Linuxydable/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity
suyu/src/video_core/renderer_vulkan/vk_buffer_cache.cpp

176 lines
7.7 KiB
C++
Raw Normal View History

vk_buffer_cache: Implement a buffer cache This buffer cache is just like OpenGL's buffer cache with some minor style changes. It uses VKStreamBuffer.
2019-02-28 04:49:28 +01:00
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
#include <algorithm>
#include <cstring>
#include <memory>
#include <optional>
#include <tuple>
#include "common/assert.h"
#include "common/bit_util.h"
#include "core/core.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"
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
#include "video_core/renderer_vulkan/wrapper.h"
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
namespace Vulkan {
namespace {
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
constexpr VkBufferUsageFlags BUFFER_USAGE =
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
constexpr VkPipelineStageFlags UPLOAD_PIPELINE_STAGE =
VK_PIPELINE_STAGE_TRANSFER_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
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;
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) {
return std::make_unique<VKStreamBuffer>(device, scheduler, BUFFER_USAGE);
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
}
} // Anonymous namespace
CachedBufferBlock::CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
Buffer Cache: Use vAddr instead of physical memory.
2020-04-05 23:23:49 +02:00
VAddr cpu_addr, std::size_t size)
: VideoCommon::BufferBlock{cpu_addr, size} {
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
VkBufferCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.size = static_cast<VkDeviceSize>(size);
ci.usage = BUFFER_USAGE | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
buffer.handle = device.GetLogical().CreateBuffer(ci);
buffer.commit = memory_manager.Commit(buffer.handle, false);
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
}
CachedBufferBlock::~CachedBufferBlock() = default;
VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
const VKDevice& device, VKMemoryManager& memory_manager,
VKScheduler& scheduler, VKStagingBufferPool& staging_pool)
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
: VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer, system,
CreateStreamBuffer(device,
scheduler)},
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
device{device}, memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{
staging_pool} {}
VKBufferCache::~VKBufferCache() = default;
Buffer Cache: Use vAddr instead of physical memory.
2020-04-05 23:23:49 +02:00
Buffer VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
return std::make_shared<CachedBufferBlock>(device, memory_manager, cpu_addr, size);
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
}
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
VkBuffer VKBufferCache::ToHandle(const Buffer& buffer) {
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
return buffer->GetHandle();
}
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
VkBuffer VKBufferCache::GetEmptyBuffer(std::size_t size) {
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
size = std::max(size, std::size_t(4));
const auto& empty = staging_pool.GetUnusedBuffer(size, false);
scheduler.RequestOutsideRenderPassOperationContext();
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf) {
cmdbuf.FillBuffer(buffer, 0, size, 0);
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
});
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
return *empty.handle;
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
}
void VKBufferCache::UploadBlockData(const Buffer& buffer, 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);
scheduler.RequestOutsideRenderPassOperationContext();
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
size](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(staging, buffer, VkBufferCopy{0, offset, size});
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
barrier.dstAccessMask = UPLOAD_ACCESS_BARRIERS;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = offset;
barrier.size = size;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {},
barrier, {});
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
});
}
void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
u8* data) {
const auto& staging = staging_pool.GetUnusedBuffer(size, true);
scheduler.RequestOutsideRenderPassOperationContext();
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
size](vk::CommandBuffer cmdbuf) {
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = offset;
barrier.size = 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(buffer, staging, VkBufferCopy{offset, 0, size});
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
});
scheduler.Finish();
std::memcpy(data, staging.commit->Map(size), size);
}
void VKBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
std::size_t dst_offset, std::size_t size) {
scheduler.RequestOutsideRenderPassOperationContext();
buffer_cache: Return handles instead of pointer to handles The original idea of returning pointers is that handles can be moved. The problem is that the implementation didn't take that in mind and made everything harder to work with. This commit drops pointer to handles and returns the handles themselves. While it is still true that handles can be invalidated, this way we get an old handle instead of a dangling pointer. This problem can be solved in the future with sparse buffers.
2020-04-04 07:54:55 +02:00
scheduler.Record([src_buffer = src->GetHandle(), dst_buffer = dst->GetHandle(), src_offset,
renderer_vulkan: Drop Vulkan-Hpp
2020-03-27 05:33:21 +01:00
dst_offset, size](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, size});
std::array<VkBufferMemoryBarrier, 2> barriers;
barriers[0].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barriers[0].pNext = nullptr;
barriers[0].srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
barriers[0].dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barriers[0].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[0].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[0].buffer = src_buffer;
barriers[0].offset = src_offset;
barriers[0].size = size;
barriers[1].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barriers[1].pNext = nullptr;
barriers[1].srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
barriers[1].dstAccessMask = UPLOAD_ACCESS_BARRIERS;
barriers[1].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[1].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[1].buffer = dst_buffer;
barriers[1].offset = dst_offset;
barriers[1].size = size;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {},
barriers, {});
vk_stream_buffer/vk_buffer_cache: Avoid halting and use generic cache The stream buffer before this commit once it was full (no more bytes to write before looping) waiting for all previous operations to finish. This was a temporary solution and had a noticeable performance penalty in performance (from what a profiler showed). To avoid this mark with fences usages of the stream buffer and once it loops wait for them to be signaled. On average this will never wait. Each fence knows where its usage finishes, resulting in a non-paged stream buffer. On the other side, the buffer cache is reimplemented using the generic buffer cache. It makes use of the staging buffer pool and the new stream buffer.
2020-01-06 21:59:20 +01:00
});
}
} // namespace Vulkan
Copy permalink