mirror of https://git.suyu.dev/suyu/suyu
query_cache: Abstract OpenGL implementation
Abstract the current OpenGL implementation into the VideoCommon namespace and reimplement it on top of that. Doing this avoids repeating code and logic in the Vulkan implementation.merge-requests/60/head
ReinUsesLisp
parent
73d2d3342d
commit
c31382ced5
@ -0,0 +1,323 @@
|
||||
// Copyright 2020 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <algorithm>
|
||||
#include <array>
|
||||
#include <cstring>
|
||||
#include <iterator>
|
||||
#include <memory>
|
||||
#include <optional>
|
||||
#include <unordered_map>
|
||||
#include <vector>
|
||||
|
||||
#include "common/assert.h"
|
||||
#include "core/core.h"
|
||||
#include "video_core/engines/maxwell_3d.h"
|
||||
#include "video_core/gpu.h"
|
||||
#include "video_core/memory_manager.h"
|
||||
#include "video_core/rasterizer_interface.h"
|
||||
|
||||
namespace VideoCommon {
|
||||
|
||||
template <class QueryCache, class HostCounter>
|
||||
class CounterStreamBase {
|
||||
public:
|
||||
explicit CounterStreamBase(QueryCache& cache, VideoCore::QueryType type)
|
||||
: cache{cache}, type{type} {}
|
||||
|
||||
/// Updates the state of the stream, enabling or disabling as needed.
|
||||
void Update(bool enabled) {
|
||||
if (enabled) {
|
||||
Enable();
|
||||
} else {
|
||||
Disable();
|
||||
}
|
||||
}
|
||||
|
||||
/// Resets the stream to zero. It doesn't disable the query after resetting.
|
||||
void Reset() {
|
||||
if (current) {
|
||||
current->EndQuery();
|
||||
|
||||
// Immediately start a new query to avoid disabling its state.
|
||||
current = cache.Counter(nullptr, type);
|
||||
}
|
||||
last = nullptr;
|
||||
}
|
||||
|
||||
/// Returns the current counter slicing as needed.
|
||||
std::shared_ptr<HostCounter> Current() {
|
||||
if (!current) {
|
||||
return nullptr;
|
||||
}
|
||||
current->EndQuery();
|
||||
last = std::move(current);
|
||||
current = cache.Counter(last, type);
|
||||
return last;
|
||||
}
|
||||
|
||||
/// Returns true when the counter stream is enabled.
|
||||
bool IsEnabled() const {
|
||||
return static_cast<bool>(current);
|
||||
}
|
||||
|
||||
private:
|
||||
/// Enables the stream.
|
||||
void Enable() {
|
||||
if (current) {
|
||||
return;
|
||||
}
|
||||
current = cache.Counter(last, type);
|
||||
}
|
||||
|
||||
// Disables the stream.
|
||||
void Disable() {
|
||||
if (current) {
|
||||
current->EndQuery();
|
||||
}
|
||||
last = std::exchange(current, nullptr);
|
||||
}
|
||||
|
||||
QueryCache& cache;
|
||||
const VideoCore::QueryType type;
|
||||
|
||||
std::shared_ptr<HostCounter> current;
|
||||
std::shared_ptr<HostCounter> last;
|
||||
};
|
||||
|
||||
template <class QueryCache, class CachedQuery, class CounterStream, class HostCounter>
|
||||
class QueryCacheBase {
|
||||
public:
|
||||
explicit QueryCacheBase(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
|
||||
: system{system}, rasterizer{rasterizer}, streams{{CounterStream{
|
||||
static_cast<QueryCache&>(*this),
|
||||
VideoCore::QueryType::SamplesPassed}}} {}
|
||||
|
||||
void InvalidateRegion(CacheAddr addr, std::size_t size) {
|
||||
FlushAndRemoveRegion(addr, size);
|
||||
}
|
||||
|
||||
void FlushRegion(CacheAddr addr, std::size_t size) {
|
||||
FlushAndRemoveRegion(addr, size);
|
||||
}
|
||||
|
||||
/**
|
||||
* Records a query in GPU mapped memory, potentially marked with a timestamp.
|
||||
* @param gpu_addr GPU address to flush to when the mapped memory is read.
|
||||
* @param type Query type, e.g. SamplesPassed.
|
||||
* @param timestamp Timestamp, when empty the flushed query is assumed to be short.
|
||||
*/
|
||||
void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) {
|
||||
auto& memory_manager = system.GPU().MemoryManager();
|
||||
const auto host_ptr = memory_manager.GetPointer(gpu_addr);
|
||||
|
||||
CachedQuery* query = TryGet(ToCacheAddr(host_ptr));
|
||||
if (!query) {
|
||||
const auto cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
|
||||
ASSERT_OR_EXECUTE(cpu_addr, return;);
|
||||
|
||||
query = Register(type, *cpu_addr, host_ptr, timestamp.has_value());
|
||||
}
|
||||
|
||||
query->BindCounter(Stream(type).Current(), timestamp);
|
||||
}
|
||||
|
||||
/// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch.
|
||||
void UpdateCounters() {
|
||||
const auto& regs = system.GPU().Maxwell3D().regs;
|
||||
Stream(VideoCore::QueryType::SamplesPassed).Update(regs.samplecnt_enable);
|
||||
}
|
||||
|
||||
/// Resets a counter to zero. It doesn't disable the query after resetting.
|
||||
void ResetCounter(VideoCore::QueryType type) {
|
||||
Stream(type).Reset();
|
||||
}
|
||||
|
||||
/// Returns a new host counter.
|
||||
std::shared_ptr<HostCounter> Counter(std::shared_ptr<HostCounter> dependency,
|
||||
VideoCore::QueryType type) {
|
||||
return std::make_shared<HostCounter>(static_cast<QueryCache&>(*this), std::move(dependency),
|
||||
type);
|
||||
}
|
||||
|
||||
/// Returns the counter stream of the specified type.
|
||||
CounterStream& Stream(VideoCore::QueryType type) {
|
||||
return streams[static_cast<std::size_t>(type)];
|
||||
}
|
||||
|
||||
private:
|
||||
/// Flushes a memory range to guest memory and removes it from the cache.
|
||||
void FlushAndRemoveRegion(CacheAddr addr, std::size_t size) {
|
||||
const u64 addr_begin = static_cast<u64>(addr);
|
||||
const u64 addr_end = addr_begin + static_cast<u64>(size);
|
||||
const auto in_range = [addr_begin, addr_end](CachedQuery& query) {
|
||||
const u64 cache_begin = query.CacheAddr();
|
||||
const u64 cache_end = cache_begin + query.SizeInBytes();
|
||||
return cache_begin < addr_end && addr_begin < cache_end;
|
||||
};
|
||||
|
||||
const u64 page_end = addr_end >> PAGE_SHIFT;
|
||||
for (u64 page = addr_begin >> PAGE_SHIFT; page <= page_end; ++page) {
|
||||
const auto& it = cached_queries.find(page);
|
||||
if (it == std::end(cached_queries)) {
|
||||
continue;
|
||||
}
|
||||
auto& contents = it->second;
|
||||
for (auto& query : contents) {
|
||||
if (!in_range(query)) {
|
||||
continue;
|
||||
}
|
||||
rasterizer.UpdatePagesCachedCount(query.CpuAddr(), query.SizeInBytes(), -1);
|
||||
query.Flush();
|
||||
}
|
||||
contents.erase(std::remove_if(std::begin(contents), std::end(contents), in_range),
|
||||
std::end(contents));
|
||||
}
|
||||
}
|
||||
|
||||
/// Registers the passed parameters as cached and returns a pointer to the stored cached query.
|
||||
CachedQuery* Register(VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr, bool timestamp) {
|
||||
rasterizer.UpdatePagesCachedCount(cpu_addr, CachedQuery::SizeInBytes(timestamp), 1);
|
||||
const u64 page = static_cast<u64>(ToCacheAddr(host_ptr)) >> PAGE_SHIFT;
|
||||
return &cached_queries[page].emplace_back(static_cast<QueryCache&>(*this), type, cpu_addr,
|
||||
host_ptr);
|
||||
}
|
||||
|
||||
/// Tries to a get a cached query. Returns nullptr on failure.
|
||||
CachedQuery* TryGet(CacheAddr addr) {
|
||||
const u64 page = static_cast<u64>(addr) >> PAGE_SHIFT;
|
||||
const auto it = cached_queries.find(page);
|
||||
if (it == std::end(cached_queries)) {
|
||||
return nullptr;
|
||||
}
|
||||
auto& contents = it->second;
|
||||
const auto found = std::find_if(std::begin(contents), std::end(contents),
|
||||
[addr](auto& query) { return query.CacheAddr() == addr; });
|
||||
return found != std::end(contents) ? &*found : nullptr;
|
||||
}
|
||||
|
||||
static constexpr std::uintptr_t PAGE_SIZE = 4096;
|
||||
static constexpr int PAGE_SHIFT = 12;
|
||||
|
||||
Core::System& system;
|
||||
VideoCore::RasterizerInterface& rasterizer;
|
||||
|
||||
std::unordered_map<u64, std::vector<CachedQuery>> cached_queries;
|
||||
|
||||
std::array<CounterStream, VideoCore::NumQueryTypes> streams;
|
||||
};
|
||||
|
||||
template <class QueryCache, class HostCounter>
|
||||
class HostCounterBase {
|
||||
public:
|
||||
explicit HostCounterBase(std::shared_ptr<HostCounter> dependency)
|
||||
: dependency{std::move(dependency)} {}
|
||||
|
||||
/// Returns the current value of the query.
|
||||
u64 Query() {
|
||||
if (result) {
|
||||
return *result;
|
||||
}
|
||||
|
||||
u64 value = BlockingQuery();
|
||||
if (dependency) {
|
||||
value += dependency->Query();
|
||||
}
|
||||
|
||||
return *(result = value);
|
||||
}
|
||||
|
||||
/// Returns true when flushing this query will potentially wait.
|
||||
bool WaitPending() const noexcept {
|
||||
return result.has_value();
|
||||
}
|
||||
|
||||
protected:
|
||||
/// Returns the value of query from the backend API blocking as needed.
|
||||
virtual u64 BlockingQuery() const = 0;
|
||||
|
||||
private:
|
||||
std::shared_ptr<HostCounter> dependency; ///< Counter to add to this value.
|
||||
std::optional<u64> result; ///< Filled with the already returned value.
|
||||
};
|
||||
|
||||
template <class HostCounter>
|
||||
class CachedQueryBase {
|
||||
public:
|
||||
explicit CachedQueryBase(VAddr cpu_addr, u8* host_ptr)
|
||||
: cpu_addr{cpu_addr}, host_ptr{host_ptr} {}
|
||||
|
||||
CachedQueryBase(CachedQueryBase&& rhs) noexcept
|
||||
: cpu_addr{rhs.cpu_addr}, host_ptr{rhs.host_ptr}, counter{std::move(rhs.counter)},
|
||||
timestamp{rhs.timestamp} {}
|
||||
|
||||
CachedQueryBase(const CachedQueryBase&) = delete;
|
||||
|
||||
CachedQueryBase& operator=(CachedQueryBase&& rhs) noexcept {
|
||||
cpu_addr = rhs.cpu_addr;
|
||||
host_ptr = rhs.host_ptr;
|
||||
counter = std::move(rhs.counter);
|
||||
timestamp = rhs.timestamp;
|
||||
return *this;
|
||||
}
|
||||
|
||||
/// Flushes the query to guest memory.
|
||||
virtual void Flush() {
|
||||
// When counter is nullptr it means that it's just been reseted. We are supposed to write a
|
||||
// zero in these cases.
|
||||
const u64 value = counter ? counter->Query() : 0;
|
||||
std::memcpy(host_ptr, &value, sizeof(u64));
|
||||
|
||||
if (timestamp) {
|
||||
std::memcpy(host_ptr + TIMESTAMP_OFFSET, &*timestamp, sizeof(u64));
|
||||
}
|
||||
}
|
||||
|
||||
/// Binds a counter to this query.
|
||||
void BindCounter(std::shared_ptr<HostCounter> counter_, std::optional<u64> timestamp_) {
|
||||
if (counter) {
|
||||
// If there's an old counter set it means the query is being rewritten by the game.
|
||||
// To avoid losing the data forever, flush here.
|
||||
Flush();
|
||||
}
|
||||
counter = std::move(counter_);
|
||||
timestamp = timestamp_;
|
||||
}
|
||||
|
||||
VAddr CpuAddr() const noexcept {
|
||||
return cpu_addr;
|
||||
}
|
||||
|
||||
CacheAddr CacheAddr() const noexcept {
|
||||
return ToCacheAddr(host_ptr);
|
||||
}
|
||||
|
||||
u64 SizeInBytes() const noexcept {
|
||||
return SizeInBytes(timestamp.has_value());
|
||||
}
|
||||
|
||||
static u64 SizeInBytes(bool with_timestamp) {
|
||||
return with_timestamp ? LARGE_QUERY_SIZE : SMALL_QUERY_SIZE;
|
||||
}
|
||||
|
||||
protected:
|
||||
/// Returns true when querying the counter may potentially block.
|
||||
bool WaitPending() const noexcept {
|
||||
return counter && counter->WaitPending();
|
||||
}
|
||||
|
||||
private:
|
||||
static constexpr std::size_t SMALL_QUERY_SIZE = 8; // Query size without timestamp.
|
||||
static constexpr std::size_t LARGE_QUERY_SIZE = 16; // Query size with timestamp.
|
||||
static constexpr std::intptr_t TIMESTAMP_OFFSET = 8; // Timestamp offset in a large query.
|
||||
|
||||
VAddr cpu_addr; ///< Guest CPU address.
|
||||
u8* host_ptr; ///< Writable host pointer.
|
||||
std::shared_ptr<HostCounter> counter; ///< Host counter to query, owns the dependency tree.
|
||||
std::optional<u64> timestamp; ///< Timestamp to flush to guest memory.
|
||||
};
|
||||
|
||||
} // namespace VideoCommon
|
||||
Loading…
Reference in New Issue