mirror of
https://git.h3cjp.net/H3cJP/yuzu.git
synced 2024-12-17 16:26:15 +00:00
video_core: rasterizer_accelerated: Use a flat array instead of interval_map for cached pages.
- Uses a fixed 64MB for the cache instead of an ever growing map. - Slightly faster by using atomics instead of a single mutex for access. - Thanks for Rodrigo for the idea.
This commit is contained in:
parent
cd25817938
commit
94da1e8a7e
|
@ -2,64 +2,39 @@
|
|||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#include <mutex>
|
||||
|
||||
#include <boost/icl/interval_map.hpp>
|
||||
#include <boost/range/iterator_range.hpp>
|
||||
|
||||
#include "common/assert.h"
|
||||
#include "common/common_types.h"
|
||||
#include "common/div_ceil.h"
|
||||
#include "core/memory.h"
|
||||
#include "video_core/rasterizer_accelerated.h"
|
||||
|
||||
namespace VideoCore {
|
||||
|
||||
namespace {
|
||||
|
||||
template <typename Map, typename Interval>
|
||||
constexpr auto RangeFromInterval(Map& map, const Interval& interval) {
|
||||
return boost::make_iterator_range(map.equal_range(interval));
|
||||
}
|
||||
|
||||
} // Anonymous namespace
|
||||
|
||||
RasterizerAccelerated::RasterizerAccelerated(Core::Memory::Memory& cpu_memory_)
|
||||
: cpu_memory{cpu_memory_} {}
|
||||
|
||||
RasterizerAccelerated::~RasterizerAccelerated() = default;
|
||||
|
||||
void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {
|
||||
std::lock_guard lock{pages_mutex};
|
||||
const u64 page_start{addr >> Core::Memory::PAGE_BITS};
|
||||
const u64 page_end{(addr + size + Core::Memory::PAGE_SIZE - 1) >> Core::Memory::PAGE_BITS};
|
||||
const auto page_end = Common::DivCeil(addr + size, Core::Memory::PAGE_SIZE);
|
||||
for (auto page = addr >> Core::Memory::PAGE_BITS; page != page_end; ++page) {
|
||||
auto& count = cached_pages.at(page >> 3).Count(page);
|
||||
|
||||
// Interval maps will erase segments if count reaches 0, so if delta is negative we have to
|
||||
// subtract after iterating
|
||||
const auto pages_interval = CachedPageMap::interval_type::right_open(page_start, page_end);
|
||||
if (delta > 0) {
|
||||
cached_pages.add({pages_interval, delta});
|
||||
}
|
||||
ASSERT_MSG(count < UINT8_MAX, "Count may exceed UINT8_MAX!");
|
||||
|
||||
for (const auto& pair : RangeFromInterval(cached_pages, pages_interval)) {
|
||||
const auto interval = pair.first & pages_interval;
|
||||
const int count = pair.second;
|
||||
count += delta;
|
||||
|
||||
const VAddr interval_start_addr = boost::icl::first(interval) << Core::Memory::PAGE_BITS;
|
||||
const VAddr interval_end_addr = boost::icl::last_next(interval) << Core::Memory::PAGE_BITS;
|
||||
const u64 interval_size = interval_end_addr - interval_start_addr;
|
||||
|
||||
if (delta > 0 && count == delta) {
|
||||
cpu_memory.RasterizerMarkRegionCached(interval_start_addr, interval_size, true);
|
||||
} else if (delta < 0 && count == -delta) {
|
||||
cpu_memory.RasterizerMarkRegionCached(interval_start_addr, interval_size, false);
|
||||
// Assume delta is either -1 or 1
|
||||
if (count == 0) {
|
||||
cpu_memory.RasterizerMarkRegionCached(page << Core::Memory::PAGE_BITS,
|
||||
Core::Memory::PAGE_SIZE, false);
|
||||
} else if (count == 1 && delta > 0) {
|
||||
cpu_memory.RasterizerMarkRegionCached(page << Core::Memory::PAGE_BITS,
|
||||
Core::Memory::PAGE_SIZE, true);
|
||||
} else {
|
||||
ASSERT(count >= 0);
|
||||
}
|
||||
}
|
||||
|
||||
if (delta < 0) {
|
||||
cached_pages.add({pages_interval, delta});
|
||||
}
|
||||
}
|
||||
|
||||
} // namespace VideoCore
|
||||
|
|
|
@ -4,9 +4,8 @@
|
|||
|
||||
#pragma once
|
||||
|
||||
#include <mutex>
|
||||
|
||||
#include <boost/icl/interval_map.hpp>
|
||||
#include <array>
|
||||
#include <atomic>
|
||||
|
||||
#include "common/common_types.h"
|
||||
#include "video_core/rasterizer_interface.h"
|
||||
|
@ -26,10 +25,24 @@ public:
|
|||
void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) override;
|
||||
|
||||
private:
|
||||
using CachedPageMap = boost::icl::interval_map<u64, int>;
|
||||
CachedPageMap cached_pages;
|
||||
std::mutex pages_mutex;
|
||||
class CacheEntry final {
|
||||
public:
|
||||
CacheEntry() = default;
|
||||
|
||||
std::atomic_uint8_t& Count(std::size_t page) {
|
||||
return values[page & 7];
|
||||
}
|
||||
|
||||
const std::atomic_uint8_t& Count(std::size_t page) const {
|
||||
return values[page & 7];
|
||||
}
|
||||
|
||||
private:
|
||||
std::array<std::atomic_uint8_t, 8> values{};
|
||||
};
|
||||
static_assert(sizeof(CacheEntry) == 8, "CacheEntry should be 8 bytes!");
|
||||
|
||||
std::array<CacheEntry, 0x800000> cached_pages;
|
||||
Core::Memory::Memory& cpu_memory;
|
||||
};
|
||||
|
||||
|
|
Loading…
Reference in a new issue