Refactor VideoCore to use AS sepparate from Channel.

master
Fernando Sahmkow 2021-12-17 16:45:06 +07:00
parent bb74973bba
commit e462191482
10 changed files with 172 additions and 153 deletions

@ -18,4 +18,11 @@ struct PairHash {
} }
}; };
template <typename T>
struct IdentityHash {
[[nodiscard]] size_t operator()(T value) const noexcept {
return static_cast<size_t>(value);
}
};
} // namespace Common } // namespace Common

@ -203,6 +203,7 @@ add_library(video_core STATIC
texture_cache/render_targets.h texture_cache/render_targets.h
texture_cache/samples_helper.h texture_cache/samples_helper.h
texture_cache/slot_vector.h texture_cache/slot_vector.h
texture_cache/texture_cache.cpp
texture_cache/texture_cache.h texture_cache/texture_cache.h
texture_cache/texture_cache_base.h texture_cache/texture_cache_base.h
texture_cache/types.h texture_cache/types.h

@ -1,5 +1,11 @@
#include "video_core/control/channel_state_cache.inc" #include "video_core/control/channel_state_cache.inc"
namespace VideoCommon { namespace VideoCommon {
ChannelInfo::ChannelInfo(Tegra::Control::ChannelState& channel_state)
: maxwell3d{*channel_state.maxwell_3d}, kepler_compute{*channel_state.kepler_compute},
gpu_memory{*channel_state.memory_manager} {}
template class VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo>; template class VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo>;
}
} // namespace VideoCommon

@ -2,6 +2,7 @@
#include <deque> #include <deque>
#include <limits> #include <limits>
#include <mutex>
#include <unordered_map> #include <unordered_map>
#include "common/common_types.h" #include "common/common_types.h"
@ -41,9 +42,10 @@ template <class P>
class ChannelSetupCaches { class ChannelSetupCaches {
public: public:
/// Operations for seting the channel of execution. /// Operations for seting the channel of execution.
virtual ~ChannelSetupCaches();
/// Create channel state. /// Create channel state.
void CreateChannel(Tegra::Control::ChannelState& channel); virtual void CreateChannel(Tegra::Control::ChannelState& channel);
/// Bind a channel for execution. /// Bind a channel for execution.
void BindToChannel(s32 id); void BindToChannel(s32 id);
@ -51,18 +53,34 @@ public:
/// Erase channel's state. /// Erase channel's state.
void EraseChannel(s32 id); void EraseChannel(s32 id);
Tegra::MemoryManager* GetFromID(size_t id) const {
std::unique_lock<std::mutex> lk(config_mutex);
const auto ref = address_spaces.find(id);
return ref->second.gpu_memory;
}
protected: protected:
static constexpr size_t UNSET_CHANNEL{std::numeric_limits<size_t>::max()}; static constexpr size_t UNSET_CHANNEL{std::numeric_limits<size_t>::max()};
P* channel_state;
size_t current_channel_id{UNSET_CHANNEL};
size_t current_address_space{};
Tegra::Engines::Maxwell3D* maxwell3d;
Tegra::Engines::KeplerCompute* kepler_compute;
Tegra::MemoryManager* gpu_memory;
std::deque<P> channel_storage; std::deque<P> channel_storage;
std::deque<size_t> free_channel_ids; std::deque<size_t> free_channel_ids;
std::unordered_map<s32, size_t> channel_map; std::unordered_map<s32, size_t> channel_map;
struct AddresSpaceRef {
P* channel_state; size_t ref_count;
size_t current_channel_id{UNSET_CHANNEL}; size_t storage_id;
Tegra::Engines::Maxwell3D* maxwell3d;
Tegra::Engines::KeplerCompute* kepler_compute;
Tegra::MemoryManager* gpu_memory; Tegra::MemoryManager* gpu_memory;
}; };
std::unordered_map<size_t, AddresSpaceRef> address_spaces;
mutable std::mutex config_mutex;
virtual void OnGPUASRegister([[maybe_unused]] size_t map_id) {}
};
} // namespace VideoCommon } // namespace VideoCommon

@ -6,18 +6,18 @@
namespace VideoCommon { namespace VideoCommon {
ChannelInfo::ChannelInfo(Tegra::Control::ChannelState& channel_state) template <class P>
: maxwell3d{*channel_state.maxwell_3d}, kepler_compute{*channel_state.kepler_compute}, ChannelSetupCaches<P>::~ChannelSetupCaches() = default;
gpu_memory{*channel_state.memory_manager} {}
template <class P> template <class P>
void ChannelSetupCaches<P>::CreateChannel(struct Tegra::Control::ChannelState& channel) { void ChannelSetupCaches<P>::CreateChannel(struct Tegra::Control::ChannelState& channel) {
std::unique_lock<std::mutex> lk(config_mutex);
ASSERT(channel_map.find(channel.bind_id) == channel_map.end() && channel.bind_id >= 0); ASSERT(channel_map.find(channel.bind_id) == channel_map.end() && channel.bind_id >= 0);
auto new_id = [this, &channel]() { auto new_id = [this, &channel]() {
if (!free_channel_ids.empty()) { if (!free_channel_ids.empty()) {
auto id = free_channel_ids.front(); auto id = free_channel_ids.front();
free_channel_ids.pop_front(); free_channel_ids.pop_front();
new (&channel_storage[id]) ChannelInfo(channel); new (&channel_storage[id]) P(channel);
return id; return id;
} }
channel_storage.emplace_back(channel); channel_storage.emplace_back(channel);
@ -27,11 +27,24 @@ void ChannelSetupCaches<P>::CreateChannel(struct Tegra::Control::ChannelState& c
if (current_channel_id != UNSET_CHANNEL) { if (current_channel_id != UNSET_CHANNEL) {
channel_state = &channel_storage[current_channel_id]; channel_state = &channel_storage[current_channel_id];
} }
auto as_it = address_spaces.find(channel.memory_manager->GetID());
if (as_it != address_spaces.end()) {
as_it->second.ref_count++;
return;
}
AddresSpaceRef new_gpu_mem_ref{
.ref_count = 1,
.storage_id = address_spaces.size(),
.gpu_memory = channel.memory_manager.get(),
};
address_spaces.emplace(channel.memory_manager->GetID(), new_gpu_mem_ref);
OnGPUASRegister(channel.memory_manager->GetID());
} }
/// Bind a channel for execution. /// Bind a channel for execution.
template <class P> template <class P>
void ChannelSetupCaches<P>::BindToChannel(s32 id) { void ChannelSetupCaches<P>::BindToChannel(s32 id) {
std::unique_lock<std::mutex> lk(config_mutex);
auto it = channel_map.find(id); auto it = channel_map.find(id);
ASSERT(it != channel_map.end() && id >= 0); ASSERT(it != channel_map.end() && id >= 0);
current_channel_id = it->second; current_channel_id = it->second;
@ -39,11 +52,13 @@ void ChannelSetupCaches<P>::BindToChannel(s32 id) {
maxwell3d = &channel_state->maxwell3d; maxwell3d = &channel_state->maxwell3d;
kepler_compute = &channel_state->kepler_compute; kepler_compute = &channel_state->kepler_compute;
gpu_memory = &channel_state->gpu_memory; gpu_memory = &channel_state->gpu_memory;
current_address_space = gpu_memory->GetID();
} }
/// Erase channel's channel_state. /// Erase channel's channel_state.
template <class P> template <class P>
void ChannelSetupCaches<P>::EraseChannel(s32 id) { void ChannelSetupCaches<P>::EraseChannel(s32 id) {
std::unique_lock<std::mutex> lk(config_mutex);
const auto it = channel_map.find(id); const auto it = channel_map.find(id);
ASSERT(it != channel_map.end() && id >= 0); ASSERT(it != channel_map.end() && id >= 0);
const auto this_id = it->second; const auto this_id = it->second;

@ -16,9 +16,12 @@
namespace Tegra { namespace Tegra {
std::atomic<size_t> MemoryManager::unique_identifier_generator{};
MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64 page_bits_) MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, u64 page_bits_)
: system{system_}, address_space_bits{address_space_bits_}, page_bits{page_bits_}, entries{}, : system{system_}, address_space_bits{address_space_bits_}, page_bits{page_bits_}, entries{},
page_table{address_space_bits, address_space_bits + page_bits - 38, page_bits} { page_table{address_space_bits, address_space_bits + page_bits - 38, page_bits},
unique_identifier{unique_identifier_generator.fetch_add(1, std::memory_order_acq_rel)} {
address_space_size = 1ULL << address_space_bits; address_space_size = 1ULL << address_space_bits;
allocate_start = address_space_bits > 32 ? 1ULL << 32 : 0; allocate_start = address_space_bits > 32 ? 1ULL << 32 : 0;
page_size = 1ULL << page_bits; page_size = 1ULL << page_bits;

@ -3,6 +3,7 @@
#pragma once #pragma once
#include <atomic>
#include <map> #include <map>
#include <optional> #include <optional>
#include <vector> #include <vector>
@ -26,6 +27,10 @@ public:
u64 page_bits_ = 16); u64 page_bits_ = 16);
~MemoryManager(); ~MemoryManager();
size_t GetID() const {
return unique_identifier;
}
/// Binds a renderer to the memory manager. /// Binds a renderer to the memory manager.
void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); void BindRasterizer(VideoCore::RasterizerInterface* rasterizer);
@ -140,6 +145,10 @@ private:
void SetEntry(size_t position, EntryType entry); void SetEntry(size_t position, EntryType entry);
Common::MultiLevelPageTable<u32> page_table; Common::MultiLevelPageTable<u32> page_table;
const size_t unique_identifier;
static std::atomic<size_t> unique_identifier_generator;
}; };
} // namespace Tegra } // namespace Tegra

@ -0,0 +1,16 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv3 or any later version
// Refer to the license.txt file included.
#include "video_core/control/channel_state_cache.inc"
#include "video_core/texture_cache/texture_cache_base.h"
namespace VideoCommon {
TextureCacheChannelInfo::TextureCacheChannelInfo(Tegra::Control::ChannelState& state) noexcept
: ChannelInfo(state), graphics_image_table{gpu_memory}, graphics_sampler_table{gpu_memory},
compute_image_table{gpu_memory}, compute_sampler_table{gpu_memory} {}
template class VideoCommon::ChannelSetupCaches<VideoCommon::TextureCacheChannelInfo>;
} // namespace VideoCommon

@ -1,5 +1,7 @@
// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project // SPDX-FileCopyrightText: 2021 yuzu emulator team
// SPDX-License-Identifier: GPL-2.0-or-later // (https://github.com/skyline-emu/)
// SPDX-License-Identifier: GPL-3.0-or-later Licensed under GPLv3
// or any later version Refer to the license.txt file included.
#pragma once #pragma once
@ -41,10 +43,6 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
// Setup channels // Setup channels
current_channel_id = UNSET_CHANNEL; current_channel_id = UNSET_CHANNEL;
state = nullptr;
maxwell3d = nullptr;
kepler_compute = nullptr;
gpu_memory = nullptr;
// Make sure the first index is reserved for the null resources // Make sure the first index is reserved for the null resources
// This way the null resource becomes a compile time constant // This way the null resource becomes a compile time constant
@ -156,23 +154,24 @@ void TextureCache<P>::MarkModification(ImageId id) noexcept {
template <class P> template <class P>
template <bool has_blacklists> template <bool has_blacklists>
void TextureCache<P>::FillGraphicsImageViews(std::span<ImageViewInOut> views) { void TextureCache<P>::FillGraphicsImageViews(std::span<ImageViewInOut> views) {
FillImageViews<has_blacklists>(state->graphics_image_table, state->graphics_image_view_ids, FillImageViews<has_blacklists>(channel_state->graphics_image_table,
views); channel_state->graphics_image_view_ids, views);
} }
template <class P> template <class P>
void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) { void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) {
FillImageViews<true>(state->compute_image_table, state->compute_image_view_ids, views); FillImageViews<true>(channel_state->compute_image_table, channel_state->compute_image_view_ids,
views);
} }
template <class P> template <class P>
typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) { typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) {
if (index > state->graphics_sampler_table.Limit()) { if (index > channel_state->graphics_sampler_table.Limit()) {
LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index);
return &slot_samplers[NULL_SAMPLER_ID]; return &slot_samplers[NULL_SAMPLER_ID];
} }
const auto [descriptor, is_new] = state->graphics_sampler_table.Read(index); const auto [descriptor, is_new] = channel_state->graphics_sampler_table.Read(index);
SamplerId& id = state->graphics_sampler_ids[index]; SamplerId& id = channel_state->graphics_sampler_ids[index];
if (is_new) { if (is_new) {
id = FindSampler(descriptor); id = FindSampler(descriptor);
} }
@ -181,12 +180,12 @@ typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) {
template <class P> template <class P>
typename P::Sampler* TextureCache<P>::GetComputeSampler(u32 index) { typename P::Sampler* TextureCache<P>::GetComputeSampler(u32 index) {
if (index > state->compute_sampler_table.Limit()) { if (index > channel_state->compute_sampler_table.Limit()) {
LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index);
return &slot_samplers[NULL_SAMPLER_ID]; return &slot_samplers[NULL_SAMPLER_ID];
} }
const auto [descriptor, is_new] = state->compute_sampler_table.Read(index); const auto [descriptor, is_new] = channel_state->compute_sampler_table.Read(index);
SamplerId& id = state->compute_sampler_ids[index]; SamplerId& id = channel_state->compute_sampler_ids[index];
if (is_new) { if (is_new) {
id = FindSampler(descriptor); id = FindSampler(descriptor);
} }
@ -199,11 +198,12 @@ void TextureCache<P>::SynchronizeGraphicsDescriptors() {
const bool linked_tsc = maxwell3d->regs.sampler_index == SamplerIndex::ViaHeaderIndex; const bool linked_tsc = maxwell3d->regs.sampler_index == SamplerIndex::ViaHeaderIndex;
const u32 tic_limit = maxwell3d->regs.tic.limit; const u32 tic_limit = maxwell3d->regs.tic.limit;
const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d->regs.tsc.limit; const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d->regs.tsc.limit;
if (state->graphics_sampler_table.Synchornize(maxwell3d->regs.tsc.Address(), tsc_limit)) { if (channel_state->graphics_sampler_table.Synchornize(maxwell3d->regs.tsc.Address(),
state->graphics_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); tsc_limit)) {
channel_state->graphics_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID);
} }
if (state->graphics_image_table.Synchornize(maxwell3d->regs.tic.Address(), tic_limit)) { if (channel_state->graphics_image_table.Synchornize(maxwell3d->regs.tic.Address(), tic_limit)) {
state->graphics_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); channel_state->graphics_image_view_ids.resize(tic_limit + 1, CORRUPT_ID);
} }
} }
@ -213,11 +213,12 @@ void TextureCache<P>::SynchronizeComputeDescriptors() {
const u32 tic_limit = kepler_compute->regs.tic.limit; const u32 tic_limit = kepler_compute->regs.tic.limit;
const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute->regs.tsc.limit; const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute->regs.tsc.limit;
const GPUVAddr tsc_gpu_addr = kepler_compute->regs.tsc.Address(); const GPUVAddr tsc_gpu_addr = kepler_compute->regs.tsc.Address();
if (state->compute_sampler_table.Synchornize(tsc_gpu_addr, tsc_limit)) { if (channel_state->compute_sampler_table.Synchornize(tsc_gpu_addr, tsc_limit)) {
state->compute_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); channel_state->compute_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID);
} }
if (state->compute_image_table.Synchornize(kepler_compute->regs.tic.Address(), tic_limit)) { if (channel_state->compute_image_table.Synchornize(kepler_compute->regs.tic.Address(),
state->compute_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); tic_limit)) {
channel_state->compute_image_view_ids.resize(tic_limit + 1, CORRUPT_ID);
} }
} }
@ -738,7 +739,7 @@ ImageViewId TextureCache<P>::FindImageView(const TICEntry& config) {
if (!IsValidEntry(*gpu_memory, config)) { if (!IsValidEntry(*gpu_memory, config)) {
return NULL_IMAGE_VIEW_ID; return NULL_IMAGE_VIEW_ID;
} }
const auto [pair, is_new] = state->image_views.try_emplace(config); const auto [pair, is_new] = channel_state->image_views.try_emplace(config);
ImageViewId& image_view_id = pair->second; ImageViewId& image_view_id = pair->second;
if (is_new) { if (is_new) {
image_view_id = CreateImageView(config); image_view_id = CreateImageView(config);
@ -1198,7 +1199,7 @@ SamplerId TextureCache<P>::FindSampler(const TSCEntry& config) {
if (std::ranges::all_of(config.raw, [](u64 value) { return value == 0; })) { if (std::ranges::all_of(config.raw, [](u64 value) { return value == 0; })) {
return NULL_SAMPLER_ID; return NULL_SAMPLER_ID;
} }
const auto [pair, is_new] = state->samplers.try_emplace(config); const auto [pair, is_new] = channel_state->samplers.try_emplace(config);
if (is_new) { if (is_new) {
pair->second = slot_samplers.insert(runtime, config); pair->second = slot_samplers.insert(runtime, config);
} }
@ -1327,8 +1328,8 @@ void TextureCache<P>::ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Fu
static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>; static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;
boost::container::small_vector<ImageId, 8> images; boost::container::small_vector<ImageId, 8> images;
ForEachGPUPage(gpu_addr, size, [this, &images, gpu_addr, size, func](u64 page) { ForEachGPUPage(gpu_addr, size, [this, &images, gpu_addr, size, func](u64 page) {
const auto it = state->gpu_page_table.find(page); const auto it = channel_state->gpu_page_table->find(page);
if (it == state->gpu_page_table.end()) { if (it == channel_state->gpu_page_table->end()) {
if constexpr (BOOL_BREAK) { if constexpr (BOOL_BREAK) {
return false; return false;
} else { } else {
@ -1454,8 +1455,9 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
} }
image.lru_index = lru_cache.Insert(image_id, frame_tick); image.lru_index = lru_cache.Insert(image_id, frame_tick);
ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
[this, image_id](u64 page) { state->gpu_page_table[page].push_back(image_id); }); (*channel_state->gpu_page_table)[page].push_back(image_id);
});
if (False(image.flags & ImageFlagBits::Sparse)) { if (False(image.flags & ImageFlagBits::Sparse)) {
auto map_id = auto map_id =
slot_map_views.insert(image.gpu_addr, image.cpu_addr, image.guest_size_bytes, image_id); slot_map_views.insert(image.gpu_addr, image.cpu_addr, image.guest_size_bytes, image_id);
@ -1486,9 +1488,9 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
image.flags &= ~ImageFlagBits::BadOverlap; image.flags &= ~ImageFlagBits::BadOverlap;
lru_cache.Free(image.lru_index); lru_cache.Free(image.lru_index);
const auto& clear_page_table = const auto& clear_page_table =
[this, image_id]( [this, image_id](u64 page,
u64 page, std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>>&
std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>>& selected_page_table) { selected_page_table) {
const auto page_it = selected_page_table.find(page); const auto page_it = selected_page_table.find(page);
if (page_it == selected_page_table.end()) { if (page_it == selected_page_table.end()) {
ASSERT_MSG(false, "Unregistering unregistered page=0x{:x}", page << YUZU_PAGEBITS); ASSERT_MSG(false, "Unregistering unregistered page=0x{:x}", page << YUZU_PAGEBITS);
@ -1504,7 +1506,7 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
image_ids.erase(vector_it); image_ids.erase(vector_it);
}; };
ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, &clear_page_table](u64 page) { ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, &clear_page_table](u64 page) {
clear_page_table(page, state->gpu_page_table); clear_page_table(page, (*channel_state->gpu_page_table));
}); });
if (False(image.flags & ImageFlagBits::Sparse)) { if (False(image.flags & ImageFlagBits::Sparse)) {
const auto map_id = image.map_view_id; const auto map_id = image.map_view_id;
@ -1701,11 +1703,11 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
template <class P> template <class P>
void TextureCache<P>::RemoveImageViewReferences(std::span<const ImageViewId> removed_views) { void TextureCache<P>::RemoveImageViewReferences(std::span<const ImageViewId> removed_views) {
auto it = state->image_views.begin(); auto it = channel_state->image_views.begin();
while (it != state->image_views.end()) { while (it != channel_state->image_views.end()) {
const auto found = std::ranges::find(removed_views, it->second); const auto found = std::ranges::find(removed_views, it->second);
if (found != removed_views.end()) { if (found != removed_views.end()) {
it = state->image_views.erase(it); it = channel_state->image_views.erase(it);
} else { } else {
++it; ++it;
} }
@ -1967,61 +1969,19 @@ bool TextureCache<P>::IsFullClear(ImageViewId id) {
scissor.max_y >= size.height; scissor.max_y >= size.height;
} }
template <class P>
TextureCache<P>::ChannelInfo::ChannelInfo(Tegra::Control::ChannelState& state) noexcept
: maxwell3d{*state.maxwell_3d}, kepler_compute{*state.kepler_compute},
gpu_memory{*state.memory_manager}, graphics_image_table{gpu_memory},
graphics_sampler_table{gpu_memory}, compute_image_table{gpu_memory}, compute_sampler_table{
gpu_memory} {}
template <class P> template <class P>
void TextureCache<P>::CreateChannel(struct Tegra::Control::ChannelState& channel) { void TextureCache<P>::CreateChannel(struct Tegra::Control::ChannelState& channel) {
ASSERT(channel_map.find(channel.bind_id) == channel_map.end() && channel.bind_id >= 0); VideoCommon::ChannelSetupCaches<TextureCacheChannelInfo>::CreateChannel(channel);
auto new_id = [this, &channel]() { const auto it = channel_map.find(channel.bind_id);
if (!free_channel_ids.empty()) { auto* this_state = &channel_storage[it->second];
auto id = free_channel_ids.front(); const auto& this_as_ref = address_spaces[channel.memory_manager->GetID()];
free_channel_ids.pop_front(); this_state->gpu_page_table = &gpu_page_table_storage[this_as_ref.storage_id];
new (&channel_storage[id]) ChannelInfo(channel);
return id;
}
channel_storage.emplace_back(channel);
return channel_storage.size() - 1;
}();
channel_map.emplace(channel.bind_id, new_id);
if (current_channel_id != UNSET_CHANNEL) {
state = &channel_storage[current_channel_id];
}
} }
/// Bind a channel for execution. /// Bind a channel for execution.
template <class P> template <class P>
void TextureCache<P>::BindToChannel(s32 id) { void TextureCache<P>::OnGPUASRegister([[maybe_unused]] size_t map_id) {
auto it = channel_map.find(id); gpu_page_table_storage.emplace_back();
ASSERT(it != channel_map.end() && id >= 0);
current_channel_id = it->second;
state = &channel_storage[current_channel_id];
maxwell3d = &state->maxwell3d;
kepler_compute = &state->kepler_compute;
gpu_memory = &state->gpu_memory;
}
/// Erase channel's state.
template <class P>
void TextureCache<P>::EraseChannel(s32 id) {
const auto it = channel_map.find(id);
ASSERT(it != channel_map.end() && id >= 0);
const auto this_id = it->second;
free_channel_ids.push_back(this_id);
channel_map.erase(it);
if (this_id == current_channel_id) {
current_channel_id = UNSET_CHANNEL;
state = nullptr;
maxwell3d = nullptr;
kepler_compute = nullptr;
gpu_memory = nullptr;
} else if (current_channel_id != UNSET_CHANNEL) {
state = &channel_storage[current_channel_id];
}
} }
} // namespace VideoCommon } // namespace VideoCommon

@ -1,5 +1,7 @@
// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project // SPDX-FileCopyrightText: 2021 yuzu emulator team
// SPDX-License-Identifier: GPL-2.0-or-later // (https://github.com/skyline-emu/)
// SPDX-License-Identifier: GPL-3.0-or-later Licensed under GPLv3
// or any later version Refer to the license.txt file included.
#pragma once #pragma once
@ -13,9 +15,11 @@
#include <queue> #include <queue>
#include "common/common_types.h" #include "common/common_types.h"
#include "common/hash.h"
#include "common/literals.h" #include "common/literals.h"
#include "common/lru_cache.h" #include "common/lru_cache.h"
#include "video_core/compatible_formats.h" #include "video_core/compatible_formats.h"
#include "video_core/control/channel_state_cache.h"
#include "video_core/delayed_destruction_ring.h" #include "video_core/delayed_destruction_ring.h"
#include "video_core/engines/fermi_2d.h" #include "video_core/engines/fermi_2d.h"
#include "video_core/surface.h" #include "video_core/surface.h"
@ -50,8 +54,35 @@ struct ImageViewInOut {
ImageViewId id{}; ImageViewId id{};
}; };
using TextureCacheGPUMap = std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>>;
class TextureCacheChannelInfo : public ChannelInfo {
public:
TextureCacheChannelInfo() = delete;
TextureCacheChannelInfo(Tegra::Control::ChannelState& state) noexcept;
TextureCacheChannelInfo(const TextureCacheChannelInfo& state) = delete;
TextureCacheChannelInfo& operator=(const TextureCacheChannelInfo&) = delete;
TextureCacheChannelInfo(TextureCacheChannelInfo&& other) noexcept = default;
TextureCacheChannelInfo& operator=(TextureCacheChannelInfo&& other) noexcept = default;
DescriptorTable<TICEntry> graphics_image_table{gpu_memory};
DescriptorTable<TSCEntry> graphics_sampler_table{gpu_memory};
std::vector<SamplerId> graphics_sampler_ids;
std::vector<ImageViewId> graphics_image_view_ids;
DescriptorTable<TICEntry> compute_image_table{gpu_memory};
DescriptorTable<TSCEntry> compute_sampler_table{gpu_memory};
std::vector<SamplerId> compute_sampler_ids;
std::vector<ImageViewId> compute_image_view_ids;
std::unordered_map<TICEntry, ImageViewId> image_views;
std::unordered_map<TSCEntry, SamplerId> samplers;
TextureCacheGPUMap* gpu_page_table;
};
template <class P> template <class P>
class TextureCache { class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelInfo> {
/// Address shift for caching images into a hash table /// Address shift for caching images into a hash table
static constexpr u64 YUZU_PAGEBITS = 20; static constexpr u64 YUZU_PAGEBITS = 20;
@ -85,13 +116,6 @@ class TextureCache {
PixelFormat src_format; PixelFormat src_format;
}; };
template <typename T>
struct IdentityHash {
[[nodiscard]] size_t operator()(T value) const noexcept {
return static_cast<size_t>(value);
}
};
public: public:
explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&); explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&);
@ -179,13 +203,7 @@ public:
[[nodiscard]] bool IsRescaling(const ImageViewBase& image_view) const noexcept; [[nodiscard]] bool IsRescaling(const ImageViewBase& image_view) const noexcept;
/// Create channel state. /// Create channel state.
void CreateChannel(struct Tegra::Control::ChannelState& channel); void CreateChannel(Tegra::Control::ChannelState& channel) final override;
/// Bind a channel for execution.
void BindToChannel(s32 id);
/// Erase channel's state.
void EraseChannel(s32 id);
std::mutex mutex; std::mutex mutex;
@ -221,6 +239,8 @@ private:
} }
} }
void OnGPUASRegister(size_t map_id) final override;
/// Runs the Garbage Collector. /// Runs the Garbage Collector.
void RunGarbageCollector(); void RunGarbageCollector();
@ -355,51 +375,15 @@ private:
Runtime& runtime; Runtime& runtime;
struct ChannelInfo {
ChannelInfo() = delete;
ChannelInfo(struct Tegra::Control::ChannelState& state) noexcept;
ChannelInfo(const ChannelInfo& state) = delete;
ChannelInfo& operator=(const ChannelInfo&) = delete;
ChannelInfo(ChannelInfo&& other) noexcept = default;
ChannelInfo& operator=(ChannelInfo&& other) noexcept = default;
Tegra::Engines::Maxwell3D& maxwell3d;
Tegra::Engines::KeplerCompute& kepler_compute;
Tegra::MemoryManager& gpu_memory;
DescriptorTable<TICEntry> graphics_image_table{gpu_memory};
DescriptorTable<TSCEntry> graphics_sampler_table{gpu_memory};
std::vector<SamplerId> graphics_sampler_ids;
std::vector<ImageViewId> graphics_image_view_ids;
DescriptorTable<TICEntry> compute_image_table{gpu_memory};
DescriptorTable<TSCEntry> compute_sampler_table{gpu_memory};
std::vector<SamplerId> compute_sampler_ids;
std::vector<ImageViewId> compute_image_view_ids;
std::unordered_map<TICEntry, ImageViewId> image_views;
std::unordered_map<TSCEntry, SamplerId> samplers;
std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> gpu_page_table;
};
std::deque<ChannelInfo> channel_storage;
std::deque<size_t> free_channel_ids;
std::unordered_map<s32, size_t> channel_map;
ChannelInfo* state;
size_t current_channel_id{UNSET_CHANNEL};
VideoCore::RasterizerInterface& rasterizer; VideoCore::RasterizerInterface& rasterizer;
Tegra::Engines::Maxwell3D* maxwell3d; std::deque<TextureCacheGPUMap> gpu_page_table_storage;
Tegra::Engines::KeplerCompute* kepler_compute;
Tegra::MemoryManager* gpu_memory;
RenderTargets render_targets; RenderTargets render_targets;
std::unordered_map<RenderTargets, FramebufferId> framebuffers; std::unordered_map<RenderTargets, FramebufferId> framebuffers;
std::unordered_map<u64, std::vector<ImageMapId>, IdentityHash<u64>> page_table; std::unordered_map<u64, std::vector<ImageMapId>, Common::IdentityHash<u64>> page_table;
std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> sparse_page_table; std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>> sparse_page_table;
std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views; std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views;
VAddr virtual_invalid_space{}; VAddr virtual_invalid_space{};