(), + RELEASE_VERSION); + } + progress.setValue(1000); + progress.close(); + Refresh(); + QMessageBox::information(this, Tr("NVIDIA DLSS"), + Tr("NVIDIA DLSS is ready. It is used the next time a game starts.")); +} + +#endif diff --git a/src/yuzu/configuration/nvidia_dlss_panel.h b/src/yuzu/configuration/nvidia_dlss_panel.h new file mode 100644 index 0000000000..1ab39178eb --- /dev/null +++ b/src/yuzu/configuration/nvidia_dlss_panel.h @@ -0,0 +1,34 @@ +// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +// Shown under the NVIDIA DLSS setting: which of NVIDIA's files were found, and +// a button that fetches missing ones from NVIDIA's own Streamline release. + +#pragma once + +#ifdef EGLANTINE_STREAMLINE + +#include + +class QLabel; +class QPushButton; + +class NvidiaDlssPanel : public QWidget { +public: + explicit NvidiaDlssPanel(QWidget* parent); + + void Refresh(); + +protected: + void showEvent(QShowEvent* event) override; + +private: + void Download(); + + QLabel* icon_label; + QLabel* status_label; + QPushButton* download_button; + QPushButton* folder_button; +}; + +#endif -- cgit v1.2.3 From b1179a44fbd1cf514c2bed56bf235fbe456c8c7b Mon Sep 17 00:00:00 2001 From: Duck912 <350528965+Duck912@users.noreply.github.com> Date: Tue, 3 Nov 2026 12:56:12 -0500 Subject: video_core: clear the DLSS output when Streamline writes nothing to it The upscaler records its work on the worker thread and reports success once it is queued. When Streamline then rejects the DLSS options, gives no frame token or fails the evaluation, nothing is written to the output image, but the caller still draws it into the game's image that frame: undefined contents after a resize, the previous frame otherwise. FSR 3.1 takes over from the next frame. Clear the output on those paths, with barriers around the clear, so that frame is black instead. Found by CodeRabbit review. Assisted-by: Claude Opus 5.5 (implementation) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com> --- .../renderer_vulkan/vk_streamline_dlss.cpp | 39 ++++++++++++++++++++++ 1 file changed, 39 insertions(+) diff --git a/src/video_core/renderer_vulkan/vk_streamline_dlss.cpp b/src/video_core/renderer_vulkan/vk_streamline_dlss.cpp index aef32735f9..4fa9e240c4 100644 --- a/src/video_core/renderer_vulkan/vk_streamline_dlss.cpp +++ b/src/video_core/renderer_vulkan/vk_streamline_dlss.cpp @@ -260,6 +260,40 @@ bool StreamlineDlssUpscaler::Upscale(const Inputs& in) { output_view_handle = *output_view](vk::CommandBuffer cmdbuf) { const Streamline::Api& api = Streamline::GetApi(); const sl::ViewportHandle viewport{0u}; + // The caller draws the output this frame even when DLSS writes nothing to it: clear it + // then, rather than show undefined contents (FSR 3.1 takes over from the next frame). + const auto clear_output = [&] { + const VkImageSubresourceRange range{ + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }; + const auto barrier = [&](VkAccessFlags src_access, VkAccessFlags dst_access) { + return VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = src_access, + .dstAccessMask = dst_access, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = output_image_handle, + .subresourceRange = range, + }; + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_TRANSFER_BIT, 0, + barrier(VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, + VK_ACCESS_TRANSFER_WRITE_BIT)); + cmdbuf.ClearColorImage(output_image_handle, VK_IMAGE_LAYOUT_GENERAL, + VkClearColorValue{}, range); + cmdbuf.PipelineBarrier( + VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, + barrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT)); + }; if (options_changed) { const auto set = reinterpret_cast(set_options); if (const sl::Result result = set(viewport, options); result != sl::Result::eOk) { @@ -268,6 +302,7 @@ bool StreamlineDlssUpscaler::Upscale(const Inputs& in) { "Streamline rejected the DLSS options (result {}); using FSR 3.1", static_cast(result)); } + clear_output(); return; } } @@ -278,6 +313,7 @@ bool StreamlineDlssUpscaler::Upscale(const Inputs& in) { LOG_ERROR(Render_Vulkan, "Streamline gave no frame token (result {}); using FSR 3.1", static_cast(result)); } + clear_output(); return; } @@ -377,6 +413,9 @@ bool StreamlineDlssUpscaler::Upscale(const Inputs& in) { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, output_written, {}, to_general); + if (result != sl::Result::eOk) { + clear_output(); + } }); // DLSS binds its own pipelines and descriptors on the command buffer. scheduler.InvalidateState(); -- cgit v1.2.3 From f4f076fe447b2ddc158fcc12ee81ed2af2aecad7 Mon Sep 17 00:00:00 2001 From: Duck912 <350528965+Duck912@users.noreply.github.com> Date: Tue, 3 Nov 2026 12:56:27 -0500 Subject: video_core, yuzu: write NVIDIA DLSS file paths to logs as UTF-8 The Streamline start-up and DLSS download code formatted paths with std::filesystem::path::string(), which on MSVC converts to the active code page and throws when a character has no mapping there, for example a user name in another script under %APPDATA%. The start-up log runs inside the renderer's constructor, which only catches vk::Exception, so such a user would crash at game start with the setting on. Use Common::FS::PathToUTF8String for every path in these logs and messages. Found by CodeRabbit review. Assisted-by: Claude Opus 5.5 (implementation) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com> --- src/video_core/vulkan_common/streamline.cpp | 17 ++++++++++------- src/yuzu/configuration/nvidia_dlss_panel.cpp | 14 ++++++++------ 2 files changed, 18 insertions(+), 13 deletions(-) diff --git a/src/video_core/vulkan_common/streamline.cpp b/src/video_core/vulkan_common/streamline.cpp index 966de82cb5..4f8fd796c1 100644 --- a/src/video_core/vulkan_common/streamline.cpp +++ b/src/video_core/vulkan_common/streamline.cpp @@ -15,6 +15,7 @@ #include +#include "common/fs/fs_util.h" #include "common/fs/path_util.h" #include "common/logging.h" #include "common/settings.h" @@ -168,7 +169,7 @@ std::filesystem::path UseAppModel(const DlssFiles& files) { if (!current) { if (!IsSignedByNvidia(files.app_model)) { LOG_WARNING(Render_Vulkan, "The NVIDIA app's DLSS {} is not signed by NVIDIA", - files.app_model.string()); + Common::FS::PathToUTF8String(files.app_model)); return {}; } std::filesystem::create_directories(dir, ec); @@ -176,7 +177,7 @@ std::filesystem::path UseAppModel(const DlssFiles& files) { std::filesystem::copy_options::overwrite_existing, ec); if (ec) { LOG_WARNING(Render_Vulkan, "Could not copy the NVIDIA app's DLSS to {}: {}", - target.string(), ec.message()); + Common::FS::PathToUTF8String(target), ec.message()); return {}; } } @@ -186,12 +187,13 @@ std::filesystem::path UseAppModel(const DlssFiles& files) { bool Load(State& state, const std::filesystem::path& path) { if (!IsStreamlineModule(path)) { LOG_ERROR(Render_Vulkan, "{} is not signed by NVIDIA; not loading it", - path.string()); + Common::FS::PathToUTF8String(path)); return false; } state.module = LoadLibraryW(path.c_str()); if (!state.module) { - LOG_ERROR(Render_Vulkan, "Could not load {} (error {})", path.string(), GetLastError()); + LOG_ERROR(Render_Vulkan, "Could not load {} (error {})", Common::FS::PathToUTF8String(path), + GetLastError()); return false; } Api& api = state.api; @@ -298,7 +300,7 @@ bool Start() { LOG_WARNING(Render_Vulkan, "NVIDIA DLSS is on, but sl.interposer.dll (NVIDIA Streamline) is not in " "{} (the Graphics settings can download it); using FSR 3.1", - files.folder.string()); + Common::FS::PathToUTF8String(files.folder)); return false; } if (!Load(state, files.interposer)) { @@ -313,7 +315,7 @@ bool Start() { } } else if (files.model.empty()) { LOG_WARNING(Render_Vulkan, "No nvngx_dlss.dll in {} and no DLSS from the NVIDIA app", - files.folder.string()); + Common::FS::PathToUTF8String(files.folder)); } } @@ -371,7 +373,8 @@ bool Start() { sl::FeatureVersion version{}; api.get_feature_version(sl::kFeatureDLSS, version); LOG_INFO(Render_Vulkan, "NVIDIA Streamline {} started from {} (DLSS {})", - version.versionSL.toStr(), std::filesystem::path{state.plugin_dir}.string(), + version.versionSL.toStr(), + Common::FS::PathToUTF8String(std::filesystem::path{state.plugin_dir}), version.versionNGX.toStr()); state.started = true; return true; diff --git a/src/yuzu/configuration/nvidia_dlss_panel.cpp b/src/yuzu/configuration/nvidia_dlss_panel.cpp index fa1af23c4a..2329693a03 100644 --- a/src/yuzu/configuration/nvidia_dlss_panel.cpp +++ b/src/yuzu/configuration/nvidia_dlss_panel.cpp @@ -33,6 +33,7 @@ #include #include "common/common_types.h" +#include "common/fs/fs_util.h" #include "common/logging.h" #include "video_core/vulkan_common/streamline.h" #include "yuzu/configuration/nvidia_dlss_panel.h" @@ -441,7 +442,8 @@ void NvidiaDlssPanel::Download() { QStringList names; for (const std::string& entry : wanted) { - names << QString::fromStdString(std::filesystem::path{entry}.filename().string()); + names << QString::fromStdString( + Common::FS::PathToUTF8String(std::filesystem::path{entry}.filename())); } QDialog dialog(this); dialog.setWindowTitle(Tr("Download NVIDIA DLSS")); @@ -517,7 +519,7 @@ void NvidiaDlssPanel::Download() { file.write(contents->data(), static_cast(contents->size())); if (!file) { progress.close(); - fail("could not write " + partial.string()); + fail("could not write " + Common::FS::PathToUTF8String(partial)); return; } } @@ -526,18 +528,18 @@ void NvidiaDlssPanel::Download() { if (!signed_ok) { std::filesystem::remove(partial, ec); progress.close(); - fail(target.filename().string() + " is not signed by NVIDIA"); + fail(Common::FS::PathToUTF8String(target.filename()) + " is not signed by NVIDIA"); return; } std::filesystem::rename(partial, target, ec); if (ec) { std::filesystem::remove(partial, ec); progress.close(); - fail("could not save " + target.string()); + fail("could not save " + Common::FS::PathToUTF8String(target)); return; } - LOG_INFO(Frontend, "Downloaded {} from NVIDIA's Streamline {} release", target.string(), - RELEASE_VERSION); + LOG_INFO(Frontend, "Downloaded {} from NVIDIA's Streamline {} release", + Common::FS::PathToUTF8String(target), RELEASE_VERSION); } progress.setValue(1000); progress.close(); -- cgit v1.2.3 From f9d0c802e8e4e71644bf4b7e6cceca8e6df22464 Mon Sep 17 00:00:00 2001 From: Duck912 <350528965+Duck912@users.noreply.github.com> Date: Tue, 3 Nov 2026 12:56:43 -0500 Subject: yuzu: bound the zip directory reads in the DLSS download The central-directory parser trusted each entry's name, extra and comment lengths and the Zip64 extra field's length, so a malformed directory could make it read past the downloaded buffer. The data comes from NVIDIA's GitHub release over TLS and the saved files are signature-checked, so the risk is low. Stop with an error when an entry runs past the directory, and read Zip64 fields only inside their record. Found by CodeRabbit review. Assisted-by: Claude Opus 5.5 (implementation) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com> --- src/yuzu/configuration/nvidia_dlss_panel.cpp | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/src/yuzu/configuration/nvidia_dlss_panel.cpp b/src/yuzu/configuration/nvidia_dlss_panel.cpp index 2329693a03..040ea90557 100644 --- a/src/yuzu/configuration/nvidia_dlss_panel.cpp +++ b/src/yuzu/configuration/nvidia_dlss_panel.cpp @@ -166,16 +166,21 @@ public: .size = U32(*directory, at + 24), .offset = U32(*directory, at + 42), }; + const size_t extra_end = at + 46 + name_length + extra_length; + if (extra_end + comment_length > directory->size()) { + error = "bad zip directory"; + return false; + } const std::string name = directory->substr(at + 46, name_length); // Zip64 sizes and offset, in this order, for the fields that overflowed. - const size_t extra_end = at + 46 + name_length + extra_length; for (size_t extra = at + 46 + name_length; extra + 4 <= extra_end;) { const u16 id = U16(*directory, extra); const u16 length = U16(*directory, extra + 2); + const size_t record_end = std::min(extra + 4 + length, extra_end); if (id == 0x0001) { size_t field = extra + 4; for (u64* value : {&entry.size, &entry.compressed, &entry.offset}) { - if (*value == 0xFFFFFFFF) { + if (*value == 0xFFFFFFFF && field + 8 <= record_end) { *value = U64(*directory, field); field += 8; } -- cgit v1.2.3 From 1889602c7a0dc17850692594eae2e77e16ac13fc Mon Sep 17 00:00:00 2001 From: Duck912 <350528965+Duck912@users.noreply.github.com> Date: Tue, 3 Nov 2026 13:18:12 -0500 Subject: video_core: accept an empty Streamline device-extension list Streamline may report that DLSS needs no extra device extensions. The check required a non-empty list, so that result counted as "extensions unavailable": Streamline was stopped and FSR 3.1 used. Only require that every listed extension is supported. Found by CodeRabbit review. Assisted-by: Claude Opus 5.5 (implementation) Signed-off-by: Duck912 <350528965+Duck912@users.noreply.github.com> --- src/video_core/vulkan_common/vulkan_device.cpp | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 6309599877..2499f46f87 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -1153,10 +1153,9 @@ bool Device::GetSuitability(bool requires_swapchain) { // The extensions NVIDIA's DLSS asks for, where the device has them all. if (Streamline::WantsDevice()) { const std::vector wanted = Streamline::DeviceExtensions(); - const bool all = - !wanted.empty() && std::ranges::all_of(wanted, [&](const std::string& name) { - return supported_extensions.contains(name); - }); + // Streamline may ask for no extensions at all. + const bool all = std::ranges::all_of( + wanted, [&](const std::string& name) { return supported_extensions.contains(name); }); std::string names; for (const std::string& name : wanted) { if (all) { @@ -1166,7 +1165,7 @@ bool Device::GetSuitability(bool requires_swapchain) { } streamline_extensions = all; LOG_INFO(Render_Vulkan, "NVIDIA DLSS device extensions: {} ({})", - all ? "enabled" : "unavailable", names); + all ? "enabled" : "unavailable", names.empty() ? "none" : names); } #endif -- cgit v1.2.3