diff options
Diffstat (limited to 'src/video_core/renderer_vulkan')
8 files changed, 63 insertions, 54 deletions
diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index 59e963263..f93447610 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -257,6 +257,8 @@ vk::ShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage) { return vk::ShaderStageFlagBits::eGeometry; case Tegra::Engines::ShaderType::Fragment: return vk::ShaderStageFlagBits::eFragment; + case Tegra::Engines::ShaderType::Compute: + return vk::ShaderStageFlagBits::eCompute; } UNIMPLEMENTED_MSG("Unimplemented shader stage={}", static_cast<u32>(stage)); return {}; @@ -367,6 +369,10 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr return vk::Format::eR8G8B8A8Uint; case Maxwell::VertexAttribute::Size::Size_32: return vk::Format::eR32Uint; + case Maxwell::VertexAttribute::Size::Size_32_32: + return vk::Format::eR32G32Uint; + case Maxwell::VertexAttribute::Size::Size_32_32_32: + return vk::Format::eR32G32B32Uint; case Maxwell::VertexAttribute::Size::Size_32_32_32_32: return vk::Format::eR32G32B32A32Uint; default: diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 056ef495c..557b9d662 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -179,10 +179,11 @@ Tegra::Engines::ConstBufferEngineInterface& CachedShader::GetEngine( VKPipelineCache::VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer, const VKDevice& device, VKScheduler& scheduler, VKDescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue) + VKUpdateDescriptorQueue& update_descriptor_queue, + VKRenderPassCache& renderpass_cache) : RasterizerCache{rasterizer}, system{system}, device{device}, scheduler{scheduler}, descriptor_pool{descriptor_pool}, update_descriptor_queue{update_descriptor_queue}, - renderpass_cache(device) {} + renderpass_cache{renderpass_cache} {} VKPipelineCache::~VKPipelineCache() = default; @@ -191,7 +192,6 @@ std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() { std::array<Shader, Maxwell::MaxShaderProgram> shaders; for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { - const auto& shader_config = gpu.regs.shader_config[index]; const auto program{static_cast<Maxwell::ShaderProgram>(index)}; // Skip stages that are not enabled diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h index 21340c9a4..c4c112290 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h @@ -161,7 +161,8 @@ public: explicit VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer, const VKDevice& device, VKScheduler& scheduler, VKDescriptorPool& descriptor_pool, - VKUpdateDescriptorQueue& update_descriptor_queue); + VKUpdateDescriptorQueue& update_descriptor_queue, + VKRenderPassCache& renderpass_cache); ~VKPipelineCache(); std::array<Shader, Maxwell::MaxShaderProgram> GetShaders(); @@ -184,8 +185,7 @@ private: VKScheduler& scheduler; VKDescriptorPool& descriptor_pool; VKUpdateDescriptorQueue& update_descriptor_queue; - - VKRenderPassCache renderpass_cache; + VKRenderPassCache& renderpass_cache; std::array<Shader, Maxwell::MaxShaderProgram> last_shaders; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index c9886cc16..58c69b786 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -287,12 +287,13 @@ RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWind screen_info{screen_info}, device{device}, resource_manager{resource_manager}, memory_manager{memory_manager}, state_tracker{state_tracker}, scheduler{scheduler}, staging_pool(device, memory_manager, scheduler), descriptor_pool(device), - update_descriptor_queue(device, scheduler), + update_descriptor_queue(device, scheduler), renderpass_cache(device), quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), texture_cache(system, *this, device, resource_manager, memory_manager, scheduler, staging_pool), - pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue), + pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue, + renderpass_cache), buffer_cache(*this, system, device, memory_manager, scheduler, staging_pool), sampler_cache(device), query_cache(system, *this, device, scheduler) { scheduler.SetQueryCache(query_cache); @@ -365,13 +366,16 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { void RasterizerVulkan::Clear() { MICROPROFILE_SCOPE(Vulkan_Clearing); - query_cache.UpdateCounters(); - const auto& gpu = system.GPU().Maxwell3D(); if (!system.GPU().Maxwell3D().ShouldExecute()) { return; } + sampled_views.clear(); + image_views.clear(); + + query_cache.UpdateCounters(); + const auto& regs = gpu.regs; const bool use_color = regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B || regs.clear_buffers.A; @@ -380,52 +384,54 @@ void RasterizerVulkan::Clear() { if (!use_color && !use_depth && !use_stencil) { return; } - // Clearing images requires to be out of a renderpass - scheduler.RequestOutsideRenderPassOperationContext(); - // TODO(Rodrigo): Implement clears rendering a quad or using beginning a renderpass. + [[maybe_unused]] const auto texceptions = UpdateAttachments(); + DEBUG_ASSERT(texceptions.none()); + SetupImageTransitions(0, color_attachments, zeta_attachment); - if (use_color) { - View color_view; - { - MICROPROFILE_SCOPE(Vulkan_RenderTargets); - color_view = texture_cache.GetColorBufferSurface(regs.clear_buffers.RT.Value(), false); - } + const vk::RenderPass renderpass = renderpass_cache.GetRenderPass(GetRenderPassParams(0)); + const auto [framebuffer, render_area] = ConfigureFramebuffers(renderpass); + scheduler.RequestRenderpass({renderpass, framebuffer, {{0, 0}, render_area}, 0, nullptr}); + + const auto& scissor = regs.scissor_test[0]; + const vk::Offset2D scissor_offset(scissor.min_x, scissor.min_y); + vk::Extent2D scissor_extent{scissor.max_x - scissor.min_x, scissor.max_y - scissor.min_y}; + scissor_extent.width = std::min(scissor_extent.width, render_area.width); + scissor_extent.height = std::min(scissor_extent.height, render_area.height); - color_view->Transition(vk::ImageLayout::eTransferDstOptimal, - vk::PipelineStageFlagBits::eTransfer, - vk::AccessFlagBits::eTransferWrite); + const u32 layer = regs.clear_buffers.layer; + const vk::ClearRect clear_rect({scissor_offset, scissor_extent}, layer, 1); + if (use_color) { const std::array clear_color = {regs.clear_color[0], regs.clear_color[1], regs.clear_color[2], regs.clear_color[3]}; - const vk::ClearColorValue clear(clear_color); - scheduler.Record([image = color_view->GetImage(), - subresource = color_view->GetImageSubresourceRange(), - clear](auto cmdbuf, auto& dld) { - cmdbuf.clearColorImage(image, vk::ImageLayout::eTransferDstOptimal, clear, subresource, - dld); + const vk::ClearValue clear_value{clear_color}; + const u32 color_attachment = regs.clear_buffers.RT; + scheduler.Record([color_attachment, clear_value, clear_rect](auto cmdbuf, auto& dld) { + const vk::ClearAttachment attachment(vk::ImageAspectFlagBits::eColor, color_attachment, + clear_value); + cmdbuf.clearAttachments(1, &attachment, 1, &clear_rect, dld); }); } - if (use_depth || use_stencil) { - View zeta_surface; - { - MICROPROFILE_SCOPE(Vulkan_RenderTargets); - zeta_surface = texture_cache.GetDepthBufferSurface(false); - } - zeta_surface->Transition(vk::ImageLayout::eTransferDstOptimal, - vk::PipelineStageFlagBits::eTransfer, - vk::AccessFlagBits::eTransferWrite); - - const vk::ClearDepthStencilValue clear(regs.clear_depth, - static_cast<u32>(regs.clear_stencil)); - scheduler.Record([image = zeta_surface->GetImage(), - subresource = zeta_surface->GetImageSubresourceRange(), - clear](auto cmdbuf, auto& dld) { - cmdbuf.clearDepthStencilImage(image, vk::ImageLayout::eTransferDstOptimal, clear, - subresource, dld); - }); + if (!use_depth && !use_stencil) { + return; + } + vk::ImageAspectFlags aspect_flags; + if (use_depth) { + aspect_flags |= vk::ImageAspectFlagBits::eDepth; + } + if (use_stencil) { + aspect_flags |= vk::ImageAspectFlagBits::eStencil; } + + scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil, + clear_rect, aspect_flags](auto cmdbuf, auto& dld) { + const vk::ClearDepthStencilValue clear_zeta(clear_depth, clear_stencil); + const vk::ClearValue clear_value{clear_zeta}; + const vk::ClearAttachment attachment(aspect_flags, 0, clear_value); + cmdbuf.clearAttachments(1, &attachment, 1, &clear_rect, dld); + }); } void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) { @@ -542,8 +548,6 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config, // Verify that the cached surface is the same size and format as the requested framebuffer const auto& params{surface->GetSurfaceParams()}; - const auto& pixel_format{ - VideoCore::Surface::PixelFormatFromGPUPixelFormat(config.pixel_format)}; ASSERT_MSG(params.width == config.width, "Framebuffer width is different"); ASSERT_MSG(params.height == config.height, "Framebuffer height is different"); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index b2e73d98d..3185868e9 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -253,6 +253,7 @@ private: VKStagingBufferPool staging_pool; VKDescriptorPool descriptor_pool; VKUpdateDescriptorQueue update_descriptor_queue; + VKRenderPassCache renderpass_cache; QuadArrayPass quad_array_pass; Uint8Pass uint8_pass; diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index d9ea3cc21..374959f82 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -100,7 +100,6 @@ void VKStagingBufferPool::ReleaseCache(bool host_visible) { } u64 VKStagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, std::size_t log2) { - static constexpr u64 epochs_to_destroy = 180; static constexpr std::size_t deletions_per_tick = 16; auto& staging = cache[log2]; @@ -108,6 +107,7 @@ u64 VKStagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, std::size_t lo const std::size_t old_size = entries.size(); const auto is_deleteable = [this](const auto& entry) { + static constexpr u64 epochs_to_destroy = 180; return entry.last_epoch + epochs_to_destroy < epoch && !entry.watch.IsUsed(); }; const std::size_t begin_offset = staging.delete_index; diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp index d74e68b63..94a89e388 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp +++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp @@ -90,8 +90,6 @@ void StateTracker::Initialize() { SetupDirtyBlendConstants(tables); SetupDirtyDepthBounds(tables); SetupDirtyStencilProperties(tables); - - SetupCommonOnWriteStores(dirty.on_write_stores); } void StateTracker::InvalidateCommandBufferState() { diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 73d92a5ae..26175921b 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -52,6 +52,9 @@ vk::ImageType SurfaceTargetToImage(SurfaceTarget target) { return vk::ImageType::e2D; case SurfaceTarget::Texture3D: return vk::ImageType::e3D; + case SurfaceTarget::TextureBuffer: + UNREACHABLE(); + return {}; } UNREACHABLE_MSG("Unknown texture target={}", static_cast<u32>(target)); return {}; @@ -273,7 +276,6 @@ void CachedSurface::UploadImage(const std::vector<u8>& staging_buffer) { 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; @@ -422,7 +424,6 @@ void VKTextureCache::ImageCopy(Surface& src_surface, Surface& dst_surface, 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( @@ -458,7 +459,6 @@ void VKTextureCache::ImageBlit(View& src_view, View& dst_view, 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, |