From 09e17fbb0ff70a62b4cff1cafae0ab579956c648 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Mon, 13 Jan 2020 20:36:03 -0300 Subject: [PATCH] vk_texture_cache: Implement generic texture cache on Vulkan It currently ignores PBO linearizations since these should be dropped as soon as possible on OpenGL. --- src/video_core/CMakeLists.txt | 5 +- .../renderer_vulkan/vk_staging_buffer_pool.h | 1 + .../renderer_vulkan/vk_texture_cache.cpp | 484 ++++++++++++++++++ .../renderer_vulkan/vk_texture_cache.h | 244 +++++++++ 4 files changed, 733 insertions(+), 1 deletion(-) create mode 100644 src/video_core/renderer_vulkan/vk_texture_cache.cpp create mode 100644 src/video_core/renderer_vulkan/vk_texture_cache.h diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 142852082..729ee4a01 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -190,8 +190,11 @@ if (ENABLE_VULKAN) renderer_vulkan/vk_stream_buffer.h renderer_vulkan/vk_swapchain.cpp renderer_vulkan/vk_swapchain.h + renderer_vulkan/vk_texture_cache.cpp + renderer_vulkan/vk_texture_cache.h renderer_vulkan/vk_update_descriptor.cpp - renderer_vulkan/vk_update_descriptor.h) + renderer_vulkan/vk_update_descriptor.h + ) target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include) target_compile_definitions(video_core PRIVATE HAS_VULKAN) diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h index 02310375f..4d9488f49 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h @@ -13,6 +13,7 @@ #include "video_core/renderer_vulkan/declarations.h" #include "video_core/renderer_vulkan/vk_memory_manager.h" +#include "video_core/renderer_vulkan/vk_resource_manager.h" namespace Vulkan { diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp new file mode 100644 index 000000000..aa088f8d0 --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -0,0 +1,484 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include +#include +#include +#include +#include +#include +#include + +#include "common/alignment.h" +#include "common/assert.h" +#include "common/common_types.h" +#include "core/core.h" +#include "core/memory.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/morton.h" +#include "video_core/renderer_vulkan/declarations.h" +#include "video_core/renderer_vulkan/maxwell_to_vk.h" +#include "video_core/renderer_vulkan/vk_device.h" +#include "video_core/renderer_vulkan/vk_memory_manager.h" +#include "video_core/renderer_vulkan/vk_rasterizer.h" +#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" +#include "video_core/renderer_vulkan/vk_texture_cache.h" +#include "video_core/surface.h" +#include "video_core/textures/convert.h" + +namespace Vulkan { + +using VideoCore::MortonSwizzle; +using VideoCore::MortonSwizzleMode; + +using Tegra::Texture::SwizzleSource; +using VideoCore::Surface::PixelFormat; +using VideoCore::Surface::SurfaceCompression; +using VideoCore::Surface::SurfaceTarget; + +namespace { + +vk::ImageType SurfaceTargetToImage(SurfaceTarget target) { + switch (target) { + case SurfaceTarget::Texture1D: + case SurfaceTarget::Texture1DArray: + return vk::ImageType::e1D; + case SurfaceTarget::Texture2D: + case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubemap: + case SurfaceTarget::TextureCubeArray: + return vk::ImageType::e2D; + case SurfaceTarget::Texture3D: + return vk::ImageType::e3D; + } + UNREACHABLE_MSG("Unknown texture target={}", static_cast(target)); + return {}; +} + +vk::ImageAspectFlags PixelFormatToImageAspect(PixelFormat pixel_format) { + if (pixel_format < PixelFormat::MaxColorFormat) { + return vk::ImageAspectFlagBits::eColor; + } else if (pixel_format < PixelFormat::MaxDepthFormat) { + return vk::ImageAspectFlagBits::eDepth; + } else if (pixel_format < PixelFormat::MaxDepthStencilFormat) { + return vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil; + } else { + UNREACHABLE_MSG("Invalid pixel format={}", static_cast(pixel_format)); + return vk::ImageAspectFlagBits::eColor; + } +} + +vk::ImageViewType GetImageViewType(SurfaceTarget target) { + switch (target) { + case SurfaceTarget::Texture1D: + return vk::ImageViewType::e1D; + case SurfaceTarget::Texture2D: + return vk::ImageViewType::e2D; + case SurfaceTarget::Texture3D: + return vk::ImageViewType::e3D; + case SurfaceTarget::Texture1DArray: + return vk::ImageViewType::e1DArray; + case SurfaceTarget::Texture2DArray: + return vk::ImageViewType::e2DArray; + case SurfaceTarget::TextureCubemap: + return vk::ImageViewType::eCube; + case SurfaceTarget::TextureCubeArray: + return vk::ImageViewType::eCubeArray; + case SurfaceTarget::TextureBuffer: + break; + } + UNREACHABLE(); + return {}; +} + +UniqueBuffer CreateBuffer(const VKDevice& device, const SurfaceParams& params) { + // TODO(Rodrigo): Move texture buffer creation to the buffer cache + const vk::BufferCreateInfo buffer_ci({}, params.GetHostSizeInBytes(), + vk::BufferUsageFlagBits::eUniformTexelBuffer | + vk::BufferUsageFlagBits::eTransferSrc | + vk::BufferUsageFlagBits::eTransferDst, + vk::SharingMode::eExclusive, 0, nullptr); + const auto dev = device.GetLogical(); + const auto& dld = device.GetDispatchLoader(); + return dev.createBufferUnique(buffer_ci, nullptr, dld); +} + +vk::BufferViewCreateInfo GenerateBufferViewCreateInfo(const VKDevice& device, + const SurfaceParams& params, + vk::Buffer buffer) { + ASSERT(params.IsBuffer()); + + const auto format = + MaxwellToVK::SurfaceFormat(device, FormatType::Buffer, params.pixel_format).format; + return vk::BufferViewCreateInfo({}, buffer, format, 0, params.GetHostSizeInBytes()); +} + +vk::ImageCreateInfo GenerateImageCreateInfo(const VKDevice& device, const SurfaceParams& params) { + constexpr auto sample_count = vk::SampleCountFlagBits::e1; + constexpr auto tiling = vk::ImageTiling::eOptimal; + + ASSERT(!params.IsBuffer()); + + const auto [format, attachable, storage] = + MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, params.pixel_format); + + auto image_usage = vk::ImageUsageFlagBits::eSampled | vk::ImageUsageFlagBits::eTransferDst | + vk::ImageUsageFlagBits::eTransferSrc; + if (attachable) { + image_usage |= params.IsPixelFormatZeta() ? vk::ImageUsageFlagBits::eDepthStencilAttachment + : vk::ImageUsageFlagBits::eColorAttachment; + } + if (storage) { + image_usage |= vk::ImageUsageFlagBits::eStorage; + } + + vk::ImageCreateFlags flags; + vk::Extent3D extent; + switch (params.target) { + case SurfaceTarget::TextureCubemap: + case SurfaceTarget::TextureCubeArray: + flags |= vk::ImageCreateFlagBits::eCubeCompatible; + [[fallthrough]]; + case SurfaceTarget::Texture1D: + case SurfaceTarget::Texture1DArray: + case SurfaceTarget::Texture2D: + case SurfaceTarget::Texture2DArray: + extent = vk::Extent3D(params.width, params.height, 1); + break; + case SurfaceTarget::Texture3D: + extent = vk::Extent3D(params.width, params.height, params.depth); + break; + case SurfaceTarget::TextureBuffer: + UNREACHABLE(); + } + + return vk::ImageCreateInfo(flags, SurfaceTargetToImage(params.target), format, extent, + params.num_levels, static_cast(params.GetNumLayers()), + sample_count, tiling, image_usage, vk::SharingMode::eExclusive, 0, + nullptr, vk::ImageLayout::eUndefined); +} + +} // Anonymous namespace + +CachedSurface::CachedSurface(Core::System& system, const VKDevice& device, + VKResourceManager& resource_manager, VKMemoryManager& memory_manager, + VKScheduler& scheduler, VKStagingBufferPool& staging_pool, + GPUVAddr gpu_addr, const SurfaceParams& params) + : SurfaceBase{gpu_addr, params}, system{system}, device{device}, + resource_manager{resource_manager}, memory_manager{memory_manager}, scheduler{scheduler}, + staging_pool{staging_pool} { + if (params.IsBuffer()) { + buffer = CreateBuffer(device, params); + commit = memory_manager.Commit(*buffer, false); + + const auto buffer_view_ci = GenerateBufferViewCreateInfo(device, params, *buffer); + format = buffer_view_ci.format; + + const auto dev = device.GetLogical(); + const auto& dld = device.GetDispatchLoader(); + buffer_view = dev.createBufferViewUnique(buffer_view_ci, nullptr, dld); + } else { + const auto image_ci = GenerateImageCreateInfo(device, params); + format = image_ci.format; + + image.emplace(device, scheduler, image_ci, PixelFormatToImageAspect(params.pixel_format)); + commit = memory_manager.Commit(image->GetHandle(), false); + } + + // TODO(Rodrigo): Move this to a virtual function. + main_view = CreateViewInner( + ViewParams(params.target, 0, static_cast(params.GetNumLayers()), 0, params.num_levels), + true); +} + +CachedSurface::~CachedSurface() = default; + +void CachedSurface::UploadTexture(const std::vector& staging_buffer) { + // To upload data we have to be outside of a renderpass + scheduler.RequestOutsideRenderPassOperationContext(); + + if (params.IsBuffer()) { + UploadBuffer(staging_buffer); + } else { + UploadImage(staging_buffer); + } +} + +void CachedSurface::DownloadTexture(std::vector& staging_buffer) { + UNIMPLEMENTED_IF(params.IsBuffer()); + + if (params.pixel_format == VideoCore::Surface::PixelFormat::A1B5G5R5U) { + LOG_WARNING(Render_Vulkan, "A1B5G5R5 flushing is stubbed"); + } + + // We can't copy images to buffers inside a renderpass + scheduler.RequestOutsideRenderPassOperationContext(); + + FullTransition(vk::PipelineStageFlagBits::eTransfer, vk::AccessFlagBits::eTransferRead, + vk::ImageLayout::eTransferSrcOptimal); + + const auto& buffer = staging_pool.GetUnusedBuffer(host_memory_size, true); + // TODO(Rodrigo): Do this in a single copy + for (u32 level = 0; level < params.num_levels; ++level) { + scheduler.Record([image = image->GetHandle(), buffer = *buffer.handle, + copy = GetBufferImageCopy(level)](auto cmdbuf, auto& dld) { + cmdbuf.copyImageToBuffer(image, vk::ImageLayout::eTransferSrcOptimal, buffer, {copy}, + dld); + }); + } + scheduler.Finish(); + + // TODO(Rodrigo): Use an intern buffer for staging buffers and avoid this unnecesary memcpy. + std::memcpy(staging_buffer.data(), buffer.commit->Map(host_memory_size), host_memory_size); +} + +void CachedSurface::DecorateSurfaceName() { + // TODO(Rodrigo): Add name decorations +} + +View CachedSurface::CreateView(const ViewParams& params) { + return CreateViewInner(params, false); +} + +View CachedSurface::CreateViewInner(const ViewParams& params, bool is_proxy) { + auto view = std::make_shared(device, *this, params, is_proxy); + views[params] = view; + // TODO(Rodrigo): Add name decorations + return view; +} + +void CachedSurface::UploadBuffer(const std::vector& staging_buffer) { + const auto& src_buffer = staging_pool.GetUnusedBuffer(host_memory_size, true); + std::memcpy(src_buffer.commit->Map(host_memory_size), staging_buffer.data(), host_memory_size); + + scheduler.Record([src_buffer = *src_buffer.handle, dst_buffer = *buffer, + size = params.GetHostSizeInBytes()](auto cmdbuf, auto& dld) { + const vk::BufferCopy copy(0, 0, size); + cmdbuf.copyBuffer(src_buffer, dst_buffer, {copy}, dld); + + cmdbuf.pipelineBarrier( + vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eVertexShader, {}, {}, + {vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferWrite, + vk::AccessFlagBits::eShaderRead, 0, 0, dst_buffer, 0, size)}, + {}, dld); + }); +} + +void CachedSurface::UploadImage(const std::vector& staging_buffer) { + const auto& src_buffer = staging_pool.GetUnusedBuffer(host_memory_size, true); + std::memcpy(src_buffer.commit->Map(host_memory_size), staging_buffer.data(), host_memory_size); + + FullTransition(vk::PipelineStageFlagBits::eTransfer, vk::AccessFlagBits::eTransferWrite, + vk::ImageLayout::eTransferDstOptimal); + + for (u32 level = 0; level < params.num_levels; ++level) { + vk::BufferImageCopy copy = GetBufferImageCopy(level); + const auto& dld = device.GetDispatchLoader(); + if (image->GetAspectMask() == + (vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil)) { + vk::BufferImageCopy depth = copy; + vk::BufferImageCopy stencil = copy; + depth.imageSubresource.aspectMask = vk::ImageAspectFlagBits::eDepth; + stencil.imageSubresource.aspectMask = vk::ImageAspectFlagBits::eStencil; + scheduler.Record([buffer = *src_buffer.handle, image = image->GetHandle(), depth, + stencil](auto cmdbuf, auto& dld) { + cmdbuf.copyBufferToImage(buffer, image, vk::ImageLayout::eTransferDstOptimal, + {depth, stencil}, dld); + }); + } else { + scheduler.Record([buffer = *src_buffer.handle, image = image->GetHandle(), + copy](auto cmdbuf, auto& dld) { + cmdbuf.copyBufferToImage(buffer, image, vk::ImageLayout::eTransferDstOptimal, + {copy}, dld); + }); + } + } +} + +vk::BufferImageCopy CachedSurface::GetBufferImageCopy(u32 level) const { + const u32 vk_depth = params.target == SurfaceTarget::Texture3D ? params.GetMipDepth(level) : 1; + const auto compression_type = params.GetCompressionType(); + const std::size_t mip_offset = compression_type == SurfaceCompression::Converted + ? params.GetConvertedMipmapOffset(level) + : params.GetHostMipmapLevelOffset(level); + + return vk::BufferImageCopy( + mip_offset, 0, 0, + {image->GetAspectMask(), level, 0, static_cast(params.GetNumLayers())}, {0, 0, 0}, + {params.GetMipWidth(level), params.GetMipHeight(level), vk_depth}); +} + +vk::ImageSubresourceRange CachedSurface::GetImageSubresourceRange() const { + return {image->GetAspectMask(), 0, params.num_levels, 0, + static_cast(params.GetNumLayers())}; +} + +CachedSurfaceView::CachedSurfaceView(const VKDevice& device, CachedSurface& surface, + const ViewParams& params, bool is_proxy) + : VideoCommon::ViewBase{params}, params{surface.GetSurfaceParams()}, + image{surface.GetImageHandle()}, buffer_view{surface.GetBufferViewHandle()}, + aspect_mask{surface.GetAspectMask()}, device{device}, surface{surface}, + base_layer{params.base_layer}, num_layers{params.num_layers}, base_level{params.base_level}, + num_levels{params.num_levels} { + if (image) { + image_view_type = GetImageViewType(params.target); + } +} + +CachedSurfaceView::~CachedSurfaceView() = default; + +vk::ImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y_source, + SwizzleSource z_source, SwizzleSource w_source) { + const u32 swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source); + if (last_image_view && last_swizzle == swizzle) { + return last_image_view; + } + last_swizzle = swizzle; + + const auto [entry, is_cache_miss] = view_cache.try_emplace(swizzle); + auto& image_view = entry->second; + if (!is_cache_miss) { + return last_image_view = *image_view; + } + + auto swizzle_x = MaxwellToVK::SwizzleSource(x_source); + auto swizzle_y = MaxwellToVK::SwizzleSource(y_source); + auto swizzle_z = MaxwellToVK::SwizzleSource(z_source); + auto swizzle_w = MaxwellToVK::SwizzleSource(w_source); + + if (params.pixel_format == VideoCore::Surface::PixelFormat::A1B5G5R5U) { + // A1B5G5R5 is implemented as A1R5G5B5, we have to change the swizzle here. + std::swap(swizzle_x, swizzle_z); + } + + // Games can sample depth or stencil values on textures. This is decided by the swizzle value on + // hardware. To emulate this on Vulkan we specify it in the aspect. + vk::ImageAspectFlags aspect = aspect_mask; + if (aspect == (vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil)) { + UNIMPLEMENTED_IF(x_source != SwizzleSource::R && x_source != SwizzleSource::G); + const bool is_first = x_source == SwizzleSource::R; + switch (params.pixel_format) { + case VideoCore::Surface::PixelFormat::Z24S8: + case VideoCore::Surface::PixelFormat::Z32FS8: + aspect = is_first ? vk::ImageAspectFlagBits::eDepth : vk::ImageAspectFlagBits::eStencil; + break; + case VideoCore::Surface::PixelFormat::S8Z24: + aspect = is_first ? vk::ImageAspectFlagBits::eStencil : vk::ImageAspectFlagBits::eDepth; + break; + default: + aspect = vk::ImageAspectFlagBits::eDepth; + UNIMPLEMENTED(); + } + + // Vulkan doesn't seem to understand swizzling of a depth stencil image, use identity + swizzle_x = vk::ComponentSwizzle::eR; + swizzle_y = vk::ComponentSwizzle::eG; + swizzle_z = vk::ComponentSwizzle::eB; + swizzle_w = vk::ComponentSwizzle::eA; + } + + const vk::ImageViewCreateInfo image_view_ci( + {}, surface.GetImageHandle(), image_view_type, surface.GetImage().GetFormat(), + {swizzle_x, swizzle_y, swizzle_z, swizzle_w}, + {aspect, base_level, num_levels, base_layer, num_layers}); + + const auto dev = device.GetLogical(); + image_view = dev.createImageViewUnique(image_view_ci, nullptr, device.GetDispatchLoader()); + return last_image_view = *image_view; +} + +bool CachedSurfaceView::IsOverlapping(const View& rhs) const { + // TODO(Rodrigo): Also test for layer and mip level overlaps. + return &surface == &rhs->surface; +} + +VKTextureCache::VKTextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer, + const VKDevice& device, VKResourceManager& resource_manager, + VKMemoryManager& memory_manager, VKScheduler& scheduler, + VKStagingBufferPool& staging_pool) + : TextureCache(system, rasterizer), device{device}, resource_manager{resource_manager}, + memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{staging_pool} {} + +VKTextureCache::~VKTextureCache() = default; + +Surface VKTextureCache::CreateSurface(GPUVAddr gpu_addr, const SurfaceParams& params) { + return std::make_shared(system, device, resource_manager, memory_manager, + scheduler, staging_pool, gpu_addr, params); +} + +void VKTextureCache::ImageCopy(Surface& src_surface, Surface& dst_surface, + const VideoCommon::CopyParams& copy_params) { + const bool src_3d = src_surface->GetSurfaceParams().target == SurfaceTarget::Texture3D; + const bool dst_3d = dst_surface->GetSurfaceParams().target == SurfaceTarget::Texture3D; + UNIMPLEMENTED_IF(src_3d); + + // The texture cache handles depth in OpenGL terms, we have to handle it as subresource and + // dimension respectively. + const u32 dst_base_layer = dst_3d ? 0 : copy_params.dest_z; + const u32 dst_offset_z = dst_3d ? copy_params.dest_z : 0; + + const u32 extent_z = dst_3d ? copy_params.depth : 1; + const u32 num_layers = dst_3d ? 1 : copy_params.depth; + + // We can't copy inside a renderpass + scheduler.RequestOutsideRenderPassOperationContext(); + + src_surface->Transition(copy_params.source_z, copy_params.depth, copy_params.source_level, 1, + vk::PipelineStageFlagBits::eTransfer, vk::AccessFlagBits::eTransferRead, + vk::ImageLayout::eTransferSrcOptimal); + dst_surface->Transition( + dst_base_layer, num_layers, copy_params.dest_level, 1, vk::PipelineStageFlagBits::eTransfer, + vk::AccessFlagBits::eTransferWrite, vk::ImageLayout::eTransferDstOptimal); + + const auto& dld{device.GetDispatchLoader()}; + const vk::ImageSubresourceLayers src_subresource( + src_surface->GetAspectMask(), copy_params.source_level, copy_params.source_z, num_layers); + const vk::ImageSubresourceLayers dst_subresource( + dst_surface->GetAspectMask(), copy_params.dest_level, dst_base_layer, num_layers); + const vk::Offset3D src_offset(copy_params.source_x, copy_params.source_y, 0); + const vk::Offset3D dst_offset(copy_params.dest_x, copy_params.dest_y, dst_offset_z); + const vk::Extent3D extent(copy_params.width, copy_params.height, extent_z); + const vk::ImageCopy copy(src_subresource, src_offset, dst_subresource, dst_offset, extent); + const vk::Image src_image = src_surface->GetImageHandle(); + const vk::Image dst_image = dst_surface->GetImageHandle(); + scheduler.Record([src_image, dst_image, copy](auto cmdbuf, auto& dld) { + cmdbuf.copyImage(src_image, vk::ImageLayout::eTransferSrcOptimal, dst_image, + vk::ImageLayout::eTransferDstOptimal, {copy}, dld); + }); +} + +void VKTextureCache::ImageBlit(View& src_view, View& dst_view, + const Tegra::Engines::Fermi2D::Config& copy_config) { + // We can't blit inside a renderpass + scheduler.RequestOutsideRenderPassOperationContext(); + + src_view->Transition(vk::ImageLayout::eTransferSrcOptimal, vk::PipelineStageFlagBits::eTransfer, + vk::AccessFlagBits::eTransferRead); + dst_view->Transition(vk::ImageLayout::eTransferDstOptimal, vk::PipelineStageFlagBits::eTransfer, + vk::AccessFlagBits::eTransferWrite); + + const auto& cfg = copy_config; + const auto src_top_left = vk::Offset3D(cfg.src_rect.left, cfg.src_rect.top, 0); + const auto src_bot_right = vk::Offset3D(cfg.src_rect.right, cfg.src_rect.bottom, 1); + const auto dst_top_left = vk::Offset3D(cfg.dst_rect.left, cfg.dst_rect.top, 0); + const auto dst_bot_right = vk::Offset3D(cfg.dst_rect.right, cfg.dst_rect.bottom, 1); + const vk::ImageBlit blit(src_view->GetImageSubresourceLayers(), {src_top_left, src_bot_right}, + dst_view->GetImageSubresourceLayers(), {dst_top_left, dst_bot_right}); + const bool is_linear = copy_config.filter == Tegra::Engines::Fermi2D::Filter::Linear; + + const auto& dld{device.GetDispatchLoader()}; + scheduler.Record([src_image = src_view->GetImage(), dst_image = dst_view->GetImage(), blit, + is_linear](auto cmdbuf, auto& dld) { + cmdbuf.blitImage(src_image, vk::ImageLayout::eTransferSrcOptimal, dst_image, + vk::ImageLayout::eTransferDstOptimal, {blit}, + is_linear ? vk::Filter::eLinear : vk::Filter::eNearest, dld); + }); +} + +void VKTextureCache::BufferCopy(Surface& src_surface, Surface& dst_surface) { + // Currently unimplemented. PBO copies should be dropped and we should use a render pass to + // convert from color to depth and viceversa. +} + +} // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h new file mode 100644 index 000000000..4bc330c9f --- /dev/null +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -0,0 +1,244 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include +#include +#include + +#include +#include + +#include "common/assert.h" +#include "common/common_types.h" +#include "common/hash.h" +#include "common/logging/log.h" +#include "common/math_util.h" +#include "video_core/gpu.h" +#include "video_core/rasterizer_cache.h" +#include "video_core/renderer_vulkan/declarations.h" +#include "video_core/renderer_vulkan/vk_image.h" +#include "video_core/renderer_vulkan/vk_memory_manager.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/surface.h" +#include "video_core/texture_cache/surface_base.h" +#include "video_core/texture_cache/texture_cache.h" +#include "video_core/textures/decoders.h" +#include "video_core/textures/texture.h" + +namespace Core { +class System; +} + +namespace VideoCore { +class RasterizerInterface; +} + +namespace Vulkan { + +class RasterizerVulkan; +class VKDevice; +class VKResourceManager; +class VKScheduler; +class VKStagingBufferPool; + +class CachedSurfaceView; +class CachedSurface; + +using Surface = std::shared_ptr; +using View = std::shared_ptr; +using TextureCacheBase = VideoCommon::TextureCache; + +using VideoCommon::SurfaceParams; +using VideoCommon::ViewParams; + +class CachedSurface final : public VideoCommon::SurfaceBase { + friend CachedSurfaceView; + +public: + explicit CachedSurface(Core::System& system, const VKDevice& device, + VKResourceManager& resource_manager, VKMemoryManager& memory_manager, + VKScheduler& scheduler, VKStagingBufferPool& staging_pool, + GPUVAddr gpu_addr, const SurfaceParams& params); + ~CachedSurface(); + + void UploadTexture(const std::vector& staging_buffer) override; + void DownloadTexture(std::vector& staging_buffer) override; + + void FullTransition(vk::PipelineStageFlags new_stage_mask, vk::AccessFlags new_access, + vk::ImageLayout new_layout) { + image->Transition(0, static_cast(params.GetNumLayers()), 0, params.num_levels, + new_stage_mask, new_access, new_layout); + } + + void Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels, + vk::PipelineStageFlags new_stage_mask, vk::AccessFlags new_access, + vk::ImageLayout new_layout) { + image->Transition(base_layer, num_layers, base_level, num_levels, new_stage_mask, + new_access, new_layout); + } + + VKImage& GetImage() { + return *image; + } + + const VKImage& GetImage() const { + return *image; + } + + vk::Image GetImageHandle() const { + return image->GetHandle(); + } + + vk::ImageAspectFlags GetAspectMask() const { + return image->GetAspectMask(); + } + + vk::BufferView GetBufferViewHandle() const { + return *buffer_view; + } + +protected: + void DecorateSurfaceName(); + + View CreateView(const ViewParams& params) override; + View CreateViewInner(const ViewParams& params, bool is_proxy); + +private: + void UploadBuffer(const std::vector& staging_buffer); + + void UploadImage(const std::vector& staging_buffer); + + vk::BufferImageCopy GetBufferImageCopy(u32 level) const; + + vk::ImageSubresourceRange GetImageSubresourceRange() const; + + Core::System& system; + const VKDevice& device; + VKResourceManager& resource_manager; + VKMemoryManager& memory_manager; + VKScheduler& scheduler; + VKStagingBufferPool& staging_pool; + + std::optional image; + UniqueBuffer buffer; + UniqueBufferView buffer_view; + VKMemoryCommit commit; + + vk::Format format; +}; + +class CachedSurfaceView final : public VideoCommon::ViewBase { +public: + explicit CachedSurfaceView(const VKDevice& device, CachedSurface& surface, + const ViewParams& params, bool is_proxy); + ~CachedSurfaceView(); + + vk::ImageView GetHandle(Tegra::Texture::SwizzleSource x_source, + Tegra::Texture::SwizzleSource y_source, + Tegra::Texture::SwizzleSource z_source, + Tegra::Texture::SwizzleSource w_source); + + bool IsOverlapping(const View& rhs) const; + + vk::ImageView GetHandle() { + return GetHandle(Tegra::Texture::SwizzleSource::R, Tegra::Texture::SwizzleSource::G, + Tegra::Texture::SwizzleSource::B, Tegra::Texture::SwizzleSource::A); + } + + u32 GetWidth() const { + return params.GetMipWidth(base_level); + } + + u32 GetHeight() const { + return params.GetMipHeight(base_level); + } + + bool IsBufferView() const { + return buffer_view; + } + + vk::Image GetImage() const { + return image; + } + + vk::BufferView GetBufferView() const { + return buffer_view; + } + + vk::ImageSubresourceRange GetImageSubresourceRange() const { + return {aspect_mask, base_level, num_levels, base_layer, num_layers}; + } + + vk::ImageSubresourceLayers GetImageSubresourceLayers() const { + return {surface.GetAspectMask(), base_level, base_layer, num_layers}; + } + + void Transition(vk::ImageLayout new_layout, vk::PipelineStageFlags new_stage_mask, + vk::AccessFlags new_access) const { + surface.Transition(base_layer, num_layers, base_level, num_levels, new_stage_mask, + new_access, new_layout); + } + + void MarkAsModified(u64 tick) { + surface.MarkAsModified(true, tick); + } + +private: + static u32 EncodeSwizzle(Tegra::Texture::SwizzleSource x_source, + Tegra::Texture::SwizzleSource y_source, + Tegra::Texture::SwizzleSource z_source, + Tegra::Texture::SwizzleSource w_source) { + return (static_cast(x_source) << 24) | (static_cast(y_source) << 16) | + (static_cast(z_source) << 8) | static_cast(w_source); + } + + // Store a copy of these values to avoid double dereference when reading them + const SurfaceParams params; + const vk::Image image; + const vk::BufferView buffer_view; + const vk::ImageAspectFlags aspect_mask; + + const VKDevice& device; + CachedSurface& surface; + const u32 base_layer; + const u32 num_layers; + const u32 base_level; + const u32 num_levels; + vk::ImageViewType image_view_type{}; + + vk::ImageView last_image_view; + u32 last_swizzle{}; + + std::unordered_map view_cache; +}; + +class VKTextureCache final : public TextureCacheBase { +public: + explicit VKTextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer, + const VKDevice& device, VKResourceManager& resource_manager, + VKMemoryManager& memory_manager, VKScheduler& scheduler, + VKStagingBufferPool& staging_pool); + ~VKTextureCache(); + +private: + Surface CreateSurface(GPUVAddr gpu_addr, const SurfaceParams& params) override; + + void ImageCopy(Surface& src_surface, Surface& dst_surface, + const VideoCommon::CopyParams& copy_params) override; + + void ImageBlit(View& src_view, View& dst_view, + const Tegra::Engines::Fermi2D::Config& copy_config) override; + + void BufferCopy(Surface& src_surface, Surface& dst_surface) override; + + const VKDevice& device; + VKResourceManager& resource_manager; + VKMemoryManager& memory_manager; + VKScheduler& scheduler; + VKStagingBufferPool& staging_pool; +}; + +} // namespace Vulkan