Split FindOrCreate functionality across BufferManager

`FindOrCreate` ended up being monolithic function with poor readability, this commit addresses those concerns by refactoring the function to split it up into multiple member functions of `BufferManager`, while some of these member functions may only have a single call-site they are important to logically categorize tasks into individual functions. The end result is far neater logic which is far more readable and slightly better optimized by virtue of being abstracted better.
This commit is contained in:
PixelyIon 2022-07-19 22:42:43 +05:30
parent d2a34b5f7a
commit 3e9d84b0c3
No known key found for this signature in database
GPG Key ID: 11BC6C3201BC2C05
2 changed files with 114 additions and 48 deletions

View File

@ -24,49 +24,43 @@ namespace skyline::gpu {
return mutex.try_lock(); return mutex.try_lock();
} }
BufferView BufferManager::FindOrCreate(GuestBuffer guestMapping, ContextTag tag, const std::function<void(std::shared_ptr<Buffer>, ContextLock<Buffer> &&)> &attachBuffer) { BufferManager::LockedBuffer::LockedBuffer(std::shared_ptr<Buffer> pBuffer, ContextTag tag) : buffer{std::move(pBuffer)}, lock{tag, *buffer} {}
/*
* We align the buffer to the page boundary to ensure that:
* 1) Any buffer view has the same alignment guarantees as on the guest, this is required for UBOs, SSBOs and Texel buffers
* 2) We can coalesce a lot of tiny buffers into a single large buffer covering an entire page, this is often the case for index buffers and vertex buffers
*/
auto alignedStart{util::AlignDown(guestMapping.begin().base(), PAGE_SIZE)}, alignedEnd{util::AlignUp(guestMapping.end().base(), PAGE_SIZE)};
vk::DeviceSize offset{static_cast<size_t>(guestMapping.begin().base() - alignedStart)}, size{guestMapping.size()};
guestMapping = span<u8>{alignedStart, alignedEnd};
struct LockedBuffer { Buffer *BufferManager::LockedBuffer::operator->() const {
std::shared_ptr<Buffer> buffer;
ContextLock<Buffer> lock;
LockedBuffer(std::shared_ptr<Buffer> pBuffer, ContextTag tag) : buffer{std::move(pBuffer)}, lock{tag, *buffer} {}
Buffer *operator->() const {
return buffer.get(); return buffer.get();
} }
std::shared_ptr<Buffer> &operator*() { std::shared_ptr<Buffer> &BufferManager::LockedBuffer::operator*() {
return buffer; return buffer;
} }
};
// Lookup for any buffers overlapping with the supplied guest mapping BufferManager::LockedBuffers BufferManager::Lookup(span<u8> range, ContextTag tag) {
boost::container::small_vector<LockedBuffer, 4> overlaps; LockedBuffers overlaps;
for (auto entryIt{std::lower_bound(buffers.begin(), buffers.end(), guestMapping.end().base(), BufferLessThan)}; entryIt != buffers.begin() && (*--entryIt)->guest->begin() <= guestMapping.end();) auto entryIt{std::lower_bound(buffers.begin(), buffers.end(), range.end().base(), BufferLessThan)};
if ((*entryIt)->guest->end() > guestMapping.begin()) while (entryIt != buffers.begin() && (*--entryIt)->guest->begin() <= range.end())
if ((*entryIt)->guest->end() > range.begin())
overlaps.emplace_back(*entryIt, tag); overlaps.emplace_back(*entryIt, tag);
if (overlaps.size() == 1) [[likely]] { return overlaps;
auto &buffer{overlaps.front()};
if (buffer->guest->begin() <= guestMapping.begin() && buffer->guest->end() >= guestMapping.end()) {
// If we find a buffer which can entirely fit the guest mapping, we can just return a view into it
return buffer->GetView(static_cast<vk::DeviceSize>(guestMapping.begin() - buffer->guest->begin()) + offset, size);
}
} }
void BufferManager::InsertBuffer(std::shared_ptr<Buffer> buffer) {
auto bufferEnd{buffer->guest->end().base()};
buffers.insert(std::lower_bound(buffers.begin(), buffers.end(), bufferEnd, BufferLessThan), std::move(buffer));
}
void BufferManager::DeleteBuffer(const std::shared_ptr<Buffer> &buffer) {
buffers.erase(std::find(buffers.begin(), buffers.end(), buffer));
}
BufferManager::LockedBuffer BufferManager::CoalesceBuffers(span<u8> range, const LockedBuffers &srcBuffers, ContextTag tag) {
if (!range.valid())
range = span<u8>{srcBuffers.front().buffer->guest->begin(), srcBuffers.back().buffer->guest->end()};
auto lowestAddress{range.begin().base()}, highestAddress{range.end().base()};
for (const auto &srcBuffer : srcBuffers) {
// Find the extents of the new buffer we want to create that can hold all overlapping buffers // Find the extents of the new buffer we want to create that can hold all overlapping buffers
auto lowestAddress{guestMapping.begin().base()}, highestAddress{guestMapping.end().base()}; auto mapping{*srcBuffer->guest};
for (const auto &overlap : overlaps) {
auto mapping{*overlap->guest};
if (mapping.begin().base() < lowestAddress) if (mapping.begin().base() < lowestAddress)
lowestAddress = mapping.begin().base(); lowestAddress = mapping.begin().base();
if (mapping.end().base() > highestAddress) if (mapping.end().base() > highestAddress)
@ -89,12 +83,7 @@ namespace skyline::gpu {
} }
}}; //!< Copies between two buffers based off of their mappings in guest memory }}; //!< Copies between two buffers based off of their mappings in guest memory
bool shouldAttach{}; //!< If the new buffer should be attached to the current context for (auto &srcBuffer : srcBuffers) {
for (auto &srcBuffer : overlaps) {
if (!srcBuffer.lock.IsFirstUsage())
// If any overlapping buffer was already attached to the current context, we should also attach the current context
shouldAttach = true;
// All newly created buffers that have this set are guaranteed to be attached in buffer FindOrCreate, attach will then lock the buffer without resetting this flag, which will only finally be reset when the lock is released // All newly created buffers that have this set are guaranteed to be attached in buffer FindOrCreate, attach will then lock the buffer without resetting this flag, which will only finally be reset when the lock is released
newBuffer->usedByContext |= srcBuffer->usedByContext; newBuffer->usedByContext |= srcBuffer->usedByContext;
newBuffer->everHadInlineUpdate |= srcBuffer->everHadInlineUpdate; newBuffer->everHadInlineUpdate |= srcBuffer->everHadInlineUpdate;
@ -149,15 +138,54 @@ namespace skyline::gpu {
newBuffer->delegates.splice(newBuffer->delegates.end(), srcBuffer->delegates); newBuffer->delegates.splice(newBuffer->delegates.end(), srcBuffer->delegates);
srcBuffer->Invalidate(); // Invalidate the overlapping buffer so it can't be synced in the future srcBuffer->Invalidate(); // Invalidate the overlapping buffer so it can't be synced in the future
buffers.erase(std::find(buffers.begin(), buffers.end(), srcBuffer.buffer));
} }
if (shouldAttach) return newBuffer;
attachBuffer(*newBuffer, std::move(newBuffer.lock)); }
buffers.insert(std::lower_bound(buffers.begin(), buffers.end(), newBuffer->guest->end().base(), BufferLessThan), *newBuffer); BufferView BufferManager::FindOrCreate(GuestBuffer guestMapping, ContextTag tag, const std::function<void(std::shared_ptr<Buffer>, ContextLock<Buffer> &&)> &attachBuffer) {
/*
* We align the buffer to the page boundary to ensure that:
* 1) Any buffer view has the same alignment guarantees as on the guest, this is required for UBOs, SSBOs and Texel buffers
* 2) We can coalesce a lot of tiny buffers into a single large buffer covering an entire page, this is often the case for index buffers and vertex buffers
*/
auto alignedStart{util::AlignDown(guestMapping.begin().base(), PAGE_SIZE)}, alignedEnd{util::AlignUp(guestMapping.end().base(), PAGE_SIZE)};
vk::DeviceSize offset{static_cast<size_t>(guestMapping.begin().base() - alignedStart)}, size{guestMapping.size()};
guestMapping = span<u8>{alignedStart, alignedEnd};
return newBuffer->GetView(static_cast<vk::DeviceSize>(guestMapping.begin() - newBuffer->guest->begin()) + offset, size); auto overlaps{Lookup(guestMapping, tag)};
if (overlaps.size() == 1) [[likely]] {
// If we find a buffer which can entirely fit the guest mapping, we can just return a view into it
auto &firstOverlap{overlaps.front()};
if (firstOverlap->guest->begin() <= guestMapping.begin() && firstOverlap->guest->end() >= guestMapping.end())
return firstOverlap->GetView(static_cast<vk::DeviceSize>(guestMapping.begin() - firstOverlap->guest->begin()) + offset, size);
}
if (overlaps.empty()) {
// If we couldn't find any overlapping buffers, create a new buffer without coalescing
LockedBuffer buffer{std::make_shared<Buffer>(gpu, guestMapping), tag};
buffer->SynchronizeHost();
InsertBuffer(*buffer);
return buffer->GetView(offset, size);
} else {
// If the new buffer overlaps other buffers, we need to create a new buffer and coalesce all overlapping buffers into one
auto buffer{CoalesceBuffers(guestMapping, overlaps, tag)};
// If any overlapping buffer was already attached to the current context, we should also attach the new buffer
for (auto &srcBuffer : overlaps) {
if (!srcBuffer.lock.IsFirstUsage()) {
attachBuffer(*buffer, std::move(buffer.lock));
break;
}
}
// Delete older overlapping buffers and insert the new buffer into the map
for (auto &overlap : overlaps)
DeleteBuffer(*overlap);
InsertBuffer(*buffer);
return buffer->GetView(static_cast<vk::DeviceSize>(guestMapping.begin() - buffer->guest->begin()) + offset, size);
}
} }
constexpr static vk::DeviceSize MegaBufferSize{100 * 1024 * 1024}; //!< Size in bytes of the megabuffer (100MiB) constexpr static vk::DeviceSize MegaBufferSize{100 * 1024 * 1024}; //!< Size in bytes of the megabuffer (100MiB)

View File

@ -33,6 +33,44 @@ namespace skyline::gpu {
MegaBufferSlot(GPU &gpu); MegaBufferSlot(GPU &gpu);
}; };
/**
* @brief A wrapper around a Buffer which locks it with the specified ContextTag
*/
struct LockedBuffer {
std::shared_ptr<Buffer> buffer;
ContextLock<Buffer> lock;
LockedBuffer(std::shared_ptr<Buffer> pBuffer, ContextTag tag);
Buffer *operator->() const;
std::shared_ptr<Buffer> &operator*();
};
using LockedBuffers = boost::container::small_vector<LockedBuffer, 4>;
/**
* @return A vector of buffers locked with the supplied tag which are contained within the supplied range
*/
LockedBuffers Lookup(span<u8> range, ContextTag tag);
/**
* @brief Inserts the supplied buffer into the map based on its guest address
*/
void InsertBuffer(std::shared_ptr<Buffer> buffer);
/**
* @brief Deletes the supplied buffer from the map, the lifetime of the buffer will no longer be extended by the map
*/
void DeleteBuffer(const std::shared_ptr<Buffer> &buffer);
/**
* @brief Coalesce the supplied buffers into a single buffer encompassing the specified range and locks it with the supplied tag
* @param range The range of memory that the newly created buffer will cover, this will be extended to cover the entirety of the supplied buffers automatically and can be null
* @note The supplied buffers **must** be in the map and locked with the supplied tag
*/
LockedBuffer CoalesceBuffers(span<u8> range, const LockedBuffers &srcBuffers, ContextTag tag);
/** /**
* @return If the end of the supplied buffer is less than the supplied pointer * @return If the end of the supplied buffer is less than the supplied pointer
*/ */