raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

gpu: address extended guest memory without widening dense cache tables

Browse Source
This commit is contained in:
nmzik
2026-09-28 16:32:32 +04:00
parent c91fe80d64
commit 2016f8e2fe
12 changed files with 81 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/regionDefinitions.h
+8 -3
View File
@@ -3,6 +3,7 @@
#include "common/bitArray.h"
#include "common/common.h"
#include "kernel/memory.h"
#include <compare>
@@ -10,7 +11,8 @@ namespace Libs::Graphics {
constexpr uint64_t TRACKER_PAGE_SIZE = 4ull * 1024ull;
constexpr uint64_t TRACKER_REGION_SIZE = 4ull * 1024ull * 1024ull;
constexpr uint64_t TRACKER_ADDRESS_SIZE = 1ull << 40u;
constexpr uint64_t TRACKER_ADDRESS_SIZE = 1ull << 44u;
constexpr uint64_t LOWER_ADDRESS_SIZE = 1ull << 40u;
constexpr size_t TRACKER_REGION_PAGES = TRACKER_REGION_SIZE / TRACKER_PAGE_SIZE;
struct GuestRange {
@@ -19,8 +21,11 @@ struct GuestRange {
[[nodiscard]] constexpr bool Empty() const noexcept { return address == 0 && size == 0; }
[[nodiscard]] constexpr bool Valid() const noexcept {
return address != 0 && size != 0 && address < TRACKER_ADDRESS_SIZE &&
size <= TRACKER_ADDRESS_SIZE - address;
if (address == 0 || size == 0) return false;
if (address < LOWER_ADDRESS_SIZE) return size <= LOWER_ADDRESS_SIZE - address;
constexpr auto base = LibKernel::Memory::kExtendedMemoryBase;
constexpr auto length = LibKernel::Memory::kExtendedMemorySize;
return address >= base && address - base < length && size <= length - (address - base);
}
[[nodiscard]] constexpr bool ValidOrEmpty() const noexcept { return Empty() || Valid(); }
[[nodiscard]] constexpr uint64_t End() const noexcept { return address + size; }
src/graphics/host_gpu/renderer/cache/bufferCache.cpp
Vendored
+11 -5
View File
@@ -55,7 +55,8 @@ template <bool insert>
void BufferCache::ChangeRegister(BufferId id) {
auto& buffer = m_slot_buffers[id];
PageTable::PageRange pages {};
EXIT_IF(!PageTable::TryGetPageRange(buffer.CpuAddress(), buffer.Size(), pages));
EXIT_IF(!(GuestRange {buffer.CpuAddress(), buffer.Size()}.Valid()) ||
!PageTable::TryGetPageRange(buffer.CpuAddress(), buffer.Size(), pages));
for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
if constexpr (insert) {
m_page_table[page] = id;
@@ -64,6 +65,7 @@ void BufferCache::ChangeRegister(BufferId id) {
}
}
const auto size_pages = pages.last_exclusive - pages.first;
const auto table_offset = PageIndex(buffer.CpuAddress()) * sizeof(vk::DeviceAddress);
if constexpr (insert) {
const auto [it, inserted] = m_buffers.emplace(buffer.CpuAddress(), id);
(void)it;
@@ -75,7 +77,7 @@ void BufferCache::ChangeRegister(BufferId id) {
for (uint64_t i = 0; i < size_pages; ++i) {
addresses.push_back(buffer.BufferDeviceAddress() + (i << CACHING_PAGEBITS));
}
WriteDataBuffer(m_bda_pagetable_buffer, pages.first * sizeof(vk::DeviceAddress),
WriteDataBuffer(m_bda_pagetable_buffer, table_offset,
addresses.data(), addresses.size() * sizeof(vk::DeviceAddress));
} else {
const auto found = m_buffers.find(buffer.CpuAddress());
@@ -84,7 +86,7 @@ void BufferCache::ChangeRegister(BufferId id) {
EXIT_IF(buffer.Size() > m_total_used_memory);
m_total_used_memory -= buffer.Size();
m_lru_cache.Free(buffer.lru_id);
m_bda_pagetable_buffer.Fill(pages.first * sizeof(vk::DeviceAddress),
m_bda_pagetable_buffer.Fill(table_offset,
size_pages * sizeof(vk::DeviceAddress), 0);
buffer.is_deleted = true;
}
@@ -312,10 +314,14 @@ BufferCache::OverlapResult BufferCache::ResolveOverlaps(uint64_t vaddr, uint64_t
// Reserve space in the incoming stream's direction of growth.
// The old buffer extending left of the request predicts growth to the right, and vice versa.
if (expands_left) {
end += std::min(StreamLeapSize, PageTable::kAddressSpaceSize - end);
end += std::min(StreamLeapSize, (vaddr < LOWER_ADDRESS_SIZE ? LOWER_ADDRESS_SIZE
: LibKernel::Memory::kExtendedMemoryBase +
LibKernel::Memory::kExtendedMemorySize) - end);
}
if (expands_right) {
const auto minimum = CACHING_PAGESIZE * 2;
const auto minimum = vaddr < LOWER_ADDRESS_SIZE
? CACHING_PAGESIZE * 2
: LibKernel::Memory::kExtendedMemoryBase;
if (begin > minimum) {
begin -= std::min(StreamLeapSize, begin - minimum);
}
src/graphics/host_gpu/renderer/cache/bufferCache.h
Vendored
+14 -2
View File
@@ -29,10 +29,22 @@ class BufferCache {
public:
static constexpr uint32_t CACHING_PAGEBITS = 14;
static constexpr uint64_t CACHING_PAGESIZE = uint64_t {1} << CACHING_PAGEBITS;
static constexpr uint64_t CACHING_NUMPAGES = uint64_t {1} << (40 - CACHING_PAGEBITS);
static constexpr uint64_t CACHING_NUMPAGES = (LOWER_ADDRESS_SIZE + LibKernel::Memory::kExtendedMemorySize) >> CACHING_PAGEBITS;
static constexpr uint64_t BDA_PAGETABLE_SIZE =
CACHING_NUMPAGES * sizeof(vk::DeviceAddress);
static constexpr uint64_t PageIndex(uint64_t address) {
return (address < LOWER_ADDRESS_SIZE
? address
: address - LibKernel::Memory::kExtendedMemoryBase + LOWER_ADDRESS_SIZE) >>
CACHING_PAGEBITS;
}
static constexpr uint64_t GuestAddress(uint64_t offset) {
return offset < LOWER_ADDRESS_SIZE
? offset
: offset - LOWER_ADDRESS_SIZE + LibKernel::Memory::kExtendedMemoryBase;
}
BufferCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
TextureCache& texture_cache);
~BufferCache();
@@ -88,7 +100,7 @@ private:
bool has_stream_leap;
};
using PageTable = MultiLevelPageTable<BufferId, CACHING_PAGEBITS, 40, 16>;
using PageTable = MultiLevelPageTable<BufferId, CACHING_PAGEBITS, 44, 20>;
static_assert(CACHING_PAGESIZE == (uint64_t {1} << PageTable::kPageBits));
void WriteDataBuffer(Buffer& buffer, uint64_t address, const void* source, uint64_t size);
void TouchBuffer(const Buffer& buffer);
src/graphics/host_gpu/renderer/cache/faultManager.cpp
Vendored
+3 -2
View File
@@ -134,8 +134,9 @@ void FaultManager::ProcessFaultBuffer() {
const auto* faults = std::bit_cast<const uint64_t*>(mapped);
const auto count = static_cast<uint32_t>(faults[0]);
for (uint32_t index = 1; index <= count; ++index) {
fault_ranges.Add(faults[index], BufferCache::CACHING_PAGESIZE);
LOGF("Accessed non-GPU cached memory at 0x%016" PRIx64 "\n", faults[index]);
const auto address = BufferCache::GuestAddress(faults[index]);
fault_ranges.Add(address, BufferCache::CACHING_PAGESIZE);
LOGF("Accessed non-GPU cached memory at 0x%016" PRIx64 "\n", address);
}
fault_ranges.ForEach([this](uint64_t start, uint64_t end) {
EXIT_IF(end - start > std::numeric_limits<uint32_t>::max());
src/graphics/host_gpu/renderer/cache/textureCache.h
Vendored
+2 -2
View File
@@ -91,9 +91,9 @@ private:
};
using ImageIds = InlinePageOwnerList<ImageId, 16>;
using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 40, 10>;
using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 44, 14>;
// Callers have validated the nonempty 40-bit range with TryGetPageRange.
// Callers have validated the nonempty 44-bit range with TryGetPageRange.
template <typename Func>
static void ForEachPage(uint64_t address, size_t size, Func&& func) {
using FuncReturn = typename std::invoke_result<Func, uint64_t>::type;
src/graphics/host_gpu/renderer/image/image.cpp
+2 -4
View File
@@ -628,10 +628,8 @@ void Validate(const ImageInfo& info) {
break;
case ImageMetadataKind::Dcc:
case ImageMetadataKind::Cmask:
if (info.metadata.range.address == 0 ||
info.metadata.range.address >= TRACKER_ADDRESS_SIZE ||
(info.metadata.range.size != 0 &&
info.metadata.range.size > TRACKER_ADDRESS_SIZE - info.metadata.range.address) ||
if (!GuestRange {info.metadata.range.address,
std::max<uint64_t>(info.metadata.range.size, 1)}.Valid() ||
info.metadata.compression == VideoOutCompression::Unsupported) {
EXIT("invalid color metadata\n");
}
src/graphics/host_gpu/renderer/pipeline/descriptors.cpp
+2 -2
View File
@@ -196,8 +196,8 @@ bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, bo
constexpr uint32_t htile_control = 0x00280000u;
const uint32_t expected_control = htile_control | (descriptor.MsaaDepth() ? (1u << 10u) : 0u);
const auto metadata_addr = descriptor.MetaAddr() << 8u;
return metadata_control == expected_control && metadata_addr != 0 &&
metadata_addr < TRACKER_ADDRESS_SIZE && (metadata_addr & 0x7fffu) == 0 &&
return metadata_control == expected_control && GuestRange {metadata_addr, 1}.Valid() &&
(metadata_addr & 0x7fffu) == 0 &&
descriptor.TileMode() == Prospero::TileMode::kDepth;
}
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp
+8 -1
View File
@@ -975,7 +975,14 @@ void DefineGetBdaPointer(EmitterState& state) {
state.builder.AddFunction(spv::OpFunctionParameter, type, address);
EmitLabel(state, entry_label);
const auto page64 = Binary(state, spv::OpShiftRightLogical, type, address,
const auto extended = Binary(state, spv::OpUGreaterThanEqual, TypeBool(state), address,
ConstantDeviceAddress(state, LibKernel::Memory::kExtendedMemoryBase));
const auto packed = Select(state, type, extended,
Binary(state, spv::OpISub, type, address,
ConstantDeviceAddress(state,
LibKernel::Memory::kExtendedMemoryBase - LOWER_ADDRESS_SIZE)),
address);
const auto page64 = Binary(state, spv::OpShiftRightLogical, type, packed,
ConstantDeviceAddress(state, BufferCache::CACHING_PAGEBITS));
const auto page = Unary(state, spv::OpUConvert, TypeU32(state), page64);
const auto entry_pointer = state.builder.AllocateId();