mirror of
https://github.com/hrydgard/ppsspp.git
synced 2026-10-21 18:29:05 +02:00
Henrik RydgårdandClaude Opus 5.5
821fd301e6
Saving and loading stalled the emu thread for 30-75 ms, long enough to glitch audio. Now only the snapshot and the apply run there; compression and file I/O run on a savestate worker thread, in order. - Save: snapshot in Process(), then compress, write and (for a slot) move the old state to its undo name on the worker. The callback runs once the file is written. Saving again to a path with a save pending is a repeat key press, and gets that save's result without a message. - Load: read and decompress on the worker while the game runs, then apply. A slot load keeps the state it replaces for load undo at that point, rather than saving it in full before starting. - Rewind: compression and decoding are worker jobs, and the deltas get zstd level 1 on top, which makes them about a third of the size. - At most two save snapshots are in memory at once, and buffers are freed when their operation is done. - Shutdown waits for pending writes, and drops callbacks as before. On desktop, a save now costs the emu thread 2-3 ms, a load or rewind 12-15 ms. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
111 lines
3.9 KiB
C++
111 lines
3.9 KiB
C++
#pragma once
|
|
|
|
#include <atomic>
|
|
#include "Common/Serialize/Serializer.h"
|
|
#include "Common/CommonTypes.h"
|
|
#include "Common/TimeUtil.h"
|
|
|
|
namespace SaveState {
|
|
|
|
// This ring buffer of states is for rewind save states, which are kept in RAM.
|
|
// Save states are compressed against one of two reference saves (bases_), and the reference
|
|
// is switched to a fresh save every N saves, where N is BASE_USAGE_INTERVAL.
|
|
// The compression is a simple block based scheme where 0 means to copy a block from the base,
|
|
// and 1 means that the following bytes are the next block, then zstd on top. See Compress/ApplyDelta.
|
|
//
|
|
// The emu thread owns the ring itself (which slots hold what). Compressing and decoding run as jobs
|
|
// on the savestate worker, and the emu thread leaves the buffers a job uses alone until it's done.
|
|
class StateRingbuffer {
|
|
public:
|
|
StateRingbuffer() {
|
|
size_ = REWIND_NUM_STATES;
|
|
states_.resize(size_);
|
|
baseMapping_.resize(size_);
|
|
}
|
|
|
|
CChunkFileReader::Error Save();
|
|
// A rewind in two halves: StartRestore picks the state and has the worker decode it (false if
|
|
// there's nothing to rewind to), and once RestoreDecoded, FinishRestore applies it.
|
|
bool StartRestore();
|
|
bool RestoreDecoded() const { return restoreDecoded_; }
|
|
CChunkFileReader::Error FinishRestore(std::string *errorString, std::string *metadata);
|
|
// Both halves, waiting for the worker in between.
|
|
CChunkFileReader::Error Restore(std::string *errorString, std::string *metadata);
|
|
void Clear();
|
|
|
|
bool Empty() const {
|
|
return count_ == 0;
|
|
}
|
|
|
|
void Process();
|
|
void NotifyState();
|
|
|
|
double NextStateTimestamp() const;
|
|
|
|
private:
|
|
bool BaseAlive(int generation) const;
|
|
void DropNewest();
|
|
void Compress(std::vector<u8> &result, const std::vector<u8> &state, const std::vector<u8> &base);
|
|
bool Decompress(std::vector<u8> &result, const std::vector<u8> &compressed, const std::vector<u8> &base);
|
|
void ApplyDelta(std::vector<u8> &result, const std::vector<u8> &delta, const std::vector<u8> &base);
|
|
|
|
const int BLOCK_SIZE = 8192;
|
|
const int REWIND_NUM_STATES = 20;
|
|
// TODO: Instead, based on size of compressed state?
|
|
const int BASE_USAGE_INTERVAL = 15;
|
|
// A rewind this soon after the last one steps back past the state that one restored.
|
|
const double STEP_BACK_SECONDS = 0.8;
|
|
// Fast. Higher levels shrink the deltas a little more for a lot more time.
|
|
const int ZSTD_COMPRESSION_LEVEL = 1;
|
|
|
|
typedef std::vector<u8> StateBuffer;
|
|
|
|
struct RewindState {
|
|
StateBuffer stateBuffer;
|
|
// CoreTiming's clock when the state was taken, in seconds.
|
|
double emuTime = 0.0;
|
|
|
|
bool empty() const { return stateBuffer.empty(); }
|
|
void clear() { stateBuffer.clear(); emuTime = 0.0; }
|
|
};
|
|
|
|
// The slot the next save goes in, and how many slots before it hold states we can restore.
|
|
int next_ = 0;
|
|
int count_ = 0;
|
|
int size_;
|
|
|
|
std::vector<RewindState> states_;
|
|
StateBuffer bases_[2];
|
|
// Which generation each base slot currently holds, and which generation each state was
|
|
// compressed against. There are more states than bases, so states do go stale.
|
|
int baseGeneration_[2] = {-1, -1};
|
|
int nextBaseGeneration_ = 0;
|
|
std::vector<int> baseMapping_;
|
|
std::vector<u8> buffer_;
|
|
|
|
int base_ = -1;
|
|
int baseUsage_ = 0;
|
|
|
|
double rewindLastTime_ = 0.0f;
|
|
|
|
// A snapshot being compressed on the worker, from buffer_ or a base.
|
|
std::atomic<bool> compressPending_{ false };
|
|
|
|
// The rewind under way: its slot, when it started, and the state the worker decodes it into.
|
|
bool restoreActive_ = false;
|
|
int restoreSlot_ = -1;
|
|
double restoreTime_ = 0.0;
|
|
std::atomic<bool> restoreDecoded_{ false };
|
|
CChunkFileReader::Error restoreError_ = CChunkFileReader::ERROR_NONE;
|
|
std::vector<u8> restoreBuffer_;
|
|
|
|
// Whether the newest state is the one the last rewind restored, and when that was.
|
|
bool newestRestored_ = false;
|
|
double lastRestoreTime_ = 0.0;
|
|
// CoreTiming's clock when the current run of rewinds started, for "seconds ago".
|
|
double rewindFromEmuTime_ = 0.0;
|
|
bool retryingAfterFailure_ = false;
|
|
};
|
|
|
|
} // namespace SaveState
|