Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #12579 from FernandoS27/smmu

Browse source 
Core: Implement Device Mapping & GPU SMMU
This commit is contained in:
liamwhite 2024-01-22 10:55:39 -05:00 • committed by GitHub
commit 8bd10473d6
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
121 changed files with 2742 additions and 1415 deletions
Download patch file Download diff file Expand all files Collapse all files

5
src/video_core/CMakeLists.txt
View file

@ -71,6 +71,8 @@ add_library(video_core STATIC
host1x/ffmpeg/ffmpeg.h
host1x/control.cpp
host1x/control.h
host1x/gpu_device_memory_manager.cpp
host1x/gpu_device_memory_manager.h
host1x/host1x.cpp
host1x/host1x.h
host1x/nvdec.cpp
@ -93,6 +95,7 @@ add_library(video_core STATIC
gpu.h
gpu_thread.cpp
gpu_thread.h
guest_memory.h
invalidation_accumulator.h
memory_manager.cpp
memory_manager.h
@ -105,8 +108,6 @@ add_library(video_core STATIC
query_cache/query_stream.h
query_cache/types.h
query_cache.h
rasterizer_accelerated.cpp
rasterizer_accelerated.h
rasterizer_interface.h
renderer_base.cpp
renderer_base.h

3
src/video_core/buffer_cache/buffer_base.h
View file

@ -33,13 +33,12 @@ struct NullBufferParams {};
*
* The buffer size and address is forcefully aligned to CPU page boundaries.
*/
template <class RasterizerInterface>
class BufferBase {
public:
static constexpr u64 BASE_PAGE_BITS = 16;
static constexpr u64 BASE_PAGE_SIZE = 1ULL << BASE_PAGE_BITS;
explicit BufferBase(RasterizerInterface& rasterizer_, VAddr cpu_addr_, u64 size_bytes_)
explicit BufferBase(VAddr cpu_addr_, u64 size_bytes_)
: cpu_addr{cpu_addr_}, size_bytes{size_bytes_} {}
explicit BufferBase(NullBufferParams) {}

451
src/video_core/buffer_cache/buffer_cache.h
View file

File diff suppressed because it is too large Load diff

98
src/video_core/buffer_cache/buffer_cache_base.h
View file

@ -32,7 +32,6 @@
#include "common/microprofile.h"
#include "common/scope_exit.h"
#include "common/settings.h"
#include "core/memory.h"
#include "video_core/buffer_cache/buffer_base.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/delayed_destruction_ring.h"
@ -41,7 +40,6 @@
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/slot_vector.h"
#include "video_core/texture_cache/types.h"
@ -94,7 +92,7 @@ static constexpr BufferId NULL_BUFFER_ID{0};
static constexpr u32 DEFAULT_SKIP_CACHE_SIZE = static_cast<u32>(4_KiB);
struct Binding {
VAddr cpu_addr{};
DAddr device_addr{};
u32 size{};
BufferId buffer_id;
};
@ -104,7 +102,7 @@ struct TextureBufferBinding : Binding {
};
static constexpr Binding NULL_BINDING{
.cpu_addr = 0,
.device_addr = 0,
.size = 0,
.buffer_id = NULL_BUFFER_ID,
};
@ -204,10 +202,10 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
using Async_Buffer = typename P::Async_Buffer;
using MemoryTracker = typename P::MemoryTracker;
using IntervalCompare = std::less<VAddr>;
using IntervalInstance = boost::icl::interval_type_default<VAddr, std::less>;
using IntervalAllocator = boost::fast_pool_allocator<VAddr>;
using IntervalSet = boost::icl::interval_set<VAddr>;
using IntervalCompare = std::less<DAddr>;
using IntervalInstance = boost::icl::interval_type_default<DAddr, std::less>;
using IntervalAllocator = boost::fast_pool_allocator<DAddr>;
using IntervalSet = boost::icl::interval_set<DAddr>;
using IntervalType = typename IntervalSet::interval_type;
template <typename Type>
@ -230,32 +228,31 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
using OverlapCombine = counter_add_functor<int>;
using OverlapSection = boost::icl::inter_section<int>;
using OverlapCounter = boost::icl::split_interval_map<VAddr, int>;
using OverlapCounter = boost::icl::split_interval_map<DAddr, int>;
struct OverlapResult {
boost::container::small_vector<BufferId, 16> ids;
VAddr begin;
VAddr end;
DAddr begin;
DAddr end;
bool has_stream_leap = false;
};
public:
explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_,
Core::Memory::Memory& cpu_memory_, Runtime& runtime_);
explicit BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, Runtime& runtime_);
void TickFrame();
void WriteMemory(VAddr cpu_addr, u64 size);
void WriteMemory(DAddr device_addr, u64 size);
void CachedWriteMemory(VAddr cpu_addr, u64 size);
void CachedWriteMemory(DAddr device_addr, u64 size);
bool OnCPUWrite(VAddr cpu_addr, u64 size);
bool OnCPUWrite(DAddr device_addr, u64 size);
void DownloadMemory(VAddr cpu_addr, u64 size);
void DownloadMemory(DAddr device_addr, u64 size);
std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size);
std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(DAddr device_addr, u64 size);
bool InlineMemory(VAddr dest_address, size_t copy_size, std::span<const u8> inlined_buffer);
bool InlineMemory(DAddr dest_address, size_t copy_size, std::span<const u8> inlined_buffer);
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size);
@ -300,7 +297,7 @@ public:
ObtainBufferSynchronize sync_info,
ObtainBufferOperation post_op);
[[nodiscard]] std::pair<Buffer*, u32> ObtainCPUBuffer(VAddr gpu_addr, u32 size,
[[nodiscard]] std::pair<Buffer*, u32> ObtainCPUBuffer(DAddr gpu_addr, u32 size,
ObtainBufferSynchronize sync_info,
ObtainBufferOperation post_op);
void FlushCachedWrites();
@ -326,13 +323,13 @@ public:
bool DMAClear(GPUVAddr src_address, u64 amount, u32 value);
/// Return true when a CPU region is modified from the GPU
[[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
[[nodiscard]] bool IsRegionGpuModified(DAddr addr, size_t size);
/// Return true when a region is registered on the cache
[[nodiscard]] bool IsRegionRegistered(VAddr addr, size_t size);
[[nodiscard]] bool IsRegionRegistered(DAddr addr, size_t size);
/// Return true when a CPU region is modified from the CPU
[[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size);
[[nodiscard]] bool IsRegionCpuModified(DAddr addr, size_t size);
void SetDrawIndirect(
const Tegra::Engines::DrawManager::IndirectParams* current_draw_indirect_) {
@ -366,9 +363,9 @@ private:
}
template <typename Func>
void ForEachBufferInRange(VAddr cpu_addr, u64 size, Func&& func) {
const u64 page_end = Common::DivCeil(cpu_addr + size, CACHING_PAGESIZE);
for (u64 page = cpu_addr >> CACHING_PAGEBITS; page < page_end;) {
void ForEachBufferInRange(DAddr device_addr, u64 size, Func&& func) {
const u64 page_end = Common::DivCeil(device_addr + size, CACHING_PAGESIZE);
for (u64 page = device_addr >> CACHING_PAGEBITS; page < page_end;) {
const BufferId buffer_id = page_table[page];
if (!buffer_id) {
++page;
@ -377,15 +374,15 @@ private:
Buffer& buffer = slot_buffers[buffer_id];
func(buffer_id, buffer);
const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes();
const DAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes();
page = Common::DivCeil(end_addr, CACHING_PAGESIZE);
}
}
template <typename Func>
void ForEachInRangeSet(IntervalSet& current_range, VAddr cpu_addr, u64 size, Func&& func) {
const VAddr start_address = cpu_addr;
const VAddr end_address = start_address + size;
void ForEachInRangeSet(IntervalSet& current_range, DAddr device_addr, u64 size, Func&& func) {
const DAddr start_address = device_addr;
const DAddr end_address = start_address + size;
const IntervalType search_interval{start_address, end_address};
auto it = current_range.lower_bound(search_interval);
if (it == current_range.end()) {
@ -393,8 +390,8 @@ private:
}
auto end_it = current_range.upper_bound(search_interval);
for (; it != end_it; it++) {
VAddr inter_addr_end = it->upper();
VAddr inter_addr = it->lower();
DAddr inter_addr_end = it->upper();
DAddr inter_addr = it->lower();
if (inter_addr_end > end_address) {
inter_addr_end = end_address;
}
@ -406,10 +403,10 @@ private:
}
template <typename Func>
void ForEachInOverlapCounter(OverlapCounter& current_range, VAddr cpu_addr, u64 size,
void ForEachInOverlapCounter(OverlapCounter& current_range, DAddr device_addr, u64 size,
Func&& func) {
const VAddr start_address = cpu_addr;
const VAddr end_address = start_address + size;
const DAddr start_address = device_addr;
const DAddr end_address = start_address + size;
const IntervalType search_interval{start_address, end_address};
auto it = current_range.lower_bound(search_interval);
if (it == current_range.end()) {
@ -418,8 +415,8 @@ private:
auto end_it = current_range.upper_bound(search_interval);
for (; it != end_it; it++) {
auto& inter = it->first;
VAddr inter_addr_end = inter.upper();
VAddr inter_addr = inter.lower();
DAddr inter_addr_end = inter.upper();
DAddr inter_addr = inter.lower();
if (inter_addr_end > end_address) {
inter_addr_end = end_address;
}
@ -451,9 +448,9 @@ private:
} while (any_removals);
}
static bool IsRangeGranular(VAddr cpu_addr, size_t size) {
return (cpu_addr & ~Core::Memory::YUZU_PAGEMASK) ==
((cpu_addr + size) & ~Core::Memory::YUZU_PAGEMASK);
static bool IsRangeGranular(DAddr device_addr, size_t size) {
return (device_addr & ~Core::DEVICE_PAGEMASK) ==
((device_addr + size) & ~Core::DEVICE_PAGEMASK);
}
void RunGarbageCollector();
@ -508,15 +505,15 @@ private:
void UpdateComputeTextureBuffers();
void MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size);
void MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u32 size);
[[nodiscard]] BufferId FindBuffer(VAddr cpu_addr, u32 size);
[[nodiscard]] BufferId FindBuffer(DAddr device_addr, u32 size);
[[nodiscard]] OverlapResult ResolveOverlaps(VAddr cpu_addr, u32 wanted_size);
[[nodiscard]] OverlapResult ResolveOverlaps(DAddr device_addr, u32 wanted_size);
void JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, bool accumulate_stream_score);
[[nodiscard]] BufferId CreateBuffer(VAddr cpu_addr, u32 wanted_size);
[[nodiscard]] BufferId CreateBuffer(DAddr device_addr, u32 wanted_size);
void Register(BufferId buffer_id);
@ -527,7 +524,7 @@ private:
void TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept;
bool SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size);
bool SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 size);
void UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy,
std::span<BufferCopy> copies);
@ -539,7 +536,7 @@ private:
void DownloadBufferMemory(Buffer& buffer_id);
void DownloadBufferMemory(Buffer& buffer_id, VAddr cpu_addr, u64 size);
void DownloadBufferMemory(Buffer& buffer_id, DAddr device_addr, u64 size);
void DeleteBuffer(BufferId buffer_id, bool do_not_mark = false);
@ -549,7 +546,7 @@ private:
[[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size,
PixelFormat format);
[[nodiscard]] std::span<const u8> ImmediateBufferWithData(VAddr cpu_addr, size_t size);
[[nodiscard]] std::span<const u8> ImmediateBufferWithData(DAddr device_addr, size_t size);
[[nodiscard]] std::span<u8> ImmediateBuffer(size_t wanted_capacity);
@ -557,11 +554,10 @@ private:
void ClearDownload(IntervalType subtract_interval);
void InlineMemoryImplementation(VAddr dest_address, size_t copy_size,
void InlineMemoryImplementation(DAddr dest_address, size_t copy_size,
std::span<const u8> inlined_buffer);
VideoCore::RasterizerInterface& rasterizer;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
SlotVector<Buffer> slot_buffers;
DelayedDestructionRing<Buffer, 8> delayed_destruction_ring;
@ -598,7 +594,7 @@ private:
u64 critical_memory = 0;
BufferId inline_buffer_id;
std::array<BufferId, ((1ULL << 39) >> CACHING_PAGEBITS)> page_table;
std::array<BufferId, ((1ULL << 34) >> CACHING_PAGEBITS)> page_table;
Common::ScratchBuffer<u8> tmp_buffer;
};

18
src/video_core/buffer_cache/memory_tracker_base.h
View file

@ -17,19 +17,19 @@
namespace VideoCommon {
template <class RasterizerInterface>
template <typename DeviceTracker>
class MemoryTrackerBase {
static constexpr size_t MAX_CPU_PAGE_BITS = 39;
static constexpr size_t MAX_CPU_PAGE_BITS = 34;
static constexpr size_t HIGHER_PAGE_BITS = 22;
static constexpr size_t HIGHER_PAGE_SIZE = 1ULL << HIGHER_PAGE_BITS;
static constexpr size_t HIGHER_PAGE_MASK = HIGHER_PAGE_SIZE - 1ULL;
static constexpr size_t NUM_HIGH_PAGES = 1ULL << (MAX_CPU_PAGE_BITS - HIGHER_PAGE_BITS);
static constexpr size_t MANAGER_POOL_SIZE = 32;
static constexpr size_t WORDS_STACK_NEEDED = HIGHER_PAGE_SIZE / BYTES_PER_WORD;
using Manager = WordManager<RasterizerInterface, WORDS_STACK_NEEDED>;
using Manager = WordManager<DeviceTracker, WORDS_STACK_NEEDED>;
public:
MemoryTrackerBase(RasterizerInterface& rasterizer_) : rasterizer{&rasterizer_} {}
MemoryTrackerBase(DeviceTracker& device_tracker_) : device_tracker{&device_tracker_} {}
~MemoryTrackerBase() = default;
/// Returns the inclusive CPU modified range in a begin end pair
@ -74,7 +74,7 @@ public:
});
}
/// Mark region as CPU modified, notifying the rasterizer about this change
/// Mark region as CPU modified, notifying the device_tracker about this change
void MarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 query_size) {
IteratePages<true>(dirty_cpu_addr, query_size,
[](Manager* manager, u64 offset, size_t size) {
@ -83,7 +83,7 @@ public:
});
}
/// Unmark region as CPU modified, notifying the rasterizer about this change
/// Unmark region as CPU modified, notifying the device_tracker about this change
void UnmarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 query_size) {
IteratePages<true>(dirty_cpu_addr, query_size,
[](Manager* manager, u64 offset, size_t size) {
@ -139,7 +139,7 @@ public:
});
}
/// Flushes cached CPU writes, and notify the rasterizer about the deltas
/// Flushes cached CPU writes, and notify the device_tracker about the deltas
void FlushCachedWrites(VAddr query_cpu_addr, u64 query_size) noexcept {
IteratePages<false>(query_cpu_addr, query_size,
[](Manager* manager, [[maybe_unused]] u64 offset,
@ -280,7 +280,7 @@ private:
manager_pool.emplace_back();
auto& last_pool = manager_pool.back();
for (size_t i = 0; i < MANAGER_POOL_SIZE; i++) {
new (&last_pool[i]) Manager(0, *rasterizer, HIGHER_PAGE_SIZE);
new (&last_pool[i]) Manager(0, *device_tracker, HIGHER_PAGE_SIZE);
free_managers.push_back(&last_pool[i]);
}
return on_return();
@ -293,7 +293,7 @@ private:
std::unordered_set<u32> cached_pages;
RasterizerInterface* rasterizer = nullptr;
DeviceTracker* device_tracker = nullptr;
};
} // namespace VideoCommon

28
src/video_core/buffer_cache/word_manager.h
View file

@ -13,12 +13,12 @@
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/div_ceil.h"
#include "core/memory.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
namespace VideoCommon {
constexpr u64 PAGES_PER_WORD = 64;
constexpr u64 BYTES_PER_PAGE = Core::Memory::YUZU_PAGESIZE;
constexpr u64 BYTES_PER_PAGE = Core::DEVICE_PAGESIZE;
constexpr u64 BYTES_PER_WORD = PAGES_PER_WORD * BYTES_PER_PAGE;
enum class Type {
@ -163,11 +163,11 @@ struct Words {
WordsArray<stack_words> preflushable;
};
template <class RasterizerInterface, size_t stack_words = 1>
template <class DeviceTracker, size_t stack_words = 1>
class WordManager {
public:
explicit WordManager(VAddr cpu_addr_, RasterizerInterface& rasterizer_, u64 size_bytes)
: cpu_addr{cpu_addr_}, rasterizer{&rasterizer_}, words{size_bytes} {}
explicit WordManager(VAddr cpu_addr_, DeviceTracker& tracker_, u64 size_bytes)
: cpu_addr{cpu_addr_}, tracker{&tracker_}, words{size_bytes} {}
explicit WordManager() = default;
@ -279,7 +279,7 @@ public:
}
/**
* Loop over each page in the given range, turn off those bits and notify the rasterizer if
* Loop over each page in the given range, turn off those bits and notify the tracker if
* needed. Call the given function on each turned off range.
*
* @param query_cpu_range Base CPU address to loop over
@ -459,26 +459,26 @@ private:
}
/**
* Notify rasterizer about changes in the CPU tracking state of a word in the buffer
* Notify tracker about changes in the CPU tracking state of a word in the buffer
*
* @param word_index Index to the word to notify to the rasterizer
* @param word_index Index to the word to notify to the tracker
* @param current_bits Current state of the word
* @param new_bits New state of the word
*
* @tparam add_to_rasterizer True when the rasterizer should start tracking the new pages
* @tparam add_to_tracker True when the tracker should start tracking the new pages
*/
template <bool add_to_rasterizer>
template <bool add_to_tracker>
void NotifyRasterizer(u64 word_index, u64 current_bits, u64 new_bits) const {
u64 changed_bits = (add_to_rasterizer ? current_bits : ~current_bits) & new_bits;
u64 changed_bits = (add_to_tracker ? current_bits : ~current_bits) & new_bits;
VAddr addr = cpu_addr + word_index * BYTES_PER_WORD;
IteratePages(changed_bits, [&](size_t offset, size_t size) {
rasterizer->UpdatePagesCachedCount(addr + offset * BYTES_PER_PAGE,
size * BYTES_PER_PAGE, add_to_rasterizer ? 1 : -1);
tracker->UpdatePagesCachedCount(addr + offset * BYTES_PER_PAGE, size * BYTES_PER_PAGE,
add_to_tracker ? 1 : -1);
});
}
VAddr cpu_addr = 0;
RasterizerInterface* rasterizer = nullptr;
DeviceTracker* tracker = nullptr;
Words<stack_words> words;
};

10
src/video_core/dma_pusher.cpp
View file

@ -5,10 +5,10 @@
#include "common/microprofile.h"
#include "common/settings.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/dma_pusher.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/gpu.h"
#include "video_core/guest_memory.h"
#include "video_core/memory_manager.h"
namespace Tegra {
@ -85,15 +85,15 @@ bool DmaPusher::Step() {
}
}
const auto safe_process = [&] {
Core::Memory::GpuGuestMemory<Tegra::CommandHeader,
Core::Memory::GuestMemoryFlags::SafeRead>
Tegra::Memory::GpuGuestMemory<Tegra::CommandHeader,
Tegra::Memory::GuestMemoryFlags::SafeRead>
headers(memory_manager, dma_state.dma_get, command_list_header.size,
&command_headers);
ProcessCommands(headers);
};
const auto unsafe_process = [&] {
Core::Memory::GpuGuestMemory<Tegra::CommandHeader,
Core::Memory::GuestMemoryFlags::UnsafeRead>
Tegra::Memory::GpuGuestMemory<Tegra::CommandHeader,
Tegra::Memory::GuestMemoryFlags::UnsafeRead>
headers(memory_manager, dma_state.dma_get, command_list_header.size,
&command_headers);
ProcessCommands(headers);

5
src/video_core/engines/engine_upload.cpp
View file

@ -5,8 +5,8 @@
#include "common/algorithm.h"
#include "common/assert.h"
#include "core/memory.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/guest_memory.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/textures/decoders.h"
@ -68,7 +68,8 @@ void State::ProcessData(std::span<const u8> read_buffer) {
true, bytes_per_pixel, width, regs.dest.height, regs.dest.depth,
regs.dest.BlockHeight(), regs.dest.BlockDepth());
Core::Memory::GpuGuestMemoryScoped<u8, Core::Memory::GuestMemoryFlags::SafeReadCachedWrite>
Tegra::Memory::GpuGuestMemoryScoped<u8,
Tegra::Memory::GuestMemoryFlags::SafeReadCachedWrite>
tmp(memory_manager, address, dst_size, &tmp_buffer);
Tegra::Texture::SwizzleSubrect(tmp, read_buffer, bytes_per_pixel, width, regs.dest.height,

1
src/video_core/engines/maxwell_3d.cpp
View file

@ -9,7 +9,6 @@
#include "common/settings.h"
#include "core/core.h"
#include "core/core_timing.h"
#include "core/memory.h"
#include "video_core/dirty_flags.h"
#include "video_core/engines/draw_manager.h"
#include "video_core/engines/maxwell_3d.h"

26
src/video_core/engines/maxwell_dma.cpp
View file

@ -8,9 +8,9 @@
#include "common/polyfill_ranges.h"
#include "common/settings.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/maxwell_dma.h"
#include "video_core/guest_memory.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_base.h"
#include "video_core/textures/decoders.h"
@ -133,8 +133,8 @@ void MaxwellDMA::Launch() {
UNIMPLEMENTED_IF(regs.offset_out % 16 != 0);
read_buffer.resize_destructive(16);
for (u32 offset = 0; offset < regs.line_length_in; offset += 16) {
Core::Memory::GpuGuestMemoryScoped<
u8, Core::Memory::GuestMemoryFlags::SafeReadCachedWrite>
Tegra::Memory::GpuGuestMemoryScoped<
u8, Tegra::Memory::GuestMemoryFlags::SafeReadCachedWrite>
tmp_write_buffer(memory_manager,
convert_linear_2_blocklinear_addr(regs.offset_in + offset),
16, &read_buffer);
@ -146,16 +146,16 @@ void MaxwellDMA::Launch() {
UNIMPLEMENTED_IF(regs.offset_out % 16 != 0);
read_buffer.resize_destructive(16);
for (u32 offset = 0; offset < regs.line_length_in; offset += 16) {
Core::Memory::GpuGuestMemoryScoped<
u8, Core::Memory::GuestMemoryFlags::SafeReadCachedWrite>
Tegra::Memory::GpuGuestMemoryScoped<
u8, Tegra::Memory::GuestMemoryFlags::SafeReadCachedWrite>
tmp_write_buffer(memory_manager, regs.offset_in + offset, 16, &read_buffer);
tmp_write_buffer.SetAddressAndSize(
convert_linear_2_blocklinear_addr(regs.offset_out + offset), 16);
}
} else {
if (!accelerate.BufferCopy(regs.offset_in, regs.offset_out, regs.line_length_in)) {
Core::Memory::GpuGuestMemoryScoped<
u8, Core::Memory::GuestMemoryFlags::SafeReadCachedWrite>
Tegra::Memory::GpuGuestMemoryScoped<
u8, Tegra::Memory::GuestMemoryFlags::SafeReadCachedWrite>
tmp_write_buffer(memory_manager, regs.offset_in, regs.line_length_in,
&read_buffer);
tmp_write_buffer.SetAddressAndSize(regs.offset_out, regs.line_length_in);
@ -226,9 +226,9 @@ void MaxwellDMA::CopyBlockLinearToPitch() {
const size_t dst_size = dst_operand.pitch * regs.line_count;
Core::Memory::GpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead> tmp_read_buffer(
Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::SafeRead> tmp_read_buffer(
memory_manager, src_operand.address, src_size, &read_buffer);
Core::Memory::GpuGuestMemoryScoped<u8, Core::Memory::GuestMemoryFlags::UnsafeReadCachedWrite>
Tegra::Memory::GpuGuestMemoryScoped<u8, Tegra::Memory::GuestMemoryFlags::UnsafeReadCachedWrite>
tmp_write_buffer(memory_manager, dst_operand.address, dst_size, &write_buffer);
UnswizzleSubrect(tmp_write_buffer, tmp_read_buffer, bytes_per_pixel, width, height, depth,
@ -290,9 +290,9 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
GPUVAddr src_addr = regs.offset_in;
GPUVAddr dst_addr = regs.offset_out;
Core::Memory::GpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead> tmp_read_buffer(
Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::SafeRead> tmp_read_buffer(
memory_manager, src_addr, src_size, &read_buffer);
Core::Memory::GpuGuestMemoryScoped<u8, Core::Memory::GuestMemoryFlags::UnsafeReadCachedWrite>
Tegra::Memory::GpuGuestMemoryScoped<u8, Tegra::Memory::GuestMemoryFlags::UnsafeReadCachedWrite>
tmp_write_buffer(memory_manager, dst_addr, dst_size, &write_buffer);
// If the input is linear and the output is tiled, swizzle the input and copy it over.
@ -344,9 +344,9 @@ void MaxwellDMA::CopyBlockLinearToBlockLinear() {
intermediate_buffer.resize_destructive(mid_buffer_size);
Core::Memory::GpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead> tmp_read_buffer(
Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::SafeRead> tmp_read_buffer(
memory_manager, regs.offset_in, src_size, &read_buffer);
Core::Memory::GpuGuestMemoryScoped<u8, Core::Memory::GuestMemoryFlags::SafeReadCachedWrite>
Tegra::Memory::GpuGuestMemoryScoped<u8, Tegra::Memory::GuestMemoryFlags::SafeReadCachedWrite>
tmp_write_buffer(memory_manager, regs.offset_out, dst_size, &write_buffer);
UnswizzleSubrect(intermediate_buffer, tmp_read_buffer, bytes_per_pixel, src_width, src.height,

5
src/video_core/engines/sw_blitter/blitter.cpp
View file

@ -8,6 +8,7 @@
#include "common/scratch_buffer.h"
#include "video_core/engines/sw_blitter/blitter.h"
#include "video_core/engines/sw_blitter/converter.h"
#include "video_core/guest_memory.h"
#include "video_core/memory_manager.h"
#include "video_core/surface.h"
#include "video_core/textures/decoders.h"
@ -160,7 +161,7 @@ bool SoftwareBlitEngine::Blit(Fermi2D::Surface& src, Fermi2D::Surface& dst,
const auto dst_bytes_per_pixel = BytesPerBlock(PixelFormatFromRenderTargetFormat(dst.format));
const size_t src_size = get_surface_size(src, src_bytes_per_pixel);
Core::Memory::GpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::SafeRead> tmp_buffer(
Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::SafeRead> tmp_buffer(
memory_manager, src.Address(), src_size, &impl->tmp_buffer);
const size_t src_copy_size = src_extent_x * src_extent_y * src_bytes_per_pixel;
@ -220,7 +221,7 @@ bool SoftwareBlitEngine::Blit(Fermi2D::Surface& src, Fermi2D::Surface& dst,
}
const size_t dst_size = get_surface_size(dst, dst_bytes_per_pixel);
Core::Memory::GpuGuestMemoryScoped<u8, Core::Memory::GuestMemoryFlags::SafeReadWrite>
Tegra::Memory::GpuGuestMemoryScoped<u8, Tegra::Memory::GuestMemoryFlags::SafeReadWrite>
tmp_buffer2(memory_manager, dst.Address(), dst_size, &impl->tmp_buffer);
if (dst.linear == Fermi2D::MemoryLayout::BlockLinear) {

2
src/video_core/framebuffer_config.h
View file

@ -14,7 +14,7 @@ namespace Tegra {
* Struct describing framebuffer configuration
*/
struct FramebufferConfig {
VAddr address{};
DAddr address{};
u32 offset{};
u32 width{};
u32 height{};

29
src/video_core/gpu.cpp
View file

@ -85,7 +85,8 @@ struct GPU::Impl {
void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) {
renderer = std::move(renderer_);
rasterizer = renderer->ReadRasterizer();
host1x.MemoryManager().BindRasterizer(rasterizer);
host1x.MemoryManager().BindInterface(rasterizer);
host1x.GMMU().BindRasterizer(rasterizer);
}
/// Flush all current written commands into the host GPU for execution.
@ -95,8 +96,8 @@ struct GPU::Impl {
/// Synchronizes CPU writes with Host GPU memory.
void InvalidateGPUCache() {
std::function<void(VAddr, size_t)> callback_writes(
[this](VAddr address, size_t size) { rasterizer->OnCacheInvalidation(address, size); });
std::function<void(PAddr, size_t)> callback_writes(
[this](PAddr address, size_t size) { rasterizer->OnCacheInvalidation(address, size); });
system.GatherGPUDirtyMemory(callback_writes);
}
@ -279,11 +280,11 @@ struct GPU::Impl {
}
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
void FlushRegion(VAddr addr, u64 size) {
void FlushRegion(DAddr addr, u64 size) {
gpu_thread.FlushRegion(addr, size);
}
VideoCore::RasterizerDownloadArea OnCPURead(VAddr addr, u64 size) {
VideoCore::RasterizerDownloadArea OnCPURead(DAddr addr, u64 size) {
auto raster_area = rasterizer->GetFlushArea(addr, size);
if (raster_area.preemtive) {
return raster_area;
@ -299,16 +300,16 @@ struct GPU::Impl {
}
/// Notify rasterizer that any caches of the specified region should be invalidated
void InvalidateRegion(VAddr addr, u64 size) {
void InvalidateRegion(DAddr addr, u64 size) {
gpu_thread.InvalidateRegion(addr, size);
}
bool OnCPUWrite(VAddr addr, u64 size) {
bool OnCPUWrite(DAddr addr, u64 size) {
return rasterizer->OnCPUWrite(addr, size);
}
/// Notify rasterizer that any caches of the specified region should be flushed and invalidated
void FlushAndInvalidateRegion(VAddr addr, u64 size) {
void FlushAndInvalidateRegion(DAddr addr, u64 size) {
gpu_thread.FlushAndInvalidateRegion(addr, size);
}
@ -437,7 +438,7 @@ void GPU::OnCommandListEnd() {
impl->OnCommandListEnd();
}
u64 GPU::RequestFlush(VAddr addr, std::size_t size) {
u64 GPU::RequestFlush(DAddr addr, std::size_t size) {
return impl->RequestSyncOperation(
[this, addr, size]() { impl->rasterizer->FlushRegion(addr, size); });
}
@ -557,23 +558,23 @@ void GPU::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
impl->SwapBuffers(framebuffer);
}
VideoCore::RasterizerDownloadArea GPU::OnCPURead(VAddr addr, u64 size) {
VideoCore::RasterizerDownloadArea GPU::OnCPURead(PAddr addr, u64 size) {
return impl->OnCPURead(addr, size);
}
void GPU::FlushRegion(VAddr addr, u64 size) {
void GPU::FlushRegion(DAddr addr, u64 size) {
impl->FlushRegion(addr, size);
}
void GPU::InvalidateRegion(VAddr addr, u64 size) {
void GPU::InvalidateRegion(DAddr addr, u64 size) {
impl->InvalidateRegion(addr, size);
}
bool GPU::OnCPUWrite(VAddr addr, u64 size) {
bool GPU::OnCPUWrite(DAddr addr, u64 size) {
return impl->OnCPUWrite(addr, size);
}
void GPU::FlushAndInvalidateRegion(VAddr addr, u64 size) {
void GPU::FlushAndInvalidateRegion(DAddr addr, u64 size) {
impl->FlushAndInvalidateRegion(addr, size);
}

12
src/video_core/gpu.h
View file

@ -158,7 +158,7 @@ public:
void InitAddressSpace(Tegra::MemoryManager& memory_manager);
/// Request a host GPU memory flush from the CPU.
[[nodiscard]] u64 RequestFlush(VAddr addr, std::size_t size);
[[nodiscard]] u64 RequestFlush(DAddr addr, std::size_t size);
/// Obtains current flush request fence id.
[[nodiscard]] u64 CurrentSyncRequestFence() const;
@ -242,20 +242,20 @@ public:
void SwapBuffers(const Tegra::FramebufferConfig* framebuffer);
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
[[nodiscard]] VideoCore::RasterizerDownloadArea OnCPURead(VAddr addr, u64 size);
[[nodiscard]] VideoCore::RasterizerDownloadArea OnCPURead(DAddr addr, u64 size);
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
void FlushRegion(VAddr addr, u64 size);
void FlushRegion(DAddr addr, u64 size);
/// Notify rasterizer that any caches of the specified region should be invalidated
void InvalidateRegion(VAddr addr, u64 size);
void InvalidateRegion(DAddr addr, u64 size);
/// Notify rasterizer that CPU is trying to write this area. It returns true if the area is
/// sensible, false otherwise
bool OnCPUWrite(VAddr addr, u64 size);
bool OnCPUWrite(DAddr addr, u64 size);
/// Notify rasterizer that any caches of the specified region should be flushed and invalidated
void FlushAndInvalidateRegion(VAddr addr, u64 size);
void FlushAndInvalidateRegion(DAddr addr, u64 size);
private:
struct Impl;

6
src/video_core/gpu_thread.cpp
View file

@ -82,7 +82,7 @@ void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
PushCommand(SwapBuffersCommand(framebuffer ? std::make_optional(*framebuffer) : std::nullopt));
}
void ThreadManager::FlushRegion(VAddr addr, u64 size) {
void ThreadManager::FlushRegion(DAddr addr, u64 size) {
if (!is_async) {
// Always flush with synchronous GPU mode
PushCommand(FlushRegionCommand(addr, size));
@ -101,11 +101,11 @@ void ThreadManager::TickGPU() {
PushCommand(GPUTickCommand());
}
void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
void ThreadManager::InvalidateRegion(DAddr addr, u64 size) {
rasterizer->OnCacheInvalidation(addr, size);
}
void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
void ThreadManager::FlushAndInvalidateRegion(DAddr addr, u64 size) {
// Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important
rasterizer->OnCacheInvalidation(addr, size);
}

18
src/video_core/gpu_thread.h
View file

@ -54,26 +54,26 @@ struct SwapBuffersCommand final {
/// Command to signal to the GPU thread to flush a region
struct FlushRegionCommand final {
explicit constexpr FlushRegionCommand(VAddr addr_, u64 size_) : addr{addr_}, size{size_} {}
explicit constexpr FlushRegionCommand(DAddr addr_, u64 size_) : addr{addr_}, size{size_} {}
VAddr addr;
DAddr addr;
u64 size;
};
/// Command to signal to the GPU thread to invalidate a region
struct InvalidateRegionCommand final {
explicit constexpr InvalidateRegionCommand(VAddr addr_, u64 size_) : addr{addr_}, size{size_} {}
explicit constexpr InvalidateRegionCommand(DAddr addr_, u64 size_) : addr{addr_}, size{size_} {}
VAddr addr;
DAddr addr;
u64 size;
};
/// Command to signal to the GPU thread to flush and invalidate a region
struct FlushAndInvalidateRegionCommand final {
explicit constexpr FlushAndInvalidateRegionCommand(VAddr addr_, u64 size_)
explicit constexpr FlushAndInvalidateRegionCommand(DAddr addr_, u64 size_)
: addr{addr_}, size{size_} {}
VAddr addr;
DAddr addr;
u64 size;
};
@ -122,13 +122,13 @@ public:
void SwapBuffers(const Tegra::FramebufferConfig* framebuffer);
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
void FlushRegion(VAddr addr, u64 size);
void FlushRegion(DAddr addr, u64 size);
/// Notify rasterizer that any caches of the specified region should be invalidated
void InvalidateRegion(VAddr addr, u64 size);
void InvalidateRegion(DAddr addr, u64 size);
/// Notify rasterizer that any caches of the specified region should be flushed and invalidated
void FlushAndInvalidateRegion(VAddr addr, u64 size);
void FlushAndInvalidateRegion(DAddr addr, u64 size);
void TickGPU();

30
src/video_core/guest_memory.h Normal file
View file

@ -0,0 +1,30 @@
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <iterator>
#include <memory>
#include <optional>
#include <span>
#include <vector>
#include "common/scratch_buffer.h"
#include "core/guest_memory.h"
#include "video_core/memory_manager.h"
namespace Tegra::Memory {
using GuestMemoryFlags = Core::Memory::GuestMemoryFlags;
template <typename T, GuestMemoryFlags FLAGS>
using DeviceGuestMemory = Core::Memory::GuestMemory<Tegra::MaxwellDeviceMemoryManager, T, FLAGS>;
template <typename T, GuestMemoryFlags FLAGS>
using DeviceGuestMemoryScoped =
Core::Memory::GuestMemoryScoped<Tegra::MaxwellDeviceMemoryManager, T, FLAGS>;
template <typename T, GuestMemoryFlags FLAGS>
using GpuGuestMemory = Core::Memory::GuestMemory<Tegra::MemoryManager, T, FLAGS>;
template <typename T, GuestMemoryFlags FLAGS>
using GpuGuestMemoryScoped = Core::Memory::GuestMemoryScoped<Tegra::MemoryManager, T, FLAGS>;
} // namespace Tegra::Memory

9
src/video_core/host1x/codecs/h264.cpp
View file

@ -32,13 +32,12 @@ H264::~H264() = default;
std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state,
size_t* out_configuration_size, bool is_first_frame) {
H264DecoderContext context;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &context,
sizeof(H264DecoderContext));
host1x.GMMU().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext));
const s64 frame_number = context.h264_parameter_set.frame_number.Value();
if (!is_first_frame && frame_number != 0) {
frame.resize_destructive(context.stream_len);
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
host1x.GMMU().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
*out_configuration_size = 0;
return frame;
}
@ -159,8 +158,8 @@ std::span<const u8> H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters
std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
*out_configuration_size = encoded_header.size();
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset,
frame.data() + encoded_header.size(), context.stream_len);
host1x.GMMU().ReadBlock(state.frame_bitstream_offset, frame.data() + encoded_header.size(),
context.stream_len);
return frame;
}

4
src/video_core/host1x/codecs/vp8.cpp
View file

@ -14,7 +14,7 @@ VP8::~VP8() = default;
std::span<const u8> VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) {
VP8PictureInfo info;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo));
host1x.GMMU().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo));
const bool is_key_frame = info.key_frame == 1u;
const auto bitstream_size = static_cast<size_t>(info.vld_buffer_size);
@ -45,7 +45,7 @@ std::span<const u8> VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters&
frame[9] = static_cast<u8>(((info.frame_height >> 8) & 0x3f));
}
const u64 bitstream_offset = state.frame_bitstream_offset;
host1x.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size);
host1x.GMMU().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size);
return frame;
}

9
src/video_core/host1x/codecs/vp9.cpp
View file

@ -358,7 +358,7 @@ void VP9::WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_
Vp9PictureInfo VP9::GetVp9PictureInfo(const Host1x::NvdecCommon::NvdecRegisters& state) {
PictureInfo picture_info;
host1x.MemoryManager().ReadBlock(state.picture_info_offset, &picture_info, sizeof(PictureInfo));
host1x.GMMU().ReadBlock(state.picture_info_offset, &picture_info, sizeof(PictureInfo));
Vp9PictureInfo vp9_info = picture_info.Convert();
InsertEntropy(state.vp9_entropy_probs_offset, vp9_info.entropy);
@ -373,7 +373,7 @@ Vp9PictureInfo VP9::GetVp9PictureInfo(const Host1x::NvdecCommon::NvdecRegisters&
void VP9::InsertEntropy(u64 offset, Vp9EntropyProbs& dst) {
EntropyProbs entropy;
host1x.MemoryManager().ReadBlock(offset, &entropy, sizeof(EntropyProbs));
host1x.GMMU().ReadBlock(offset, &entropy, sizeof(EntropyProbs));
entropy.Convert(dst);
}
@ -383,9 +383,8 @@ Vp9FrameContainer VP9::GetCurrentFrame(const Host1x::NvdecCommon::NvdecRegisters
// gpu.SyncGuestHost(); epic, why?
current_frame.info = GetVp9PictureInfo(state);
current_frame.bit_stream.resize(current_frame.info.bitstream_size);
host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset,
current_frame.bit_stream.data(),
current_frame.info.bitstream_size);
host1x.GMMU().ReadBlock(state.frame_bitstream_offset, current_frame.bit_stream.data(),
current_frame.info.bitstream_size);
}
if (!next_frame.bit_stream.empty()) {
Vp9FrameContainer temp{

32
src/video_core/host1x/gpu_device_memory_manager.cpp Normal file
View file

@ -0,0 +1,32 @@
// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "core/device_memory_manager.inc"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/rasterizer_interface.h"
namespace Tegra {
struct MaxwellDeviceMethods {
static inline void MarkRegionCaching(Core::Memory::Memory* interface, VAddr address,
size_t size, bool caching) {
interface->RasterizerMarkRegionCached(address, size, caching);
}
};
} // namespace Tegra
template struct Core::DeviceMemoryManagerAllocator<Tegra::MaxwellDeviceTraits>;
template class Core::DeviceMemoryManager<Tegra::MaxwellDeviceTraits>;
template const u8* Tegra::MaxwellDeviceMemoryManager::GetPointer<u8>(DAddr addr) const;
template u8* Tegra::MaxwellDeviceMemoryManager::GetPointer<u8>(DAddr addr);
template u8 Tegra::MaxwellDeviceMemoryManager::Read<u8>(DAddr addr) const;
template u16 Tegra::MaxwellDeviceMemoryManager::Read<u16>(DAddr addr) const;
template u32 Tegra::MaxwellDeviceMemoryManager::Read<u32>(DAddr addr) const;
template u64 Tegra::MaxwellDeviceMemoryManager::Read<u64>(DAddr addr) const;
template void Tegra::MaxwellDeviceMemoryManager::Write<u8>(DAddr addr, u8 data);
template void Tegra::MaxwellDeviceMemoryManager::Write<u16>(DAddr addr, u16 data);
template void Tegra::MaxwellDeviceMemoryManager::Write<u32>(DAddr addr, u32 data);
template void Tegra::MaxwellDeviceMemoryManager::Write<u64>(DAddr addr, u64 data);

24
src/video_core/host1x/gpu_device_memory_manager.h Normal file
View file

@ -0,0 +1,24 @@
// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include "core/device_memory_manager.h"
namespace VideoCore {
class RasterizerInterface;
}
namespace Tegra {
struct MaxwellDeviceMethods;
struct MaxwellDeviceTraits {
static constexpr size_t device_virtual_bits = 34;
using DeviceInterface = typename VideoCore::RasterizerInterface;
using DeviceMethods = MaxwellDeviceMethods;
};
using MaxwellDeviceMemoryManager = Core::DeviceMemoryManager<MaxwellDeviceTraits>;
} // namespace Tegra

5
src/video_core/host1x/host1x.cpp
View file

@ -9,9 +9,12 @@ namespace Tegra {
namespace Host1x {
Host1x::Host1x(Core::System& system_)
: system{system_}, syncpoint_manager{}, memory_manager{system, 32, 12},
: system{system_}, syncpoint_manager{},
memory_manager(system.DeviceMemory()), gmmu_manager{system, memory_manager, 32, 12},
allocator{std::make_unique<Common::FlatAllocator<u32, 0, 32>>(1 << 12)} {}
Host1x::~Host1x() = default;
} // namespace Host1x
} // namespace Tegra

17
src/video_core/host1x/host1x.h
View file

@ -6,6 +6,7 @@
#include "common/common_types.h"
#include "common/address_space.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/host1x/syncpoint_manager.h"
#include "video_core/memory_manager.h"
@ -20,6 +21,7 @@ namespace Host1x {
class Host1x {
public:
explicit Host1x(Core::System& system);
~Host1x();
SyncpointManager& GetSyncpointManager() {
return syncpoint_manager;
@ -29,14 +31,22 @@ public:
return syncpoint_manager;
}
Tegra::MemoryManager& MemoryManager() {
Tegra::MaxwellDeviceMemoryManager& MemoryManager() {
return memory_manager;
}
const Tegra::MemoryManager& MemoryManager() const {
const Tegra::MaxwellDeviceMemoryManager& MemoryManager() const {
return memory_manager;
}
Tegra::MemoryManager& GMMU() {
return gmmu_manager;
}
const Tegra::MemoryManager& GMMU() const {
return gmmu_manager;
}
Common::FlatAllocator<u32, 0, 32>& Allocator() {
return *allocator;
}
@ -48,7 +58,8 @@ public:
private:
Core::System& system;
SyncpointManager syncpoint_manager;
Tegra::MemoryManager memory_manager;
Tegra::MaxwellDeviceMemoryManager memory_manager;
Tegra::MemoryManager gmmu_manager;
std::unique_ptr<Common::FlatAllocator<u32, 0, 32>> allocator;
};

15
src/video_core/host1x/vic.cpp
View file

@ -81,7 +81,7 @@ void Vic::Execute() {
LOG_ERROR(Service_NVDRV, "VIC Luma address not set.");
return;
}
const VicConfig config{host1x.MemoryManager().Read<u64>(config_struct_address + 0x20)};
const VicConfig config{host1x.GMMU().Read<u64>(config_struct_address + 0x20)};
auto frame = nvdec_processor->GetFrame();
if (!frame) {
return;
@ -162,12 +162,12 @@ void Vic::WriteRGBFrame(std::unique_ptr<FFmpeg::Frame> frame, const VicConfig& c
Texture::SwizzleSubrect(luma_buffer, frame_buff, 4, width, height, 1, 0, 0, width, height,
block_height, 0, width * 4);
host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size);
host1x.GMMU().WriteBlock(output_surface_luma_address, luma_buffer.data(), size);
} else {
// send pitch linear frame
const size_t linear_size = width * height * 4;
host1x.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
linear_size);
host1x.GMMU().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
linear_size);
}
}
@ -193,8 +193,7 @@ void Vic::WriteYUVFrame(std::unique_ptr<FFmpeg::Frame> frame, const VicConfig& c
const std::size_t dst = y * aligned_width;
std::memcpy(luma_buffer.data() + dst, luma_src + src, frame_width);
}
host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),
luma_buffer.size());
host1x.GMMU().WriteBlock(output_surface_luma_address, luma_buffer.data(), luma_buffer.size());
// Chroma
const std::size_t half_height = frame_height / 2;
@ -233,8 +232,8 @@ void Vic::WriteYUVFrame(std::unique_ptr<FFmpeg::Frame> frame, const VicConfig& c
ASSERT(false);
break;
}
host1x.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(),
chroma_buffer.size());
host1x.GMMU().WriteBlock(output_surface_chroma_address, chroma_buffer.data(),
chroma_buffer.size());
}
} // namespace Host1x

229
src/video_core/memory_manager.cpp
View file

@ -7,25 +7,26 @@
#include "common/assert.h"
#include "common/logging/log.h"
#include "core/core.h"
#include "core/device_memory.h"
#include "core/hle/kernel/k_page_table.h"
#include "core/hle/kernel/k_process.h"
#include "video_core/guest_memory.h"
#include "video_core/host1x/host1x.h"
#include "video_core/invalidation_accumulator.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_base.h"
namespace Tegra {
using Core::Memory::GuestMemoryFlags;
using Tegra::Memory::GuestMemoryFlags;
std::atomic<size_t> MemoryManager::unique_identifier_generator{};
MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64 big_page_bits_,
u64 page_bits_)
: system{system_}, memory{system.ApplicationMemory()}, device_memory{system.DeviceMemory()},
address_space_bits{address_space_bits_}, page_bits{page_bits_}, big_page_bits{big_page_bits_},
entries{}, big_entries{}, page_table{address_space_bits, address_space_bits + page_bits - 38,
page_bits != big_page_bits ? page_bits : 0},
MemoryManager::MemoryManager(Core::System& system_, MaxwellDeviceMemoryManager& memory_,
u64 address_space_bits_, u64 big_page_bits_, u64 page_bits_)
: system{system_}, memory{memory_}, address_space_bits{address_space_bits_},
page_bits{page_bits_}, big_page_bits{big_page_bits_}, entries{}, big_entries{},
page_table{address_space_bits, address_space_bits + page_bits - 38,
page_bits != big_page_bits ? page_bits : 0},
kind_map{PTEKind::INVALID}, unique_identifier{unique_identifier_generator.fetch_add(
1, std::memory_order_acq_rel)},
accumulator{std::make_unique<VideoCommon::InvalidationAccumulator>()} {
@ -42,11 +43,16 @@ MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64
big_page_table_mask = big_page_table_size - 1;
big_entries.resize(big_page_table_size / 32, 0);
big_page_table_cpu.resize(big_page_table_size);
big_page_table_dev.resize(big_page_table_size);
big_page_continuous.resize(big_page_table_size / continuous_bits, 0);
entries.resize(page_table_size / 32, 0);
}
MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64 big_page_bits_,
u64 page_bits_)
: MemoryManager(system_, system_.Host1x().MemoryManager(), address_space_bits_, big_page_bits_,
page_bits_) {}
MemoryManager::~MemoryManager() = default;
template <bool is_big_page>
@ -100,7 +106,7 @@ inline void MemoryManager::SetBigPageContinuous(size_t big_page_index, bool valu
}
template <MemoryManager::EntryType entry_type>
GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size,
GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size,
PTEKind kind) {
[[maybe_unused]] u64 remaining_size{size};
if constexpr (entry_type == EntryType::Mapped) {
@ -114,9 +120,9 @@ GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cp
rasterizer->ModifyGPUMemory(unique_identifier, current_gpu_addr, page_size);
}
if constexpr (entry_type == EntryType::Mapped) {
const VAddr current_cpu_addr = cpu_addr + offset;
const DAddr current_dev_addr = dev_addr + offset;
const auto index = PageEntryIndex<false>(current_gpu_addr);
const u32 sub_value = static_cast<u32>(current_cpu_addr >> cpu_page_bits);
const u32 sub_value = static_cast<u32>(current_dev_addr >> cpu_page_bits);
page_table[index] = sub_value;
}
remaining_size -= page_size;
@ -126,7 +132,7 @@ GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cp
}
template <MemoryManager::EntryType entry_type>
GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr,
GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr,
size_t size, PTEKind kind) {
[[maybe_unused]] u64 remaining_size{size};
for (u64 offset{}; offset < size; offset += big_page_size) {
@ -137,20 +143,20 @@ GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr
rasterizer->ModifyGPUMemory(unique_identifier, current_gpu_addr, big_page_size);
}
if constexpr (entry_type == EntryType::Mapped) {
const VAddr current_cpu_addr = cpu_addr + offset;
const DAddr current_dev_addr = dev_addr + offset;
const auto index = PageEntryIndex<true>(current_gpu_addr);
const u32 sub_value = static_cast<u32>(current_cpu_addr >> cpu_page_bits);
big_page_table_cpu[index] = sub_value;
const u32 sub_value = static_cast<u32>(current_dev_addr >> cpu_page_bits);
big_page_table_dev[index] = sub_value;
const bool is_continuous = ([&] {
uintptr_t base_ptr{
reinterpret_cast<uintptr_t>(memory.GetPointerSilent(current_cpu_addr))};
reinterpret_cast<uintptr_t>(memory.GetPointer<u8>(current_dev_addr))};
if (base_ptr == 0) {
return false;
}
for (VAddr start_cpu = current_cpu_addr + page_size;
start_cpu < current_cpu_addr + big_page_size; start_cpu += page_size) {
for (DAddr start_cpu = current_dev_addr + page_size;
start_cpu < current_dev_addr + big_page_size; start_cpu += page_size) {
base_ptr += page_size;
auto next_ptr = reinterpret_cast<uintptr_t>(memory.GetPointerSilent(start_cpu));
auto next_ptr = reinterpret_cast<uintptr_t>(memory.GetPointer<u8>(start_cpu));
if (next_ptr == 0 || base_ptr != next_ptr) {
return false;
}
@ -172,12 +178,12 @@ void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_)
rasterizer = rasterizer_;
}
GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size, PTEKind kind,
GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, DAddr dev_addr, std::size_t size, PTEKind kind,
bool is_big_pages) {
if (is_big_pages) [[likely]] {
return BigPageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size, kind);
return BigPageTableOp<EntryType::Mapped>(gpu_addr, dev_addr, size, kind);
}
return PageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size, kind);
return PageTableOp<EntryType::Mapped>(gpu_addr, dev_addr, size, kind);
}
GPUVAddr MemoryManager::MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages) {
@ -202,7 +208,7 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
PageTableOp<EntryType::Free>(gpu_addr, 0, size, PTEKind::INVALID);
}
std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) const {
std::optional<DAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) const {
if (!IsWithinGPUAddressRange(gpu_addr)) [[unlikely]] {
return std::nullopt;
}
@ -211,17 +217,17 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) const {
return std::nullopt;
}
const VAddr cpu_addr_base = static_cast<VAddr>(page_table[PageEntryIndex<false>(gpu_addr)])
const DAddr dev_addr_base = static_cast<DAddr>(page_table[PageEntryIndex<false>(gpu_addr)])
<< cpu_page_bits;
return cpu_addr_base + (gpu_addr & page_mask);
return dev_addr_base + (gpu_addr & page_mask);
}
const VAddr cpu_addr_base =
static_cast<VAddr>(big_page_table_cpu[PageEntryIndex<true>(gpu_addr)]) << cpu_page_bits;
return cpu_addr_base + (gpu_addr & big_page_mask);
const DAddr dev_addr_base =
static_cast<DAddr>(big_page_table_dev[PageEntryIndex<true>(gpu_addr)]) << cpu_page_bits;
return dev_addr_base + (gpu_addr & big_page_mask);
}
std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr, std::size_t size) const {
std::optional<DAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr, std::size_t size) const {
size_t page_index{addr >> page_bits};
const size_t page_last{(addr + size + page_size - 1) >> page_bits};
while (page_index < page_last) {
@ -274,7 +280,7 @@ u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) {
return {};
}
return memory.GetPointer(*address);
return memory.GetPointer<u8>(*address);
}
const u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) const {
@ -283,7 +289,7 @@ const u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) const {
return {};
}
return memory.GetPointer(*address);
return memory.GetPointer<u8>(*address);
}
#ifdef _MSC_VER // no need for gcc / clang but msvc's compiler is more conservative with inlining.
@ -367,25 +373,25 @@ void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std:
dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
};
auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
if constexpr (is_safe) {
rasterizer->FlushRegion(cpu_addr_base, copy_amount, which);
rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
}
u8* physical = memory.GetPointer(cpu_addr_base);
u8* physical = memory.GetPointer<u8>(dev_addr_base);
std::memcpy(dest_buffer, physical, copy_amount);
dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
};
auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
if constexpr (is_safe) {
rasterizer->FlushRegion(cpu_addr_base, copy_amount, which);
rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
}
if (!IsBigPageContinuous(page_index)) [[unlikely]] {
memory.ReadBlockUnsafe(cpu_addr_base, dest_buffer, copy_amount);
memory.ReadBlockUnsafe(dev_addr_base, dest_buffer, copy_amount);
} else {
u8* physical = memory.GetPointer(cpu_addr_base);
u8* physical = memory.GetPointer<u8>(dev_addr_base);
std::memcpy(dest_buffer, physical, copy_amount);
}
dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
@ -416,25 +422,25 @@ void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffe
src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
};
auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
if constexpr (is_safe) {
rasterizer->InvalidateRegion(cpu_addr_base, copy_amount, which);
rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
}
u8* physical = memory.GetPointer(cpu_addr_base);
u8* physical = memory.GetPointer<u8>(dev_addr_base);
std::memcpy(physical, src_buffer, copy_amount);
src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
};
auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
if constexpr (is_safe) {
rasterizer->InvalidateRegion(cpu_addr_base, copy_amount, which);
rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
}
if (!IsBigPageContinuous(page_index)) [[unlikely]] {
memory.WriteBlockUnsafe(cpu_addr_base, src_buffer, copy_amount);
memory.WriteBlockUnsafe(dev_addr_base, src_buffer, copy_amount);
} else {
u8* physical = memory.GetPointer(cpu_addr_base);
u8* physical = memory.GetPointer<u8>(dev_addr_base);
std::memcpy(physical, src_buffer, copy_amount);
}
src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
@ -470,14 +476,14 @@ void MemoryManager::FlushRegion(GPUVAddr gpu_addr, size_t size,
[[maybe_unused]] std::size_t copy_amount) {};
auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
rasterizer->FlushRegion(cpu_addr_base, copy_amount, which);
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
};
auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
rasterizer->FlushRegion(cpu_addr_base, copy_amount, which);
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
};
auto flush_short_pages = [&](std::size_t page_index, std::size_t offset,
std::size_t copy_amount) {
@ -495,15 +501,15 @@ bool MemoryManager::IsMemoryDirty(GPUVAddr gpu_addr, size_t size,
[[maybe_unused]] std::size_t copy_amount) { return false; };
auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
result |= rasterizer->MustFlushRegion(cpu_addr_base, copy_amount, which);
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
result |= rasterizer->MustFlushRegion(dev_addr_base, copy_amount, which);
return result;
};
auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
result |= rasterizer->MustFlushRegion(cpu_addr_base, copy_amount, which);
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
result |= rasterizer->MustFlushRegion(dev_addr_base, copy_amount, which);
return result;
};
auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
@ -517,7 +523,7 @@ bool MemoryManager::IsMemoryDirty(GPUVAddr gpu_addr, size_t size,
}
size_t MemoryManager::MaxContinuousRange(GPUVAddr gpu_addr, size_t size) const {
std::optional<VAddr> old_page_addr{};
std::optional<DAddr> old_page_addr{};
size_t range_so_far = 0;
bool result{false};
auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset,
@ -526,24 +532,24 @@ size_t MemoryManager::MaxContinuousRange(GPUVAddr gpu_addr, size_t size) const {
return true;
};
auto short_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != cpu_addr_base) {
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != dev_addr_base) {
result = true;
return true;
}
range_so_far += copy_amount;
old_page_addr = {cpu_addr_base + copy_amount};
old_page_addr = {dev_addr_base + copy_amount};
return false;
};
auto big_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != cpu_addr_base) {
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != dev_addr_base) {
return true;
}
range_so_far += copy_amount;
old_page_addr = {cpu_addr_base + copy_amount};
old_page_addr = {dev_addr_base + copy_amount};
return false;
};
auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
@ -568,14 +574,14 @@ void MemoryManager::InvalidateRegion(GPUVAddr gpu_addr, size_t size,
[[maybe_unused]] std::size_t copy_amount) {};
auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
rasterizer->InvalidateRegion(cpu_addr_base, copy_amount, which);
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
};
auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
rasterizer->InvalidateRegion(cpu_addr_base, copy_amount, which);
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
};
auto invalidate_short_pages = [&](std::size_t page_index, std::size_t offset,
std::size_t copy_amount) {
@ -587,7 +593,7 @@ void MemoryManager::InvalidateRegion(GPUVAddr gpu_addr, size_t size,
void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size,
VideoCommon::CacheType which) {
Core::Memory::GpuGuestMemoryScoped<u8, GuestMemoryFlags::SafeReadWrite> data(
Tegra::Memory::GpuGuestMemoryScoped<u8, GuestMemoryFlags::SafeReadWrite> data(
*this, gpu_src_addr, size);
data.SetAddressAndSize(gpu_dest_addr, size);
FlushRegion(gpu_dest_addr, size, which);
@ -600,18 +606,18 @@ bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const {
const std::size_t page{(page_index & big_page_mask) + size};
return page <= big_page_size;
}
const std::size_t page{(gpu_addr & Core::Memory::YUZU_PAGEMASK) + size};
return page <= Core::Memory::YUZU_PAGESIZE;
const std::size_t page{(gpu_addr & Core::DEVICE_PAGEMASK) + size};
return page <= Core::DEVICE_PAGESIZE;
}
if (GetEntry<false>(gpu_addr) != EntryType::Mapped) {
return false;
}
const std::size_t page{(gpu_addr & Core::Memory::YUZU_PAGEMASK) + size};
return page <= Core::Memory::YUZU_PAGESIZE;
const std::size_t page{(gpu_addr & Core::DEVICE_PAGEMASK) + size};
return page <= Core::DEVICE_PAGESIZE;
}
bool MemoryManager::IsContinuousRange(GPUVAddr gpu_addr, std::size_t size) const {
std::optional<VAddr> old_page_addr{};
std::optional<DAddr> old_page_addr{};
bool result{true};
auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset,
std::size_t copy_amount) {
@ -619,23 +625,23 @@ bool MemoryManager::IsContinuousRange(GPUVAddr gpu_addr, std::size_t size) const
return true;
};
auto short_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != cpu_addr_base) {
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != dev_addr_base) {
result = false;
return true;
}
old_page_addr = {cpu_addr_base + copy_amount};
old_page_addr = {dev_addr_base + copy_amount};
return false;
};
auto big_check = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != cpu_addr_base) {
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
if (old_page_addr && *old_page_addr != dev_addr_base) {
result = false;
return true;
}
old_page_addr = {cpu_addr_base + copy_amount};
old_page_addr = {dev_addr_base + copy_amount};
return false;
};
auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
@ -678,11 +684,11 @@ template <bool is_gpu_address>
void MemoryManager::GetSubmappedRangeImpl(
GPUVAddr gpu_addr, std::size_t size,
boost::container::small_vector<
std::pair<std::conditional_t<is_gpu_address, GPUVAddr, VAddr>, std::size_t>, 32>& result)
std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>, 32>& result)
const {
std::optional<std::pair<std::conditional_t<is_gpu_address, GPUVAddr, VAddr>, std::size_t>>
std::optional<std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>>
last_segment{};
std::optional<VAddr> old_page_addr{};
std::optional<DAddr> old_page_addr{};
const auto split = [&last_segment, &result]([[maybe_unused]] std::size_t page_index,
[[maybe_unused]] std::size_t offset,
[[maybe_unused]] std::size_t copy_amount) {
@ -694,20 +700,20 @@ void MemoryManager::GetSubmappedRangeImpl(
const auto extend_size_big = [this, &split, &old_page_addr,
&last_segment](std::size_t page_index, std::size_t offset,
std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(big_page_table_cpu[page_index]) << cpu_page_bits) + offset;
const DAddr dev_addr_base =
(static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
if (old_page_addr) {
if (*old_page_addr != cpu_addr_base) {
if (*old_page_addr != dev_addr_base) {
split(0, 0, 0);
}
}
old_page_addr = {cpu_addr_base + copy_amount};
old_page_addr = {dev_addr_base + copy_amount};
if (!last_segment) {
if constexpr (is_gpu_address) {
const GPUVAddr new_base_addr = (page_index << big_page_bits) + offset;
last_segment = {new_base_addr, copy_amount};
} else {
last_segment = {cpu_addr_base, copy_amount};
last_segment = {dev_addr_base, copy_amount};
}
} else {
last_segment->second += copy_amount;
@ -716,20 +722,20 @@ void MemoryManager::GetSubmappedRangeImpl(
const auto extend_size_short = [this, &split, &old_page_addr,
&last_segment](std::size_t page_index, std::size_t offset,
std::size_t copy_amount) {
const VAddr cpu_addr_base =
(static_cast<VAddr>(page_table[page_index]) << cpu_page_bits) + offset;
const DAddr dev_addr_base =
(static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
if (old_page_addr) {
if (*old_page_addr != cpu_addr_base) {
if (*old_page_addr != dev_addr_base) {
split(0, 0, 0);
}
}
old_page_addr = {cpu_addr_base + copy_amount};
old_page_addr = {dev_addr_base + copy_amount};
if (!last_segment) {
if constexpr (is_gpu_address) {
const GPUVAddr new_base_addr = (page_index << page_bits) + offset;
last_segment = {new_base_addr, copy_amount};
} else {
last_segment = {cpu_addr_base, copy_amount};
last_segment = {dev_addr_base, copy_amount};
}
} else {
last_segment->second += copy_amount;
@ -756,9 +762,12 @@ void MemoryManager::FlushCaching() {
}
const u8* MemoryManager::GetSpan(const GPUVAddr src_addr, const std::size_t size) const {
auto cpu_addr = GpuToCpuAddress(src_addr);
if (cpu_addr) {
return memory.GetSpan(*cpu_addr, size);
if (!IsContinuousRange(src_addr, size)) {
return nullptr;
}
auto dev_addr = GpuToCpuAddress(src_addr);
if (dev_addr) {
return memory.GetSpan(*dev_addr, size);
}
return nullptr;
}
@ -767,9 +776,9 @@ u8* MemoryManager::GetSpan(const GPUVAddr src_addr, const std::size_t size) {
if (!IsContinuousRange(src_addr, size)) {
return nullptr;
}
auto cpu_addr = GpuToCpuAddress(src_addr);
if (cpu_addr) {
return memory.GetSpan(*cpu_addr, size);
auto dev_addr = GpuToCpuAddress(src_addr);
if (dev_addr) {
return memory.GetSpan(*dev_addr, size);
}
return nullptr;
}

36
src/video_core/memory_manager.h
View file

@ -15,8 +15,8 @@
#include "common/range_map.h"
#include "common/scratch_buffer.h"
#include "common/virtual_buffer.h"
#include "core/memory.h"
#include "video_core/cache_types.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/pte_kind.h"
namespace VideoCore {
@ -28,10 +28,6 @@ class InvalidationAccumulator;
}
namespace Core {
class DeviceMemory;
namespace Memory {
class Memory;
} // namespace Memory
class System;
} // namespace Core
@ -41,6 +37,9 @@ class MemoryManager final {
public:
explicit MemoryManager(Core::System& system_, u64 address_space_bits_ = 40,
u64 big_page_bits_ = 16, u64 page_bits_ = 12);
explicit MemoryManager(Core::System& system_, MaxwellDeviceMemoryManager& memory_,
u64 address_space_bits_ = 40, u64 big_page_bits_ = 16,
u64 page_bits_ = 12);
~MemoryManager();
size_t GetID() const {
@ -50,9 +49,9 @@ public:
/// Binds a renderer to the memory manager.
void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
[[nodiscard]] std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr) const;
[[nodiscard]] std::optional<DAddr> GpuToCpuAddress(GPUVAddr addr) const;
[[nodiscard]] std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr, std::size_t size) const;
[[nodiscard]] std::optional<DAddr> GpuToCpuAddress(GPUVAddr addr, std::size_t size) const;
template <typename T>
[[nodiscard]] T Read(GPUVAddr addr) const;
@ -69,7 +68,7 @@ public:
if (!address) {
return {};
}
return memory.GetPointer(*address);
return memory.GetPointer<T>(*address);
}
template <typename T>
@ -110,7 +109,7 @@ public:
[[nodiscard]] bool IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const;
/**
* Checks if a gpu region is mapped by a single range of cpu addresses.
* Checks if a gpu region is mapped by a single range of device addresses.
*/
[[nodiscard]] bool IsContinuousRange(GPUVAddr gpu_addr, std::size_t size) const;
@ -120,14 +119,14 @@ public:
[[nodiscard]] bool IsFullyMappedRange(GPUVAddr gpu_addr, std::size_t size) const;
/**
* Returns a vector with all the subranges of cpu addresses mapped beneath.
* Returns a vector with all the subranges of device addresses mapped beneath.
* if the region is continuous, a single pair will be returned. If it's unmapped, an empty
* vector will be returned;
*/
boost::container::small_vector<std::pair<GPUVAddr, std::size_t>, 32> GetSubmappedRange(
GPUVAddr gpu_addr, std::size_t size) const;
GPUVAddr Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size,
GPUVAddr Map(GPUVAddr gpu_addr, DAddr dev_addr, std::size_t size,
PTEKind kind = PTEKind::INVALID, bool is_big_pages = true);
GPUVAddr MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages = true);
void Unmap(GPUVAddr gpu_addr, std::size_t size);
@ -186,12 +185,11 @@ private:
void GetSubmappedRangeImpl(
GPUVAddr gpu_addr, std::size_t size,
boost::container::small_vector<
std::pair<std::conditional_t<is_gpu_address, GPUVAddr, VAddr>, std::size_t>, 32>&
std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>, 32>&
result) const;
Core::System& system;
Core::Memory::Memory& memory;
Core::DeviceMemory& device_memory;
MaxwellDeviceMemoryManager& memory;
const u64 address_space_bits;
const u64 page_bits;
@ -218,11 +216,11 @@ private:
std::vector<u64> big_entries;
template <EntryType entry_type>
GPUVAddr PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size,
GPUVAddr PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size,
PTEKind kind);
template <EntryType entry_type>
GPUVAddr BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size,
GPUVAddr BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size,
PTEKind kind);
template <bool is_big_page>
@ -233,11 +231,11 @@ private:
Common::MultiLevelPageTable<u32> page_table;
Common::RangeMap<GPUVAddr, PTEKind> kind_map;
Common::VirtualBuffer<u32> big_page_table_cpu;
Common::VirtualBuffer<u32> big_page_table_dev;
std::vector<u64> big_page_continuous;
boost::container::small_vector<std::pair<VAddr, std::size_t>, 32> page_stash{};
boost::container::small_vector<std::pair<VAddr, std::size_t>, 32> page_stash2{};
boost::container::small_vector<std::pair<DAddr, std::size_t>, 32> page_stash{};
boost::container::small_vector<std::pair<DAddr, std::size_t>, 32> page_stash2{};
mutable std::mutex guard;

30
src/video_core/query_cache.h
View file

@ -18,9 +18,9 @@
#include "common/assert.h"
#include "common/settings.h"
#include "core/memory.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/texture_cache/slot_vector.h"
@ -102,18 +102,19 @@ template <class QueryCache, class CachedQuery, class CounterStream, class HostCo
class QueryCacheLegacy : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
public:
explicit QueryCacheLegacy(VideoCore::RasterizerInterface& rasterizer_,
Core::Memory::Memory& cpu_memory_)
Tegra::MaxwellDeviceMemoryManager& device_memory_)
: rasterizer{rasterizer_},
// Use reinterpret_cast instead of static_cast as workaround for
// UBSan bug (https://github.com/llvm/llvm-project/issues/59060)
cpu_memory{cpu_memory_}, streams{{
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::SamplesPassed}},
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::PrimitivesGenerated}},
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::TfbPrimitivesWritten}},
}} {
device_memory{device_memory_},
streams{{
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::SamplesPassed}},
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::PrimitivesGenerated}},
{CounterStream{reinterpret_cast<QueryCache&>(*this),
VideoCore::QueryType::TfbPrimitivesWritten}},
}} {
(void)slot_async_jobs.insert(); // Null value
}
@ -322,13 +323,14 @@ private:
local_lock.unlock();
if (timestamp) {
u64 timestamp_value = *timestamp;
cpu_memory.WriteBlockUnsafe(address + sizeof(u64), &timestamp_value, sizeof(u64));
cpu_memory.WriteBlockUnsafe(address, &value, sizeof(u64));
device_memory.WriteBlockUnsafe(address + sizeof(u64), &timestamp_value,
sizeof(u64));
device_memory.WriteBlockUnsafe(address, &value, sizeof(u64));
rasterizer.InvalidateRegion(address, sizeof(u64) * 2,
VideoCommon::CacheType::NoQueryCache);
} else {
u32 small_value = static_cast<u32>(value);
cpu_memory.WriteBlockUnsafe(address, &small_value, sizeof(u32));
device_memory.WriteBlockUnsafe(address, &small_value, sizeof(u32));
rasterizer.InvalidateRegion(address, sizeof(u32),
VideoCommon::CacheType::NoQueryCache);
}
@ -342,7 +344,7 @@ private:
SlotVector<AsyncJob> slot_async_jobs;
VideoCore::RasterizerInterface& rasterizer;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
mutable std::recursive_mutex mutex;

4
src/video_core/query_cache/query_base.h
View file

@ -23,7 +23,7 @@ DECLARE_ENUM_FLAG_OPERATORS(QueryFlagBits)
class QueryBase {
public:
VAddr guest_address{};
DAddr guest_address{};
QueryFlagBits flags{};
u64 value{};
@ -32,7 +32,7 @@ protected:
QueryBase() = default;
// Parameterized constructor
QueryBase(VAddr address, QueryFlagBits flags_, u64 value_)
QueryBase(DAddr address, QueryFlagBits flags_, u64 value_)
: guest_address(address), flags(flags_), value{value_} {}
};

37
src/video_core/query_cache/query_cache.h
View file

@ -15,9 +15,9 @@
#include "common/logging/log.h"
#include "common/scope_exit.h"
#include "common/settings.h"
#include "core/memory.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/gpu.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/memory_manager.h"
#include "video_core/query_cache/bank_base.h"
#include "video_core/query_cache/query_base.h"
@ -113,9 +113,10 @@ struct QueryCacheBase<Traits>::QueryCacheBaseImpl {
using RuntimeType = typename Traits::RuntimeType;
QueryCacheBaseImpl(QueryCacheBase<Traits>* owner_, VideoCore::RasterizerInterface& rasterizer_,
Core::Memory::Memory& cpu_memory_, RuntimeType& runtime_, Tegra::GPU& gpu_)
Tegra::MaxwellDeviceMemoryManager& device_memory_, RuntimeType& runtime_,
Tegra::GPU& gpu_)
: owner{owner_}, rasterizer{rasterizer_},
cpu_memory{cpu_memory_}, runtime{runtime_}, gpu{gpu_} {
device_memory{device_memory_}, runtime{runtime_}, gpu{gpu_} {
streamer_mask = 0;
for (size_t i = 0; i < static_cast<size_t>(QueryType::MaxQueryTypes); i++) {
streamers[i] = runtime.GetStreamerInterface(static_cast<QueryType>(i));
@ -158,7 +159,7 @@ struct QueryCacheBase<Traits>::QueryCacheBaseImpl {
QueryCacheBase<Traits>* owner;
VideoCore::RasterizerInterface& rasterizer;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
RuntimeType& runtime;
Tegra::GPU& gpu;
std::array<StreamerInterface*, static_cast<size_t>(QueryType::MaxQueryTypes)> streamers;
@ -171,10 +172,11 @@ struct QueryCacheBase<Traits>::QueryCacheBaseImpl {
template <typename Traits>
QueryCacheBase<Traits>::QueryCacheBase(Tegra::GPU& gpu_,
VideoCore::RasterizerInterface& rasterizer_,
Core::Memory::Memory& cpu_memory_, RuntimeType& runtime_)
Tegra::MaxwellDeviceMemoryManager& device_memory_,
RuntimeType& runtime_)
: cached_queries{} {
impl = std::make_unique<QueryCacheBase<Traits>::QueryCacheBaseImpl>(
this, rasterizer_, cpu_memory_, runtime_, gpu_);
this, rasterizer_, device_memory_, runtime_, gpu_);
}
template <typename Traits>
@ -240,7 +242,7 @@ void QueryCacheBase<Traits>::CounterReport(GPUVAddr addr, QueryType counter_type
if (!cpu_addr_opt) [[unlikely]] {
return;
}
VAddr cpu_addr = *cpu_addr_opt;
DAddr cpu_addr = *cpu_addr_opt;
const size_t new_query_id = streamer->WriteCounter(cpu_addr, has_timestamp, payload, subreport);
auto* query = streamer->GetQuery(new_query_id);
if (is_fence) {
@ -250,13 +252,12 @@ void QueryCacheBase<Traits>::CounterReport(GPUVAddr addr, QueryType counter_type
query_location.stream_id.Assign(static_cast<u32>(streamer_id));
query_location.query_id.Assign(static_cast<u32>(new_query_id));
const auto gen_caching_indexing = [](VAddr cur_addr) {
return std::make_pair<u64, u32>(cur_addr >> Core::Memory::YUZU_PAGEBITS,
static_cast<u32>(cur_addr & Core::Memory::YUZU_PAGEMASK));
return std::make_pair<u64, u32>(cur_addr >> Core::DEVICE_PAGEBITS,
static_cast<u32>(cur_addr & Core::DEVICE_PAGEMASK));
};
u8* pointer = impl->cpu_memory.GetPointer(cpu_addr);
u8* pointer_timestamp = impl->cpu_memory.GetPointer(cpu_addr + 8);
u8* pointer = impl->device_memory.template GetPointer<u8>(cpu_addr);
u8* pointer_timestamp = impl->device_memory.template GetPointer<u8>(cpu_addr + 8);
bool is_synced = !Settings::IsGPULevelHigh() && is_fence;
std::function<void()> operation([this, is_synced, streamer, query_base = query, query_location,
pointer, pointer_timestamp] {
if (True(query_base->flags & QueryFlagBits::IsInvalidated)) {
@ -323,8 +324,8 @@ void QueryCacheBase<Traits>::CounterReport(GPUVAddr addr, QueryType counter_type
template <typename Traits>
void QueryCacheBase<Traits>::UnregisterPending() {
const auto gen_caching_indexing = [](VAddr cur_addr) {
return std::make_pair<u64, u32>(cur_addr >> Core::Memory::YUZU_PAGEBITS,
static_cast<u32>(cur_addr & Core::Memory::YUZU_PAGEMASK));
return std::make_pair<u64, u32>(cur_addr >> Core::DEVICE_PAGEBITS,
static_cast<u32>(cur_addr & Core::DEVICE_PAGEMASK));
};
std::scoped_lock lock(cache_mutex);
for (QueryLocation loc : impl->pending_unregister) {
@ -388,7 +389,7 @@ bool QueryCacheBase<Traits>::AccelerateHostConditionalRendering() {
}
VAddr cpu_addr = *cpu_addr_opt;
std::scoped_lock lock(cache_mutex);
auto it1 = cached_queries.find(cpu_addr >> Core::Memory::YUZU_PAGEBITS);
auto it1 = cached_queries.find(cpu_addr >> Core::DEVICE_PAGEBITS);
if (it1 == cached_queries.end()) {
return VideoCommon::LookupData{
.address = cpu_addr,
@ -396,10 +397,10 @@ bool QueryCacheBase<Traits>::AccelerateHostConditionalRendering() {
};
}
auto& sub_container = it1->second;
auto it_current = sub_container.find(cpu_addr & Core::Memory::YUZU_PAGEMASK);
auto it_current = sub_container.find(cpu_addr & Core::DEVICE_PAGEMASK);
if (it_current == sub_container.end()) {
auto it_current_2 = sub_container.find((cpu_addr & Core::Memory::YUZU_PAGEMASK) + 4);
auto it_current_2 = sub_container.find((cpu_addr & Core::DEVICE_PAGEMASK) + 4);
if (it_current_2 == sub_container.end()) {
return VideoCommon::LookupData{
.address = cpu_addr,
@ -559,7 +560,7 @@ bool QueryCacheBase<Traits>::SemiFlushQueryDirty(QueryCacheBase<Traits>::QueryLo
}
if (True(query_base->flags & QueryFlagBits::IsFinalValueSynced) &&
False(query_base->flags & QueryFlagBits::IsGuestSynced)) {
auto* ptr = impl->cpu_memory.GetPointer(query_base->guest_address);
auto* ptr = impl->device_memory.template GetPointer<u8>(query_base->guest_address);
if (True(query_base->flags & QueryFlagBits::HasTimestamp)) {
std::memcpy(ptr, &query_base->value, sizeof(query_base->value));
return false;

15
src/video_core/query_cache/query_cache_base.h
View file

@ -13,15 +13,11 @@
#include "common/assert.h"
#include "common/bit_field.h"
#include "common/common_types.h"
#include "core/memory.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/query_cache/query_base.h"
#include "video_core/query_cache/types.h"
namespace Core::Memory {
class Memory;
}
namespace VideoCore {
class RasterizerInterface;
}
@ -53,7 +49,8 @@ public:
};
explicit QueryCacheBase(Tegra::GPU& gpu, VideoCore::RasterizerInterface& rasterizer_,
Core::Memory::Memory& cpu_memory_, RuntimeType& runtime_);
Tegra::MaxwellDeviceMemoryManager& device_memory_,
RuntimeType& runtime_);
~QueryCacheBase();
@ -125,10 +122,10 @@ protected:
const u64 addr_begin = addr;
const u64 addr_end = addr_begin + size;
const u64 page_end = addr_end >> Core::Memory::YUZU_PAGEBITS;
const u64 page_end = addr_end >> Core::DEVICE_PAGEBITS;
std::scoped_lock lock(cache_mutex);
for (u64 page = addr_begin >> Core::Memory::YUZU_PAGEBITS; page <= page_end; ++page) {
const u64 page_start = page << Core::Memory::YUZU_PAGEBITS;
for (u64 page = addr_begin >> Core::DEVICE_PAGEBITS; page <= page_end; ++page) {
const u64 page_start = page << Core::DEVICE_PAGEBITS;
const auto in_range = [page_start, addr_begin, addr_end](const u32 query_location) {
const u64 cache_begin = page_start + query_location;
const u64 cache_end = cache_begin + sizeof(u32);

72
src/video_core/rasterizer_accelerated.cpp
View file

@ -1,72 +0,0 @@
// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <atomic>
#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 {
using namespace Core::Memory;
RasterizerAccelerated::RasterizerAccelerated(Memory& cpu_memory_)
: cached_pages(std::make_unique<CachedPages>()), cpu_memory{cpu_memory_} {}
RasterizerAccelerated::~RasterizerAccelerated() = default;
void RasterizerAccelerated::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {
u64 uncache_begin = 0;
u64 cache_begin = 0;
u64 uncache_bytes = 0;
u64 cache_bytes = 0;
std::atomic_thread_fence(std::memory_order_acquire);
const u64 page_end = Common::DivCeil(addr + size, YUZU_PAGESIZE);
for (u64 page = addr >> YUZU_PAGEBITS; page != page_end; ++page) {
std::atomic_uint16_t& count = cached_pages->at(page >> 2).Count(page);
if (delta > 0) {
ASSERT_MSG(count.load(std::memory_order::relaxed) < UINT16_MAX, "Count may overflow!");
} else if (delta < 0) {
ASSERT_MSG(count.load(std::memory_order::relaxed) > 0, "Count may underflow!");
} else {
ASSERT_MSG(false, "Delta must be non-zero!");
}
// Adds or subtracts 1, as count is a unsigned 8-bit value
count.fetch_add(static_cast<u16>(delta), std::memory_order_release);
// Assume delta is either -1 or 1
if (count.load(std::memory_order::relaxed) == 0) {
if (uncache_bytes == 0) {
uncache_begin = page;
}
uncache_bytes += YUZU_PAGESIZE;
} else if (uncache_bytes > 0) {
cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes,
false);
uncache_bytes = 0;
}
if (count.load(std::memory_order::relaxed) == 1 && delta > 0) {
if (cache_bytes == 0) {
cache_begin = page;
}
cache_bytes += YUZU_PAGESIZE;
} else if (cache_bytes > 0) {
cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true);
cache_bytes = 0;
}
}
if (uncache_bytes > 0) {
cpu_memory.RasterizerMarkRegionCached(uncache_begin << YUZU_PAGEBITS, uncache_bytes, false);
}
if (cache_bytes > 0) {
cpu_memory.RasterizerMarkRegionCached(cache_begin << YUZU_PAGEBITS, cache_bytes, true);
}
}
} // namespace VideoCore

49
src/video_core/rasterizer_accelerated.h
View file

@ -1,49 +0,0 @@
// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#pragma once
#include <array>
#include <atomic>
#include "common/common_types.h"
#include "video_core/rasterizer_interface.h"
namespace Core::Memory {
class Memory;
}
namespace VideoCore {
/// Implements the shared part in GPU accelerated rasterizers in RasterizerInterface.
class RasterizerAccelerated : public RasterizerInterface {
public:
explicit RasterizerAccelerated(Core::Memory::Memory& cpu_memory_);
~RasterizerAccelerated() override;
void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) override;
private:
class CacheEntry final {
public:
CacheEntry() = default;
std::atomic_uint16_t& Count(std::size_t page) {
return values[page & 3];
}
const std::atomic_uint16_t& Count(std::size_t page) const {
return values[page & 3];
}
private:
std::array<std::atomic_uint16_t, 4> values{};
};
static_assert(sizeof(CacheEntry) == 8, "CacheEntry should be 8 bytes!");
using CachedPages = std::array<CacheEntry, 0x2000000>;
std::unique_ptr<CachedPages> cached_pages;
Core::Memory::Memory& cpu_memory;
};
} // namespace VideoCore

23
src/video_core/rasterizer_interface.h
View file

@ -86,35 +86,35 @@ public:
virtual void FlushAll() = 0;
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
virtual void FlushRegion(VAddr addr, u64 size,
virtual void FlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) = 0;
/// Check if the the specified memory area requires flushing to CPU Memory.
virtual bool MustFlushRegion(VAddr addr, u64 size,
virtual bool MustFlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) = 0;
virtual RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) = 0;
virtual RasterizerDownloadArea GetFlushArea(DAddr addr, u64 size) = 0;
/// Notify rasterizer that any caches of the specified region should be invalidated
virtual void InvalidateRegion(VAddr addr, u64 size,
virtual void InvalidateRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) = 0;
virtual void InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) {
virtual void InnerInvalidation(std::span<const std::pair<DAddr, std::size_t>> sequences) {
for (const auto& [cpu_addr, size] : sequences) {
InvalidateRegion(cpu_addr, size);
}
}
/// Notify rasterizer that any caches of the specified region are desync with guest
virtual void OnCacheInvalidation(VAddr addr, u64 size) = 0;
virtual void OnCacheInvalidation(PAddr addr, u64 size) = 0;
virtual bool OnCPUWrite(VAddr addr, u64 size) = 0;
virtual bool OnCPUWrite(PAddr addr, u64 size) = 0;
/// Sync memory between guest and host.
virtual void InvalidateGPUCache() = 0;
/// Unmap memory range
virtual void UnmapMemory(VAddr addr, u64 size) = 0;
virtual void UnmapMemory(DAddr addr, u64 size) = 0;
/// Remap GPU memory range. This means underneath backing memory changed
virtual void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) = 0;
@ -122,7 +122,7 @@ public:
/// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
/// and invalidated
virtual void FlushAndInvalidateRegion(
VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) = 0;
DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) = 0;
/// Notify the host renderer to wait for previous primitive and compute operations.
virtual void WaitForIdle() = 0;
@ -157,13 +157,10 @@ public:
/// Attempt to use a faster method to display the framebuffer to screen
[[nodiscard]] virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config,
VAddr framebuffer_addr, u32 pixel_stride) {
DAddr framebuffer_addr, u32 pixel_stride) {
return false;
}
/// Increase/decrease the number of object in pages touching the specified region
virtual void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {}
/// Initialize disk cached resources for the game being emulated
virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const DiskResourceLoadCallback& callback) {}

26
src/video_core/renderer_null/null_rasterizer.cpp
View file

@ -2,7 +2,6 @@
// SPDX-License-Identifier: GPL-2.0-or-later
#include "common/alignment.h"
#include "core/memory.h"
#include "video_core/control/channel_state.h"
#include "video_core/host1x/host1x.h"
#include "video_core/memory_manager.h"
@ -19,8 +18,7 @@ bool AccelerateDMA::BufferClear(GPUVAddr src_address, u64 amount, u32 value) {
return true;
}
RasterizerNull::RasterizerNull(Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu)
: RasterizerAccelerated(cpu_memory_), m_gpu{gpu} {}
RasterizerNull::RasterizerNull(Tegra::GPU& gpu) : m_gpu{gpu} {}
RasterizerNull::~RasterizerNull() = default;
void RasterizerNull::Draw(bool is_indexed, u32 instance_count) {}
@ -45,25 +43,25 @@ void RasterizerNull::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr
u32 size) {}
void RasterizerNull::DisableGraphicsUniformBuffer(size_t stage, u32 index) {}
void RasterizerNull::FlushAll() {}
void RasterizerNull::FlushRegion(VAddr addr, u64 size, VideoCommon::CacheType) {}
bool RasterizerNull::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType) {
void RasterizerNull::FlushRegion(DAddr addr, u64 size, VideoCommon::CacheType) {}
bool RasterizerNull::MustFlushRegion(DAddr addr, u64 size, VideoCommon::CacheType) {
return false;
}
void RasterizerNull::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType) {}
bool RasterizerNull::OnCPUWrite(VAddr addr, u64 size) {
void RasterizerNull::InvalidateRegion(DAddr addr, u64 size, VideoCommon::CacheType) {}
bool RasterizerNull::OnCPUWrite(PAddr addr, u64 size) {
return false;
}
void RasterizerNull::OnCacheInvalidation(VAddr addr, u64 size) {}
VideoCore::RasterizerDownloadArea RasterizerNull::GetFlushArea(VAddr addr, u64 size) {
void RasterizerNull::OnCacheInvalidation(PAddr addr, u64 size) {}
VideoCore::RasterizerDownloadArea RasterizerNull::GetFlushArea(PAddr addr, u64 size) {
VideoCore::RasterizerDownloadArea new_area{
.start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE),
.end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE),
.start_address = Common::AlignDown(addr, Core::DEVICE_PAGESIZE),
.end_address = Common::AlignUp(addr + size, Core::DEVICE_PAGESIZE),
.preemtive = true,
};
return new_area;
}
void RasterizerNull::InvalidateGPUCache() {}
void RasterizerNull::UnmapMemory(VAddr addr, u64 size) {}
void RasterizerNull::UnmapMemory(DAddr addr, u64 size) {}
void RasterizerNull::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) {}
void RasterizerNull::SignalFence(std::function<void()>&& func) {
func();
@ -78,7 +76,7 @@ void RasterizerNull::SignalSyncPoint(u32 value) {
}
void RasterizerNull::SignalReference() {}
void RasterizerNull::ReleaseFences(bool) {}
void RasterizerNull::FlushAndInvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType) {}
void RasterizerNull::FlushAndInvalidateRegion(DAddr addr, u64 size, VideoCommon::CacheType) {}
void RasterizerNull::WaitForIdle() {}
void RasterizerNull::FragmentBarrier() {}
void RasterizerNull::TiledCacheBarrier() {}
@ -95,7 +93,7 @@ bool RasterizerNull::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surfac
void RasterizerNull::AccelerateInlineToMemory(GPUVAddr address, size_t copy_size,
std::span<const u8> memory) {}
bool RasterizerNull::AccelerateDisplay(const Tegra::FramebufferConfig& config,
VAddr framebuffer_addr, u32 pixel_stride) {
DAddr framebuffer_addr, u32 pixel_stride) {
return true;
}
void RasterizerNull::LoadDiskResources(u64 title_id, std::stop_token stop_loading,

23
src/video_core/renderer_null/null_rasterizer.h
View file

@ -6,7 +6,6 @@
#include "common/common_types.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/engines/maxwell_dma.h"
#include "video_core/rasterizer_accelerated.h"
#include "video_core/rasterizer_interface.h"
namespace Core {
@ -32,10 +31,10 @@ public:
}
};
class RasterizerNull final : public VideoCore::RasterizerAccelerated,
class RasterizerNull final : public VideoCore::RasterizerInterface,
protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
public:
explicit RasterizerNull(Core::Memory::Memory& cpu_memory, Tegra::GPU& gpu);
explicit RasterizerNull(Tegra::GPU& gpu);
~RasterizerNull() override;
void Draw(bool is_indexed, u32 instance_count) override;
@ -48,17 +47,17 @@ public:
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
void DisableGraphicsUniformBuffer(size_t stage, u32 index) override;
void FlushAll() override;
void FlushRegion(VAddr addr, u64 size,
void FlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
bool MustFlushRegion(VAddr addr, u64 size,
bool MustFlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void InvalidateRegion(VAddr addr, u64 size,
void InvalidateRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void OnCacheInvalidation(VAddr addr, u64 size) override;
bool OnCPUWrite(VAddr addr, u64 size) override;
VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override;
void OnCacheInvalidation(DAddr addr, u64 size) override;
bool OnCPUWrite(DAddr addr, u64 size) override;
VideoCore::RasterizerDownloadArea GetFlushArea(DAddr addr, u64 size) override;
void InvalidateGPUCache() override;
void UnmapMemory(VAddr addr, u64 size) override;
void UnmapMemory(DAddr addr, u64 size) override;
void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override;
void SignalFence(std::function<void()>&& func) override;
void SyncOperation(std::function<void()>&& func) override;
@ -66,7 +65,7 @@ public:
void SignalReference() override;
void ReleaseFences(bool force) override;
void FlushAndInvalidateRegion(
VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void WaitForIdle() override;
void FragmentBarrier() override;
void TiledCacheBarrier() override;
@ -78,7 +77,7 @@ public:
Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override;
void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size,
std::span<const u8> memory) override;
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, DAddr framebuffer_addr,
u32 pixel_stride) override;
void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) override;

5
src/video_core/renderer_null/renderer_null.cpp
View file

@ -7,10 +7,9 @@
namespace Null {
RendererNull::RendererNull(Core::Frontend::EmuWindow& emu_window, Core::Memory::Memory& cpu_memory,
Tegra::GPU& gpu,
RendererNull::RendererNull(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu,
std::unique_ptr<Core::Frontend::GraphicsContext> context_)
: RendererBase(emu_window, std::move(context_)), m_gpu(gpu), m_rasterizer(cpu_memory, gpu) {}
: RendererBase(emu_window, std::move(context_)), m_gpu(gpu), m_rasterizer(gpu) {}
RendererNull::~RendererNull() = default;

3
src/video_core/renderer_null/renderer_null.h
View file

@ -13,8 +13,7 @@ namespace Null {
class RendererNull final : public VideoCore::RendererBase {
public:
explicit RendererNull(Core::Frontend::EmuWindow& emu_window, Core::Memory::Memory& cpu_memory,
Tegra::GPU& gpu,
explicit RendererNull(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu,
std::unique_ptr<Core::Frontend::GraphicsContext> context);
~RendererNull() override;

7
src/video_core/renderer_opengl/gl_buffer_cache.cpp
View file

@ -47,11 +47,10 @@ constexpr std::array PROGRAM_LUT{
} // Anonymous namespace
Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {}
: VideoCommon::BufferBase(null_params) {}
Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
VAddr cpu_addr_, u64 size_bytes_)
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) {
Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_)
: VideoCommon::BufferBase(cpu_addr_, size_bytes_) {
buffer.Create();
if (runtime.device.HasDebuggingToolAttached()) {
const std::string name = fmt::format("Buffer 0x{:x}", CpuAddr());

8
src/video_core/renderer_opengl/gl_buffer_cache.h
View file

@ -10,7 +10,6 @@
#include "common/common_types.h"
#include "video_core/buffer_cache/buffer_cache_base.h"
#include "video_core/buffer_cache/memory_tracker_base.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_staging_buffer_pool.h"
@ -19,10 +18,9 @@ namespace OpenGL {
class BufferCacheRuntime;
class Buffer : public VideoCommon::BufferBase<VideoCore::RasterizerInterface> {
class Buffer : public VideoCommon::BufferBase {
public:
explicit Buffer(BufferCacheRuntime&, VideoCore::RasterizerInterface& rasterizer, VAddr cpu_addr,
u64 size_bytes);
explicit Buffer(BufferCacheRuntime&, DAddr cpu_addr, u64 size_bytes);
explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams);
void ImmediateUpload(size_t offset, std::span<const u8> data) noexcept;
@ -244,7 +242,7 @@ struct BufferCacheParams {
using Runtime = OpenGL::BufferCacheRuntime;
using Buffer = OpenGL::Buffer;
using Async_Buffer = OpenGL::StagingBufferMap;
using MemoryTracker = VideoCommon::MemoryTrackerBase<VideoCore::RasterizerInterface>;
using MemoryTracker = VideoCommon::MemoryTrackerBase<Tegra::MaxwellDeviceMemoryManager>;
static constexpr bool IS_OPENGL = true;
static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = true;

5
src/video_core/renderer_opengl/gl_query_cache.cpp
View file

@ -35,8 +35,9 @@ constexpr GLenum GetTarget(VideoCore::QueryType type) {
} // Anonymous namespace
QueryCache::QueryCache(RasterizerOpenGL& rasterizer_, Core::Memory::Memory& cpu_memory_)
: QueryCacheLegacy(rasterizer_, cpu_memory_), gl_rasterizer{rasterizer_} {
QueryCache::QueryCache(RasterizerOpenGL& rasterizer_,
Tegra::MaxwellDeviceMemoryManager& device_memory_)
: QueryCacheLegacy(rasterizer_, device_memory_), gl_rasterizer{rasterizer_} {
EnableCounters();
}

4
src/video_core/renderer_opengl/gl_query_cache.h
View file

@ -8,6 +8,7 @@
#include <vector>
#include "common/common_types.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/query_cache.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_opengl/gl_resource_manager.h"
@ -28,7 +29,8 @@ using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>;
class QueryCache final
: public VideoCommon::QueryCacheLegacy<QueryCache, CachedQuery, CounterStream, HostCounter> {
public:
explicit QueryCache(RasterizerOpenGL& rasterizer_, Core::Memory::Memory& cpu_memory_);
explicit QueryCache(RasterizerOpenGL& rasterizer_,
Tegra::MaxwellDeviceMemoryManager& device_memory_);
~QueryCache();
OGLQuery AllocateQuery(VideoCore::QueryType type);

41
src/video_core/renderer_opengl/gl_rasterizer.cpp
View file

@ -70,18 +70,18 @@ std::optional<VideoCore::QueryType> MaxwellToVideoCoreQuery(VideoCommon::QueryTy
} // Anonymous namespace
RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
Core::Memory::Memory& cpu_memory_, const Device& device_,
ScreenInfo& screen_info_, ProgramManager& program_manager_,
StateTracker& state_tracker_)
: RasterizerAccelerated(cpu_memory_), gpu(gpu_), device(device_), screen_info(screen_info_),
Tegra::MaxwellDeviceMemoryManager& device_memory_,
const Device& device_, ScreenInfo& screen_info_,
ProgramManager& program_manager_, StateTracker& state_tracker_)
: gpu(gpu_), device_memory(device_memory_), device(device_), screen_info(screen_info_),
program_manager(program_manager_), state_tracker(state_tracker_),
texture_cache_runtime(device, program_manager, state_tracker, staging_buffer_pool),
texture_cache(texture_cache_runtime, *this),
texture_cache(texture_cache_runtime, device_memory_),
buffer_cache_runtime(device, staging_buffer_pool),
buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager,
state_tracker, gpu.ShaderNotify()),
query_cache(*this, cpu_memory_), accelerate_dma(buffer_cache, texture_cache),
buffer_cache(device_memory_, buffer_cache_runtime),
shader_cache(device_memory_, emu_window_, device, texture_cache, buffer_cache,
program_manager, state_tracker, gpu.ShaderNotify()),
query_cache(*this, device_memory_), accelerate_dma(buffer_cache, texture_cache),
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
blit_image(program_manager_) {}
@ -475,7 +475,7 @@ void RasterizerOpenGL::DisableGraphicsUniformBuffer(size_t stage, u32 index) {
void RasterizerOpenGL::FlushAll() {}
void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
void RasterizerOpenGL::FlushRegion(DAddr addr, u64 size, VideoCommon::CacheType which) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
if (addr == 0 || size == 0) {
return;
@ -493,7 +493,7 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size, VideoCommon::CacheType
}
}
bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
bool RasterizerOpenGL::MustFlushRegion(DAddr addr, u64 size, VideoCommon::CacheType which) {
if ((True(which & VideoCommon::CacheType::BufferCache))) {
std::scoped_lock lock{buffer_cache.mutex};
if (buffer_cache.IsRegionGpuModified(addr, size)) {
@ -510,7 +510,7 @@ bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheT
return false;
}
VideoCore::RasterizerDownloadArea RasterizerOpenGL::GetFlushArea(VAddr addr, u64 size) {
VideoCore::RasterizerDownloadArea RasterizerOpenGL::GetFlushArea(DAddr addr, u64 size) {
{
std::scoped_lock lock{texture_cache.mutex};
auto area = texture_cache.GetFlushArea(addr, size);
@ -526,14 +526,14 @@ VideoCore::RasterizerDownloadArea RasterizerOpenGL::GetFlushArea(VAddr addr, u64
}
}
VideoCore::RasterizerDownloadArea new_area{
.start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE),
.end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE),
.start_address = Common::AlignDown(addr, Core::DEVICE_PAGESIZE),
.end_address = Common::AlignUp(addr + size, Core::DEVICE_PAGESIZE),
.preemtive = true,
};
return new_area;
}
void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
void RasterizerOpenGL::InvalidateRegion(DAddr addr, u64 size, VideoCommon::CacheType which) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
if (addr == 0 || size == 0) {
return;
@ -554,7 +554,7 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache
}
}
bool RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
bool RasterizerOpenGL::OnCPUWrite(DAddr addr, u64 size) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
if (addr == 0 || size == 0) {
return false;
@ -576,8 +576,9 @@ bool RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
return false;
}
void RasterizerOpenGL::OnCacheInvalidation(VAddr addr, u64 size) {
void RasterizerOpenGL::OnCacheInvalidation(DAddr addr, u64 size) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
if (addr == 0 || size == 0) {
return;
}
@ -596,7 +597,7 @@ void RasterizerOpenGL::InvalidateGPUCache() {
gpu.InvalidateGPUCache();
}
void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) {
void RasterizerOpenGL::UnmapMemory(DAddr addr, u64 size) {
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.UnmapMemory(addr, size);
@ -635,7 +636,7 @@ void RasterizerOpenGL::ReleaseFences(bool force) {
fence_manager.WaitPendingFences(force);
}
void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size,
void RasterizerOpenGL::FlushAndInvalidateRegion(DAddr addr, u64 size,
VideoCommon::CacheType which) {
if (Settings::IsGPULevelExtreme()) {
FlushRegion(addr, size, which);
@ -739,7 +740,7 @@ void RasterizerOpenGL::AccelerateInlineToMemory(GPUVAddr address, size_t copy_si
}
bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
VAddr framebuffer_addr, u32 pixel_stride) {
DAddr framebuffer_addr, u32 pixel_stride) {
if (framebuffer_addr == 0) {
return false;
}

28
src/video_core/renderer_opengl/gl_rasterizer.h
View file

@ -14,7 +14,6 @@
#include "common/common_types.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/engines/maxwell_dma.h"
#include "video_core/rasterizer_accelerated.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_opengl/blit_image.h"
#include "video_core/renderer_opengl/gl_buffer_cache.h"
@ -72,13 +71,13 @@ private:
TextureCache& texture_cache;
};
class RasterizerOpenGL : public VideoCore::RasterizerAccelerated,
class RasterizerOpenGL : public VideoCore::RasterizerInterface,
protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
public:
explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
Core::Memory::Memory& cpu_memory_, const Device& device_,
ScreenInfo& screen_info_, ProgramManager& program_manager_,
StateTracker& state_tracker_);
Tegra::MaxwellDeviceMemoryManager& device_memory_,
const Device& device_, ScreenInfo& screen_info_,
ProgramManager& program_manager_, StateTracker& state_tracker_);
~RasterizerOpenGL() override;
void Draw(bool is_indexed, u32 instance_count) override;
@ -92,17 +91,17 @@ public:
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
void DisableGraphicsUniformBuffer(size_t stage, u32 index) override;
void FlushAll() override;
void FlushRegion(VAddr addr, u64 size,
void FlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
bool MustFlushRegion(VAddr addr, u64 size,
bool MustFlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override;
void InvalidateRegion(VAddr addr, u64 size,
VideoCore::RasterizerDownloadArea GetFlushArea(PAddr addr, u64 size) override;
void InvalidateRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void OnCacheInvalidation(VAddr addr, u64 size) override;
bool OnCPUWrite(VAddr addr, u64 size) override;
void OnCacheInvalidation(PAddr addr, u64 size) override;
bool OnCPUWrite(PAddr addr, u64 size) override;
void InvalidateGPUCache() override;
void UnmapMemory(VAddr addr, u64 size) override;
void UnmapMemory(DAddr addr, u64 size) override;
void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override;
void SignalFence(std::function<void()>&& func) override;
void SyncOperation(std::function<void()>&& func) override;
@ -110,7 +109,7 @@ public:
void SignalReference() override;
void ReleaseFences(bool force = true) override;
void FlushAndInvalidateRegion(
VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void WaitForIdle() override;
void FragmentBarrier() override;
void TiledCacheBarrier() override;
@ -123,7 +122,7 @@ public:
Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override;
void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size,
std::span<const u8> memory) override;
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, DAddr framebuffer_addr,
u32 pixel_stride) override;
void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) override;
@ -235,6 +234,7 @@ private:
VideoCommon::QueryPropertiesFlags flags, u32 payload, u32 subreport);
Tegra::GPU& gpu;
Tegra::MaxwellDeviceMemoryManager& device_memory;
const Device& device;
ScreenInfo& screen_info;

11
src/video_core/renderer_opengl/gl_shader_cache.cpp
View file

@ -168,11 +168,12 @@ void SetXfbState(VideoCommon::TransformFeedbackState& state, const Maxwell& regs
}
} // Anonymous namespace
ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_,
const Device& device_, TextureCache& texture_cache_,
BufferCache& buffer_cache_, ProgramManager& program_manager_,
StateTracker& state_tracker_, VideoCore::ShaderNotify& shader_notify_)
: VideoCommon::ShaderCache{rasterizer_}, emu_window{emu_window_}, device{device_},
ShaderCache::ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
Core::Frontend::EmuWindow& emu_window_, const Device& device_,
TextureCache& texture_cache_, BufferCache& buffer_cache_,
ProgramManager& program_manager_, StateTracker& state_tracker_,
VideoCore::ShaderNotify& shader_notify_)
: VideoCommon::ShaderCache{device_memory_}, emu_window{emu_window_}, device{device_},
texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, program_manager{program_manager_},
state_tracker{state_tracker_}, shader_notify{shader_notify_},
use_asynchronous_shaders{device.UseAsynchronousShaders()},

11
src/video_core/renderer_opengl/gl_shader_cache.h
View file

@ -17,7 +17,7 @@
namespace Tegra {
class MemoryManager;
}
} // namespace Tegra
namespace OpenGL {
@ -28,10 +28,11 @@ using ShaderWorker = Common::StatefulThreadWorker<ShaderContext::Context>;
class ShaderCache : public VideoCommon::ShaderCache {
public:
explicit ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_,
const Device& device_, TextureCache& texture_cache_,
BufferCache& buffer_cache_, ProgramManager& program_manager_,
StateTracker& state_tracker_, VideoCore::ShaderNotify& shader_notify_);
explicit ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
Core::Frontend::EmuWindow& emu_window_, const Device& device_,
TextureCache& texture_cache_, BufferCache& buffer_cache_,
ProgramManager& program_manager_, StateTracker& state_tracker_,
VideoCore::ShaderNotify& shader_notify_);
~ShaderCache();
void LoadDiskResources(u64 title_id, std::stop_token stop_loading,

10
src/video_core/renderer_opengl/renderer_opengl.cpp
View file

@ -15,7 +15,6 @@
#include "common/telemetry.h"
#include "core/core_timing.h"
#include "core/frontend/emu_window.h"
#include "core/memory.h"
#include "core/telemetry_session.h"
#include "video_core/host_shaders/ffx_a_h.h"
#include "video_core/host_shaders/ffx_fsr1_h.h"
@ -144,12 +143,13 @@ void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum severit
RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_,
Core::Frontend::EmuWindow& emu_window_,
Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_,
Tegra::MaxwellDeviceMemoryManager& device_memory_, Tegra::GPU& gpu_,
std::unique_ptr<Core::Frontend::GraphicsContext> context_)
: RendererBase{emu_window_, std::move(context_)}, telemetry_session{telemetry_session_},
emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, device{emu_window_},
emu_window{emu_window_}, device_memory{device_memory_}, gpu{gpu_}, device{emu_window_},
state_tracker{}, program_manager{device},
rasterizer(emu_window, gpu, cpu_memory, device, screen_info, program_manager, state_tracker) {
rasterizer(emu_window, gpu, device_memory, device, screen_info, program_manager,
state_tracker) {
if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) {
glEnable(GL_DEBUG_OUTPUT);
glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
@ -242,7 +242,7 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
const u32 bytes_per_pixel{VideoCore::Surface::BytesPerBlock(pixel_format)};
const u64 size_in_bytes{Tegra::Texture::CalculateSize(
true, bytes_per_pixel, framebuffer.stride, framebuffer.height, 1, block_height_log2, 0)};
const u8* const host_ptr{cpu_memory.GetPointer(framebuffer_addr)};
const u8* const host_ptr{device_memory.GetPointer<u8>(framebuffer_addr)};
const std::span<const u8> input_data(host_ptr, size_in_bytes);
Tegra::Texture::UnswizzleTexture(gl_framebuffer_data, input_data, bytes_per_pixel,
framebuffer.width, framebuffer.height, 1, block_height_log2,

4
src/video_core/renderer_opengl/renderer_opengl.h
View file

@ -61,7 +61,7 @@ class RendererOpenGL final : public VideoCore::RendererBase {
public:
explicit RendererOpenGL(Core::TelemetrySession& telemetry_session_,
Core::Frontend::EmuWindow& emu_window_,
Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_,
Tegra::MaxwellDeviceMemoryManager& device_memory_, Tegra::GPU& gpu_,
std::unique_ptr<Core::Frontend::GraphicsContext> context_);
~RendererOpenGL() override;
@ -101,7 +101,7 @@ private:
Core::TelemetrySession& telemetry_session;
Core::Frontend::EmuWindow& emu_window;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
Tegra::GPU& gpu;
Device device;

1
src/video_core/renderer_vulkan/pipeline_helper.h
View file

@ -12,7 +12,6 @@
#include "shader_recompiler/shader_info.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/texture_cache/texture_cache.h"
#include "video_core/texture_cache/types.h"
#include "video_core/vulkan_common/vulkan_device.h"

12
src/video_core/renderer_vulkan/renderer_vulkan.cpp
View file

@ -82,10 +82,10 @@ Device CreateDevice(const vk::Instance& instance, const vk::InstanceDispatch& dl
RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
Core::Frontend::EmuWindow& emu_window,
Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_,
Tegra::MaxwellDeviceMemoryManager& device_memory_, Tegra::GPU& gpu_,
std::unique_ptr<Core::Frontend::GraphicsContext> context_) try
: RendererBase(emu_window, std::move(context_)), telemetry_session(telemetry_session_),
cpu_memory(cpu_memory_), gpu(gpu_), library(OpenLibrary(context.get())),
device_memory(device_memory_), gpu(gpu_), library(OpenLibrary(context.get())),
instance(CreateInstance(*library, dld, VK_API_VERSION_1_1, render_window.GetWindowInfo().type,
Settings::values.renderer_debug.GetValue())),
debug_messenger(Settings::values.renderer_debug ? CreateDebugUtilsCallback(instance)
@ -97,9 +97,9 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
render_window.GetFramebufferLayout().height),
present_manager(instance, render_window, device, memory_allocator, scheduler, swapchain,
surface),
blit_screen(cpu_memory, render_window, device, memory_allocator, swapchain, present_manager,
scheduler, screen_info),
rasterizer(render_window, gpu, cpu_memory, screen_info, device, memory_allocator,
blit_screen(device_memory, render_window, device, memory_allocator, swapchain,
present_manager, scheduler, screen_info),
rasterizer(render_window, gpu, device_memory, screen_info, device, memory_allocator,
state_tracker, scheduler) {
if (Settings::values.renderer_force_max_clock.GetValue() && device.ShouldBoostClocks()) {
turbo_mode.emplace(instance, dld);
@ -128,7 +128,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
screen_info.width = framebuffer->width;
screen_info.height = framebuffer->height;
const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset;
const DAddr framebuffer_addr = framebuffer->address + framebuffer->offset;
const bool use_accelerated =
rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride);
RenderScreenshot(*framebuffer, use_accelerated);

8
src/video_core/renderer_vulkan/renderer_vulkan.h
View file

@ -7,12 +7,12 @@
#include <string>
#include <variant>
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "common/dynamic_library.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/renderer_base.h"
#include "video_core/renderer_vulkan/vk_blit_screen.h"
#include "video_core/renderer_vulkan/vk_present_manager.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_state_tracker.h"
#include "video_core/renderer_vulkan/vk_swapchain.h"
@ -42,7 +42,7 @@ class RendererVulkan final : public VideoCore::RendererBase {
public:
explicit RendererVulkan(Core::TelemetrySession& telemtry_session,
Core::Frontend::EmuWindow& emu_window,
Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_,
Tegra::MaxwellDeviceMemoryManager& device_memory_, Tegra::GPU& gpu_,
std::unique_ptr<Core::Frontend::GraphicsContext> context_);
~RendererVulkan() override;
@ -62,7 +62,7 @@ private:
void RenderScreenshot(const Tegra::FramebufferConfig& framebuffer, bool use_accelerated);
Core::TelemetrySession& telemetry_session;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
Tegra::GPU& gpu;
std::shared_ptr<Common::DynamicLibrary> library;

17
src/video_core/renderer_vulkan/vk_blit_screen.cpp
View file

@ -14,8 +14,8 @@
#include "common/settings.h"
#include "core/core.h"
#include "core/frontend/emu_window.h"
#include "core/memory.h"
#include "video_core/gpu.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/host_shaders/fxaa_frag_spv.h"
#include "video_core/host_shaders/fxaa_vert_spv.h"
#include "video_core/host_shaders/present_bicubic_frag_spv.h"
@ -121,11 +121,12 @@ struct BlitScreen::BufferData {
// Unaligned image data goes here
};
BlitScreen::BlitScreen(Core::Memory::Memory& cpu_memory_, Core::Frontend::EmuWindow& render_window_,
const Device& device_, MemoryAllocator& memory_allocator_,
Swapchain& swapchain_, PresentManager& present_manager_,
Scheduler& scheduler_, const ScreenInfo& screen_info_)
: cpu_memory{cpu_memory_}, render_window{render_window_}, device{device_},
BlitScreen::BlitScreen(Tegra::MaxwellDeviceMemoryManager& device_memory_,
Core::Frontend::EmuWindow& render_window_, const Device& device_,
MemoryAllocator& memory_allocator_, Swapchain& swapchain_,
PresentManager& present_manager_, Scheduler& scheduler_,
const ScreenInfo& screen_info_)
: device_memory{device_memory_}, render_window{render_window_}, device{device_},
memory_allocator{memory_allocator_}, swapchain{swapchain_}, present_manager{present_manager_},
scheduler{scheduler_}, image_count{swapchain.GetImageCount()}, screen_info{screen_info_} {
resource_ticks.resize(image_count);
@ -219,8 +220,8 @@ void BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
if (!use_accelerated) {
const u64 image_offset = GetRawImageOffset(framebuffer);
const VAddr framebuffer_addr = framebuffer.address + framebuffer.offset;
const u8* const host_ptr = cpu_memory.GetPointer(framebuffer_addr);
const DAddr framebuffer_addr = framebuffer.address + framebuffer.offset;
const u8* const host_ptr = device_memory.GetPointer<u8>(framebuffer_addr);
// TODO(Rodrigo): Read this from HLE
constexpr u32 block_height_log2 = 4;

12
src/video_core/renderer_vulkan/vk_blit_screen.h
View file

@ -6,6 +6,7 @@
#include <memory>
#include "core/frontend/framebuffer_layout.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@ -13,10 +14,6 @@ namespace Core {
class System;
}
namespace Core::Memory {
class Memory;
}
namespace Core::Frontend {
class EmuWindow;
}
@ -56,8 +53,9 @@ struct ScreenInfo {
class BlitScreen {
public:
explicit BlitScreen(Core::Memory::Memory& cpu_memory, Core::Frontend::EmuWindow& render_window,
const Device& device, MemoryAllocator& memory_manager, Swapchain& swapchain,
explicit BlitScreen(Tegra::MaxwellDeviceMemoryManager& device_memory,
Core::Frontend::EmuWindow& render_window, const Device& device,
MemoryAllocator& memory_manager, Swapchain& swapchain,
PresentManager& present_manager, Scheduler& scheduler,
const ScreenInfo& screen_info);
~BlitScreen();
@ -109,7 +107,7 @@ private:
u64 CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const;
u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer) const;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
Core::Frontend::EmuWindow& render_window;
const Device& device;
MemoryAllocator& memory_allocator;

10
src/video_core/renderer_vulkan/vk_buffer_cache.cpp
View file

@ -79,7 +79,7 @@ vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allo
} // Anonymous namespace
Buffer::Buffer(BufferCacheRuntime& runtime, VideoCommon::NullBufferParams null_params)
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params), tracker{4096} {
: VideoCommon::BufferBase(null_params), tracker{4096} {
if (runtime.device.HasNullDescriptor()) {
return;
}
@ -88,11 +88,9 @@ Buffer::Buffer(BufferCacheRuntime& runtime, VideoCommon::NullBufferParams null_p
is_null = true;
}
Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
VAddr cpu_addr_, u64 size_bytes_)
: VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_),
device{&runtime.device}, buffer{CreateBuffer(*device, runtime.memory_allocator, SizeBytes())},
tracker{SizeBytes()} {
Buffer::Buffer(BufferCacheRuntime& runtime, DAddr cpu_addr_, u64 size_bytes_)
: VideoCommon::BufferBase(cpu_addr_, size_bytes_), device{&runtime.device},
buffer{CreateBuffer(*device, runtime.memory_allocator, SizeBytes())}, tracker{SizeBytes()} {
if (runtime.device.HasDebuggingToolAttached()) {
buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str());
}

7
src/video_core/renderer_vulkan/vk_buffer_cache.h
View file

@ -23,11 +23,10 @@ struct HostVertexBinding;
class BufferCacheRuntime;
class Buffer : public VideoCommon::BufferBase<VideoCore::RasterizerInterface> {
class Buffer : public VideoCommon::BufferBase {
public:
explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params);
explicit Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
VAddr cpu_addr_, u64 size_bytes_);
explicit Buffer(BufferCacheRuntime& runtime, VAddr cpu_addr_, u64 size_bytes_);
[[nodiscard]] VkBufferView View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format);
@ -173,7 +172,7 @@ struct BufferCacheParams {
using Runtime = Vulkan::BufferCacheRuntime;
using Buffer = Vulkan::Buffer;
using Async_Buffer = Vulkan::StagingBufferRef;
using MemoryTracker = VideoCommon::MemoryTrackerBase<VideoCore::RasterizerInterface>;
using MemoryTracker = VideoCommon::MemoryTrackerBase<Tegra::MaxwellDeviceMemoryManager>;
static constexpr bool IS_OPENGL = false;
static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = false;

1
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View file

@ -19,6 +19,7 @@
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/shader_notify.h"
#include "video_core/texture_cache/texture_cache.h"
#include "video_core/vulkan_common/vulkan_device.h"
#if defined(_MSC_VER) && defined(NDEBUG)

8
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View file

@ -30,7 +30,6 @@
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
@ -299,12 +298,13 @@ bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) c
return std::memcmp(&rhs, this, Size()) == 0;
}
PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device_,
Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_,
GuestDescriptorQueue& guest_descriptor_queue_,
RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_,
TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_)
: VideoCommon::ShaderCache{rasterizer_}, device{device_}, scheduler{scheduler_},
: VideoCommon::ShaderCache{device_memory_}, device{device_}, scheduler{scheduler_},
descriptor_pool{descriptor_pool_}, guest_descriptor_queue{guest_descriptor_queue_},
render_pass_cache{render_pass_cache_}, buffer_cache{buffer_cache_},
texture_cache{texture_cache_}, shader_notify{shader_notify_},

6
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View file

@ -20,6 +20,7 @@
#include "shader_recompiler/object_pool.h"
#include "shader_recompiler/profile.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
@ -79,7 +80,6 @@ class ComputePipeline;
class DescriptorPool;
class Device;
class PipelineStatistics;
class RasterizerVulkan;
class RenderPassCache;
class Scheduler;
@ -99,8 +99,8 @@ struct ShaderPools {
class PipelineCache : public VideoCommon::ShaderCache {
public:
explicit PipelineCache(RasterizerVulkan& rasterizer, const Device& device, Scheduler& scheduler,
DescriptorPool& descriptor_pool,
explicit PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, const Device& device,
Scheduler& scheduler, DescriptorPool& descriptor_pool,
GuestDescriptorQueue& guest_descriptor_queue,
RenderPassCache& render_pass_cache, BufferCache& buffer_cache,
TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_);

62
src/video_core/renderer_vulkan/vk_query_cache.cpp
View file

@ -13,9 +13,10 @@
#include "common/bit_util.h"
#include "common/common_types.h"
#include "core/memory.h"
#include "video_core/engines/draw_manager.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/query_cache/query_cache.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_query_cache.h"
@ -102,7 +103,7 @@ private:
using BaseStreamer = VideoCommon::SimpleStreamer<VideoCommon::HostQueryBase>;
struct HostSyncValues {
VAddr address;
DAddr address;
size_t size;
size_t offset;
@ -317,7 +318,7 @@ public:
pending_sync.clear();
}
size_t WriteCounter(VAddr address, bool has_timestamp, u32 value,
size_t WriteCounter(DAddr address, bool has_timestamp, u32 value,
[[maybe_unused]] std::optional<u32> subreport) override {
PauseCounter();
auto index = BuildQuery();
@ -738,7 +739,7 @@ public:
pending_sync.clear();
}
size_t WriteCounter(VAddr address, bool has_timestamp, u32 value,
size_t WriteCounter(DAddr address, bool has_timestamp, u32 value,
std::optional<u32> subreport_) override {
auto index = BuildQuery();
auto* new_query = GetQuery(index);
@ -769,9 +770,9 @@ public:
return index;
}
std::optional<std::pair<VAddr, size_t>> GetLastQueryStream(size_t stream) {
std::optional<std::pair<DAddr, size_t>> GetLastQueryStream(size_t stream) {
if (last_queries[stream] != 0) {
std::pair<VAddr, size_t> result(last_queries[stream], last_queries_stride[stream]);
std::pair<DAddr, size_t> result(last_queries[stream], last_queries_stride[stream]);
return result;
}
return std::nullopt;
@ -974,7 +975,7 @@ private:
size_t buffers_count{};
std::array<VkBuffer, NUM_STREAMS> counter_buffers{};
std::array<VkDeviceSize, NUM_STREAMS> offsets{};
std::array<VAddr, NUM_STREAMS> last_queries;
std::array<DAddr, NUM_STREAMS> last_queries;
std::array<size_t, NUM_STREAMS> last_queries_stride;
Maxwell3D::Regs::PrimitiveTopology out_topology;
u64 streams_mask;
@ -987,7 +988,7 @@ public:
: VideoCommon::QueryBase(0, VideoCommon::QueryFlagBits::IsHostManaged, 0) {}
// Parameterized constructor
PrimitivesQueryBase(bool has_timestamp, VAddr address)
PrimitivesQueryBase(bool has_timestamp, DAddr address)
: VideoCommon::QueryBase(address, VideoCommon::QueryFlagBits::IsHostManaged, 0) {
if (has_timestamp) {
flags |= VideoCommon::QueryFlagBits::HasTimestamp;
@ -995,7 +996,7 @@ public:
}
u64 stride{};
VAddr dependant_address{};
DAddr dependant_address{};
Maxwell3D::Regs::PrimitiveTopology topology{Maxwell3D::Regs::PrimitiveTopology::Points};
size_t dependant_index{};
bool dependant_manage{};
@ -1005,15 +1006,15 @@ class PrimitivesSucceededStreamer : public VideoCommon::SimpleStreamer<Primitive
public:
explicit PrimitivesSucceededStreamer(size_t id_, QueryCacheRuntime& runtime_,
TFBCounterStreamer& tfb_streamer_,
Core::Memory::Memory& cpu_memory_)
Tegra::MaxwellDeviceMemoryManager& device_memory_)
: VideoCommon::SimpleStreamer<PrimitivesQueryBase>(id_), runtime{runtime_},
tfb_streamer{tfb_streamer_}, cpu_memory{cpu_memory_} {
tfb_streamer{tfb_streamer_}, device_memory{device_memory_} {
MakeDependent(&tfb_streamer);
}
~PrimitivesSucceededStreamer() = default;
size_t WriteCounter(VAddr address, bool has_timestamp, u32 value,
size_t WriteCounter(DAddr address, bool has_timestamp, u32 value,
std::optional<u32> subreport_) override {
auto index = BuildQuery();
auto* new_query = GetQuery(index);
@ -1063,6 +1064,8 @@ public:
}
});
}
auto* ptr = device_memory.GetPointer<u8>(new_query->dependant_address);
ASSERT(ptr != nullptr);
new_query->dependant_manage = must_manage_dependance;
pending_flush_queries.push_back(index);
@ -1100,7 +1103,7 @@ public:
num_vertices = dependant_query->value / query->stride;
tfb_streamer.Free(query->dependant_index);
} else {
u8* pointer = cpu_memory.GetPointer(query->dependant_address);
u8* pointer = device_memory.GetPointer<u8>(query->dependant_address);
u32 result;
std::memcpy(&result, pointer, sizeof(u32));
num_vertices = static_cast<u64>(result) / query->stride;
@ -1137,7 +1140,7 @@ public:
private:
QueryCacheRuntime& runtime;
TFBCounterStreamer& tfb_streamer;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
// syncing queue
std::vector<size_t> pending_sync;
@ -1152,12 +1155,13 @@ private:
struct QueryCacheRuntimeImpl {
QueryCacheRuntimeImpl(QueryCacheRuntime& runtime, VideoCore::RasterizerInterface* rasterizer_,
Core::Memory::Memory& cpu_memory_, Vulkan::BufferCache& buffer_cache_,
const Device& device_, const MemoryAllocator& memory_allocator_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_,
Tegra::MaxwellDeviceMemoryManager& device_memory_,
Vulkan::BufferCache& buffer_cache_, const Device& device_,
const MemoryAllocator& memory_allocator_, Scheduler& scheduler_,
StagingBufferPool& staging_pool_,
ComputePassDescriptorQueue& compute_pass_descriptor_queue,
DescriptorPool& descriptor_pool)
: rasterizer{rasterizer_}, cpu_memory{cpu_memory_},
: rasterizer{rasterizer_}, device_memory{device_memory_},
buffer_cache{buffer_cache_}, device{device_},
memory_allocator{memory_allocator_}, scheduler{scheduler_}, staging_pool{staging_pool_},
guest_streamer(0, runtime),
@ -1168,7 +1172,7 @@ struct QueryCacheRuntimeImpl {
scheduler, memory_allocator, staging_pool),
primitives_succeeded_streamer(
static_cast<size_t>(QueryType::StreamingPrimitivesSucceeded), runtime, tfb_streamer,
cpu_memory_),
device_memory_),
primitives_needed_minus_succeeded_streamer(
static_cast<size_t>(QueryType::StreamingPrimitivesNeededMinusSucceeded), runtime, 0u),
hcr_setup{}, hcr_is_set{}, is_hcr_running{}, maxwell3d{} {
@ -1195,7 +1199,7 @@ struct QueryCacheRuntimeImpl {
}
VideoCore::RasterizerInterface* rasterizer;
Core::Memory::Memory& cpu_memory;
Tegra::MaxwellDeviceMemoryManager& device_memory;
Vulkan::BufferCache& buffer_cache;
const Device& device;
@ -1210,7 +1214,7 @@ struct QueryCacheRuntimeImpl {
PrimitivesSucceededStreamer primitives_succeeded_streamer;
VideoCommon::StubStreamer<QueryCacheParams> primitives_needed_minus_succeeded_streamer;
std::vector<std::pair<VAddr, VAddr>> little_cache;
std::vector<std::pair<DAddr, DAddr>> little_cache;
std::vector<std::pair<VkBuffer, VkDeviceSize>> buffers_to_upload_to;
std::vector<size_t> redirect_cache;
std::vector<std::vector<VkBufferCopy>> copies_setup;
@ -1229,14 +1233,14 @@ struct QueryCacheRuntimeImpl {
};
QueryCacheRuntime::QueryCacheRuntime(VideoCore::RasterizerInterface* rasterizer,
Core::Memory::Memory& cpu_memory_,
Tegra::MaxwellDeviceMemoryManager& device_memory_,
Vulkan::BufferCache& buffer_cache_, const Device& device_,
const MemoryAllocator& memory_allocator_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_,
ComputePassDescriptorQueue& compute_pass_descriptor_queue,
DescriptorPool& descriptor_pool) {
impl = std::make_unique<QueryCacheRuntimeImpl>(
*this, rasterizer, cpu_memory_, buffer_cache_, device_, memory_allocator_, scheduler_,
*this, rasterizer, device_memory_, buffer_cache_, device_, memory_allocator_, scheduler_,
staging_pool_, compute_pass_descriptor_queue, descriptor_pool);
}
@ -1309,7 +1313,7 @@ void QueryCacheRuntime::HostConditionalRenderingCompareValueImpl(VideoCommon::Lo
ResumeHostConditionalRendering();
}
void QueryCacheRuntime::HostConditionalRenderingCompareBCImpl(VAddr address, bool is_equal) {
void QueryCacheRuntime::HostConditionalRenderingCompareBCImpl(DAddr address, bool is_equal) {
VkBuffer to_resolve;
u32 to_resolve_offset;
{
@ -1350,11 +1354,11 @@ bool QueryCacheRuntime::HostConditionalRenderingCompareValues(VideoCommon::Looku
return false;
}
const auto check_in_bc = [&](VAddr address) {
const auto check_in_bc = [&](DAddr address) {
return impl->buffer_cache.IsRegionGpuModified(address, 8);
};
const auto check_value = [&](VAddr address) {
u8* ptr = impl->cpu_memory.GetPointer(address);
const auto check_value = [&](DAddr address) {
u8* ptr = impl->device_memory.GetPointer<u8>(address);
u64 value{};
std::memcpy(&value, ptr, sizeof(value));
return value == 0;
@ -1477,8 +1481,8 @@ void QueryCacheRuntime::SyncValues(std::span<SyncValuesType> values, VkBuffer ba
for (auto& sync_val : values) {
total_size += sync_val.size;
bool found = false;
VAddr base = Common::AlignDown(sync_val.address, Core::Memory::YUZU_PAGESIZE);
VAddr base_end = base + Core::Memory::YUZU_PAGESIZE;
DAddr base = Common::AlignDown(sync_val.address, Core::DEVICE_PAGESIZE);
DAddr base_end = base + Core::DEVICE_PAGESIZE;
for (size_t i = 0; i < impl->little_cache.size(); i++) {
const auto set_found = [&] {
impl->redirect_cache.push_back(i);

4
src/video_core/renderer_vulkan/vk_query_cache.h
View file

@ -27,7 +27,7 @@ struct QueryCacheRuntimeImpl;
class QueryCacheRuntime {
public:
explicit QueryCacheRuntime(VideoCore::RasterizerInterface* rasterizer,
Core::Memory::Memory& cpu_memory_,
Tegra::MaxwellDeviceMemoryManager& device_memory_,
Vulkan::BufferCache& buffer_cache_, const Device& device_,
const MemoryAllocator& memory_allocator_, Scheduler& scheduler_,
StagingBufferPool& staging_pool_,
@ -61,7 +61,7 @@ public:
private:
void HostConditionalRenderingCompareValueImpl(VideoCommon::LookupData object, bool is_equal);
void HostConditionalRenderingCompareBCImpl(VAddr address, bool is_equal);
void HostConditionalRenderingCompareBCImpl(DAddr address, bool is_equal);
friend struct QueryCacheRuntimeImpl;
std::unique_ptr<QueryCacheRuntimeImpl> impl;
};

44
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -18,6 +18,7 @@
#include "video_core/engines/draw_manager.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/renderer_vulkan/blit_image.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
@ -163,10 +164,11 @@ DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances,
} // Anonymous namespace
RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
Core::Memory::Memory& cpu_memory_, ScreenInfo& screen_info_,
const Device& device_, MemoryAllocator& memory_allocator_,
StateTracker& state_tracker_, Scheduler& scheduler_)
: RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, screen_info{screen_info_}, device{device_},
Tegra::MaxwellDeviceMemoryManager& device_memory_,
ScreenInfo& screen_info_, const Device& device_,
MemoryAllocator& memory_allocator_, StateTracker& state_tracker_,
Scheduler& scheduler_)
: gpu{gpu_}, device_memory{device_memory_}, screen_info{screen_info_}, device{device_},
memory_allocator{memory_allocator_}, state_tracker{state_tracker_}, scheduler{scheduler_},
staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler),
guest_descriptor_queue(device, scheduler), compute_pass_descriptor_queue(device, scheduler),
@ -174,14 +176,14 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
texture_cache_runtime{
device, scheduler, memory_allocator, staging_pool,
blit_image, render_pass_cache, descriptor_pool, compute_pass_descriptor_queue},
texture_cache(texture_cache_runtime, *this),
texture_cache(texture_cache_runtime, device_memory),
buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool,
guest_descriptor_queue, compute_pass_descriptor_queue, descriptor_pool),
buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
query_cache_runtime(this, cpu_memory_, buffer_cache, device, memory_allocator, scheduler,
buffer_cache(device_memory, buffer_cache_runtime),
query_cache_runtime(this, device_memory, buffer_cache, device, memory_allocator, scheduler,
staging_pool, compute_pass_descriptor_queue, descriptor_pool),
query_cache(gpu, *this, cpu_memory_, query_cache_runtime),
pipeline_cache(*this, device, scheduler, descriptor_pool, guest_descriptor_queue,
query_cache(gpu, *this, device_memory, query_cache_runtime),
pipeline_cache(device_memory, device, scheduler, descriptor_pool, guest_descriptor_queue,
render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()),
accelerate_dma(buffer_cache, texture_cache, scheduler),
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
@ -508,7 +510,7 @@ void Vulkan::RasterizerVulkan::DisableGraphicsUniformBuffer(size_t stage, u32 in
void RasterizerVulkan::FlushAll() {}
void RasterizerVulkan::FlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
void RasterizerVulkan::FlushRegion(DAddr addr, u64 size, VideoCommon::CacheType which) {
if (addr == 0 || size == 0) {
return;
}
@ -525,7 +527,7 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size, VideoCommon::CacheType
}
}
bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
bool RasterizerVulkan::MustFlushRegion(DAddr addr, u64 size, VideoCommon::CacheType which) {
if ((True(which & VideoCommon::CacheType::BufferCache))) {
std::scoped_lock lock{buffer_cache.mutex};
if (buffer_cache.IsRegionGpuModified(addr, size)) {
@ -542,7 +544,7 @@ bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheT
return false;
}
VideoCore::RasterizerDownloadArea RasterizerVulkan::GetFlushArea(VAddr addr, u64 size) {
VideoCore::RasterizerDownloadArea RasterizerVulkan::GetFlushArea(DAddr addr, u64 size) {
{
std::scoped_lock lock{texture_cache.mutex};
auto area = texture_cache.GetFlushArea(addr, size);
@ -551,14 +553,14 @@ VideoCore::RasterizerDownloadArea RasterizerVulkan::GetFlushArea(VAddr addr, u64
}
}
VideoCore::RasterizerDownloadArea new_area{
.start_address = Common::AlignDown(addr, Core::Memory::YUZU_PAGESIZE),
.end_address = Common::AlignUp(addr + size, Core::Memory::YUZU_PAGESIZE),
.start_address = Common::AlignDown(addr, Core::DEVICE_PAGESIZE),
.end_address = Common::AlignUp(addr + size, Core::DEVICE_PAGESIZE),
.preemtive = true,
};
return new_area;
}
void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
void RasterizerVulkan::InvalidateRegion(DAddr addr, u64 size, VideoCommon::CacheType which) {
if (addr == 0 || size == 0) {
return;
}
@ -578,7 +580,7 @@ void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache
}
}
void RasterizerVulkan::InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) {
void RasterizerVulkan::InnerInvalidation(std::span<const std::pair<DAddr, std::size_t>> sequences) {
{
std::scoped_lock lock{texture_cache.mutex};
for (const auto& [addr, size] : sequences) {
@ -599,7 +601,7 @@ void RasterizerVulkan::InnerInvalidation(std::span<const std::pair<VAddr, std::s
}
}
bool RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
bool RasterizerVulkan::OnCPUWrite(DAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return false;
}
@ -620,7 +622,7 @@ bool RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
return false;
}
void RasterizerVulkan::OnCacheInvalidation(VAddr addr, u64 size) {
void RasterizerVulkan::OnCacheInvalidation(DAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return;
}
@ -640,7 +642,7 @@ void RasterizerVulkan::InvalidateGPUCache() {
gpu.InvalidateGPUCache();
}
void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) {
void RasterizerVulkan::UnmapMemory(DAddr addr, u64 size) {
{
std::scoped_lock lock{texture_cache.mutex};
texture_cache.UnmapMemory(addr, size);
@ -679,7 +681,7 @@ void RasterizerVulkan::ReleaseFences(bool force) {
fence_manager.WaitPendingFences(force);
}
void RasterizerVulkan::FlushAndInvalidateRegion(VAddr addr, u64 size,
void RasterizerVulkan::FlushAndInvalidateRegion(DAddr addr, u64 size,
VideoCommon::CacheType which) {
if (Settings::IsGPULevelExtreme()) {
FlushRegion(addr, size, which);
@ -782,7 +784,7 @@ void RasterizerVulkan::AccelerateInlineToMemory(GPUVAddr address, size_t copy_si
}
bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config,
VAddr framebuffer_addr, u32 pixel_stride) {
DAddr framebuffer_addr, u32 pixel_stride) {
if (!framebuffer_addr) {
return false;
}

41
src/video_core/renderer_vulkan/vk_rasterizer.h
View file

@ -7,14 +7,13 @@
#include <boost/container/static_vector.hpp>
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "common/common_types.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/engines/maxwell_dma.h"
#include "video_core/rasterizer_accelerated.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_vulkan/blit_image.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_fence_manager.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
@ -34,10 +33,14 @@ namespace Core::Frontend {
class EmuWindow;
}
namespace Tegra::Engines {
namespace Tegra {
namespace Engines {
class Maxwell3D;
}
} // namespace Tegra
namespace Vulkan {
struct ScreenInfo;
@ -70,13 +73,14 @@ private:
Scheduler& scheduler;
};
class RasterizerVulkan final : public VideoCore::RasterizerAccelerated,
class RasterizerVulkan final : public VideoCore::RasterizerInterface,
protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
public:
explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
Core::Memory::Memory& cpu_memory_, ScreenInfo& screen_info_,
const Device& device_, MemoryAllocator& memory_allocator_,
StateTracker& state_tracker_, Scheduler& scheduler_);
Tegra::MaxwellDeviceMemoryManager& device_memory_,
ScreenInfo& screen_info_, const Device& device_,
MemoryAllocator& memory_allocator_, StateTracker& state_tracker_,
Scheduler& scheduler_);
~RasterizerVulkan() override;
void Draw(bool is_indexed, u32 instance_count) override;
@ -90,18 +94,18 @@ public:
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
void DisableGraphicsUniformBuffer(size_t stage, u32 index) override;
void FlushAll() override;
void FlushRegion(VAddr addr, u64 size,
void FlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
bool MustFlushRegion(VAddr addr, u64 size,
bool MustFlushRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
VideoCore::RasterizerDownloadArea GetFlushArea(VAddr addr, u64 size) override;
void InvalidateRegion(VAddr addr, u64 size,
VideoCore::RasterizerDownloadArea GetFlushArea(DAddr addr, u64 size) override;
void InvalidateRegion(DAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) override;
void OnCacheInvalidation(VAddr addr, u64 size) override;
bool OnCPUWrite(VAddr addr, u64 size) override;
void InnerInvalidation(std::span<const std::pair<DAddr, std::size_t>> sequences) override;
void OnCacheInvalidation(DAddr addr, u64 size) override;
bool OnCPUWrite(DAddr addr, u64 size) override;
void InvalidateGPUCache() override;
void UnmapMemory(VAddr addr, u64 size) override;
void UnmapMemory(DAddr addr, u64 size) override;
void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) override;
void SignalFence(std::function<void()>&& func) override;
void SyncOperation(std::function<void()>&& func) override;
@ -109,7 +113,7 @@ public:
void SignalReference() override;
void ReleaseFences(bool force = true) override;
void FlushAndInvalidateRegion(
VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
DAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void WaitForIdle() override;
void FragmentBarrier() override;
void TiledCacheBarrier() override;
@ -122,7 +126,7 @@ public:
Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override;
void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size,
std::span<const u8> memory) override;
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
bool AccelerateDisplay(const Tegra::FramebufferConfig& config, DAddr framebuffer_addr,
u32 pixel_stride) override;
void LoadDiskResources(u64 title_id, std::stop_token stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) override;
@ -176,6 +180,7 @@ private:
void UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs);
Tegra::GPU& gpu;
Tegra::MaxwellDeviceMemoryManager& device_memory;
ScreenInfo& screen_info;
const Device& device;

8
src/video_core/shader_cache.cpp
View file

@ -12,6 +12,7 @@
#include "video_core/dirty_flags.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/memory_manager.h"
#include "video_core/shader_cache.h"
#include "video_core/shader_environment.h"
@ -34,7 +35,8 @@ void ShaderCache::SyncGuestHost() {
RemovePendingShaders();
}
ShaderCache::ShaderCache(VideoCore::RasterizerInterface& rasterizer_) : rasterizer{rasterizer_} {}
ShaderCache::ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory_)
: device_memory{device_memory_} {}
bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) {
auto& dirty{maxwell3d->dirty.flags};
@ -132,7 +134,7 @@ void ShaderCache::Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t
storage.push_back(std::move(data));
rasterizer.UpdatePagesCachedCount(addr, size, 1);
device_memory.UpdatePagesCachedCount(addr, size, 1);
}
void ShaderCache::InvalidatePagesInRegion(VAddr addr, size_t size) {
@ -209,7 +211,7 @@ void ShaderCache::UnmarkMemory(Entry* entry) {
const VAddr addr = entry->addr_start;
const size_t size = entry->addr_end - addr;
rasterizer.UpdatePagesCachedCount(addr, size, -1);
device_memory.UpdatePagesCachedCount(addr, size, -1);
}
void ShaderCache::RemoveShadersFromStorage(std::span<ShaderInfo*> removed_shaders) {

5
src/video_core/shader_cache.h
View file

@ -14,6 +14,7 @@
#include "common/common_types.h"
#include "common/polyfill_ranges.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/shader_environment.h"
@ -77,7 +78,7 @@ protected:
}
};
explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_);
explicit ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory);
/// @brief Update the hashes and information of shader stages
/// @param unique_hashes Shader hashes to store into when a stage is enabled
@ -145,7 +146,7 @@ private:
/// @brief Create a new shader entry and register it
const ShaderInfo* MakeShaderInfo(GenericEnvironment& env, VAddr cpu_addr);
VideoCore::RasterizerInterface& rasterizer;
Tegra::MaxwellDeviceMemoryManager& device_memory;
mutable std::mutex lookup_mutex;
std::mutex invalidation_mutex;

75
src/video_core/texture_cache/texture_cache.h
View file

@ -8,10 +8,11 @@
#include "common/alignment.h"
#include "common/settings.h"
#include "core/memory.h"
#include "video_core/control/channel_state.h"
#include "video_core/dirty_flags.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/guest_memory.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/texture_cache/image_view_base.h"
#include "video_core/texture_cache/samples_helper.h"
#include "video_core/texture_cache/texture_cache_base.h"
@ -27,8 +28,8 @@ using VideoCore::Surface::SurfaceType;
using namespace Common::Literals;
template <class P>
TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& rasterizer_)
: runtime{runtime_}, rasterizer{rasterizer_} {
TextureCache<P>::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManager& device_memory_)
: runtime{runtime_}, device_memory{device_memory_} {
// Configure null sampler
TSCEntry sampler_descriptor{};
sampler_descriptor.min_filter.Assign(Tegra::Texture::TextureFilter::Linear);
@ -49,19 +50,19 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
void(slot_samplers.insert(runtime, sampler_descriptor));
if constexpr (HAS_DEVICE_MEMORY_INFO) {
const s64 device_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
const s64 min_spacing_expected = device_memory - 1_GiB;
const s64 min_spacing_critical = device_memory - 512_MiB;
const s64 mem_threshold = std::min(device_memory, TARGET_THRESHOLD);
const s64 device_local_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
const s64 min_spacing_expected = device_local_memory - 1_GiB;
const s64 min_spacing_critical = device_local_memory - 512_MiB;
const s64 mem_threshold = std::min(device_local_memory, TARGET_THRESHOLD);
const s64 min_vacancy_expected = (6 * mem_threshold) / 10;
const s64 min_vacancy_critical = (3 * mem_threshold) / 10;
expected_memory = static_cast<u64>(
std::max(std::min(device_memory - min_vacancy_expected, min_spacing_expected),
std::max(std::min(device_local_memory - min_vacancy_expected, min_spacing_expected),
DEFAULT_EXPECTED_MEMORY));
critical_memory = static_cast<u64>(
std::max(std::min(device_memory - min_vacancy_critical, min_spacing_critical),
std::max(std::min(device_local_memory - min_vacancy_critical, min_spacing_critical),
DEFAULT_CRITICAL_MEMORY));
minimum_memory = static_cast<u64>((device_memory - mem_threshold) / 2);
minimum_memory = static_cast<u64>((device_local_memory - mem_threshold) / 2);
} else {
expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB;
critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB;
@ -513,7 +514,7 @@ FramebufferId TextureCache<P>::GetFramebufferId(const RenderTargets& key) {
}
template <class P>
void TextureCache<P>::WriteMemory(VAddr cpu_addr, size_t size) {
void TextureCache<P>::WriteMemory(DAddr cpu_addr, size_t size) {
ForEachImageInRegion(cpu_addr, size, [this](ImageId image_id, Image& image) {
if (True(image.flags & ImageFlagBits::CpuModified)) {
return;
@ -526,7 +527,7 @@ void TextureCache<P>::WriteMemory(VAddr cpu_addr, size_t size) {
}
template <class P>
void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) {
void TextureCache<P>::DownloadMemory(DAddr cpu_addr, size_t size) {
boost::container::small_vector<ImageId, 16> images;
ForEachImageInRegion(cpu_addr, size, [&images](ImageId image_id, ImageBase& image) {
if (!image.IsSafeDownload()) {
@ -553,7 +554,7 @@ void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) {
}
template <class P>
std::optional<VideoCore::RasterizerDownloadArea> TextureCache<P>::GetFlushArea(VAddr cpu_addr,
std::optional<VideoCore::RasterizerDownloadArea> TextureCache<P>::GetFlushArea(DAddr cpu_addr,
u64 size) {
std::optional<VideoCore::RasterizerDownloadArea> area{};
ForEachImageInRegion(cpu_addr, size, [&](ImageId, ImageBase& image) {
@ -579,7 +580,7 @@ std::optional<VideoCore::RasterizerDownloadArea> TextureCache<P>::GetFlushArea(V
}
template <class P>
void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) {
void TextureCache<P>::UnmapMemory(DAddr cpu_addr, size_t size) {
boost::container::small_vector<ImageId, 16> deleted_images;
ForEachImageInRegion(cpu_addr, size, [&](ImageId id, Image&) { deleted_images.push_back(id); });
for (const ImageId id : deleted_images) {
@ -713,7 +714,7 @@ bool TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
template <class P>
typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(
const Tegra::FramebufferConfig& config, VAddr cpu_addr) {
const Tegra::FramebufferConfig& config, DAddr cpu_addr) {
// TODO: Properly implement this
const auto it = page_table.find(cpu_addr >> YUZU_PAGEBITS);
if (it == page_table.end()) {
@ -940,7 +941,7 @@ bool TextureCache<P>::IsRescaling(const ImageViewBase& image_view) const noexcep
}
template <class P>
bool TextureCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
bool TextureCache<P>::IsRegionGpuModified(DAddr addr, size_t size) {
bool is_modified = false;
ForEachImageInRegion(addr, size, [&is_modified](ImageId, ImageBase& image) {
if (False(image.flags & ImageFlagBits::GpuModified)) {
@ -1059,7 +1060,7 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
return;
}
Core::Memory::GpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::UnsafeRead> swizzle_data(
Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::UnsafeRead> swizzle_data(
*gpu_memory, gpu_addr, image.guest_size_bytes, &swizzle_data_buffer);
if (True(image.flags & ImageFlagBits::Converted)) {
@ -1124,7 +1125,7 @@ ImageId TextureCache<P>::FindOrInsertImage(const ImageInfo& info, GPUVAddr gpu_a
template <class P>
ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
RelaxedOptions options) {
std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
std::optional<DAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
if (!cpu_addr) {
cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, CalculateGuestSizeInBytes(info));
if (!cpu_addr) {
@ -1265,7 +1266,7 @@ void TextureCache<P>::QueueAsyncDecode(Image& image, ImageId image_id) {
static Common::ScratchBuffer<u8> local_unswizzle_data_buffer;
local_unswizzle_data_buffer.resize_destructive(image.unswizzled_size_bytes);
Core::Memory::GpuGuestMemory<u8, Core::Memory::GuestMemoryFlags::UnsafeRead> swizzle_data(
Tegra::Memory::GpuGuestMemory<u8, Tegra::Memory::GuestMemoryFlags::UnsafeRead> swizzle_data(
*gpu_memory, image.gpu_addr, image.guest_size_bytes, &swizzle_data_buffer);
auto copies = UnswizzleImage(*gpu_memory, image.gpu_addr, image.info, swizzle_data,
@ -1339,14 +1340,14 @@ bool TextureCache<P>::ScaleDown(Image& image) {
template <class P>
ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
RelaxedOptions options) {
std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
std::optional<DAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
if (!cpu_addr) {
const auto size = CalculateGuestSizeInBytes(info);
cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, size);
if (!cpu_addr) {
const VAddr fake_addr = ~(1ULL << 40ULL) + virtual_invalid_space;
const DAddr fake_addr = ~(1ULL << 40ULL) + virtual_invalid_space;
virtual_invalid_space += Common::AlignUp(size, 32);
cpu_addr = std::optional<VAddr>(fake_addr);
cpu_addr = std::optional<DAddr>(fake_addr);
}
}
ASSERT_MSG(cpu_addr, "Tried to insert an image to an invalid gpu_addr=0x{:x}", gpu_addr);
@ -1362,7 +1363,7 @@ ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
}
template <class P>
ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr) {
ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DAddr cpu_addr) {
ImageInfo new_info = info;
const size_t size_bytes = CalculateGuestSizeInBytes(new_info);
const bool broken_views = runtime.HasBrokenTextureViewFormats();
@ -1650,7 +1651,7 @@ std::optional<typename TextureCache<P>::BlitImages> TextureCache<P>::GetBlitImag
template <class P>
ImageId TextureCache<P>::FindDMAImage(const ImageInfo& info, GPUVAddr gpu_addr) {
std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
std::optional<DAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
if (!cpu_addr) {
cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, CalculateGuestSizeInBytes(info));
if (!cpu_addr) {
@ -1780,7 +1781,7 @@ ImageViewId TextureCache<P>::FindRenderTargetView(const ImageInfo& info, GPUVAdd
template <class P>
template <typename Func>
void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func) {
void TextureCache<P>::ForEachImageInRegion(DAddr cpu_addr, size_t size, Func&& func) {
using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type;
static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;
boost::container::small_vector<ImageId, 32> images;
@ -1924,11 +1925,11 @@ void TextureCache<P>::ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size,
template <class P>
template <typename Func>
void TextureCache<P>::ForEachSparseSegment(ImageBase& image, Func&& func) {
using FuncReturn = typename std::invoke_result<Func, GPUVAddr, VAddr, size_t>::type;
using FuncReturn = typename std::invoke_result<Func, GPUVAddr, DAddr, size_t>::type;
static constexpr bool RETURNS_BOOL = std::is_same_v<FuncReturn, bool>;
const auto segments = gpu_memory->GetSubmappedRange(image.gpu_addr, image.guest_size_bytes);
for (const auto& [gpu_addr, size] : segments) {
std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
std::optional<DAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr);
if constexpr (RETURNS_BOOL) {
if (func(gpu_addr, *cpu_addr, size)) {
@ -1980,7 +1981,7 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
}
boost::container::small_vector<ImageViewId, 16> sparse_maps;
ForEachSparseSegment(
image, [this, image_id, &sparse_maps](GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) {
image, [this, image_id, &sparse_maps](GPUVAddr gpu_addr, DAddr cpu_addr, size_t size) {
auto map_id = slot_map_views.insert(gpu_addr, cpu_addr, size, image_id);
ForEachCPUPage(cpu_addr, size,
[this, map_id](u64 page) { page_table[page].push_back(map_id); });
@ -2048,7 +2049,7 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
auto& sparse_maps = it->second;
for (auto& map_view_id : sparse_maps) {
const auto& map_range = slot_map_views[map_view_id];
const VAddr cpu_addr = map_range.cpu_addr;
const DAddr cpu_addr = map_range.cpu_addr;
const std::size_t size = map_range.size;
ForEachCPUPage(cpu_addr, size, [this, image_id](u64 page) {
const auto page_it = page_table.find(page);
@ -2080,7 +2081,7 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
ASSERT(False(image.flags & ImageFlagBits::Tracked));
image.flags |= ImageFlagBits::Tracked;
if (False(image.flags & ImageFlagBits::Sparse)) {
rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
device_memory.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
return;
}
if (True(image.flags & ImageFlagBits::Registered)) {
@ -2089,15 +2090,15 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
auto& sparse_maps = it->second;
for (auto& map_view_id : sparse_maps) {
const auto& map = slot_map_views[map_view_id];
const VAddr cpu_addr = map.cpu_addr;
const DAddr cpu_addr = map.cpu_addr;
const std::size_t size = map.size;
rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1);
device_memory.UpdatePagesCachedCount(cpu_addr, size, 1);
}
return;
}
ForEachSparseSegment(image,
[this]([[maybe_unused]] GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) {
rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1);
[this]([[maybe_unused]] GPUVAddr gpu_addr, DAddr cpu_addr, size_t size) {
device_memory.UpdatePagesCachedCount(cpu_addr, size, 1);
});
}
@ -2106,7 +2107,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
ASSERT(True(image.flags & ImageFlagBits::Tracked));
image.flags &= ~ImageFlagBits::Tracked;
if (False(image.flags & ImageFlagBits::Sparse)) {
rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1);
device_memory.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1);
return;
}
ASSERT(True(image.flags & ImageFlagBits::Registered));
@ -2115,9 +2116,9 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
auto& sparse_maps = it->second;
for (auto& map_view_id : sparse_maps) {
const auto& map = slot_map_views[map_view_id];
const VAddr cpu_addr = map.cpu_addr;
const DAddr cpu_addr = map.cpu_addr;
const std::size_t size = map.size;
rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1);
device_memory.UpdatePagesCachedCount(cpu_addr, size, -1);
}
}

28
src/video_core/texture_cache/texture_cache_base.h
View file

@ -36,9 +36,11 @@
#include "video_core/texture_cache/types.h"
#include "video_core/textures/texture.h"
namespace Tegra::Control {
namespace Tegra {
namespace Control {
struct ChannelState;
}
} // namespace Tegra
namespace VideoCommon {
@ -126,7 +128,7 @@ class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelI
};
public:
explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&);
explicit TextureCache(Runtime&, Tegra::MaxwellDeviceMemoryManager&);
/// Notify the cache that a new frame has been queued
void TickFrame();
@ -190,15 +192,15 @@ public:
Framebuffer* GetFramebuffer();
/// Mark images in a range as modified from the CPU
void WriteMemory(VAddr cpu_addr, size_t size);
void WriteMemory(DAddr cpu_addr, size_t size);
/// Download contents of host images to guest memory in a region
void DownloadMemory(VAddr cpu_addr, size_t size);
void DownloadMemory(DAddr cpu_addr, size_t size);
std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size);
std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(DAddr cpu_addr, u64 size);
/// Remove images in a region
void UnmapMemory(VAddr cpu_addr, size_t size);
void UnmapMemory(DAddr cpu_addr, size_t size);
/// Remove images in a region
void UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size);
@ -210,7 +212,7 @@ public:
/// Try to find a cached image view in the given CPU address
[[nodiscard]] ImageView* TryFindFramebufferImageView(const Tegra::FramebufferConfig& config,
VAddr cpu_addr);
DAddr cpu_addr);
/// Return true when there are uncommitted images to be downloaded
[[nodiscard]] bool HasUncommittedFlushes() const noexcept;
@ -235,7 +237,7 @@ public:
GPUVAddr address = 0, size_t size = 0);
/// Return true when a CPU region is modified from the GPU
[[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
[[nodiscard]] bool IsRegionGpuModified(DAddr addr, size_t size);
[[nodiscard]] bool IsRescaling() const noexcept;
@ -252,7 +254,7 @@ public:
private:
/// Iterate over all page indices in a range
template <typename Func>
static void ForEachCPUPage(VAddr addr, size_t size, Func&& func) {
static void ForEachCPUPage(DAddr addr, size_t size, Func&& func) {
static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
const u64 page_end = (addr + size - 1) >> YUZU_PAGEBITS;
for (u64 page = addr >> YUZU_PAGEBITS; page <= page_end; ++page) {
@ -326,7 +328,7 @@ private:
/// Create a new image and join perfectly matching existing images
/// Remove joined images from the cache
[[nodiscard]] ImageId JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr);
[[nodiscard]] ImageId JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DAddr cpu_addr);
[[nodiscard]] ImageId FindDMAImage(const ImageInfo& info, GPUVAddr gpu_addr);
@ -349,7 +351,7 @@ private:
/// Iterates over all the images in a region calling func
template <typename Func>
void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func);
void ForEachImageInRegion(DAddr cpu_addr, size_t size, Func&& func);
template <typename Func>
void ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_addr, size_t size, Func&& func);
@ -421,7 +423,7 @@ private:
Runtime& runtime;
VideoCore::RasterizerInterface& rasterizer;
Tegra::MaxwellDeviceMemoryManager& device_memory;
std::deque<TextureCacheGPUMap> gpu_page_table_storage;
RenderTargets render_targets;
@ -432,7 +434,7 @@ private:
std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>> sparse_page_table;
std::unordered_map<ImageId, boost::container::small_vector<ImageViewId, 16>> sparse_views;
VAddr virtual_invalid_space{};
DAddr virtual_invalid_space{};
bool has_deleted_images = false;
bool is_rescaling = false;

5
src/video_core/texture_cache/util.cpp
View file

@ -20,9 +20,9 @@
#include "common/div_ceil.h"
#include "common/scratch_buffer.h"
#include "common/settings.h"
#include "core/memory.h"
#include "video_core/compatible_formats.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/guest_memory.h"
#include "video_core/memory_manager.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/decode_bc.h"
@ -552,7 +552,8 @@ void SwizzleBlockLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr
for (s32 layer = 0; layer < info.resources.layers; ++layer) {
const std::span<const u8> src = input.subspan(host_offset);
{
Core::Memory::GpuGuestMemoryScoped<u8, Core::Memory::GuestMemoryFlags::UnsafeReadWrite>
Tegra::Memory::GpuGuestMemoryScoped<u8,
Tegra::Memory::GuestMemoryFlags::UnsafeReadWrite>
dst(gpu_memory, gpu_addr + guest_offset, subresource_size, &tmp_buffer);
SwizzleTexture(dst, src, bytes_per_block, num_tiles.width, num_tiles.height,

15
src/video_core/video_core.cpp
View file

@ -6,6 +6,8 @@
#include "common/logging/log.h"
#include "common/settings.h"
#include "core/core.h"
#include "video_core/host1x/gpu_device_memory_manager.h"
#include "video_core/host1x/host1x.h"
#include "video_core/renderer_base.h"
#include "video_core/renderer_null/renderer_null.h"
#include "video_core/renderer_opengl/renderer_opengl.h"
@ -18,18 +20,17 @@ std::unique_ptr<VideoCore::RendererBase> CreateRenderer(
Core::System& system, Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu,
std::unique_ptr<Core::Frontend::GraphicsContext> context) {
auto& telemetry_session = system.TelemetrySession();
auto& cpu_memory = system.ApplicationMemory();
auto& device_memory = system.Host1x().MemoryManager();
switch (Settings::values.renderer_backend.GetValue()) {
case Settings::RendererBackend::OpenGL:
return std::make_unique<OpenGL::RendererOpenGL>(telemetry_session, emu_window, cpu_memory,
gpu, std::move(context));
return std::make_unique<OpenGL::RendererOpenGL>(telemetry_session, emu_window,
device_memory, gpu, std::move(context));
case Settings::RendererBackend::Vulkan:
return std::make_unique<Vulkan::RendererVulkan>(telemetry_session, emu_window, cpu_memory,
gpu, std::move(context));
return std::make_unique<Vulkan::RendererVulkan>(telemetry_session, emu_window,
device_memory, gpu, std::move(context));
case Settings::RendererBackend::Null:
return std::make_unique<Null::RendererNull>(emu_window, cpu_memory, gpu,
std::move(context));
return std::make_unique<Null::RendererNull>(emu_window, gpu, std::move(context));
default:
return nullptr;
}