Panda3DS/include/renderer_gl/surface_cache.hpp
wheremyfoodat d06f600b3a
Some checks are pending
Android Build / x64 (release) (push) Waiting to run
Android Build / arm64 (release) (push) Waiting to run
HTTP Server Build / build (push) Waiting to run
Hydra Core Build / Windows (push) Waiting to run
Hydra Core Build / MacOS (push) Waiting to run
Hydra Core Build / Linux (push) Waiting to run
Hydra Core Build / Android-x64 (push) Waiting to run
Hydra Core Build / ARM-Libretro (push) Waiting to run
Linux AppImage Build / build (push) Waiting to run
Linux Build / build (push) Waiting to run
MacOS Build / MacOS-arm64 (push) Waiting to run
MacOS Build / MacOS-x86_64 (push) Waiting to run
MacOS Build / MacOS-Universal (push) Blocked by required conditions
Qt Build / Windows (push) Waiting to run
Qt Build / MacOS-arm64 (push) Waiting to run
Qt Build / MacOS-x86_64 (push) Waiting to run
Qt Build / MacOS-Universal (push) Blocked by required conditions
Qt Build / Linux (push) Waiting to run
Windows Build / build (push) Waiting to run
iOS Simulator Build / build (push) Waiting to run
Use multimap for indexing surfaces (#771)
* Use an std::map for faster texture cache lookup

* Surface cache: Use map instead of multimap, optimize `find` to perform tree scan

* Add comments
2025-07-04 14:23:08 +03:00

138 lines
No EOL
4.3 KiB
C++

#pragma once
#include <array>
#include <functional>
#include <map>
#include <optional>
#include "surfaces.hpp"
#include "textures.hpp"
// Surface cache class that can fit "capacity" instances of the "SurfaceType" class of surfaces
// SurfaceType *must* have all of the following.
// - An "allocate" function that allocates GL resources for the surfaces. On overflow it will panic
// if evictOnOverflow is false, or kick out the oldest item if it is true (like a ring buffer)
// - A "free" function that frees up all resources the surface is taking up
// - A "matches" function that, when provided with a SurfaceType object reference
// Will tell us if the 2 surfaces match (Only as far as location in VRAM, format, dimensions, etc)
// Are concerned. We could overload the == operator, but that implies full equality
// Including equality of the allocated OpenGL resources, which we don't want
// - A "valid" member that tells us whether the function is still valid or not
// - A "location" member which tells us which location in 3DS memory this surface occupies
template <typename SurfaceType, size_t capacity, bool evictOnOverflow = false>
class SurfaceCache {
// Vanilla std::optional can't hold actual references
using OptionalRef = std::optional<std::reference_wrapper<SurfaceType>>;
size_t size = 0;
size_t evictionIndex = 0;
std::array<SurfaceType, capacity> buffer;
// Map from address to a surface in the above buffer.
// Several cached surfaces may have the same starting address, so we use a multimap.
std::multimap<u32, SurfaceType*> surfaceMap;
// Adds a surface to our map
void indexSurface(SurfaceType& surface) { surfaceMap.emplace(surface.location, &surface); }
// Removes a surface from our map
void unindexSurface(SurfaceType& surface) {
auto range = surfaceMap.equal_range(surface.location);
for (auto it = range.first; it != range.second; ++it) {
if (it->second == &surface) {
surfaceMap.erase(it);
break;
}
}
}
public:
void reset() {
size = 0;
evictionIndex = 0;
surfaceMap.clear();
// Free the memory of all surfaces
for (auto& e : buffer) {
e.free();
e.valid = false;
}
}
// Use our map to only scan the surfaces with the same starting location
OptionalRef find(SurfaceType& other) {
auto range = surfaceMap.equal_range(other.location);
for (auto it = range.first; it != range.second; ++it) {
SurfaceType* candidate = it->second;
if (candidate->valid && candidate->matches(other)) {
return *candidate;
}
}
return std::nullopt;
}
OptionalRef findFromAddress(u32 address) {
for (auto it = surfaceMap.begin(); it != surfaceMap.end(); ++it) {
SurfaceType* surface = it->second;
if (surface->valid && surface->location <= address && surface->location + surface->sizeInBytes() > address) {
return *surface;
}
}
return std::nullopt;
}
// Adds a surface object to the cache and returns it
SurfaceType& add(const SurfaceType& surface) {
if (size >= capacity) {
if constexpr (evictOnOverflow) { // Do a ring buffer if evictOnOverflow is true
if constexpr (std::is_same<SurfaceType, ColourBuffer>() || std::is_same<SurfaceType, DepthBuffer>()) {
Helpers::panicDev("Colour/Depth buffer cache overflowed, currently stubbed to do a ring-buffer. This might snap in half");
}
auto& e = buffer[evictionIndex];
unindexSurface(e);
evictionIndex = (evictionIndex + 1) % capacity;
e.valid = false;
e.free();
e = surface;
e.allocate();
indexSurface(e);
return e;
} else {
Helpers::panic("Surface cache full! Add emptying!");
}
}
size++;
// See if any existing surface fully overlaps
for (auto& e : buffer) {
if (e.valid && e.range.lower() >= surface.range.lower() && e.range.upper() <= surface.range.upper()) {
unindexSurface(e);
e.free();
e = surface;
e.allocate();
indexSurface(e);
return e;
}
}
// Find an invalid entry in the cache and overwrite it with the new surface
for (auto& e : buffer) {
if (!e.valid) {
e = surface;
e.allocate();
indexSurface(e);
return e;
}
}
// This should be unreachable but helps to panic anyways
Helpers::panic("Couldn't add surface to cache\n");
}
SurfaceType& operator[](size_t i) { return buffer[i]; }
const SurfaceType& operator[](size_t i) const { return buffer[i]; }
};