Files
eden/src/common/sparse_large_vector.h
T
Exverge 5f142c7926 [common/core/dynarmic] Optimize page table allocations (#4219)
Reduces the page entries from 32 bytes to 8 and rewrites `VirtualBuffer` to be more efficient in memory usage and specifically for large zero regions.

The page table will now only reserve 1GiB instead of 4GiB and of this memory it should only use at most ~8MiB.

This PR has the side effect of using Eden on Windows on low memory systems much more plausible since it would previously require ~10GiB of committable memory at front (despite using ~5-6 at most, inadvertently stalling other processes) where as now it should only require around the amount that it'll actually use.

Co-authored-by: Lizzie <lizzie@eden-emu.dev>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4219
Reviewed-by: Maufeat <sahyno1996@gmail.com>
Reviewed-by: lizzie <lizzie@eden-emu.dev>
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
2026-09-09 21:35:49 +02:00

195 lines
6.2 KiB
C++

// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
/* virtual_buffer.h */
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <atomic>
#include <bit>
#include <utility>
#include <vector>
#ifndef _WIN32
#include <unistd.h>
#include <sys/mman.h>
#endif
#include "common/alignment.h"
#include "common/assert.h"
namespace Common {
#ifdef _WIN32
constexpr u64 HostPageSize = 0x1000;
constexpr u64 HostPageBits = 12;
constexpr u64 HostPageMask = ~(HostPageSize - 1);
bool CommitVectorPage(uintptr_t addr, bool write) noexcept;
#else
const u64 HostPageSize = sysconf(_SC_PAGESIZE);
const u64 HostPageBits = std::countr_zero(HostPageSize);
const u64 HostPageMask = ~(HostPageSize - 1);
#endif
void* AllocateMemoryPages(std::size_t size) noexcept;
void FreeMemoryPages(void* base, std::size_t size) noexcept;
/// A large page-aligned buffer that has optimized memory usage for zero-writes.
template <typename T>
requires std::is_trivially_copyable_v<T>
class SparseLargeVector final {
public:
constexpr SparseLargeVector() = default;
explicit SparseLargeVector(std::size_t count) noexcept
: alloc_size{count * sizeof(T)}
{
base_ptr = static_cast<T*>(AllocateMemoryPages(alloc_size));
// each item in vector holds information for 64 pages
auto denom = HostPageSize * 64;
committed_pages = std::vector<std::atomic<u64>>((alloc_size + denom - 1) / denom);
}
~SparseLargeVector() noexcept {
FreeMemoryPages(base_ptr, alloc_size);
}
SparseLargeVector(const SparseLargeVector&) = delete;
SparseLargeVector& operator=(const SparseLargeVector&) = delete;
SparseLargeVector(SparseLargeVector&& other) = delete;
SparseLargeVector& operator=(SparseLargeVector&& other) = delete;
void ResizeAndClear(std::size_t count) noexcept {
if (auto const new_size = count * sizeof(T); new_size != alloc_size) {
FreeMemoryPages(base_ptr, alloc_size);
alloc_size = new_size;
base_ptr = static_cast<T*>(AllocateMemoryPages(alloc_size));
auto denom = HostPageSize * 64;
committed_pages = std::vector<std::atomic<u64>>((alloc_size + denom - 1) / denom);
}
}
/// Returns a reference to the value of the requested index and allocates memory if needed.
T& GetAndFault(std::size_t index) noexcept {
if (index > alloc_size / sizeof(T)) {
UNREACHABLE_MSG("Out of bounds RW access on SparseLargeVector @ {}", index);
}
if (!IsCommittedPage(index)) {
CommitPage(index);
}
return base_ptr[index];
}
/// Returns a reference to the value of the requested index if initialized, or will otherwise return a zero-initialized object.
const T& GetOrDefault(std::size_t index) const {
#ifdef _WIN32
if (!IsCommittedPage(index)) {
return *reinterpret_cast<const T*>(&default_val);
}
#endif
// On non-Windows, OS page table should optimize this by pointing to a zero page if unallocated.
return base_ptr[index];
}
void Set(std::size_t index, const T& value) noexcept {
if (index > alloc_size / sizeof(T)) {
LOG_CRITICAL(Common_Memory, "Out of bounds write on SparseLargeVector @ {}", index);
return;
}
if (!IsCommittedPage(index))
CommitPage(index);
base_ptr[index] = value;
}
void ZeroRegion(std::size_t start, std::size_t end_) noexcept {
u64 base = reinterpret_cast<u64>(&base_ptr[start]);
const u64 end = reinterpret_cast<u64>(&base_ptr[end_]);
const u64 end_page = AlignUp(base, HostPageSize);
const u64 first_size = (std::min)(end_page, end) - base;
if (IsCommittedPage(start / sizeof(T))) {
std::memset(reinterpret_cast<void*>(base), 0, first_size);
}
if (end <= end_page)
return;
base = end_page;
for (u64 page = base; page < end; page += HostPageSize) {
if (!IsCommittedPage((page - reinterpret_cast<u64>(base_ptr)) / sizeof(T))) {
continue;
}
std::memset(reinterpret_cast<void*>(page), 0, (std::min)( HostPageSize, end - page));
}
}
constexpr void CommitRegion(size_t index, size_t end_) {
const u64 base = static_cast<u64>(index) * sizeof(T);
const u64 end = static_cast<u64>(end_) * sizeof(T);
for (u64 page = AlignDown(base, HostPageSize); page < end; page += HostPageSize) {
if (!IsCommittedPage(page / sizeof(T))) {
CommitPage(page / sizeof(T));
}
}
}
constexpr T& GetUnchecked(size_t index) {
return base_ptr[index];
}
[[nodiscard]] constexpr const T& operator[](std::size_t index) const noexcept {
return GetOrDefault(index);
}
[[nodiscard]] constexpr const T* data() const noexcept {
return base_ptr;
}
[[nodiscard]] constexpr std::size_t size() const noexcept {
return alloc_size / sizeof(T);
}
private:
[[nodiscard]] constexpr bool IsCommittedPage(std::size_t index) const noexcept {
if (index > alloc_size / sizeof(T)) {
LOG_CRITICAL(Common_Memory, "Out of bounds access on large vector @ {}", index);
return false;
}
auto page = (index * sizeof(T)) >> HostPageBits;
auto val = committed_pages[page >> 6].load(std::memory_order_acquire);
return (val >> (page & 63)) & 1;
}
constexpr void CommitPage(std::size_t index) noexcept {
auto page_index = (index * sizeof(T)) >> HostPageBits;
auto page = reinterpret_cast<uintptr_t>(base_ptr + index) & HostPageMask;
#if defined(_WIN32)
CommitVectorPage(page, true);
#else
mprotect(reinterpret_cast<void*>(page), HostPageSize, PROT_READ | PROT_WRITE);
#endif
committed_pages[page_index >> 6].fetch_or(1ULL << (page_index & 63), std::memory_order_release);
}
std::size_t alloc_size{};
T* base_ptr{};
std::vector<std::atomic<u64>> committed_pages{};
#ifdef _WIN32
const std::array<u8, sizeof(T)> default_val{};
#endif
};
} // namespace Common