From 9764c13d6d2977903f407761b27d847c0056e1c4 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Wed, 30 Dec 2020 02:25:23 -0300 Subject: video_core: Rewrite the texture cache The current texture cache has several points that hurt maintainability and performance. It's easy to break unrelated parts of the cache when doing minor changes. The cache can easily forget valuable information about the cached textures by CPU writes or simply by its normal usage.The current texture cache has several points that hurt maintainability and performance. It's easy to break unrelated parts of the cache when doing minor changes. The cache can easily forget valuable information about the cached textures by CPU writes or simply by its normal usage. This commit aims to address those issues. --- src/video_core/renderer_vulkan/vk_buffer_cache.cpp | 67 ++++++++++++++-------- 1 file changed, 44 insertions(+), 23 deletions(-) (limited to 'src/video_core/renderer_vulkan/vk_buffer_cache.cpp') diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 444d3fb93..10d296c2f 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -31,15 +31,19 @@ constexpr VkAccessFlags UPLOAD_ACCESS_BARRIERS = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT; +constexpr VkAccessFlags TRANSFORM_FEEDBACK_WRITE_ACCESS = + VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT | VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT; + std::unique_ptr CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) { - return std::make_unique(device, scheduler, BUFFER_USAGE); + return std::make_unique(device, scheduler); } } // Anonymous namespace -Buffer::Buffer(const VKDevice& device, VKMemoryManager& memory_manager, VKScheduler& scheduler_, +Buffer::Buffer(const VKDevice& device_, VKMemoryManager& memory_manager, VKScheduler& scheduler_, VKStagingBufferPool& staging_pool_, VAddr cpu_addr_, std::size_t size_) - : BufferBlock{cpu_addr_, size_}, scheduler{scheduler_}, staging_pool{staging_pool_} { + : BufferBlock{cpu_addr_, size_}, device{device_}, scheduler{scheduler_}, staging_pool{ + staging_pool_} { const VkBufferCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, @@ -64,24 +68,39 @@ void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) { scheduler.RequestOutsideRenderPassOperationContext(); const VkBuffer handle = Handle(); - scheduler.Record( - [staging = *staging.handle, handle, offset, data_size](vk::CommandBuffer cmdbuf) { - cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, data_size}); - - const VkBufferMemoryBarrier barrier{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = UPLOAD_ACCESS_BARRIERS, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .buffer = handle, - .offset = offset, - .size = data_size, - }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {}, - barrier, {}); - }); + scheduler.Record([staging = *staging.handle, handle, offset, data_size, + &device = device](vk::CommandBuffer cmdbuf) { + const VkBufferMemoryBarrier read_barrier{ + .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = + VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | + VK_ACCESS_HOST_WRITE_BIT | + (device.IsExtTransformFeedbackSupported() ? TRANSFORM_FEEDBACK_WRITE_ACCESS : 0), + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .buffer = handle, + .offset = offset, + .size = data_size, + }; + const VkBufferMemoryBarrier write_barrier{ + .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = UPLOAD_ACCESS_BARRIERS, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .buffer = handle, + .offset = offset, + .size = data_size, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, read_barrier); + cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, data_size}); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, + write_barrier); + }); } void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) { @@ -150,8 +169,10 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer_, Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, const VKDevice& device_, VKMemoryManager& memory_manager_, - VKScheduler& scheduler_, VKStagingBufferPool& staging_pool_) - : BufferCache{rasterizer_, gpu_memory_, cpu_memory_, CreateStreamBuffer(device_, scheduler_)}, + VKScheduler& scheduler_, VKStreamBuffer& stream_buffer_, + VKStagingBufferPool& staging_pool_) + : VideoCommon::BufferCache{rasterizer_, gpu_memory_, + cpu_memory_, stream_buffer_}, device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_}, staging_pool{ staging_pool_} {} -- cgit v1.2.3