yuzu/src/video_core/fence_manager.h
ReinUsesLisp 82c2601555 video_core: Reimplement the buffer cache
Reimplement the buffer cache using cached bindings and page level
granularity for modification tracking. This also drops the usage of
shared pointers and virtual functions from the cache.

- Bindings are cached, allowing to skip work when the game changes few
  bits between draws.
- OpenGL Assembly shaders no longer copy when a region has been modified
  from the GPU to emulate constant buffers, instead GL_EXT_memory_object
  is used to alias sub-buffers within the same allocation.
- OpenGL Assembly shaders stream constant buffer data using
  glProgramBufferParametersIuivNV, from NV_parameter_buffer_object. In
  theory this should save one hash table resolve inside the driver
  compared to glBufferSubData.
- A new OpenGL stream buffer is implemented based on fences for drivers
  that are not Nvidia's proprietary, due to their low performance on
  partial glBufferSubData calls synchronized with 3D rendering (that
  some games use a lot).
- Most optimizations are shared between APIs now, allowing Vulkan to
  cache more bindings than before, skipping unnecesarry work.

This commit adds the necessary infrastructure to use Vulkan object from
OpenGL. Overall, it improves performance and fixes some bugs present on
the old cache. There are still some edge cases hit by some games that
harm performance on some vendors, this are planned to be fixed in later
commits.
2021-02-13 02:17:22 -03:00

181 lines
5.8 KiB
C++

// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <algorithm>
#include <queue>
#include "common/common_types.h"
#include "core/core.h"
#include "video_core/delayed_destruction_ring.h"
#include "video_core/gpu.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
namespace VideoCommon {
class FenceBase {
public:
explicit FenceBase(u32 payload_, bool is_stubbed_)
: address{}, payload{payload_}, is_semaphore{false}, is_stubbed{is_stubbed_} {}
explicit FenceBase(GPUVAddr address_, u32 payload_, bool is_stubbed_)
: address{address_}, payload{payload_}, is_semaphore{true}, is_stubbed{is_stubbed_} {}
GPUVAddr GetAddress() const {
return address;
}
u32 GetPayload() const {
return payload;
}
bool IsSemaphore() const {
return is_semaphore;
}
private:
GPUVAddr address;
u32 payload;
bool is_semaphore;
protected:
bool is_stubbed;
};
template <typename TFence, typename TTextureCache, typename TTBufferCache, typename TQueryCache>
class FenceManager {
public:
/// Notify the fence manager about a new frame
void TickFrame() {
delayed_destruction_ring.Tick();
}
void SignalSemaphore(GPUVAddr addr, u32 value) {
TryReleasePendingFences();
const bool should_flush = ShouldFlush();
CommitAsyncFlushes();
TFence new_fence = CreateFence(addr, value, !should_flush);
fences.push(new_fence);
QueueFence(new_fence);
if (should_flush) {
rasterizer.FlushCommands();
}
rasterizer.SyncGuestHost();
}
void SignalSyncPoint(u32 value) {
TryReleasePendingFences();
const bool should_flush = ShouldFlush();
CommitAsyncFlushes();
TFence new_fence = CreateFence(value, !should_flush);
fences.push(new_fence);
QueueFence(new_fence);
if (should_flush) {
rasterizer.FlushCommands();
}
rasterizer.SyncGuestHost();
}
void WaitPendingFences() {
while (!fences.empty()) {
TFence& current_fence = fences.front();
if (ShouldWait()) {
WaitFence(current_fence);
}
PopAsyncFlushes();
if (current_fence->IsSemaphore()) {
gpu_memory.template Write<u32>(current_fence->GetAddress(),
current_fence->GetPayload());
} else {
gpu.IncrementSyncPoint(current_fence->GetPayload());
}
PopFence();
}
}
protected:
explicit FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
TTextureCache& texture_cache_, TTBufferCache& buffer_cache_,
TQueryCache& query_cache_)
: rasterizer{rasterizer_}, gpu{gpu_}, gpu_memory{gpu.MemoryManager()},
texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, query_cache{query_cache_} {}
virtual ~FenceManager() = default;
/// Creates a Sync Point Fence Interface, does not create a backend fence if 'is_stubbed' is
/// true
virtual TFence CreateFence(u32 value, bool is_stubbed) = 0;
/// Creates a Semaphore Fence Interface, does not create a backend fence if 'is_stubbed' is true
virtual TFence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) = 0;
/// Queues a fence into the backend if the fence isn't stubbed.
virtual void QueueFence(TFence& fence) = 0;
/// Notifies that the backend fence has been signaled/reached in host GPU.
virtual bool IsFenceSignaled(TFence& fence) const = 0;
/// Waits until a fence has been signalled by the host GPU.
virtual void WaitFence(TFence& fence) = 0;
VideoCore::RasterizerInterface& rasterizer;
Tegra::GPU& gpu;
Tegra::MemoryManager& gpu_memory;
TTextureCache& texture_cache;
TTBufferCache& buffer_cache;
TQueryCache& query_cache;
private:
void TryReleasePendingFences() {
while (!fences.empty()) {
TFence& current_fence = fences.front();
if (ShouldWait() && !IsFenceSignaled(current_fence)) {
return;
}
PopAsyncFlushes();
if (current_fence->IsSemaphore()) {
gpu_memory.template Write<u32>(current_fence->GetAddress(),
current_fence->GetPayload());
} else {
gpu.IncrementSyncPoint(current_fence->GetPayload());
}
PopFence();
}
}
bool ShouldWait() const {
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
return texture_cache.ShouldWaitAsyncFlushes() || buffer_cache.ShouldWaitAsyncFlushes() ||
query_cache.ShouldWaitAsyncFlushes();
}
bool ShouldFlush() const {
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
return texture_cache.HasUncommittedFlushes() || buffer_cache.HasUncommittedFlushes() ||
query_cache.HasUncommittedFlushes();
}
void PopAsyncFlushes() {
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
texture_cache.PopAsyncFlushes();
buffer_cache.PopAsyncFlushes();
query_cache.PopAsyncFlushes();
}
void CommitAsyncFlushes() {
std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
texture_cache.CommitAsyncFlushes();
buffer_cache.CommitAsyncFlushes();
query_cache.CommitAsyncFlushes();
}
void PopFence() {
delayed_destruction_ring.Push(std::move(fences.front()));
fences.pop();
}
std::queue<TFence> fences;
DelayedDestructionRing<TFence, 6> delayed_destruction_ring;
};
} // namespace VideoCommon