diff options
Diffstat (limited to 'src/video_core/renderer_vulkan')
19 files changed, 2095 insertions, 223 deletions
diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index 6c1b2f063..b3884a4f5 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -363,7 +363,7 @@ BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_, BlitImageHelper::~BlitImageHelper() = default; -void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, +void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView src_view, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, Tegra::Engines::Fermi2D::Operation operation) { @@ -373,9 +373,8 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageV .operation = operation, }; const VkPipelineLayout layout = *one_texture_pipeline_layout; - const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D); const VkSampler sampler = is_linear ? *linear_sampler : *nearest_sampler; - const VkPipeline pipeline = FindOrEmplacePipeline(key); + const VkPipeline pipeline = FindOrEmplaceColorPipeline(key); scheduler.RequestRenderpass(dst_framebuffer); scheduler.Record([this, dst_region, src_region, pipeline, layout, sampler, src_view](vk::CommandBuffer cmdbuf) { @@ -398,10 +397,13 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer, Tegra::Engines::Fermi2D::Operation operation) { ASSERT(filter == Tegra::Engines::Fermi2D::Filter::Point); ASSERT(operation == Tegra::Engines::Fermi2D::Operation::SrcCopy); - + const BlitImagePipelineKey key{ + .renderpass = dst_framebuffer->RenderPass(), + .operation = operation, + }; const VkPipelineLayout layout = *two_textures_pipeline_layout; const VkSampler sampler = *nearest_sampler; - const VkPipeline pipeline = BlitDepthStencilPipeline(dst_framebuffer->RenderPass()); + const VkPipeline pipeline = FindOrEmplaceDepthStencilPipeline(key); scheduler.RequestRenderpass(dst_framebuffer); scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_depth_view, src_stencil_view, this](vk::CommandBuffer cmdbuf) { @@ -419,40 +421,45 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer, } void BlitImageHelper::ConvertD32ToR32(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertDepthToColorPipeline(convert_d32_to_r32_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_d32_to_r32_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_d32_to_r32_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::ConvertR32ToD32(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertColorToDepthPipeline(convert_r32_to_d32_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::ConvertD16ToR16(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertDepthToColorPipeline(convert_d16_to_r16_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_d16_to_r16_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_d16_to_r16_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, + u32 down_shift) { ConvertColorToDepthPipeline(convert_r16_to_d16_pipeline, dst_framebuffer->RenderPass()); - Convert(*convert_r16_to_d16_pipeline, dst_framebuffer, src_image_view); + Convert(*convert_r16_to_d16_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift); } void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, - const ImageView& src_image_view) { + const ImageView& src_image_view, u32 up_scale, u32 down_shift) { const VkPipelineLayout layout = *one_texture_pipeline_layout; const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D); const VkSampler sampler = *nearest_sampler; const VkExtent2D extent{ - .width = src_image_view.size.width, - .height = src_image_view.size.height, + .width = std::max((src_image_view.size.width * up_scale) >> down_shift, 1U), + .height = std::max((src_image_view.size.height * up_scale) >> down_shift, 1U), }; scheduler.RequestRenderpass(dst_framebuffer); - scheduler.Record([pipeline, layout, sampler, src_view, extent, this](vk::CommandBuffer cmdbuf) { + scheduler.Record([pipeline, layout, sampler, src_view, extent, up_scale, down_shift, + this](vk::CommandBuffer cmdbuf) { const VkOffset2D offset{ .x = 0, .y = 0, @@ -488,7 +495,7 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb scheduler.InvalidateState(); } -VkPipeline BlitImageHelper::FindOrEmplacePipeline(const BlitImagePipelineKey& key) { +VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key) { const auto it = std::ranges::find(blit_color_keys, key); if (it != blit_color_keys.end()) { return *blit_color_pipelines[std::distance(blit_color_keys.begin(), it)]; @@ -542,12 +549,14 @@ VkPipeline BlitImageHelper::FindOrEmplacePipeline(const BlitImagePipelineKey& ke return *blit_color_pipelines.back(); } -VkPipeline BlitImageHelper::BlitDepthStencilPipeline(VkRenderPass renderpass) { - if (blit_depth_stencil_pipeline) { - return *blit_depth_stencil_pipeline; +VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key) { + const auto it = std::ranges::find(blit_depth_stencil_keys, key); + if (it != blit_depth_stencil_keys.end()) { + return *blit_depth_stencil_pipelines[std::distance(blit_depth_stencil_keys.begin(), it)]; } + blit_depth_stencil_keys.push_back(key); const std::array stages = MakeStages(*full_screen_vert, *blit_depth_stencil_frag); - blit_depth_stencil_pipeline = device.GetLogical().CreateGraphicsPipeline({ + blit_depth_stencil_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({ .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, .pNext = nullptr, .flags = 0, @@ -560,15 +569,15 @@ VkPipeline BlitImageHelper::BlitDepthStencilPipeline(VkRenderPass renderpass) { .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO, .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, .pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, - .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO, + .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO, .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO, .layout = *two_textures_pipeline_layout, - .renderPass = renderpass, + .renderPass = key.renderpass, .subpass = 0, .basePipelineHandle = VK_NULL_HANDLE, .basePipelineIndex = 0, - }); - return *blit_depth_stencil_pipeline; + })); + return *blit_depth_stencil_pipelines.back(); } void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) { diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h index 33ee095c1..d77f76678 100644 --- a/src/video_core/renderer_vulkan/blit_image.h +++ b/src/video_core/renderer_vulkan/blit_image.h @@ -34,7 +34,7 @@ public: StateTracker& state_tracker, DescriptorPool& descriptor_pool); ~BlitImageHelper(); - void BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + void BlitColor(const Framebuffer* dst_framebuffer, VkImageView src_image_view, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, Tegra::Engines::Fermi2D::Operation operation); @@ -44,21 +44,25 @@ public: const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, Tegra::Engines::Fermi2D::Operation operation); - void ConvertD32ToR32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertD32ToR32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); - void ConvertR32ToD32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertR32ToD32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); - void ConvertD16ToR16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertD16ToR16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); - void ConvertR16ToD16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view); + void ConvertR16ToD16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view, + u32 up_scale, u32 down_shift); private: void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, - const ImageView& src_image_view); + const ImageView& src_image_view, u32 up_scale, u32 down_shift); - [[nodiscard]] VkPipeline FindOrEmplacePipeline(const BlitImagePipelineKey& key); + [[nodiscard]] VkPipeline FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key); - [[nodiscard]] VkPipeline BlitDepthStencilPipeline(VkRenderPass renderpass); + [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key); void ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass); @@ -84,7 +88,8 @@ private: std::vector<BlitImagePipelineKey> blit_color_keys; std::vector<vk::Pipeline> blit_color_pipelines; - vk::Pipeline blit_depth_stencil_pipeline; + std::vector<BlitImagePipelineKey> blit_depth_stencil_keys; + std::vector<vk::Pipeline> blit_depth_stencil_pipelines; vk::Pipeline convert_d32_to_r32_pipeline; vk::Pipeline convert_r32_to_d32_pipeline; vk::Pipeline convert_d16_to_r16_pipeline; diff --git a/src/video_core/renderer_vulkan/pipeline_helper.h b/src/video_core/renderer_vulkan/pipeline_helper.h index 4847db6b6..11c160570 100644 --- a/src/video_core/renderer_vulkan/pipeline_helper.h +++ b/src/video_core/renderer_vulkan/pipeline_helper.h @@ -10,6 +10,7 @@ #include "common/assert.h" #include "common/common_types.h" +#include "shader_recompiler/backend/spirv/emit_spirv.h" #include "shader_recompiler/shader_info.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" @@ -20,6 +21,8 @@ namespace Vulkan { +using Shader::Backend::SPIRV::NUM_TEXTURE_AND_IMAGE_SCALING_WORDS; + class DescriptorLayoutBuilder { public: DescriptorLayoutBuilder(const Device& device_) : device{&device_} {} @@ -68,18 +71,28 @@ public: } vk::PipelineLayout CreatePipelineLayout(VkDescriptorSetLayout descriptor_set_layout) const { + using Shader::Backend::SPIRV::RescalingLayout; + const u32 size_offset = is_compute ? sizeof(RescalingLayout::down_factor) : 0u; + const VkPushConstantRange range{ + .stageFlags = static_cast<VkShaderStageFlags>( + is_compute ? VK_SHADER_STAGE_COMPUTE_BIT : VK_SHADER_STAGE_ALL_GRAPHICS), + .offset = 0, + .size = static_cast<u32>(sizeof(RescalingLayout)) - size_offset, + }; return device->GetLogical().CreatePipelineLayout({ .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, .pNext = nullptr, .flags = 0, .setLayoutCount = descriptor_set_layout ? 1U : 0U, .pSetLayouts = bindings.empty() ? nullptr : &descriptor_set_layout, - .pushConstantRangeCount = 0, - .pPushConstantRanges = nullptr, + .pushConstantRangeCount = 1, + .pPushConstantRanges = &range, }); } void Add(const Shader::Info& info, VkShaderStageFlags stage) { + is_compute |= (stage & VK_SHADER_STAGE_COMPUTE_BIT) != 0; + Add(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, stage, info.constant_buffer_descriptors); Add(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, stage, info.storage_buffers_descriptors); Add(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, stage, info.texture_buffer_descriptors); @@ -115,6 +128,7 @@ private: } const Device* device{}; + bool is_compute{}; boost::container::small_vector<VkDescriptorSetLayoutBinding, 32> bindings; boost::container::small_vector<VkDescriptorUpdateTemplateEntryKHR, 32> entries; u32 binding{}; @@ -122,31 +136,68 @@ private: size_t offset{}; }; -inline void PushImageDescriptors(const Shader::Info& info, const VkSampler*& samplers, - const ImageId*& image_view_ids, TextureCache& texture_cache, - VKUpdateDescriptorQueue& update_descriptor_queue) { - for (const auto& desc : info.texture_buffer_descriptors) { - image_view_ids += desc.count; +class RescalingPushConstant { +public: + explicit RescalingPushConstant() noexcept {} + + void PushTexture(bool is_rescaled) noexcept { + *texture_ptr |= is_rescaled ? texture_bit : 0u; + texture_bit <<= 1u; + if (texture_bit == 0u) { + texture_bit = 1u; + ++texture_ptr; + } } - for (const auto& desc : info.image_buffer_descriptors) { - image_view_ids += desc.count; + + void PushImage(bool is_rescaled) noexcept { + *image_ptr |= is_rescaled ? image_bit : 0u; + image_bit <<= 1u; + if (image_bit == 0u) { + image_bit = 1u; + ++image_ptr; + } } + + const std::array<u32, NUM_TEXTURE_AND_IMAGE_SCALING_WORDS>& Data() const noexcept { + return words; + } + +private: + std::array<u32, NUM_TEXTURE_AND_IMAGE_SCALING_WORDS> words{}; + u32* texture_ptr{words.data()}; + u32* image_ptr{words.data() + Shader::Backend::SPIRV::NUM_TEXTURE_SCALING_WORDS}; + u32 texture_bit{1u}; + u32 image_bit{1u}; +}; + +inline void PushImageDescriptors(TextureCache& texture_cache, + VKUpdateDescriptorQueue& update_descriptor_queue, + const Shader::Info& info, RescalingPushConstant& rescaling, + const VkSampler*& samplers, + const VideoCommon::ImageViewInOut*& views) { + const u32 num_texture_buffers = Shader::NumDescriptors(info.texture_buffer_descriptors); + const u32 num_image_buffers = Shader::NumDescriptors(info.image_buffer_descriptors); + views += num_texture_buffers; + views += num_image_buffers; for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { + const VideoCommon::ImageViewId image_view_id{(views++)->id}; const VkSampler sampler{*(samplers++)}; - ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))}; + ImageView& image_view{texture_cache.GetImageView(image_view_id)}; const VkImageView vk_image_view{image_view.Handle(desc.type)}; update_descriptor_queue.AddSampledImage(vk_image_view, sampler); + rescaling.PushTexture(texture_cache.IsRescaling(image_view)); } } for (const auto& desc : info.image_descriptors) { for (u32 index = 0; index < desc.count; ++index) { - ImageView& image_view{texture_cache.GetImageView(*(image_view_ids++))}; + ImageView& image_view{texture_cache.GetImageView((views++)->id)}; if (desc.is_written) { texture_cache.MarkModification(image_view.image_id); } const VkImageView vk_image_view{image_view.StorageView(desc.type, desc.format)}; update_descriptor_queue.AddImage(vk_image_view); + rescaling.PushImage(texture_cache.IsRescaling(image_view)); } } } diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index 888bc7392..1e447e621 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -12,14 +12,22 @@ #include "common/assert.h" #include "common/common_types.h" #include "common/math_util.h" +#include "common/settings.h" #include "core/core.h" #include "core/frontend/emu_window.h" #include "core/memory.h" #include "video_core/gpu.h" +#include "video_core/host_shaders/fxaa_frag_spv.h" +#include "video_core/host_shaders/fxaa_vert_spv.h" +#include "video_core/host_shaders/present_bicubic_frag_spv.h" +#include "video_core/host_shaders/present_gaussian_frag_spv.h" #include "video_core/host_shaders/vulkan_present_frag_spv.h" +#include "video_core/host_shaders/vulkan_present_scaleforce_fp16_frag_spv.h" +#include "video_core/host_shaders/vulkan_present_scaleforce_fp32_frag_spv.h" #include "video_core/host_shaders/vulkan_present_vert_spv.h" #include "video_core/renderer_vulkan/renderer_vulkan.h" #include "video_core/renderer_vulkan/vk_blit_screen.h" +#include "video_core/renderer_vulkan/vk_fsr.h" #include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_shader_util.h" @@ -144,8 +152,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, scheduler.Wait(resource_ticks[image_index]); resource_ticks[image_index] = scheduler.CurrentTick(); - UpdateDescriptorSet(image_index, - use_accelerated ? screen_info.image_view : *raw_image_views[image_index]); + VkImageView source_image_view = + use_accelerated ? screen_info.image_view : *raw_image_views[image_index]; BufferData data; SetUniformData(data, layout); @@ -222,9 +230,134 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, read_barrier); cmdbuf.CopyBufferToImage(*buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy); cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier); + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + 0, write_barrier); }); } + + const auto anti_alias_pass = Settings::values.anti_aliasing.GetValue(); + if (use_accelerated && anti_alias_pass != Settings::AntiAliasing::None) { + UpdateAADescriptorSet(image_index, source_image_view, false); + const u32 up_scale = Settings::values.resolution_info.up_scale; + const u32 down_shift = Settings::values.resolution_info.down_shift; + VkExtent2D size{ + .width = (up_scale * framebuffer.width) >> down_shift, + .height = (up_scale * framebuffer.height) >> down_shift, + }; + scheduler.Record([this, image_index, size, anti_alias_pass](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier base_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = 0, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = {}, + .subresourceRange = + { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }; + + { + VkImageMemoryBarrier fsr_write_barrier = base_barrier; + fsr_write_barrier.image = *aa_image; + fsr_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, fsr_write_barrier); + } + + const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f; + const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f; + const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f; + const VkClearValue clear_color{ + .color = {.float32 = {bg_red, bg_green, bg_blue, 1.0f}}, + }; + const VkRenderPassBeginInfo renderpass_bi{ + .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO, + .pNext = nullptr, + .renderPass = *aa_renderpass, + .framebuffer = *aa_framebuffer, + .renderArea = + { + .offset = {0, 0}, + .extent = size, + }, + .clearValueCount = 1, + .pClearValues = &clear_color, + }; + const VkViewport viewport{ + .x = 0.0f, + .y = 0.0f, + .width = static_cast<float>(size.width), + .height = static_cast<float>(size.height), + .minDepth = 0.0f, + .maxDepth = 1.0f, + }; + const VkRect2D scissor{ + .offset = {0, 0}, + .extent = size, + }; + cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE); + switch (anti_alias_pass) { + case Settings::AntiAliasing::Fxaa: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *aa_pipeline); + break; + default: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *aa_pipeline); + break; + } + cmdbuf.SetViewport(0, viewport); + cmdbuf.SetScissor(0, scissor); + + cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices)); + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *aa_pipeline_layout, 0, + aa_descriptor_sets[image_index], {}); + cmdbuf.Draw(4, 1, 0, 0); + cmdbuf.EndRenderPass(); + + { + VkImageMemoryBarrier blit_read_barrier = base_barrier; + blit_read_barrier.image = *aa_image; + blit_read_barrier.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + blit_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, blit_read_barrier); + } + }); + source_image_view = *aa_image_view; + } + + if (fsr) { + auto crop_rect = framebuffer.crop_rect; + if (crop_rect.GetWidth() == 0) { + crop_rect.right = framebuffer.width; + } + if (crop_rect.GetHeight() == 0) { + crop_rect.bottom = framebuffer.height; + } + crop_rect = crop_rect.Scale(Settings::values.resolution_info.up_factor); + VkExtent2D fsr_input_size{ + .width = Settings::values.resolution_info.ScaleUp(framebuffer.width), + .height = Settings::values.resolution_info.ScaleUp(framebuffer.height), + }; + VkImageView fsr_image_view = + fsr->Draw(scheduler, image_index, source_image_view, fsr_input_size, crop_rect); + UpdateDescriptorSet(image_index, fsr_image_view, true); + } else { + const bool is_nn = + Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::NearestNeighbor; + UpdateDescriptorSet(image_index, source_image_view, is_nn); + } + scheduler.Record( [this, host_framebuffer, image_index, size = render_area](vk::CommandBuffer cmdbuf) { const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f; @@ -258,8 +391,28 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, .offset = {0, 0}, .extent = size, }; + const auto filter = Settings::values.scaling_filter.GetValue(); cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE); - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline); + switch (filter) { + case Settings::ScalingFilter::NearestNeighbor: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bilinear_pipeline); + break; + case Settings::ScalingFilter::Bilinear: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bilinear_pipeline); + break; + case Settings::ScalingFilter::Bicubic: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bicubic_pipeline); + break; + case Settings::ScalingFilter::Gaussian: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *gaussian_pipeline); + break; + case Settings::ScalingFilter::ScaleForce: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *scaleforce_pipeline); + break; + default: + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *bilinear_pipeline); + break; + } cmdbuf.SetViewport(0, viewport); cmdbuf.SetScissor(0, scissor); @@ -281,11 +434,16 @@ VkSemaphore VKBlitScreen::DrawToSwapchain(const Tegra::FramebufferConfig& frameb } vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent) { + return CreateFramebuffer(image_view, extent, renderpass); +} + +vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent, + vk::RenderPass& rd) { return device.GetLogical().CreateFramebuffer(VkFramebufferCreateInfo{ .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, - .renderPass = *renderpass, + .renderPass = *rd, .attachmentCount = 1, .pAttachments = &image_view, .width = extent.width, @@ -308,9 +466,21 @@ void VKBlitScreen::CreateDynamicResources() { CreateRenderPass(); CreateFramebuffers(); CreateGraphicsPipeline(); + fsr.reset(); + if (Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::Fsr) { + CreateFSR(); + } } void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) { + if (Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::Fsr) { + if (!fsr) { + CreateFSR(); + } + } else { + fsr.reset(); + } + if (framebuffer.width == raw_width && framebuffer.height == raw_height && !raw_images.empty()) { return; } @@ -324,7 +494,16 @@ void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) void VKBlitScreen::CreateShaders() { vertex_shader = BuildShader(device, VULKAN_PRESENT_VERT_SPV); - fragment_shader = BuildShader(device, VULKAN_PRESENT_FRAG_SPV); + fxaa_vertex_shader = BuildShader(device, FXAA_VERT_SPV); + fxaa_fragment_shader = BuildShader(device, FXAA_FRAG_SPV); + bilinear_fragment_shader = BuildShader(device, VULKAN_PRESENT_FRAG_SPV); + bicubic_fragment_shader = BuildShader(device, PRESENT_BICUBIC_FRAG_SPV); + gaussian_fragment_shader = BuildShader(device, PRESENT_GAUSSIAN_FRAG_SPV); + if (device.IsFloat16Supported()) { + scaleforce_fragment_shader = BuildShader(device, VULKAN_PRESENT_SCALEFORCE_FP16_FRAG_SPV); + } else { + scaleforce_fragment_shader = BuildShader(device, VULKAN_PRESENT_SCALEFORCE_FP32_FRAG_SPV); + } } void VKBlitScreen::CreateSemaphores() { @@ -344,6 +523,13 @@ void VKBlitScreen::CreateDescriptorPool() { }, }}; + const std::array<VkDescriptorPoolSize, 1> pool_sizes_aa{{ + { + .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = static_cast<u32>(image_count * 2), + }, + }}; + const VkDescriptorPoolCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, .pNext = nullptr, @@ -353,19 +539,33 @@ void VKBlitScreen::CreateDescriptorPool() { .pPoolSizes = pool_sizes.data(), }; descriptor_pool = device.GetLogical().CreateDescriptorPool(ci); + + const VkDescriptorPoolCreateInfo ci_aa{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, + .pNext = nullptr, + .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, + .maxSets = static_cast<u32>(image_count), + .poolSizeCount = static_cast<u32>(pool_sizes_aa.size()), + .pPoolSizes = pool_sizes_aa.data(), + }; + aa_descriptor_pool = device.GetLogical().CreateDescriptorPool(ci_aa); } void VKBlitScreen::CreateRenderPass() { + renderpass = CreateRenderPassImpl(swapchain.GetImageViewFormat()); +} + +vk::RenderPass VKBlitScreen::CreateRenderPassImpl(VkFormat format, bool is_present) { const VkAttachmentDescription color_attachment{ .flags = 0, - .format = swapchain.GetImageViewFormat(), + .format = format, .samples = VK_SAMPLE_COUNT_1_BIT, .loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR, .storeOp = VK_ATTACHMENT_STORE_OP_STORE, .stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE, .stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE, .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED, - .finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, + .finalLayout = is_present ? VK_IMAGE_LAYOUT_PRESENT_SRC_KHR : VK_IMAGE_LAYOUT_GENERAL, }; const VkAttachmentReference color_attachment_ref{ @@ -408,7 +608,7 @@ void VKBlitScreen::CreateRenderPass() { .pDependencies = &dependency, }; - renderpass = device.GetLogical().CreateRenderPass(renderpass_ci); + return device.GetLogical().CreateRenderPass(renderpass_ci); } void VKBlitScreen::CreateDescriptorSetLayout() { @@ -429,6 +629,23 @@ void VKBlitScreen::CreateDescriptorSetLayout() { }, }}; + const std::array<VkDescriptorSetLayoutBinding, 2> layout_bindings_aa{{ + { + .binding = 0, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_VERTEX_BIT, + .pImmutableSamplers = nullptr, + }, + { + .binding = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT, + .pImmutableSamplers = nullptr, + }, + }}; + const VkDescriptorSetLayoutCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, .pNext = nullptr, @@ -437,11 +654,21 @@ void VKBlitScreen::CreateDescriptorSetLayout() { .pBindings = layout_bindings.data(), }; + const VkDescriptorSetLayoutCreateInfo ci_aa{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .bindingCount = static_cast<u32>(layout_bindings_aa.size()), + .pBindings = layout_bindings_aa.data(), + }; + descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci); + aa_descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci_aa); } void VKBlitScreen::CreateDescriptorSets() { const std::vector layouts(image_count, *descriptor_set_layout); + const std::vector layouts_aa(image_count, *aa_descriptor_set_layout); const VkDescriptorSetAllocateInfo ai{ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, @@ -451,7 +678,16 @@ void VKBlitScreen::CreateDescriptorSets() { .pSetLayouts = layouts.data(), }; + const VkDescriptorSetAllocateInfo ai_aa{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, + .pNext = nullptr, + .descriptorPool = *aa_descriptor_pool, + .descriptorSetCount = static_cast<u32>(image_count), + .pSetLayouts = layouts_aa.data(), + }; + descriptor_sets = descriptor_pool.Allocate(ai); + aa_descriptor_sets = aa_descriptor_pool.Allocate(ai_aa); } void VKBlitScreen::CreatePipelineLayout() { @@ -464,11 +700,63 @@ void VKBlitScreen::CreatePipelineLayout() { .pushConstantRangeCount = 0, .pPushConstantRanges = nullptr, }; + const VkPipelineLayoutCreateInfo ci_aa{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .setLayoutCount = 1, + .pSetLayouts = aa_descriptor_set_layout.address(), + .pushConstantRangeCount = 0, + .pPushConstantRanges = nullptr, + }; pipeline_layout = device.GetLogical().CreatePipelineLayout(ci); + aa_pipeline_layout = device.GetLogical().CreatePipelineLayout(ci_aa); } void VKBlitScreen::CreateGraphicsPipeline() { - const std::array<VkPipelineShaderStageCreateInfo, 2> shader_stages{{ + const std::array<VkPipelineShaderStageCreateInfo, 2> bilinear_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *bilinear_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const std::array<VkPipelineShaderStageCreateInfo, 2> bicubic_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *bicubic_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const std::array<VkPipelineShaderStageCreateInfo, 2> gaussian_shader_stages{{ { .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, .pNext = nullptr, @@ -483,7 +771,28 @@ void VKBlitScreen::CreateGraphicsPipeline() { .pNext = nullptr, .flags = 0, .stage = VK_SHADER_STAGE_FRAGMENT_BIT, - .module = *fragment_shader, + .module = *gaussian_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const std::array<VkPipelineShaderStageCreateInfo, 2> scaleforce_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *scaleforce_fragment_shader, .pName = "main", .pSpecializationInfo = nullptr, }, @@ -583,12 +892,12 @@ void VKBlitScreen::CreateGraphicsPipeline() { .pDynamicStates = dynamic_states.data(), }; - const VkGraphicsPipelineCreateInfo pipeline_ci{ + const VkGraphicsPipelineCreateInfo bilinear_pipeline_ci{ .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, .pNext = nullptr, .flags = 0, - .stageCount = static_cast<u32>(shader_stages.size()), - .pStages = shader_stages.data(), + .stageCount = static_cast<u32>(bilinear_shader_stages.size()), + .pStages = bilinear_shader_stages.data(), .pVertexInputState = &vertex_input_ci, .pInputAssemblyState = &input_assembly_ci, .pTessellationState = nullptr, @@ -605,7 +914,76 @@ void VKBlitScreen::CreateGraphicsPipeline() { .basePipelineIndex = 0, }; - pipeline = device.GetLogical().CreateGraphicsPipeline(pipeline_ci); + const VkGraphicsPipelineCreateInfo bicubic_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(bicubic_shader_stages.size()), + .pStages = bicubic_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *pipeline_layout, + .renderPass = *renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + const VkGraphicsPipelineCreateInfo gaussian_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(gaussian_shader_stages.size()), + .pStages = gaussian_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *pipeline_layout, + .renderPass = *renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + const VkGraphicsPipelineCreateInfo scaleforce_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(scaleforce_shader_stages.size()), + .pStages = scaleforce_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *pipeline_layout, + .renderPass = *renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + bilinear_pipeline = device.GetLogical().CreateGraphicsPipeline(bilinear_pipeline_ci); + bicubic_pipeline = device.GetLogical().CreateGraphicsPipeline(bicubic_pipeline_ci); + gaussian_pipeline = device.GetLogical().CreateGraphicsPipeline(gaussian_pipeline_ci); + scaleforce_pipeline = device.GetLogical().CreateGraphicsPipeline(scaleforce_pipeline_ci); } void VKBlitScreen::CreateSampler() { @@ -614,8 +992,29 @@ void VKBlitScreen::CreateSampler() { .pNext = nullptr, .flags = 0, .magFilter = VK_FILTER_LINEAR, + .minFilter = VK_FILTER_LINEAR, + .mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST, + .addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, + .addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, + .addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, + .mipLodBias = 0.0f, + .anisotropyEnable = VK_FALSE, + .maxAnisotropy = 0.0f, + .compareEnable = VK_FALSE, + .compareOp = VK_COMPARE_OP_NEVER, + .minLod = 0.0f, + .maxLod = 0.0f, + .borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK, + .unnormalizedCoordinates = VK_FALSE, + }; + + const VkSamplerCreateInfo ci_nn{ + .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .magFilter = VK_FILTER_NEAREST, .minFilter = VK_FILTER_NEAREST, - .mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR, + .mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST, .addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, .addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, .addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER, @@ -631,6 +1030,7 @@ void VKBlitScreen::CreateSampler() { }; sampler = device.GetLogical().CreateSampler(ci); + nn_sampler = device.GetLogical().CreateSampler(ci_nn); } void VKBlitScreen::CreateFramebuffers() { @@ -639,7 +1039,7 @@ void VKBlitScreen::CreateFramebuffers() { for (std::size_t i = 0; i < image_count; ++i) { const VkImageView image_view{swapchain.GetImageViewIndex(i)}; - framebuffers[i] = CreateFramebuffer(image_view, size); + framebuffers[i] = CreateFramebuffer(image_view, size, renderpass); } } @@ -649,6 +1049,11 @@ void VKBlitScreen::ReleaseRawImages() { } raw_images.clear(); raw_buffer_commits.clear(); + + aa_image_view.reset(); + aa_image.reset(); + aa_commit = MemoryCommit{}; + buffer.reset(); buffer_commit = MemoryCommit{}; } @@ -675,8 +1080,11 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) raw_image_views.resize(image_count); raw_buffer_commits.resize(image_count); - for (size_t i = 0; i < image_count; ++i) { - raw_images[i] = device.GetLogical().CreateImage(VkImageCreateInfo{ + const auto create_image = [&](bool used_on_framebuffer = false, u32 up_scale = 1, + u32 down_shift = 0) { + u32 extra_usages = used_on_framebuffer ? VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT + : VK_IMAGE_USAGE_TRANSFER_DST_BIT; + return device.GetLogical().CreateImage(VkImageCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, .pNext = nullptr, .flags = 0, @@ -684,26 +1092,30 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) .format = GetFormat(framebuffer), .extent = { - .width = framebuffer.width, - .height = framebuffer.height, + .width = (up_scale * framebuffer.width) >> down_shift, + .height = (up_scale * framebuffer.height) >> down_shift, .depth = 1, }, .mipLevels = 1, .arrayLayers = 1, .samples = VK_SAMPLE_COUNT_1_BIT, - .tiling = VK_IMAGE_TILING_LINEAR, - .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT, + .tiling = used_on_framebuffer ? VK_IMAGE_TILING_OPTIMAL : VK_IMAGE_TILING_LINEAR, + .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT | extra_usages, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED, }); - raw_buffer_commits[i] = memory_allocator.Commit(raw_images[i], MemoryUsage::DeviceLocal); - raw_image_views[i] = device.GetLogical().CreateImageView(VkImageViewCreateInfo{ + }; + const auto create_commit = [&](vk::Image& image) { + return memory_allocator.Commit(image, MemoryUsage::DeviceLocal); + }; + const auto create_image_view = [&](vk::Image& image) { + return device.GetLogical().CreateImageView(VkImageViewCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = nullptr, .flags = 0, - .image = *raw_images[i], + .image = *image, .viewType = VK_IMAGE_VIEW_TYPE_2D, .format = GetFormat(framebuffer), .components = @@ -722,10 +1134,211 @@ void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) .layerCount = 1, }, }); + }; + + for (size_t i = 0; i < image_count; ++i) { + raw_images[i] = create_image(); + raw_buffer_commits[i] = create_commit(raw_images[i]); + raw_image_views[i] = create_image_view(raw_images[i]); } + + // AA Resources + const u32 up_scale = Settings::values.resolution_info.up_scale; + const u32 down_shift = Settings::values.resolution_info.down_shift; + aa_image = create_image(true, up_scale, down_shift); + aa_commit = create_commit(aa_image); + aa_image_view = create_image_view(aa_image); + VkExtent2D size{ + .width = (up_scale * framebuffer.width) >> down_shift, + .height = (up_scale * framebuffer.height) >> down_shift, + }; + if (aa_renderpass) { + aa_framebuffer = CreateFramebuffer(*aa_image_view, size, aa_renderpass); + return; + } + aa_renderpass = CreateRenderPassImpl(GetFormat(framebuffer), false); + aa_framebuffer = CreateFramebuffer(*aa_image_view, size, aa_renderpass); + + const std::array<VkPipelineShaderStageCreateInfo, 2> fxaa_shader_stages{{ + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_VERTEX_BIT, + .module = *fxaa_vertex_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + { + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_FRAGMENT_BIT, + .module = *fxaa_fragment_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }, + }}; + + const auto vertex_binding_description = ScreenRectVertex::GetDescription(); + const auto vertex_attrs_description = ScreenRectVertex::GetAttributes(); + + const VkPipelineVertexInputStateCreateInfo vertex_input_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .vertexBindingDescriptionCount = 1, + .pVertexBindingDescriptions = &vertex_binding_description, + .vertexAttributeDescriptionCount = u32{vertex_attrs_description.size()}, + .pVertexAttributeDescriptions = vertex_attrs_description.data(), + }; + + const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP, + .primitiveRestartEnable = VK_FALSE, + }; + + const VkPipelineViewportStateCreateInfo viewport_state_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .viewportCount = 1, + .pViewports = nullptr, + .scissorCount = 1, + .pScissors = nullptr, + }; + + const VkPipelineRasterizationStateCreateInfo rasterization_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .depthClampEnable = VK_FALSE, + .rasterizerDiscardEnable = VK_FALSE, + .polygonMode = VK_POLYGON_MODE_FILL, + .cullMode = VK_CULL_MODE_NONE, + .frontFace = VK_FRONT_FACE_CLOCKWISE, + .depthBiasEnable = VK_FALSE, + .depthBiasConstantFactor = 0.0f, + .depthBiasClamp = 0.0f, + .depthBiasSlopeFactor = 0.0f, + .lineWidth = 1.0f, + }; + + const VkPipelineMultisampleStateCreateInfo multisampling_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .rasterizationSamples = VK_SAMPLE_COUNT_1_BIT, + .sampleShadingEnable = VK_FALSE, + .minSampleShading = 0.0f, + .pSampleMask = nullptr, + .alphaToCoverageEnable = VK_FALSE, + .alphaToOneEnable = VK_FALSE, + }; + + const VkPipelineColorBlendAttachmentState color_blend_attachment{ + .blendEnable = VK_FALSE, + .srcColorBlendFactor = VK_BLEND_FACTOR_ZERO, + .dstColorBlendFactor = VK_BLEND_FACTOR_ZERO, + .colorBlendOp = VK_BLEND_OP_ADD, + .srcAlphaBlendFactor = VK_BLEND_FACTOR_ZERO, + .dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO, + .alphaBlendOp = VK_BLEND_OP_ADD, + .colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | + VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT, + }; + + const VkPipelineColorBlendStateCreateInfo color_blend_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .logicOpEnable = VK_FALSE, + .logicOp = VK_LOGIC_OP_COPY, + .attachmentCount = 1, + .pAttachments = &color_blend_attachment, + .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f}, + }; + + static constexpr std::array dynamic_states{ + VK_DYNAMIC_STATE_VIEWPORT, + VK_DYNAMIC_STATE_SCISSOR, + }; + const VkPipelineDynamicStateCreateInfo dynamic_state_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .dynamicStateCount = static_cast<u32>(dynamic_states.size()), + .pDynamicStates = dynamic_states.data(), + }; + + const VkGraphicsPipelineCreateInfo fxaa_pipeline_ci{ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(fxaa_shader_stages.size()), + .pStages = fxaa_shader_stages.data(), + .pVertexInputState = &vertex_input_ci, + .pInputAssemblyState = &input_assembly_ci, + .pTessellationState = nullptr, + .pViewportState = &viewport_state_ci, + .pRasterizationState = &rasterization_ci, + .pMultisampleState = &multisampling_ci, + .pDepthStencilState = nullptr, + .pColorBlendState = &color_blend_ci, + .pDynamicState = &dynamic_state_ci, + .layout = *aa_pipeline_layout, + .renderPass = *aa_renderpass, + .subpass = 0, + .basePipelineHandle = 0, + .basePipelineIndex = 0, + }; + + // AA + aa_pipeline = device.GetLogical().CreateGraphicsPipeline(fxaa_pipeline_ci); } -void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const { +void VKBlitScreen::UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view, + bool nn) const { + const VkDescriptorImageInfo image_info{ + .sampler = nn ? *nn_sampler : *sampler, + .imageView = image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + + const VkWriteDescriptorSet sampler_write{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = aa_descriptor_sets[image_index], + .dstBinding = 0, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .pImageInfo = &image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + const VkWriteDescriptorSet sampler_write_2{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = aa_descriptor_sets[image_index], + .dstBinding = 1, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .pImageInfo = &image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, sampler_write_2}, {}); +} + +void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view, + bool nn) const { const VkDescriptorBufferInfo buffer_info{ .buffer = *buffer, .offset = offsetof(BufferData, uniform), @@ -746,7 +1359,7 @@ void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView imag }; const VkDescriptorImageInfo image_info{ - .sampler = *sampler, + .sampler = nn ? *nn_sampler : *sampler, .imageView = image_view, .imageLayout = VK_IMAGE_LAYOUT_GENERAL, }; @@ -798,17 +1411,19 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi UNIMPLEMENTED_IF(framebuffer_crop_rect.top != 0); UNIMPLEMENTED_IF(framebuffer_crop_rect.left != 0); - // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering - // (e.g. handheld mode) on a 1920x1080 framebuffer. f32 scale_u = 1.0f; f32 scale_v = 1.0f; - if (framebuffer_crop_rect.GetWidth() > 0) { - scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) / - static_cast<f32>(screen_info.width); - } - if (framebuffer_crop_rect.GetHeight() > 0) { - scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) / - static_cast<f32>(screen_info.height); + // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering + // (e.g. handheld mode) on a 1920x1080 framebuffer. + if (!fsr) { + if (framebuffer_crop_rect.GetWidth() > 0) { + scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) / + static_cast<f32>(screen_info.width); + } + if (framebuffer_crop_rect.GetHeight() > 0) { + scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) / + static_cast<f32>(screen_info.height); + } } const auto& screen = layout.screen; @@ -822,6 +1437,15 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi data.vertices[3] = ScreenRectVertex(x + w, y + h, texcoords.bottom * scale_u, right * scale_v); } +void VKBlitScreen::CreateFSR() { + const auto& layout = render_window.GetFramebufferLayout(); + const VkExtent2D fsr_size{ + .width = layout.screen.GetWidth(), + .height = layout.screen.GetHeight(), + }; + fsr = std::make_unique<FSR>(device, memory_allocator, image_count, fsr_size); +} + u64 VKBlitScreen::CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const { return sizeof(BufferData) + GetSizeInBytes(framebuffer) * image_count; } diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h index 430bcfbca..bbca71af3 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.h +++ b/src/video_core/renderer_vulkan/vk_blit_screen.h @@ -34,6 +34,7 @@ namespace Vulkan { struct ScreenInfo; class Device; +class FSR; class RasterizerVulkan; class VKScheduler; class VKSwapchain; @@ -66,6 +67,9 @@ public: [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent); + [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view, + VkExtent2D extent, vk::RenderPass& rd); + private: struct BufferData; @@ -74,6 +78,7 @@ private: void CreateSemaphores(); void CreateDescriptorPool(); void CreateRenderPass(); + vk::RenderPass CreateRenderPassImpl(VkFormat, bool is_present = true); void CreateDescriptorSetLayout(); void CreateDescriptorSets(); void CreatePipelineLayout(); @@ -88,11 +93,14 @@ private: void CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer); void CreateRawImages(const Tegra::FramebufferConfig& framebuffer); - void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const; + void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view, bool nn) const; + void UpdateAADescriptorSet(std::size_t image_index, VkImageView image_view, bool nn) const; void SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const; void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer, const Layout::FramebufferLayout layout) const; + void CreateFSR(); + u64 CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const; u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer, std::size_t image_index) const; @@ -107,14 +115,24 @@ private: const VKScreenInfo& screen_info; vk::ShaderModule vertex_shader; - vk::ShaderModule fragment_shader; + vk::ShaderModule fxaa_vertex_shader; + vk::ShaderModule fxaa_fragment_shader; + vk::ShaderModule bilinear_fragment_shader; + vk::ShaderModule bicubic_fragment_shader; + vk::ShaderModule gaussian_fragment_shader; + vk::ShaderModule scaleforce_fragment_shader; vk::DescriptorPool descriptor_pool; vk::DescriptorSetLayout descriptor_set_layout; vk::PipelineLayout pipeline_layout; - vk::Pipeline pipeline; + vk::Pipeline nearest_neightbor_pipeline; + vk::Pipeline bilinear_pipeline; + vk::Pipeline bicubic_pipeline; + vk::Pipeline gaussian_pipeline; + vk::Pipeline scaleforce_pipeline; vk::RenderPass renderpass; std::vector<vk::Framebuffer> framebuffers; vk::DescriptorSets descriptor_sets; + vk::Sampler nn_sampler; vk::Sampler sampler; vk::Buffer buffer; @@ -126,8 +144,22 @@ private: std::vector<vk::Image> raw_images; std::vector<vk::ImageView> raw_image_views; std::vector<MemoryCommit> raw_buffer_commits; + + vk::DescriptorPool aa_descriptor_pool; + vk::DescriptorSetLayout aa_descriptor_set_layout; + vk::PipelineLayout aa_pipeline_layout; + vk::Pipeline aa_pipeline; + vk::RenderPass aa_renderpass; + vk::Framebuffer aa_framebuffer; + vk::DescriptorSets aa_descriptor_sets; + vk::Image aa_image; + vk::ImageView aa_image_view; + MemoryCommit aa_commit; + u32 raw_width = 0; u32 raw_height = 0; + + std::unique_ptr<FSR> fsr; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 8ac58bc2f..5ffd93499 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -146,7 +146,7 @@ void BufferCacheRuntime::Finish() { } void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, - std::span<const VideoCommon::BufferCopy> copies) { + std::span<const VideoCommon::BufferCopy> copies, bool barrier) { static constexpr VkMemoryBarrier READ_BARRIER{ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, .pNext = nullptr, @@ -163,10 +163,42 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, boost::container::small_vector<VkBufferCopy, 3> vk_copies(copies.size()); std::ranges::transform(copies, vk_copies.begin(), MakeBufferCopy); scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) { + scheduler.Record([src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) { + if (barrier) { + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER); + } + cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies); + if (barrier) { + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER); + } + }); +} + +void BufferCacheRuntime::PreCopyBarrier() { + static constexpr VkMemoryBarrier READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + }; + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER); - cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies); + }); +} + +void BufferCacheRuntime::PostCopyBarrier() { + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + }; + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER); }); diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index c27402ff0..1ee0d8420 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -69,8 +69,12 @@ public: [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size); + void PreCopyBarrier(); + void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer, - std::span<const VideoCommon::BufferCopy> copies); + std::span<const VideoCommon::BufferCopy> copies, bool barrier = true); + + void PostCopyBarrier(); void ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value); diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index 44faf626a..de36bcdb7 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -22,6 +22,7 @@ namespace Vulkan { using Shader::ImageBufferDescriptor; +using Shader::Backend::SPIRV::RESCALING_LAYOUT_WORDS_OFFSET; using Tegra::Texture::TexturePair; ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool, @@ -108,8 +109,7 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, texture_cache.SynchronizeComputeDescriptors(); static constexpr size_t max_elements = 64; - std::array<ImageId, max_elements> image_view_ids; - boost::container::static_vector<u32, max_elements> image_view_indices; + boost::container::static_vector<VideoCommon::ImageViewInOut, max_elements> views; boost::container::static_vector<VkSampler, max_elements> samplers; const auto& qmd{kepler_compute.launch_description}; @@ -134,30 +134,37 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, } return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); }}; - const auto add_image{[&](const auto& desc) { + const auto add_image{[&](const auto& desc, bool blacklist) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices.push_back(handle.first); + views.push_back({ + .index = handle.first, + .blacklist = blacklist, + .id = {}, + }); } }}; - std::ranges::for_each(info.texture_buffer_descriptors, add_image); - std::ranges::for_each(info.image_buffer_descriptors, add_image); + for (const auto& desc : info.texture_buffer_descriptors) { + add_image(desc, false); + } + for (const auto& desc : info.image_buffer_descriptors) { + add_image(desc, false); + } for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices.push_back(handle.first); + views.push_back({handle.first}); Sampler* const sampler = texture_cache.GetComputeSampler(handle.second); samplers.push_back(sampler->Handle()); } } - std::ranges::for_each(info.image_descriptors, add_image); - - const std::span indices_span(image_view_indices.data(), image_view_indices.size()); - texture_cache.FillComputeImageViews(indices_span, image_view_ids); + for (const auto& desc : info.image_descriptors) { + add_image(desc, desc.is_written); + } + texture_cache.FillComputeImageViews(std::span(views.data(), views.size())); buffer_cache.UnbindComputeTextureBuffers(); - ImageId* texture_buffer_ids{image_view_ids.data()}; size_t index{}; const auto add_buffer{[&](const auto& desc) { constexpr bool is_image = std::is_same_v<decltype(desc), const ImageBufferDescriptor&>; @@ -166,11 +173,10 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, if constexpr (is_image) { is_written = desc.is_written; } - ImageView& image_view = texture_cache.GetImageView(*texture_buffer_ids); + ImageView& image_view = texture_cache.GetImageView(views[index].id); buffer_cache.BindComputeTextureBuffer(index, image_view.GpuAddr(), image_view.BufferSize(), image_view.format, is_written, is_image); - ++texture_buffer_ids; ++index; } }}; @@ -180,9 +186,11 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, buffer_cache.UpdateComputeBuffers(); buffer_cache.BindHostComputeBuffers(); + RescalingPushConstant rescaling; const VkSampler* samplers_it{samplers.data()}; - const ImageId* views_it{image_view_ids.data()}; - PushImageDescriptors(info, samplers_it, views_it, texture_cache, update_descriptor_queue); + const VideoCommon::ImageViewInOut* views_it{views.data()}; + PushImageDescriptors(texture_cache, update_descriptor_queue, info, rescaling, samplers_it, + views_it); if (!is_built.load(std::memory_order::relaxed)) { // Wait for the pipeline to be built @@ -192,11 +200,18 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, }); } const void* const descriptor_data{update_descriptor_queue.UpdateData()}; - scheduler.Record([this, descriptor_data](vk::CommandBuffer cmdbuf) { + const bool is_rescaling = !info.texture_descriptors.empty() || !info.image_descriptors.empty(); + scheduler.Record([this, descriptor_data, is_rescaling, + rescaling_data = rescaling.Data()](vk::CommandBuffer cmdbuf) { cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); if (!descriptor_set_layout) { return; } + if (is_rescaling) { + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, + RESCALING_LAYOUT_WORDS_OFFSET, sizeof(rescaling_data), + rescaling_data.data()); + } const VkDescriptorSet descriptor_set{descriptor_allocator.Commit()}; const vk::Device& dev{device.GetLogical()}; dev.UpdateDescriptorSet(descriptor_set, *descriptor_update_template, descriptor_data); diff --git a/src/video_core/renderer_vulkan/vk_fsr.cpp b/src/video_core/renderer_vulkan/vk_fsr.cpp new file mode 100644 index 000000000..73629d229 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_fsr.cpp @@ -0,0 +1,553 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <cmath> +#include "common/bit_cast.h" +#include "common/common_types.h" +#include "common/div_ceil.h" + +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp16_comp_spv.h" +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32_comp_spv.h" +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16_comp_spv.h" +#include "video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32_comp_spv.h" +#include "video_core/renderer_vulkan/vk_fsr.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_shader_util.h" +#include "video_core/vulkan_common/vulkan_device.h" + +namespace Vulkan { +namespace { +// Reimplementations of the constant generating functions in ffx_fsr1.h +// GCC generated a lot of warnings when using the official header. +u32 AU1_AH1_AF1(f32 f) { + static constexpr u32 base[512]{ + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, + 0x0000, 0x0000, 0x0000, 0x0000, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, + 0x0080, 0x0100, 0x0200, 0x0400, 0x0800, 0x0c00, 0x1000, 0x1400, 0x1800, 0x1c00, 0x2000, + 0x2400, 0x2800, 0x2c00, 0x3000, 0x3400, 0x3800, 0x3c00, 0x4000, 0x4400, 0x4800, 0x4c00, + 0x5000, 0x5400, 0x5800, 0x5c00, 0x6000, 0x6400, 0x6800, 0x6c00, 0x7000, 0x7400, 0x7800, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, + 0x7bff, 0x7bff, 0x7bff, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, + 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8001, 0x8002, 0x8004, 0x8008, + 0x8010, 0x8020, 0x8040, 0x8080, 0x8100, 0x8200, 0x8400, 0x8800, 0x8c00, 0x9000, 0x9400, + 0x9800, 0x9c00, 0xa000, 0xa400, 0xa800, 0xac00, 0xb000, 0xb400, 0xb800, 0xbc00, 0xc000, + 0xc400, 0xc800, 0xcc00, 0xd000, 0xd400, 0xd800, 0xdc00, 0xe000, 0xe400, 0xe800, 0xec00, + 0xf000, 0xf400, 0xf800, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, + }; + static constexpr s8 shift[512]{ + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x17, 0x16, + 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x17, + 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, + 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, + 0x18, 0x18, + }; + const u32 u = Common::BitCast<u32>(f); + const u32 i = u >> 23; + return base[i] + ((u & 0x7fffff) >> shift[i]); +} + +u32 AU1_AH2_AF2(f32 a[2]) { + return AU1_AH1_AF1(a[0]) + (AU1_AH1_AF1(a[1]) << 16); +} + +void FsrEasuCon(u32 con0[4], u32 con1[4], u32 con2[4], u32 con3[4], f32 inputViewportInPixelsX, + f32 inputViewportInPixelsY, f32 inputSizeInPixelsX, f32 inputSizeInPixelsY, + f32 outputSizeInPixelsX, f32 outputSizeInPixelsY) { + con0[0] = Common::BitCast<u32>(inputViewportInPixelsX / outputSizeInPixelsX); + con0[1] = Common::BitCast<u32>(inputViewportInPixelsY / outputSizeInPixelsY); + con0[2] = Common::BitCast<u32>(0.5f * inputViewportInPixelsX / outputSizeInPixelsX - 0.5f); + con0[3] = Common::BitCast<u32>(0.5f * inputViewportInPixelsY / outputSizeInPixelsY - 0.5f); + con1[0] = Common::BitCast<u32>(1.0f / inputSizeInPixelsX); + con1[1] = Common::BitCast<u32>(1.0f / inputSizeInPixelsY); + con1[2] = Common::BitCast<u32>(1.0f / inputSizeInPixelsX); + con1[3] = Common::BitCast<u32>(-1.0f / inputSizeInPixelsY); + con2[0] = Common::BitCast<u32>(-1.0f / inputSizeInPixelsX); + con2[1] = Common::BitCast<u32>(2.0f / inputSizeInPixelsY); + con2[2] = Common::BitCast<u32>(1.0f / inputSizeInPixelsX); + con2[3] = Common::BitCast<u32>(2.0f / inputSizeInPixelsY); + con3[0] = Common::BitCast<u32>(0.0f / inputSizeInPixelsX); + con3[1] = Common::BitCast<u32>(4.0f / inputSizeInPixelsY); + con3[2] = con3[3] = 0; +} + +void FsrEasuConOffset(u32 con0[4], u32 con1[4], u32 con2[4], u32 con3[4], + f32 inputViewportInPixelsX, f32 inputViewportInPixelsY, + f32 inputSizeInPixelsX, f32 inputSizeInPixelsY, f32 outputSizeInPixelsX, + f32 outputSizeInPixelsY, f32 inputOffsetInPixelsX, f32 inputOffsetInPixelsY) { + FsrEasuCon(con0, con1, con2, con3, inputViewportInPixelsX, inputViewportInPixelsY, + inputSizeInPixelsX, inputSizeInPixelsY, outputSizeInPixelsX, outputSizeInPixelsY); + con0[2] = Common::BitCast<u32>(0.5f * inputViewportInPixelsX / outputSizeInPixelsX - 0.5f + + inputOffsetInPixelsX); + con0[3] = Common::BitCast<u32>(0.5f * inputViewportInPixelsY / outputSizeInPixelsY - 0.5f + + inputOffsetInPixelsY); +} + +void FsrRcasCon(u32* con, f32 sharpness) { + sharpness = std::exp2f(-sharpness); + f32 hSharp[2]{sharpness, sharpness}; + con[0] = Common::BitCast<u32>(sharpness); + con[1] = AU1_AH2_AF2(hSharp); + con[2] = 0; + con[3] = 0; +} +} // Anonymous namespace + +FSR::FSR(const Device& device_, MemoryAllocator& memory_allocator_, size_t image_count_, + VkExtent2D output_size_) + : device{device_}, memory_allocator{memory_allocator_}, image_count{image_count_}, + output_size{output_size_} { + + CreateImages(); + CreateSampler(); + CreateShaders(); + CreateDescriptorPool(); + CreateDescriptorSetLayout(); + CreateDescriptorSets(); + CreatePipelineLayout(); + CreatePipeline(); +} + +VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view, + VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect) { + + UpdateDescriptorSet(image_index, image_view); + + scheduler.Record([this, image_index, input_image_extent, crop_rect](vk::CommandBuffer cmdbuf) { + const VkImageMemoryBarrier base_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = 0, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = {}, + .subresourceRange = + { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }; + + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *easu_pipeline); + + std::array<u32, 4 * 4> push_constants; + FsrEasuConOffset( + push_constants.data() + 0, push_constants.data() + 4, push_constants.data() + 8, + push_constants.data() + 12, + + static_cast<f32>(crop_rect.GetWidth()), static_cast<f32>(crop_rect.GetHeight()), + static_cast<f32>(input_image_extent.width), static_cast<f32>(input_image_extent.height), + static_cast<f32>(output_size.width), static_cast<f32>(output_size.height), + static_cast<f32>(crop_rect.left), static_cast<f32>(crop_rect.top)); + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, push_constants); + + { + VkImageMemoryBarrier fsr_write_barrier = base_barrier; + fsr_write_barrier.image = *images[image_index], + fsr_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, fsr_write_barrier); + } + + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0, + descriptor_sets[image_index * 2], {}); + cmdbuf.Dispatch(Common::DivCeil(output_size.width, 16u), + Common::DivCeil(output_size.height, 16u), 1); + + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *rcas_pipeline); + + FsrRcasCon(push_constants.data(), 0.25f); + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, push_constants); + + { + std::array<VkImageMemoryBarrier, 2> barriers; + auto& fsr_read_barrier = barriers[0]; + auto& blit_write_barrier = barriers[1]; + + fsr_read_barrier = base_barrier; + fsr_read_barrier.image = *images[image_index]; + fsr_read_barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; + fsr_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + + blit_write_barrier = base_barrier; + blit_write_barrier.image = *images[image_count + image_index]; + blit_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED; + blit_write_barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, {}, {}, barriers); + } + + cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0, + descriptor_sets[image_index * 2 + 1], {}); + cmdbuf.Dispatch(Common::DivCeil(output_size.width, 16u), + Common::DivCeil(output_size.height, 16u), 1); + + { + std::array<VkImageMemoryBarrier, 1> barriers; + auto& blit_read_barrier = barriers[0]; + + blit_read_barrier = base_barrier; + blit_read_barrier.image = *images[image_count + image_index]; + blit_read_barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; + blit_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, {}, {}, barriers); + } + }); + + return *image_views[image_count + image_index]; +} + +void FSR::CreateDescriptorPool() { + const std::array<VkDescriptorPoolSize, 2> pool_sizes{{ + { + .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = static_cast<u32>(image_count * 2), + }, + { + .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .descriptorCount = static_cast<u32>(image_count * 2), + }, + }}; + + const VkDescriptorPoolCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO, + .pNext = nullptr, + .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT, + .maxSets = static_cast<u32>(image_count * 2), + .poolSizeCount = static_cast<u32>(pool_sizes.size()), + .pPoolSizes = pool_sizes.data(), + }; + descriptor_pool = device.GetLogical().CreateDescriptorPool(ci); +} + +void FSR::CreateDescriptorSetLayout() { + const std::array<VkDescriptorSetLayoutBinding, 2> layout_bindings{{ + { + .binding = 0, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = sampler.address(), + }, + { + .binding = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .descriptorCount = 1, + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .pImmutableSamplers = sampler.address(), + }, + }}; + + const VkDescriptorSetLayoutCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .bindingCount = static_cast<u32>(layout_bindings.size()), + .pBindings = layout_bindings.data(), + }; + + descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci); +} + +void FSR::CreateDescriptorSets() { + const u32 sets = static_cast<u32>(image_count * 2); + const std::vector layouts(sets, *descriptor_set_layout); + + const VkDescriptorSetAllocateInfo ai{ + .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO, + .pNext = nullptr, + .descriptorPool = *descriptor_pool, + .descriptorSetCount = sets, + .pSetLayouts = layouts.data(), + }; + + descriptor_sets = descriptor_pool.Allocate(ai); +} + +void FSR::CreateImages() { + images.resize(image_count * 2); + image_views.resize(image_count * 2); + buffer_commits.resize(image_count * 2); + + for (size_t i = 0; i < image_count * 2; ++i) { + images[i] = device.GetLogical().CreateImage(VkImageCreateInfo{ + .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .imageType = VK_IMAGE_TYPE_2D, + .format = VK_FORMAT_R16G16B16A16_SFLOAT, + .extent = + { + .width = output_size.width, + .height = output_size.height, + .depth = 1, + }, + .mipLevels = 1, + .arrayLayers = 1, + .samples = VK_SAMPLE_COUNT_1_BIT, + .tiling = VK_IMAGE_TILING_OPTIMAL, + .usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_STORAGE_BIT | + VK_IMAGE_USAGE_SAMPLED_BIT, + .sharingMode = VK_SHARING_MODE_EXCLUSIVE, + .queueFamilyIndexCount = 0, + .pQueueFamilyIndices = nullptr, + .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED, + }); + buffer_commits[i] = memory_allocator.Commit(images[i], MemoryUsage::DeviceLocal); + image_views[i] = device.GetLogical().CreateImageView(VkImageViewCreateInfo{ + .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .image = *images[i], + .viewType = VK_IMAGE_VIEW_TYPE_2D, + .format = VK_FORMAT_R16G16B16A16_SFLOAT, + .components = + { + .r = VK_COMPONENT_SWIZZLE_IDENTITY, + .g = VK_COMPONENT_SWIZZLE_IDENTITY, + .b = VK_COMPONENT_SWIZZLE_IDENTITY, + .a = VK_COMPONENT_SWIZZLE_IDENTITY, + }, + .subresourceRange = + { + .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .baseMipLevel = 0, + .levelCount = 1, + .baseArrayLayer = 0, + .layerCount = 1, + }, + }); + } +} + +void FSR::CreatePipelineLayout() { + VkPushConstantRange push_const{ + .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, + .offset = 0, + .size = sizeof(std::array<u32, 4 * 4>), + }; + VkPipelineLayoutCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .setLayoutCount = 1, + .pSetLayouts = descriptor_set_layout.address(), + .pushConstantRangeCount = 1, + .pPushConstantRanges = &push_const, + }; + + pipeline_layout = device.GetLogical().CreatePipelineLayout(ci); +} + +void FSR::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const { + const auto fsr_image_view = *image_views[image_index]; + const auto blit_image_view = *image_views[image_count + image_index]; + + const VkDescriptorImageInfo image_info{ + .sampler = VK_NULL_HANDLE, + .imageView = image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + const VkDescriptorImageInfo fsr_image_info{ + .sampler = VK_NULL_HANDLE, + .imageView = fsr_image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + const VkDescriptorImageInfo blit_image_info{ + .sampler = VK_NULL_HANDLE, + .imageView = blit_image_view, + .imageLayout = VK_IMAGE_LAYOUT_GENERAL, + }; + + VkWriteDescriptorSet sampler_write{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = descriptor_sets[image_index * 2], + .dstBinding = 0, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .pImageInfo = &image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + VkWriteDescriptorSet output_write{ + .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET, + .pNext = nullptr, + .dstSet = descriptor_sets[image_index * 2], + .dstBinding = 1, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .pImageInfo = &fsr_image_info, + .pBufferInfo = nullptr, + .pTexelBufferView = nullptr, + }; + + device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, output_write}, {}); + + sampler_write.dstSet = descriptor_sets[image_index * 2 + 1]; + sampler_write.pImageInfo = &fsr_image_info; + output_write.dstSet = descriptor_sets[image_index * 2 + 1]; + output_write.pImageInfo = &blit_image_info; + + device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, output_write}, {}); +} + +void FSR::CreateSampler() { + const VkSamplerCreateInfo ci{ + .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .magFilter = VK_FILTER_LINEAR, + .minFilter = VK_FILTER_LINEAR, + .mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR, + .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 = 0.0f, + .compareEnable = VK_FALSE, + .compareOp = VK_COMPARE_OP_NEVER, + .minLod = 0.0f, + .maxLod = 0.0f, + .borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK, + .unnormalizedCoordinates = VK_FALSE, + }; + + sampler = device.GetLogical().CreateSampler(ci); +} + +void FSR::CreateShaders() { + if (device.IsFloat16Supported()) { + easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_FP16_COMP_SPV); + rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_RCAS_FP16_COMP_SPV); + } else { + easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_FP32_COMP_SPV); + rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_RCAS_FP32_COMP_SPV); + } +} + +void FSR::CreatePipeline() { + VkPipelineShaderStageCreateInfo shader_stage_easu{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_COMPUTE_BIT, + .module = *easu_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }; + + VkPipelineShaderStageCreateInfo shader_stage_rcas{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = VK_SHADER_STAGE_COMPUTE_BIT, + .module = *rcas_shader, + .pName = "main", + .pSpecializationInfo = nullptr, + }; + + VkComputePipelineCreateInfo pipeline_ci_easu{ + .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = shader_stage_easu, + .layout = *pipeline_layout, + .basePipelineHandle = VK_NULL_HANDLE, + .basePipelineIndex = 0, + }; + + VkComputePipelineCreateInfo pipeline_ci_rcas{ + .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stage = shader_stage_rcas, + .layout = *pipeline_layout, + .basePipelineHandle = VK_NULL_HANDLE, + .basePipelineIndex = 0, + }; + + easu_pipeline = device.GetLogical().CreateComputePipeline(pipeline_ci_easu); + rcas_pipeline = device.GetLogical().CreateComputePipeline(pipeline_ci_rcas); +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_fsr.h b/src/video_core/renderer_vulkan/vk_fsr.h new file mode 100644 index 000000000..6bbec3d36 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_fsr.h @@ -0,0 +1,54 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include "common/math_util.h" +#include "video_core/vulkan_common/vulkan_memory_allocator.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" + +namespace Vulkan { + +class Device; +class VKScheduler; + +class FSR { +public: + explicit FSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count, + VkExtent2D output_size); + VkImageView Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view, + VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect); + +private: + void CreateDescriptorPool(); + void CreateDescriptorSetLayout(); + void CreateDescriptorSets(); + void CreateImages(); + void CreateSampler(); + void CreateShaders(); + void CreatePipeline(); + void CreatePipelineLayout(); + + void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const; + + const Device& device; + MemoryAllocator& memory_allocator; + size_t image_count; + VkExtent2D output_size; + + vk::DescriptorPool descriptor_pool; + vk::DescriptorSetLayout descriptor_set_layout; + vk::DescriptorSets descriptor_sets; + vk::PipelineLayout pipeline_layout; + vk::ShaderModule easu_shader; + vk::ShaderModule rcas_shader; + vk::Pipeline easu_pipeline; + vk::Pipeline rcas_pipeline; + vk::Sampler sampler; + std::vector<vk::Image> images; + std::vector<vk::ImageView> image_views; + std::vector<MemoryCommit> buffer_commits; +}; + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 8634c3316..616a7b457 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -32,6 +32,8 @@ namespace { using boost::container::small_vector; using boost::container::static_vector; using Shader::ImageBufferDescriptor; +using Shader::Backend::SPIRV::RESCALING_LAYOUT_DOWN_FACTOR_OFFSET; +using Shader::Backend::SPIRV::RESCALING_LAYOUT_WORDS_OFFSET; using Tegra::Texture::TexturePair; using VideoCore::Surface::PixelFormat; using VideoCore::Surface::PixelFormatFromDepthFormat; @@ -235,6 +237,7 @@ GraphicsPipeline::GraphicsPipeline( stage_infos[stage] = *info; enabled_uniform_buffer_masks[stage] = info->constant_buffer_mask; std::ranges::copy(info->constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin()); + num_textures += Shader::NumDescriptors(info->texture_descriptors); } auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool, pipeline_statistics] { DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)}; @@ -277,11 +280,10 @@ void GraphicsPipeline::AddTransition(GraphicsPipeline* transition) { template <typename Spec> void GraphicsPipeline::ConfigureImpl(bool is_indexed) { - std::array<ImageId, MAX_IMAGE_ELEMENTS> image_view_ids; - std::array<u32, MAX_IMAGE_ELEMENTS> image_view_indices; + std::array<VideoCommon::ImageViewInOut, MAX_IMAGE_ELEMENTS> views; std::array<VkSampler, MAX_IMAGE_ELEMENTS> samplers; size_t sampler_index{}; - size_t image_index{}; + size_t view_index{}; texture_cache.SynchronizeGraphicsDescriptors(); @@ -322,26 +324,30 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); }}; - const auto add_image{[&](const auto& desc) { + const auto add_image{[&](const auto& desc, bool blacklist) LAMBDA_FORCEINLINE { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices[image_index++] = handle.first; + views[view_index++] = { + .index = handle.first, + .blacklist = blacklist, + .id = {}, + }; } }}; if constexpr (Spec::has_texture_buffers) { for (const auto& desc : info.texture_buffer_descriptors) { - add_image(desc); + add_image(desc, false); } } if constexpr (Spec::has_image_buffers) { for (const auto& desc : info.image_buffer_descriptors) { - add_image(desc); + add_image(desc, false); } } for (const auto& desc : info.texture_descriptors) { for (u32 index = 0; index < desc.count; ++index) { const auto handle{read_handle(desc, index)}; - image_view_indices[image_index++] = handle.first; + views[view_index++] = {handle.first}; Sampler* const sampler{texture_cache.GetGraphicsSampler(handle.second)}; samplers[sampler_index++] = sampler->Handle(); @@ -349,7 +355,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } if constexpr (Spec::has_images) { for (const auto& desc : info.image_descriptors) { - add_image(desc); + add_image(desc, desc.is_written); } } }}; @@ -368,10 +374,9 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (Spec::enabled_stages[4]) { config_stage(4); } - const std::span indices_span(image_view_indices.data(), image_index); - texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + texture_cache.FillGraphicsImageViews<Spec::has_images>(std::span(views.data(), view_index)); - ImageId* texture_buffer_index{image_view_ids.data()}; + VideoCommon::ImageViewInOut* texture_buffer_it{views.data()}; const auto bind_stage_info{[&](size_t stage) LAMBDA_FORCEINLINE { size_t index{}; const auto add_buffer{[&](const auto& desc) { @@ -381,12 +386,12 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (is_image) { is_written = desc.is_written; } - ImageView& image_view{texture_cache.GetImageView(*texture_buffer_index)}; + ImageView& image_view{texture_cache.GetImageView(texture_buffer_it->id)}; buffer_cache.BindGraphicsTextureBuffer(stage, index, image_view.GpuAddr(), image_view.BufferSize(), image_view.format, is_written, is_image); ++index; - ++texture_buffer_index; + ++texture_buffer_it; } }}; buffer_cache.UnbindGraphicsTextureBuffers(stage); @@ -402,13 +407,9 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { add_buffer(desc); } } - for (const auto& desc : info.texture_descriptors) { - texture_buffer_index += desc.count; - } + texture_buffer_it += Shader::NumDescriptors(info.texture_descriptors); if constexpr (Spec::has_images) { - for (const auto& desc : info.image_descriptors) { - texture_buffer_index += desc.count; - } + texture_buffer_it += Shader::NumDescriptors(info.image_descriptors); } }}; if constexpr (Spec::enabled_stages[0]) { @@ -432,12 +433,13 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { update_descriptor_queue.Acquire(); + RescalingPushConstant rescaling; const VkSampler* samplers_it{samplers.data()}; - const ImageId* views_it{image_view_ids.data()}; + const VideoCommon::ImageViewInOut* views_it{views.data()}; const auto prepare_stage{[&](size_t stage) LAMBDA_FORCEINLINE { buffer_cache.BindHostStageBuffers(stage); - PushImageDescriptors(stage_infos[stage], samplers_it, views_it, texture_cache, - update_descriptor_queue); + PushImageDescriptors(texture_cache, update_descriptor_queue, stage_infos[stage], rescaling, + samplers_it, views_it); }}; if constexpr (Spec::enabled_stages[0]) { prepare_stage(0); @@ -454,10 +456,10 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { if constexpr (Spec::enabled_stages[4]) { prepare_stage(4); } - ConfigureDraw(); + ConfigureDraw(rescaling); } -void GraphicsPipeline::ConfigureDraw() { +void GraphicsPipeline::ConfigureDraw(const RescalingPushConstant& rescaling) { texture_cache.UpdateRenderTargets(false); scheduler.RequestRenderpass(texture_cache.GetFramebuffer()); @@ -468,12 +470,25 @@ void GraphicsPipeline::ConfigureDraw() { build_condvar.wait(lock, [this] { return is_built.load(std::memory_order::relaxed); }); }); } + const bool is_rescaling{texture_cache.IsRescaling()}; + const bool update_rescaling{scheduler.UpdateRescaling(is_rescaling)}; const bool bind_pipeline{scheduler.UpdateGraphicsPipeline(this)}; const void* const descriptor_data{update_descriptor_queue.UpdateData()}; - scheduler.Record([this, descriptor_data, bind_pipeline](vk::CommandBuffer cmdbuf) { + scheduler.Record([this, descriptor_data, bind_pipeline, rescaling_data = rescaling.Data(), + is_rescaling, update_rescaling](vk::CommandBuffer cmdbuf) { if (bind_pipeline) { cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline); } + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_ALL_GRAPHICS, + RESCALING_LAYOUT_WORDS_OFFSET, sizeof(rescaling_data), + rescaling_data.data()); + if (update_rescaling) { + const f32 config_down_factor{Settings::values.resolution_info.down_factor}; + const f32 scale_down_factor{is_rescaling ? config_down_factor : 1.0f}; + cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_ALL_GRAPHICS, + RESCALING_LAYOUT_DOWN_FACTOR_OFFSET, sizeof(scale_down_factor), + &scale_down_factor); + } if (!descriptor_set_layout) { return; } @@ -826,18 +841,10 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) { void GraphicsPipeline::Validate() { size_t num_images{}; for (const auto& info : stage_infos) { - for (const auto& desc : info.texture_buffer_descriptors) { - num_images += desc.count; - } - for (const auto& desc : info.image_buffer_descriptors) { - num_images += desc.count; - } - for (const auto& desc : info.texture_descriptors) { - num_images += desc.count; - } - for (const auto& desc : info.image_descriptors) { - num_images += desc.count; - } + num_images += Shader::NumDescriptors(info.texture_buffer_descriptors); + num_images += Shader::NumDescriptors(info.image_buffer_descriptors); + num_images += Shader::NumDescriptors(info.texture_descriptors); + num_images += Shader::NumDescriptors(info.image_descriptors); } ASSERT(num_images <= MAX_IMAGE_ELEMENTS); } diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h index 1c780e944..a0c1d8f07 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h @@ -62,6 +62,7 @@ namespace Vulkan { class Device; class PipelineStatistics; class RenderPassCache; +class RescalingPushConstant; class VKScheduler; class VKUpdateDescriptorQueue; @@ -113,7 +114,7 @@ private: template <typename Spec> void ConfigureImpl(bool is_indexed); - void ConfigureDraw(); + void ConfigureDraw(const RescalingPushConstant& rescaling); void MakePipeline(VkRenderPass render_pass); @@ -138,6 +139,7 @@ private: std::array<Shader::Info, NUM_STAGES> stage_infos; std::array<u32, 5> enabled_uniform_buffer_masks{}; VideoCommon::UniformBufferSizes uniform_buffer_sizes{}; + u32 num_textures{}; vk::DescriptorSetLayout descriptor_set_layout; DescriptorAllocator descriptor_allocator; diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.h b/src/video_core/renderer_vulkan/vk_master_semaphore.h index 0886b7da8..9be9c9bed 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.h +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.h @@ -70,7 +70,9 @@ public: return; } // If none of the above is hit, fallback to a regular wait - semaphore.Wait(tick); + while (!semaphore.Wait(tick)) { + } + Refresh(); } private: diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 30b47a7a0..fd334a146 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -58,18 +58,28 @@ struct DrawParams { bool is_indexed; }; -VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index) { +VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index, float scale) { const auto& src = regs.viewport_transform[index]; - const float width = src.scale_x * 2.0f; - float y = src.translate_y - src.scale_y; - float height = src.scale_y * 2.0f; + const auto conv = [scale](float value) { + float new_value = value * scale; + if (scale < 1.0f) { + const bool sign = std::signbit(value); + new_value = std::round(std::abs(new_value)); + new_value = sign ? -new_value : new_value; + } + return new_value; + }; + const float x = conv(src.translate_x - src.scale_x); + const float width = conv(src.scale_x * 2.0f); + float y = conv(src.translate_y - src.scale_y); + float height = conv(src.scale_y * 2.0f); if (regs.screen_y_control.y_negate) { y += height; height = -height; } const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1.0f : 0.0f; VkViewport viewport{ - .x = src.translate_x - src.scale_x, + .x = x, .y = y, .width = width != 0.0f ? width : 1.0f, .height = height != 0.0f ? height : 1.0f, @@ -83,14 +93,27 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in return viewport; } -VkRect2D GetScissorState(const Maxwell& regs, size_t index) { +VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u32 down_shift = 0) { const auto& src = regs.scissor_test[index]; VkRect2D scissor; + const auto scale_up = [&](s32 value) -> s32 { + if (value == 0) { + return 0U; + } + const s32 upset = value * up_scale; + s32 acumm = 0; + if ((up_scale >> down_shift) == 0) { + acumm = upset % 2; + } + const s32 converted_value = (value * up_scale) >> down_shift; + return value < 0 ? std::min<s32>(converted_value - acumm, -1) + : std::max<s32>(converted_value + acumm, 1); + }; if (src.enable) { - scissor.offset.x = static_cast<s32>(src.min_x); - scissor.offset.y = static_cast<s32>(src.min_y); - scissor.extent.width = src.max_x - src.min_x; - scissor.extent.height = src.max_y - src.min_y; + scissor.offset.x = scale_up(static_cast<s32>(src.min_x)); + scissor.offset.y = scale_up(static_cast<s32>(src.min_y)); + scissor.extent.width = scale_up(src.max_x - src.min_x); + scissor.extent.height = scale_up(src.max_y - src.min_y); } else { scissor.offset.x = 0; scissor.offset.y = 0; @@ -199,7 +222,7 @@ void RasterizerVulkan::Clear() { query_cache.UpdateCounters(); - const auto& regs = maxwell3d.regs; + auto& regs = maxwell3d.regs; const bool use_color = regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B || regs.clear_buffers.A; const bool use_depth = regs.clear_buffers.Z; @@ -214,8 +237,16 @@ void RasterizerVulkan::Clear() { const VkExtent2D render_area = framebuffer->RenderArea(); scheduler.RequestRenderpass(framebuffer); + u32 up_scale = 1; + u32 down_shift = 0; + if (texture_cache.IsRescaling()) { + up_scale = Settings::values.resolution_info.up_scale; + down_shift = Settings::values.resolution_info.down_shift; + } + UpdateViewportsState(regs); + VkClearRect clear_rect{ - .rect = GetScissorState(regs, 0), + .rect = GetScissorState(regs, 0, up_scale, down_shift), .baseArrayLayer = regs.clear_buffers.layer, .layerCount = 1, }; @@ -230,7 +261,38 @@ void RasterizerVulkan::Clear() { const u32 color_attachment = regs.clear_buffers.RT; if (use_color && framebuffer->HasAspectColorBit(color_attachment)) { VkClearValue clear_value; - std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color)); + bool is_integer = false; + bool is_signed = false; + size_t int_size = 8; + for (std::size_t i = 0; i < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets; ++i) { + const auto& this_rt = regs.rt[i]; + if (this_rt.Address() == 0) { + continue; + } + if (this_rt.format == Tegra::RenderTargetFormat::NONE) { + continue; + } + const auto format = + VideoCore::Surface::PixelFormatFromRenderTargetFormat(this_rt.format); + is_integer = IsPixelFormatInteger(format); + is_signed = IsPixelFormatSignedInteger(format); + int_size = PixelComponentSizeBitsInteger(format); + break; + } + if (!is_integer) { + std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color)); + } else if (!is_signed) { + for (size_t i = 0; i < 4; i++) { + clear_value.color.uint32[i] = static_cast<u32>( + static_cast<f32>(static_cast<u64>(int_size) << 1U) * regs.clear_color[i]); + } + } else { + for (size_t i = 0; i < 4; i++) { + clear_value.color.int32[i] = + static_cast<s32>(static_cast<f32>(static_cast<s64>(int_size - 1) << 1) * + (regs.clear_color[i] - 0.5f)); + } + } scheduler.Record([color_attachment, clear_value, clear_rect](vk::CommandBuffer cmdbuf) { const VkClearAttachment attachment{ @@ -595,15 +657,17 @@ void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& reg if (!state_tracker.TouchViewports()) { return; } + const bool is_rescaling{texture_cache.IsRescaling()}; + const float scale = is_rescaling ? Settings::values.resolution_info.up_factor : 1.0f; const std::array viewports{ - GetViewportState(device, regs, 0), GetViewportState(device, regs, 1), - GetViewportState(device, regs, 2), GetViewportState(device, regs, 3), - GetViewportState(device, regs, 4), GetViewportState(device, regs, 5), - GetViewportState(device, regs, 6), GetViewportState(device, regs, 7), - GetViewportState(device, regs, 8), GetViewportState(device, regs, 9), - GetViewportState(device, regs, 10), GetViewportState(device, regs, 11), - GetViewportState(device, regs, 12), GetViewportState(device, regs, 13), - GetViewportState(device, regs, 14), GetViewportState(device, regs, 15), + GetViewportState(device, regs, 0, scale), GetViewportState(device, regs, 1, scale), + GetViewportState(device, regs, 2, scale), GetViewportState(device, regs, 3, scale), + GetViewportState(device, regs, 4, scale), GetViewportState(device, regs, 5, scale), + GetViewportState(device, regs, 6, scale), GetViewportState(device, regs, 7, scale), + GetViewportState(device, regs, 8, scale), GetViewportState(device, regs, 9, scale), + GetViewportState(device, regs, 10, scale), GetViewportState(device, regs, 11, scale), + GetViewportState(device, regs, 12, scale), GetViewportState(device, regs, 13, scale), + GetViewportState(device, regs, 14, scale), GetViewportState(device, regs, 15, scale), }; scheduler.Record([viewports](vk::CommandBuffer cmdbuf) { cmdbuf.SetViewport(0, viewports); }); } @@ -612,13 +676,29 @@ void RasterizerVulkan::UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs if (!state_tracker.TouchScissors()) { return; } + u32 up_scale = 1; + u32 down_shift = 0; + if (texture_cache.IsRescaling()) { + up_scale = Settings::values.resolution_info.up_scale; + down_shift = Settings::values.resolution_info.down_shift; + } const std::array scissors{ - GetScissorState(regs, 0), GetScissorState(regs, 1), GetScissorState(regs, 2), - GetScissorState(regs, 3), GetScissorState(regs, 4), GetScissorState(regs, 5), - GetScissorState(regs, 6), GetScissorState(regs, 7), GetScissorState(regs, 8), - GetScissorState(regs, 9), GetScissorState(regs, 10), GetScissorState(regs, 11), - GetScissorState(regs, 12), GetScissorState(regs, 13), GetScissorState(regs, 14), - GetScissorState(regs, 15), + GetScissorState(regs, 0, up_scale, down_shift), + GetScissorState(regs, 1, up_scale, down_shift), + GetScissorState(regs, 2, up_scale, down_shift), + GetScissorState(regs, 3, up_scale, down_shift), + GetScissorState(regs, 4, up_scale, down_shift), + GetScissorState(regs, 5, up_scale, down_shift), + GetScissorState(regs, 6, up_scale, down_shift), + GetScissorState(regs, 7, up_scale, down_shift), + GetScissorState(regs, 8, up_scale, down_shift), + GetScissorState(regs, 9, up_scale, down_shift), + GetScissorState(regs, 10, up_scale, down_shift), + GetScissorState(regs, 11, up_scale, down_shift), + GetScissorState(regs, 12, up_scale, down_shift), + GetScissorState(regs, 13, up_scale, down_shift), + GetScissorState(regs, 14, up_scale, down_shift), + GetScissorState(regs, 15, up_scale, down_shift), }; scheduler.Record([scissors](vk::CommandBuffer cmdbuf) { cmdbuf.SetScissor(0, scissors); }); } diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 0c11c814f..3bfdf41ba 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -128,6 +128,15 @@ bool VKScheduler::UpdateGraphicsPipeline(GraphicsPipeline* pipeline) { return true; } +bool VKScheduler::UpdateRescaling(bool is_rescaling) { + if (state.rescaling_defined && is_rescaling == state.is_rescaling) { + return false; + } + state.rescaling_defined = true; + state.is_rescaling = is_rescaling; + return true; +} + void VKScheduler::WorkerThread(std::stop_token stop_token) { Common::SetCurrentThreadName("yuzu:VulkanWorker"); do { @@ -227,6 +236,7 @@ void VKScheduler::AllocateNewContext() { void VKScheduler::InvalidateState() { state.graphics_pipeline = nullptr; + state.rescaling_defined = false; state_tracker.InvalidateCommandBufferState(); } diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 85fc1712f..1b06c9296 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -56,6 +56,9 @@ public: /// Update the pipeline to the current execution context. bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline); + /// Update the rescaling state. Returns true if the state has to be updated. + bool UpdateRescaling(bool is_rescaling); + /// Invalidates current command buffer state except for render passes void InvalidateState(); @@ -185,6 +188,8 @@ private: VkFramebuffer framebuffer = nullptr; VkExtent2D render_area = {0, 0}; GraphicsPipeline* graphics_pipeline = nullptr; + bool is_rescaling = false; + bool rescaling_defined = false; }; void WorkerThread(std::stop_token stop_token); diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.h b/src/video_core/renderer_vulkan/vk_state_tracker.h index 2f2d6b31f..40a149832 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -71,11 +71,15 @@ public: } bool TouchViewports() { - return Exchange(Dirty::Viewports, false); + const bool dirty_viewports = Exchange(Dirty::Viewports, false); + const bool rescale_viewports = Exchange(VideoCommon::Dirty::RescaleViewports, false); + return dirty_viewports || rescale_viewports; } bool TouchScissors() { - return Exchange(Dirty::Scissors, false); + const bool dirty_scissors = Exchange(Dirty::Scissors, false); + const bool rescale_scissors = Exchange(VideoCommon::Dirty::RescaleScissors, false); + return dirty_scissors || rescale_scissors; } bool TouchDepthBias() { diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 06c5fb867..407fd2a15 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -32,10 +32,12 @@ using Tegra::Engines::Fermi2D; using Tegra::Texture::SwizzleSource; using Tegra::Texture::TextureMipmapFilter; using VideoCommon::BufferImageCopy; +using VideoCommon::ImageFlagBits; using VideoCommon::ImageInfo; using VideoCommon::ImageType; using VideoCommon::SubresourceRange; using VideoCore::Surface::IsPixelFormatASTC; +using VideoCore::Surface::IsPixelFormatInteger; namespace { constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { @@ -588,8 +590,158 @@ struct RangedBarrierRange { UNREACHABLE_MSG("Invalid image format={}", format); return VK_FORMAT_R32_UINT; } + +void BlitScale(VKScheduler& scheduler, VkImage src_image, VkImage dst_image, const ImageInfo& info, + VkImageAspectFlags aspect_mask, const Settings::ResolutionScalingInfo& resolution, + bool up_scaling = true) { + const bool is_2d = info.type == ImageType::e2D; + const auto resources = info.resources; + const VkExtent2D extent{ + .width = info.size.width, + .height = info.size.height, + }; + // Depth and integer formats must use NEAREST filter for blits. + const bool is_color{aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT}; + const bool is_bilinear{is_color && !IsPixelFormatInteger(info.format)}; + const VkFilter vk_filter = is_bilinear ? VK_FILTER_LINEAR : VK_FILTER_NEAREST; + + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([dst_image, src_image, extent, resources, aspect_mask, resolution, is_2d, + vk_filter, up_scaling](vk::CommandBuffer cmdbuf) { + const VkOffset2D src_size{ + .x = static_cast<s32>(up_scaling ? extent.width : resolution.ScaleUp(extent.width)), + .y = static_cast<s32>(is_2d && up_scaling ? extent.height + : resolution.ScaleUp(extent.height)), + }; + const VkOffset2D dst_size{ + .x = static_cast<s32>(up_scaling ? resolution.ScaleUp(extent.width) : extent.width), + .y = static_cast<s32>(is_2d && up_scaling ? resolution.ScaleUp(extent.height) + : extent.height), + }; + boost::container::small_vector<VkImageBlit, 4> regions; + regions.reserve(resources.levels); + for (s32 level = 0; level < resources.levels; level++) { + regions.push_back({ + .srcSubresource{ + .aspectMask = aspect_mask, + .mipLevel = static_cast<u32>(level), + .baseArrayLayer = 0, + .layerCount = static_cast<u32>(resources.layers), + }, + .srcOffsets{ + { + .x = 0, + .y = 0, + .z = 0, + }, + { + .x = std::max(1, src_size.x >> level), + .y = std::max(1, src_size.y >> level), + .z = 1, + }, + }, + .dstSubresource{ + .aspectMask = aspect_mask, + .mipLevel = static_cast<u32>(level), + .baseArrayLayer = 0, + .layerCount = static_cast<u32>(resources.layers), + }, + .dstOffsets{ + { + .x = 0, + .y = 0, + .z = 0, + }, + { + .x = std::max(1, dst_size.x >> level), + .y = std::max(1, dst_size.y >> level), + .z = 1, + }, + }, + }); + } + const VkImageSubresourceRange subresource_range{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }; + const std::array read_barriers{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = src_image, + .subresourceRange = subresource_range, + }, + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED, // Discard contents + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst_image, + .subresourceRange = subresource_range, + }, + }; + const std::array write_barriers{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = src_image, + .subresourceRange = subresource_range, + }, + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT | VK_ACCESS_MEMORY_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst_image, + .subresourceRange = subresource_range, + }, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, nullptr, nullptr, read_barriers); + cmdbuf.BlitImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image, + VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, regions, vk_filter); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, nullptr, nullptr, write_barriers); + }); +} } // Anonymous namespace +TextureCacheRuntime::TextureCacheRuntime(const Device& device_, VKScheduler& scheduler_, + MemoryAllocator& memory_allocator_, + StagingBufferPool& staging_buffer_pool_, + BlitImageHelper& blit_image_helper_, + ASTCDecoderPass& astc_decoder_pass_, + RenderPassCache& render_pass_cache_) + : device{device_}, scheduler{scheduler_}, memory_allocator{memory_allocator_}, + staging_buffer_pool{staging_buffer_pool_}, blit_image_helper{blit_image_helper_}, + astc_decoder_pass{astc_decoder_pass_}, render_pass_cache{render_pass_cache_}, + resolution{Settings::values.resolution_info} {} + void TextureCacheRuntime::Finish() { scheduler.Finish(); } @@ -614,8 +766,8 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst return; } if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT && !is_src_msaa && !is_dst_msaa) { - blit_image_helper.BlitColor(dst_framebuffer, src, dst_region, src_region, filter, - operation); + blit_image_helper.BlitColor(dst_framebuffer, src.Handle(Shader::TextureType::Color2D), + dst_region, src_region, filter, operation); return; } if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) { @@ -719,26 +871,29 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst }); } -void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view) { +void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, + bool rescaled) { + const u32 up_scale = rescaled ? resolution.up_scale : 1; + const u32 down_shift = rescaled ? resolution.down_shift : 0; switch (dst_view.format) { case PixelFormat::R16_UNORM: if (src_view.format == PixelFormat::D16_UNORM) { - return blit_image_helper.ConvertD16ToR16(dst, src_view); + return blit_image_helper.ConvertD16ToR16(dst, src_view, up_scale, down_shift); } break; case PixelFormat::R32_FLOAT: if (src_view.format == PixelFormat::D32_FLOAT) { - return blit_image_helper.ConvertD32ToR32(dst, src_view); + return blit_image_helper.ConvertD32ToR32(dst, src_view, up_scale, down_shift); } break; case PixelFormat::D16_UNORM: if (src_view.format == PixelFormat::R16_UNORM) { - return blit_image_helper.ConvertR16ToD16(dst, src_view); + return blit_image_helper.ConvertR16ToD16(dst, src_view, up_scale, down_shift); } break; case PixelFormat::D32_FLOAT: if (src_view.format == PixelFormat::R32_FLOAT) { - return blit_image_helper.ConvertR32ToD32(dst, src_view); + return blit_image_helper.ConvertR32ToD32(dst, src_view, up_scale, down_shift); } break; default: @@ -840,36 +995,39 @@ u64 TextureCacheRuntime::GetDeviceLocalMemory() const { return device.GetDeviceLocalMemory(); } -Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_addr_, +void TextureCacheRuntime::TickFrame() {} + +Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_) - : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime.scheduler}, - image(MakeImage(runtime.device, info)), - commit(runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal)), + : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime_.scheduler}, + runtime{&runtime_}, original_image(MakeImage(runtime_.device, info)), + commit(runtime_.memory_allocator.Commit(original_image, MemoryUsage::DeviceLocal)), aspect_mask(ImageAspectMask(info.format)) { - if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) { + if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) { if (Settings::values.accelerate_astc.GetValue()) { flags |= VideoCommon::ImageFlagBits::AcceleratedUpload; } else { flags |= VideoCommon::ImageFlagBits::Converted; } } - if (runtime.device.HasDebuggingToolAttached()) { - image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); + if (runtime->device.HasDebuggingToolAttached()) { + original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str()); } static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO, .pNext = nullptr, .usage = VK_IMAGE_USAGE_STORAGE_BIT, }; - if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) { - const auto& device = runtime.device.GetLogical(); + current_image = *original_image; + if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) { + const auto& device = runtime->device.GetLogical(); storage_image_views.reserve(info.resources.levels); for (s32 level = 0; level < info.resources.levels; ++level) { storage_image_views.push_back(device.CreateImageView(VkImageViewCreateInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO, .pNext = &storage_image_view_usage_create_info, .flags = 0, - .image = *image, + .image = *original_image, .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY, .format = VK_FORMAT_A8B8G8R8_UNORM_PACK32, .components{ @@ -890,26 +1048,39 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_ } } +Image::Image(const VideoCommon::NullImageParams& params) : VideoCommon::ImageBase{params} {} + Image::~Image() = default; void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) { // TODO: Move this to another API + const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + ScaleDown(true); + } scheduler->RequestOutsideRenderPassOperationContext(); std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask); const VkBuffer src_buffer = map.buffer; - const VkImage vk_image = *image; + const VkImage vk_image = *original_image; const VkImageAspectFlags vk_aspect_mask = aspect_mask; const bool is_initialized = std::exchange(initialized, true); scheduler->Record([src_buffer, vk_image, vk_aspect_mask, is_initialized, vk_copies](vk::CommandBuffer cmdbuf) { CopyBufferToImage(cmdbuf, src_buffer, vk_image, vk_aspect_mask, is_initialized, vk_copies); }); + if (is_rescaled) { + ScaleUp(); + } } void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) { + const bool is_rescaled = True(flags & ImageFlagBits::Rescaled); + if (is_rescaled) { + ScaleDown(); + } std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask); scheduler->RequestOutsideRenderPassOperationContext(); - scheduler->Record([buffer = map.buffer, image = *image, aspect_mask = aspect_mask, + scheduler->Record([buffer = map.buffer, image = *original_image, aspect_mask = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { const VkImageMemoryBarrier read_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, @@ -959,6 +1130,146 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, memory_write_barrier, nullptr, image_write_barrier); }); + if (is_rescaled) { + ScaleUp(true); + } +} + +bool Image::ScaleUp(bool ignore) { + if (True(flags & ImageFlagBits::Rescaled)) { + return false; + } + ASSERT(info.type != ImageType::Linear); + flags |= ImageFlagBits::Rescaled; + const auto& resolution = runtime->resolution; + if (!resolution.active) { + return false; + } + has_scaled = true; + const auto& device = runtime->device; + if (!scaled_image) { + const bool is_2d = info.type == ImageType::e2D; + const u32 scaled_width = resolution.ScaleUp(info.size.width); + const u32 scaled_height = is_2d ? resolution.ScaleUp(info.size.height) : info.size.height; + auto scaled_info = info; + scaled_info.size.width = scaled_width; + scaled_info.size.height = scaled_height; + scaled_image = MakeImage(device, scaled_info); + auto& allocator = runtime->memory_allocator; + scaled_commit = MemoryCommit(allocator.Commit(scaled_image, MemoryUsage::DeviceLocal)); + ignore = false; + } + current_image = *scaled_image; + if (ignore) { + return true; + } + + if (aspect_mask == 0) { + aspect_mask = ImageAspectMask(info.format); + } + static constexpr auto OPTIMAL_FORMAT = FormatType::Optimal; + const PixelFormat format = StorageFormat(info.format); + const auto vk_format = MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, format).format; + const auto blit_usage = VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT; + if (device.IsFormatSupported(vk_format, blit_usage, OPTIMAL_FORMAT)) { + BlitScale(*scheduler, *original_image, *scaled_image, info, aspect_mask, resolution); + } else { + return BlitScaleHelper(true); + } + return true; +} + +bool Image::ScaleDown(bool ignore) { + if (False(flags & ImageFlagBits::Rescaled)) { + return false; + } + ASSERT(info.type != ImageType::Linear); + flags &= ~ImageFlagBits::Rescaled; + const auto& resolution = runtime->resolution; + if (!resolution.active) { + return false; + } + current_image = *original_image; + if (ignore) { + return true; + } + if (aspect_mask == 0) { + aspect_mask = ImageAspectMask(info.format); + } + static constexpr auto OPTIMAL_FORMAT = FormatType::Optimal; + const PixelFormat format = StorageFormat(info.format); + const auto& device = runtime->device; + const auto vk_format = MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, format).format; + const auto blit_usage = VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT; + if (device.IsFormatSupported(vk_format, blit_usage, OPTIMAL_FORMAT)) { + BlitScale(*scheduler, *scaled_image, *original_image, info, aspect_mask, resolution, false); + } else { + return BlitScaleHelper(false); + } + return true; +} + +bool Image::BlitScaleHelper(bool scale_up) { + using namespace VideoCommon; + static constexpr auto BLIT_OPERATION = Tegra::Engines::Fermi2D::Operation::SrcCopy; + const bool is_color{aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT}; + const bool is_bilinear{is_color && !IsPixelFormatInteger(info.format)}; + const auto operation = is_bilinear ? Tegra::Engines::Fermi2D::Filter::Bilinear + : Tegra::Engines::Fermi2D::Filter::Point; + + const bool is_2d = info.type == ImageType::e2D; + const auto& resolution = runtime->resolution; + const u32 scaled_width = resolution.ScaleUp(info.size.width); + const u32 scaled_height = is_2d ? resolution.ScaleUp(info.size.height) : info.size.height; + std::unique_ptr<ImageView>& blit_view = scale_up ? scale_view : normal_view; + std::unique_ptr<Framebuffer>& blit_framebuffer = + scale_up ? scale_framebuffer : normal_framebuffer; + if (!blit_view) { + const auto view_info = ImageViewInfo(ImageViewType::e2D, info.format); + blit_view = std::make_unique<ImageView>(*runtime, view_info, NULL_IMAGE_ID, *this); + } + + const u32 src_width = scale_up ? info.size.width : scaled_width; + const u32 src_height = scale_up ? info.size.height : scaled_height; + const u32 dst_width = scale_up ? scaled_width : info.size.width; + const u32 dst_height = scale_up ? scaled_height : info.size.height; + const Region2D src_region{ + .start = {0, 0}, + .end = {static_cast<s32>(src_width), static_cast<s32>(src_height)}, + }; + const Region2D dst_region{ + .start = {0, 0}, + .end = {static_cast<s32>(dst_width), static_cast<s32>(dst_height)}, + }; + const VkExtent2D extent{ + .width = std::max(scaled_width, info.size.width), + .height = std::max(scaled_height, info.size.width), + }; + + auto* view_ptr = blit_view.get(); + if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) { + if (!blit_framebuffer) { + blit_framebuffer = std::make_unique<Framebuffer>(*runtime, view_ptr, nullptr, extent); + } + const auto color_view = blit_view->Handle(Shader::TextureType::Color2D); + + runtime->blit_image_helper.BlitColor(blit_framebuffer.get(), color_view, dst_region, + src_region, operation, BLIT_OPERATION); + } else if (!runtime->device.IsBlitDepthStencilSupported() && + aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) { + if (!blit_framebuffer) { + blit_framebuffer = std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent); + } + runtime->blit_image_helper.BlitDepthStencil(blit_framebuffer.get(), blit_view->DepthView(), + blit_view->StencilView(), dst_region, + src_region, operation, BLIT_OPERATION); + } else { + // TODO: Use helper blits where applicable + flags &= ~ImageFlagBits::Rescaled; + LOG_ERROR(Render_Vulkan, "Device does not support scaling format {}", info.format); + return false; + } + return true; } ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info, @@ -1052,9 +1363,11 @@ ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, : VideoCommon::ImageViewBase{info, view_info}, gpu_addr{gpu_addr_}, buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {} -ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams& params) +ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParams& params) : VideoCommon::ImageViewBase{params} {} +ImageView::~ImageView() = default; + VkImageView ImageView::DepthView() { if (depth_view) { return *depth_view; @@ -1137,7 +1450,8 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& t LOG_WARNING(Render_Vulkan, "VK_EXT_sampler_filter_minmax is required"); } // Some games have samplers with garbage. Sanitize them here. - const float max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f); + const f32 max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f); + sampler = device.GetLogical().CreateSampler(VkSamplerCreateInfo{ .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO, .pNext = pnext, @@ -1162,7 +1476,29 @@ Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& t } Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, - ImageView* depth_buffer, const VideoCommon::RenderTargets& key) { + ImageView* depth_buffer, const VideoCommon::RenderTargets& key) + : render_area{VkExtent2D{ + .width = key.size.width, + .height = key.size.height, + }} { + CreateFramebuffer(runtime, color_buffers, depth_buffer); + if (runtime.device.HasDebuggingToolAttached()) { + framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str()); + } +} + +Framebuffer::Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, + ImageView* depth_buffer, VkExtent2D extent) + : render_area{extent} { + std::array<ImageView*, NUM_RT> color_buffers{color_buffer}; + CreateFramebuffer(runtime, color_buffers, depth_buffer); +} + +Framebuffer::~Framebuffer() = default; + +void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, + std::span<ImageView*, NUM_RT> color_buffers, + ImageView* depth_buffer) { std::vector<VkImageView> attachments; RenderPassKey renderpass_key{}; s32 num_layers = 1; @@ -1200,10 +1536,6 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM renderpass = runtime.render_pass_cache.Get(renderpass_key); - render_area = VkExtent2D{ - .width = key.size.width, - .height = key.size.height, - }; num_color_buffers = static_cast<u32>(num_colors); framebuffer = runtime.device.GetLogical().CreateFramebuffer({ .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO, @@ -1212,13 +1544,10 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM .renderPass = renderpass, .attachmentCount = static_cast<u32>(attachments.size()), .pAttachments = attachments.data(), - .width = key.size.width, - .height = key.size.height, + .width = render_area.width, + .height = render_area.height, .layers = static_cast<u32>(std::max(num_layers, 1)), }); - if (runtime.device.HasDebuggingToolAttached()) { - framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str()); - } } void TextureCacheRuntime::AccelerateImageUpload( diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index b09c468e4..ff28b4e96 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -13,6 +13,10 @@ #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" +namespace Settings { +struct ResolutionScalingInfo; +} + namespace Vulkan { using VideoCommon::ImageId; @@ -31,14 +35,14 @@ class RenderPassCache; class StagingBufferPool; class VKScheduler; -struct TextureCacheRuntime { - const Device& device; - VKScheduler& scheduler; - MemoryAllocator& memory_allocator; - StagingBufferPool& staging_buffer_pool; - BlitImageHelper& blit_image_helper; - ASTCDecoderPass& astc_decoder_pass; - RenderPassCache& render_pass_cache; +class TextureCacheRuntime { +public: + explicit TextureCacheRuntime(const Device& device_, VKScheduler& scheduler_, + MemoryAllocator& memory_allocator_, + StagingBufferPool& staging_buffer_pool_, + BlitImageHelper& blit_image_helper_, + ASTCDecoderPass& astc_decoder_pass_, + RenderPassCache& render_pass_cache_); void Finish(); @@ -46,6 +50,10 @@ struct TextureCacheRuntime { StagingBufferRef DownloadStagingBuffer(size_t size); + void TickFrame(); + + u64 GetDeviceLocalMemory() const; + void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, @@ -53,7 +61,7 @@ struct TextureCacheRuntime { void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies); - void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view); + void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, bool rescaled); bool CanAccelerateImageUpload(Image&) const noexcept { return false; @@ -74,13 +82,21 @@ struct TextureCacheRuntime { return true; } - u64 GetDeviceLocalMemory() const; + const Device& device; + VKScheduler& scheduler; + MemoryAllocator& memory_allocator; + StagingBufferPool& staging_buffer_pool; + BlitImageHelper& blit_image_helper; + ASTCDecoderPass& astc_decoder_pass; + RenderPassCache& render_pass_cache; + const Settings::ResolutionScalingInfo& resolution; }; class Image : public VideoCommon::ImageBase { public: explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr); + explicit Image(const VideoCommon::NullImageParams&); ~Image(); @@ -97,7 +113,7 @@ public: std::span<const VideoCommon::BufferImageCopy> copies); [[nodiscard]] VkImage Handle() const noexcept { - return *image; + return current_image; } [[nodiscard]] VkImageAspectFlags AspectMask() const noexcept { @@ -113,14 +129,30 @@ public: return std::exchange(initialized, true); } + bool ScaleUp(bool ignore = false); + + bool ScaleDown(bool ignore = false); + private: - VKScheduler* scheduler; - vk::Image image; + bool BlitScaleHelper(bool scale_up); + + VKScheduler* scheduler{}; + TextureCacheRuntime* runtime{}; + + vk::Image original_image; MemoryCommit commit; - vk::ImageView image_view; std::vector<vk::ImageView> storage_image_views; VkImageAspectFlags aspect_mask = 0; bool initialized = false; + vk::Image scaled_image{}; + MemoryCommit scaled_commit{}; + VkImage current_image{}; + + std::unique_ptr<Framebuffer> scale_framebuffer; + std::unique_ptr<ImageView> scale_view; + + std::unique_ptr<Framebuffer> normal_framebuffer; + std::unique_ptr<ImageView> normal_view; }; class ImageView : public VideoCommon::ImageViewBase { @@ -128,7 +160,15 @@ public: explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&); explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo&, const VideoCommon::ImageViewInfo&, GPUVAddr); - explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&); + explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParams&); + + ~ImageView(); + + ImageView(const ImageView&) = delete; + ImageView& operator=(const ImageView&) = delete; + + ImageView(ImageView&&) = default; + ImageView& operator=(ImageView&&) = default; [[nodiscard]] VkImageView DepthView(); @@ -197,9 +237,23 @@ private: class Framebuffer { public: - explicit Framebuffer(TextureCacheRuntime&, std::span<ImageView*, NUM_RT> color_buffers, + explicit Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer, const VideoCommon::RenderTargets& key); + explicit Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, + ImageView* depth_buffer, VkExtent2D extent); + + ~Framebuffer(); + + Framebuffer(const Framebuffer&) = delete; + Framebuffer& operator=(const Framebuffer&) = delete; + + Framebuffer(Framebuffer&&) = default; + Framebuffer& operator=(Framebuffer&&) = default; + + void CreateFramebuffer(TextureCacheRuntime& runtime, + std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer); + [[nodiscard]] VkFramebuffer Handle() const noexcept { return *framebuffer; } |