diff options
Diffstat (limited to 'src/video_core/renderer_vulkan/vk_buffer_cache.cpp')
-rw-r--r-- | src/video_core/renderer_vulkan/vk_buffer_cache.cpp | 125 |
1 files changed, 73 insertions, 52 deletions
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index d9d3da9ea..4d517c547 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -9,10 +9,10 @@ #include "core/core.h" #include "video_core/buffer_cache/buffer_cache.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" -#include "video_core/renderer_vulkan/vk_device.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_stream_buffer.h" -#include "video_core/renderer_vulkan/wrapper.h" +#include "video_core/vulkan_common/vulkan_device.h" +#include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { @@ -31,20 +31,24 @@ 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; -std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) { - return std::make_unique<VKStreamBuffer>(device, scheduler, BUFFER_USAGE); +constexpr VkAccessFlags TRANSFORM_FEEDBACK_WRITE_ACCESS = + VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT | VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT; + +std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const Device& device, VKScheduler& scheduler) { + return std::make_unique<VKStreamBuffer>(device, scheduler); } } // Anonymous namespace -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_} { +Buffer::Buffer(const Device& device_, VKMemoryManager& memory_manager, VKScheduler& scheduler_, + VKStagingBufferPool& staging_pool_, VAddr cpu_addr_, std::size_t size_) + : BufferBlock{cpu_addr_, size_}, device{device_}, scheduler{scheduler_}, staging_pool{ + staging_pool_} { const VkBufferCreateInfo ci{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, - .size = static_cast<VkDeviceSize>(size), + .size = static_cast<VkDeviceSize>(size_), .usage = BUFFER_USAGE | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, @@ -57,69 +61,86 @@ Buffer::Buffer(const VKDevice& device, VKMemoryManager& memory_manager, VKSchedu Buffer::~Buffer() = default; -void Buffer::Upload(std::size_t offset, std::size_t size, const u8* data) { - const auto& staging = staging_pool.GetUnusedBuffer(size, true); - std::memcpy(staging.commit->Map(size), data, size); +void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) { + const auto& staging = staging_pool.GetUnusedBuffer(data_size, true); + std::memcpy(staging.commit->Map(data_size), data, data_size); scheduler.RequestOutsideRenderPassOperationContext(); const VkBuffer handle = Handle(); - scheduler.Record([staging = *staging.handle, handle, offset, size](vk::CommandBuffer cmdbuf) { - cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, size}); - - const VkBufferMemoryBarrier 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_TRANSFER_WRITE_BIT, - .dstAccessMask = UPLOAD_ACCESS_BARRIERS, + .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 = size, + .size = data_size, }; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {}, - barrier, {}); - }); -} - -void Buffer::Download(std::size_t offset, std::size_t size, u8* data) { - const auto& staging = staging_pool.GetUnusedBuffer(size, true); - scheduler.RequestOutsideRenderPassOperationContext(); - - const VkBuffer handle = Handle(); - scheduler.Record([staging = *staging.handle, handle, offset, size](vk::CommandBuffer cmdbuf) { - const VkBufferMemoryBarrier barrier{ + const VkBufferMemoryBarrier write_barrier{ .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .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 = size, + .size = data_size, }; - - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, barrier, {}); - cmdbuf.CopyBuffer(handle, staging, VkBufferCopy{offset, 0, 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) { + const auto& staging = staging_pool.GetUnusedBuffer(data_size, true); + scheduler.RequestOutsideRenderPassOperationContext(); + + const VkBuffer handle = Handle(); + scheduler.Record( + [staging = *staging.handle, handle, offset, data_size](vk::CommandBuffer cmdbuf) { + const VkBufferMemoryBarrier barrier{ + .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .buffer = handle, + .offset = offset, + .size = data_size, + }; + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, + VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, barrier, {}); + cmdbuf.CopyBuffer(handle, staging, VkBufferCopy{offset, 0, data_size}); + }); scheduler.Finish(); - std::memcpy(data, staging.commit->Map(size), size); + std::memcpy(data, staging.commit->Map(data_size), data_size); } void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset, - std::size_t size) { + std::size_t copy_size) { scheduler.RequestOutsideRenderPassOperationContext(); const VkBuffer dst_buffer = Handle(); scheduler.Record([src_buffer = src.Handle(), dst_buffer, src_offset, dst_offset, - size](vk::CommandBuffer cmdbuf) { - cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, size}); + copy_size](vk::CommandBuffer cmdbuf) { + cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, copy_size}); std::array<VkBufferMemoryBarrier, 2> barriers; barriers[0].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; @@ -130,7 +151,7 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst barriers[0].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; barriers[0].buffer = src_buffer; barriers[0].offset = src_offset; - barriers[0].size = size; + barriers[0].size = copy_size; barriers[1].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; barriers[1].pNext = nullptr; barriers[1].srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; @@ -139,19 +160,19 @@ void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst barriers[1].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; barriers[1].buffer = dst_buffer; barriers[1].offset = dst_offset; - barriers[1].size = size; + barriers[1].size = copy_size; cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {}, barriers, {}); }); } -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_) - : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer, gpu_memory, cpu_memory, - CreateStreamBuffer(device_, - scheduler_)}, +VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer_, + Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, + const Device& device_, VKMemoryManager& memory_manager_, + VKScheduler& scheduler_, VKStreamBuffer& stream_buffer_, + VKStagingBufferPool& staging_pool_) + : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer_, gpu_memory_, + cpu_memory_, stream_buffer_}, device{device_}, memory_manager{memory_manager_}, scheduler{scheduler_}, staging_pool{ staging_pool_} {} |