diff options
Diffstat (limited to 'src')
200 files changed, 6825 insertions, 3358 deletions
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index 54de1dc94..3575a3cb3 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -121,6 +121,7 @@ else() if (ARCHITECTURE_x86_64) add_compile_options("-mcx16") + add_compile_options("-fwrapv") endif() if (APPLE AND CMAKE_CXX_COMPILER_ID STREQUAL Clang) diff --git a/src/audio_core/renderer/adsp/audio_renderer.cpp b/src/audio_core/renderer/adsp/audio_renderer.cpp index ab2257bd8..d982ef630 100644 --- a/src/audio_core/renderer/adsp/audio_renderer.cpp +++ b/src/audio_core/renderer/adsp/audio_renderer.cpp @@ -132,7 +132,7 @@ void AudioRenderer::CreateSinkStreams() { } void AudioRenderer::ThreadFunc() { - constexpr char name[]{"yuzu:AudioRenderer"}; + constexpr char name[]{"AudioRenderer"}; MicroProfileOnThreadCreate(name); Common::SetCurrentThreadName(name); Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical); diff --git a/src/audio_core/renderer/system_manager.cpp b/src/audio_core/renderer/system_manager.cpp index 9c1331e19..f66b2b890 100644 --- a/src/audio_core/renderer/system_manager.cpp +++ b/src/audio_core/renderer/system_manager.cpp @@ -94,7 +94,7 @@ bool SystemManager::Remove(System& system_) { } void SystemManager::ThreadFunc() { - constexpr char name[]{"yuzu:AudioRenderSystemManager"}; + constexpr char name[]{"AudioRenderSystemManager"}; MicroProfileOnThreadCreate(name); Common::SetCurrentThreadName(name); Common::SetCurrentThreadPriority(Common::ThreadPriority::High); diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt index 3447fabd8..a02696873 100644 --- a/src/common/CMakeLists.txt +++ b/src/common/CMakeLists.txt @@ -17,6 +17,8 @@ endif () include(GenerateSCMRev) add_library(common STATIC + address_space.cpp + address_space.h algorithm.h alignment.h announce_multiplayer_room.h @@ -81,6 +83,8 @@ add_library(common STATIC microprofile.cpp microprofile.h microprofileui.h + multi_level_page_table.cpp + multi_level_page_table.h nvidia_flags.cpp nvidia_flags.h page_table.cpp diff --git a/src/common/address_space.cpp b/src/common/address_space.cpp new file mode 100644 index 000000000..866e78dbe --- /dev/null +++ b/src/common/address_space.cpp @@ -0,0 +1,10 @@ +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/address_space.inc" + +namespace Common { + +template class Common::FlatAllocator<u32, 0, 32>; + +} diff --git a/src/common/address_space.h b/src/common/address_space.h new file mode 100644 index 000000000..9222b2fdc --- /dev/null +++ b/src/common/address_space.h @@ -0,0 +1,150 @@ +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <concepts> +#include <functional> +#include <mutex> +#include <vector> + +#include "common/common_types.h" + +namespace Common { +template <typename VaType, size_t AddressSpaceBits> +concept AddressSpaceValid = std::is_unsigned_v<VaType> && sizeof(VaType) * 8 >= AddressSpaceBits; + +struct EmptyStruct {}; + +/** + * @brief FlatAddressSpaceMap provides a generic VA->PA mapping implementation using a sorted vector + */ +template <typename VaType, VaType UnmappedVa, typename PaType, PaType UnmappedPa, + bool PaContigSplit, size_t AddressSpaceBits, typename ExtraBlockInfo = EmptyStruct> +requires AddressSpaceValid<VaType, AddressSpaceBits> +class FlatAddressSpaceMap { +public: + /// The maximum VA that this AS can technically reach + static constexpr VaType VaMaximum{(1ULL << (AddressSpaceBits - 1)) + + ((1ULL << (AddressSpaceBits - 1)) - 1)}; + + explicit FlatAddressSpaceMap(VaType va_limit, + std::function<void(VaType, VaType)> unmap_callback = {}); + + FlatAddressSpaceMap() = default; + + void Map(VaType virt, PaType phys, VaType size, ExtraBlockInfo extra_info = {}) { + std::scoped_lock lock(block_mutex); + MapLocked(virt, phys, size, extra_info); + } + + void Unmap(VaType virt, VaType size) { + std::scoped_lock lock(block_mutex); + UnmapLocked(virt, size); + } + + VaType GetVALimit() const { + return va_limit; + } + +protected: + /** + * @brief Represents a block of memory in the AS, the physical mapping is contiguous until + * another block with a different phys address is hit + */ + struct Block { + /// VA of the block + VaType virt{UnmappedVa}; + /// PA of the block, will increase 1-1 with VA until a new block is encountered + PaType phys{UnmappedPa}; + [[no_unique_address]] ExtraBlockInfo extra_info; + + Block() = default; + + Block(VaType virt_, PaType phys_, ExtraBlockInfo extra_info_) + : virt(virt_), phys(phys_), extra_info(extra_info_) {} + + bool Valid() const { + return virt != UnmappedVa; + } + + bool Mapped() const { + return phys != UnmappedPa; + } + + bool Unmapped() const { + return phys == UnmappedPa; + } + + bool operator<(const VaType& p_virt) const { + return virt < p_virt; + } + }; + + /** + * @brief Maps a PA range into the given AS region + * @note block_mutex MUST be locked when calling this + */ + void MapLocked(VaType virt, PaType phys, VaType size, ExtraBlockInfo extra_info); + + /** + * @brief Unmaps the given range and merges it with other unmapped regions + * @note block_mutex MUST be locked when calling this + */ + void UnmapLocked(VaType virt, VaType size); + + std::mutex block_mutex; + std::vector<Block> blocks{Block{}}; + + /// a soft limit on the maximum VA of the AS + VaType va_limit{VaMaximum}; + +private: + /// Callback called when the mappings in an region have changed + std::function<void(VaType, VaType)> unmap_callback{}; +}; + +/** + * @brief FlatMemoryManager specialises FlatAddressSpaceMap to work as an allocator, with an + * initial, fast linear pass and a subsequent slower pass that iterates until it finds a free block + */ +template <typename VaType, VaType UnmappedVa, size_t AddressSpaceBits> +requires AddressSpaceValid<VaType, AddressSpaceBits> +class FlatAllocator + : public FlatAddressSpaceMap<VaType, UnmappedVa, bool, false, false, AddressSpaceBits> { +private: + using Base = FlatAddressSpaceMap<VaType, UnmappedVa, bool, false, false, AddressSpaceBits>; + +public: + explicit FlatAllocator(VaType virt_start, VaType va_limit = Base::VaMaximum); + + /** + * @brief Allocates a region in the AS of the given size and returns its address + */ + VaType Allocate(VaType size); + + /** + * @brief Marks the given region in the AS as allocated + */ + void AllocateFixed(VaType virt, VaType size); + + /** + * @brief Frees an AS region so it can be used again + */ + void Free(VaType virt, VaType size); + + VaType GetVAStart() const { + return virt_start; + } + +private: + /// The base VA of the allocator, no allocations will be below this + VaType virt_start; + + /** + * The end address for the initial linear allocation pass + * Once this reaches the AS limit the slower allocation path will be used + */ + VaType current_linear_alloc_end; +}; +} // namespace Common diff --git a/src/common/address_space.inc b/src/common/address_space.inc new file mode 100644 index 000000000..2195dabd5 --- /dev/null +++ b/src/common/address_space.inc @@ -0,0 +1,366 @@ +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/address_space.h" +#include "common/assert.h" + +#define MAP_MEMBER(returnType) \ + template <typename VaType, VaType UnmappedVa, typename PaType, PaType UnmappedPa, \ + bool PaContigSplit, size_t AddressSpaceBits, typename ExtraBlockInfo> \ + requires AddressSpaceValid<VaType, AddressSpaceBits> returnType FlatAddressSpaceMap< \ + VaType, UnmappedVa, PaType, UnmappedPa, PaContigSplit, AddressSpaceBits, ExtraBlockInfo> +#define MAP_MEMBER_CONST() \ + template <typename VaType, VaType UnmappedVa, typename PaType, PaType UnmappedPa, \ + bool PaContigSplit, size_t AddressSpaceBits, typename ExtraBlockInfo> \ + requires AddressSpaceValid<VaType, AddressSpaceBits> FlatAddressSpaceMap< \ + VaType, UnmappedVa, PaType, UnmappedPa, PaContigSplit, AddressSpaceBits, ExtraBlockInfo> + +#define MM_MEMBER(returnType) \ + template <typename VaType, VaType UnmappedVa, size_t AddressSpaceBits> \ + requires AddressSpaceValid<VaType, AddressSpaceBits> returnType \ + FlatMemoryManager<VaType, UnmappedVa, AddressSpaceBits> + +#define ALLOC_MEMBER(returnType) \ + template <typename VaType, VaType UnmappedVa, size_t AddressSpaceBits> \ + requires AddressSpaceValid<VaType, AddressSpaceBits> returnType \ + FlatAllocator<VaType, UnmappedVa, AddressSpaceBits> +#define ALLOC_MEMBER_CONST() \ + template <typename VaType, VaType UnmappedVa, size_t AddressSpaceBits> \ + requires AddressSpaceValid<VaType, AddressSpaceBits> \ + FlatAllocator<VaType, UnmappedVa, AddressSpaceBits> + +namespace Common { +MAP_MEMBER_CONST()::FlatAddressSpaceMap(VaType va_limit_, + std::function<void(VaType, VaType)> unmap_callback_) + : va_limit{va_limit_}, unmap_callback{std::move(unmap_callback_)} { + if (va_limit > VaMaximum) { + ASSERT_MSG(false, "Invalid VA limit!"); + } +} + +MAP_MEMBER(void)::MapLocked(VaType virt, PaType phys, VaType size, ExtraBlockInfo extra_info) { + VaType virt_end{virt + size}; + + if (virt_end > va_limit) { + ASSERT_MSG(false, + "Trying to map a block past the VA limit: virt_end: 0x{:X}, va_limit: 0x{:X}", + virt_end, va_limit); + } + + auto block_end_successor{std::lower_bound(blocks.begin(), blocks.end(), virt_end)}; + if (block_end_successor == blocks.begin()) { + ASSERT_MSG(false, "Trying to map a block before the VA start: virt_end: 0x{:X}", virt_end); + } + + auto block_end_predecessor{std::prev(block_end_successor)}; + + if (block_end_successor != blocks.end()) { + // We have blocks in front of us, if one is directly in front then we don't have to add a + // tail + if (block_end_successor->virt != virt_end) { + PaType tailPhys{[&]() -> PaType { + if constexpr (!PaContigSplit) { + // Always propagate unmapped regions rather than calculating offset + return block_end_predecessor->phys; + } else { + if (block_end_predecessor->Unmapped()) { + // Always propagate unmapped regions rather than calculating offset + return block_end_predecessor->phys; + } else { + return block_end_predecessor->phys + virt_end - block_end_predecessor->virt; + } + } + }()}; + + if (block_end_predecessor->virt >= virt) { + // If this block's start would be overlapped by the map then reuse it as a tail + // block + block_end_predecessor->virt = virt_end; + block_end_predecessor->phys = tailPhys; + block_end_predecessor->extra_info = block_end_predecessor->extra_info; + + // No longer predecessor anymore + block_end_successor = block_end_predecessor--; + } else { + // Else insert a new one and we're done + blocks.insert(block_end_successor, + {Block(virt, phys, extra_info), + Block(virt_end, tailPhys, block_end_predecessor->extra_info)}); + if (unmap_callback) { + unmap_callback(virt, size); + } + + return; + } + } + } else { + // block_end_predecessor will always be unmapped as blocks has to be terminated by an + // unmapped chunk + if (block_end_predecessor != blocks.begin() && block_end_predecessor->virt >= virt) { + // Move the unmapped block start backwards + block_end_predecessor->virt = virt_end; + + // No longer predecessor anymore + block_end_successor = block_end_predecessor--; + } else { + // Else insert a new one and we're done + blocks.insert(block_end_successor, + {Block(virt, phys, extra_info), Block(virt_end, UnmappedPa, {})}); + if (unmap_callback) { + unmap_callback(virt, size); + } + + return; + } + } + + auto block_start_successor{block_end_successor}; + + // Walk the block vector to find the start successor as this is more efficient than another + // binary search in most scenarios + while (std::prev(block_start_successor)->virt >= virt) { + block_start_successor--; + } + + // Check that the start successor is either the end block or something in between + if (block_start_successor->virt > virt_end) { + ASSERT_MSG(false, "Unsorted block in AS map: virt: 0x{:X}", block_start_successor->virt); + } else if (block_start_successor->virt == virt_end) { + // We need to create a new block as there are none spare that we would overwrite + blocks.insert(block_start_successor, Block(virt, phys, extra_info)); + } else { + // Erase overwritten blocks + if (auto eraseStart{std::next(block_start_successor)}; eraseStart != block_end_successor) { + blocks.erase(eraseStart, block_end_successor); + } + + // Reuse a block that would otherwise be overwritten as a start block + block_start_successor->virt = virt; + block_start_successor->phys = phys; + block_start_successor->extra_info = extra_info; + } + + if (unmap_callback) { + unmap_callback(virt, size); + } +} + +MAP_MEMBER(void)::UnmapLocked(VaType virt, VaType size) { + VaType virt_end{virt + size}; + + if (virt_end > va_limit) { + ASSERT_MSG(false, + "Trying to map a block past the VA limit: virt_end: 0x{:X}, va_limit: 0x{:X}", + virt_end, va_limit); + } + + auto block_end_successor{std::lower_bound(blocks.begin(), blocks.end(), virt_end)}; + if (block_end_successor == blocks.begin()) { + ASSERT_MSG(false, "Trying to unmap a block before the VA start: virt_end: 0x{:X}", + virt_end); + } + + auto block_end_predecessor{std::prev(block_end_successor)}; + + auto walk_back_to_predecessor{[&](auto iter) { + while (iter->virt >= virt) { + iter--; + } + + return iter; + }}; + + auto erase_blocks_with_end_unmapped{[&](auto unmappedEnd) { + auto block_start_predecessor{walk_back_to_predecessor(unmappedEnd)}; + auto block_start_successor{std::next(block_start_predecessor)}; + + auto eraseEnd{[&]() { + if (block_start_predecessor->Unmapped()) { + // If the start predecessor is unmapped then we can erase everything in our region + // and be done + return std::next(unmappedEnd); + } else { + // Else reuse the end predecessor as the start of our unmapped region then erase all + // up to it + unmappedEnd->virt = virt; + return unmappedEnd; + } + }()}; + + // We can't have two unmapped regions after each other + if (eraseEnd != blocks.end() && + (eraseEnd == block_start_successor || + (block_start_predecessor->Unmapped() && eraseEnd->Unmapped()))) { + ASSERT_MSG(false, "Multiple contiguous unmapped regions are unsupported!"); + } + + blocks.erase(block_start_successor, eraseEnd); + }}; + + // We can avoid any splitting logic if these are the case + if (block_end_predecessor->Unmapped()) { + if (block_end_predecessor->virt > virt) { + erase_blocks_with_end_unmapped(block_end_predecessor); + } + + if (unmap_callback) { + unmap_callback(virt, size); + } + + return; // The region is unmapped, bail out early + } else if (block_end_successor->virt == virt_end && block_end_successor->Unmapped()) { + erase_blocks_with_end_unmapped(block_end_successor); + + if (unmap_callback) { + unmap_callback(virt, size); + } + + return; // The region is unmapped here and doesn't need splitting, bail out early + } else if (block_end_successor == blocks.end()) { + // This should never happen as the end should always follow an unmapped block + ASSERT_MSG(false, "Unexpected Memory Manager state!"); + } else if (block_end_successor->virt != virt_end) { + // If one block is directly in front then we don't have to add a tail + + // The previous block is mapped so we will need to add a tail with an offset + PaType tailPhys{[&]() { + if constexpr (PaContigSplit) { + return block_end_predecessor->phys + virt_end - block_end_predecessor->virt; + } else { + return block_end_predecessor->phys; + } + }()}; + + if (block_end_predecessor->virt >= virt) { + // If this block's start would be overlapped by the unmap then reuse it as a tail block + block_end_predecessor->virt = virt_end; + block_end_predecessor->phys = tailPhys; + + // No longer predecessor anymore + block_end_successor = block_end_predecessor--; + } else { + blocks.insert(block_end_successor, + {Block(virt, UnmappedPa, {}), + Block(virt_end, tailPhys, block_end_predecessor->extra_info)}); + if (unmap_callback) { + unmap_callback(virt, size); + } + + // The previous block is mapped and ends before + return; + } + } + + // Walk the block vector to find the start predecessor as this is more efficient than another + // binary search in most scenarios + auto block_start_predecessor{walk_back_to_predecessor(block_end_successor)}; + auto block_start_successor{std::next(block_start_predecessor)}; + + if (block_start_successor->virt > virt_end) { + ASSERT_MSG(false, "Unsorted block in AS map: virt: 0x{:X}", block_start_successor->virt); + } else if (block_start_successor->virt == virt_end) { + // There are no blocks between the start and the end that would let us skip inserting a new + // one for head + + // The previous block is may be unmapped, if so we don't need to insert any unmaps after it + if (block_start_predecessor->Mapped()) { + blocks.insert(block_start_successor, Block(virt, UnmappedPa, {})); + } + } else if (block_start_predecessor->Unmapped()) { + // If the previous block is unmapped + blocks.erase(block_start_successor, block_end_predecessor); + } else { + // Erase overwritten blocks, skipping the first one as we have written the unmapped start + // block there + if (auto eraseStart{std::next(block_start_successor)}; eraseStart != block_end_successor) { + blocks.erase(eraseStart, block_end_successor); + } + + // Add in the unmapped block header + block_start_successor->virt = virt; + block_start_successor->phys = UnmappedPa; + } + + if (unmap_callback) + unmap_callback(virt, size); +} + +ALLOC_MEMBER_CONST()::FlatAllocator(VaType virt_start_, VaType va_limit_) + : Base{va_limit_}, virt_start{virt_start_}, current_linear_alloc_end{virt_start_} {} + +ALLOC_MEMBER(VaType)::Allocate(VaType size) { + std::scoped_lock lock(this->block_mutex); + + VaType alloc_start{UnmappedVa}; + VaType alloc_end{current_linear_alloc_end + size}; + + // Avoid searching backwards in the address space if possible + if (alloc_end >= current_linear_alloc_end && alloc_end <= this->va_limit) { + auto alloc_end_successor{ + std::lower_bound(this->blocks.begin(), this->blocks.end(), alloc_end)}; + if (alloc_end_successor == this->blocks.begin()) { + ASSERT_MSG(false, "First block in AS map is invalid!"); + } + + auto alloc_end_predecessor{std::prev(alloc_end_successor)}; + if (alloc_end_predecessor->virt <= current_linear_alloc_end) { + alloc_start = current_linear_alloc_end; + } else { + // Skip over fixed any mappings in front of us + while (alloc_end_successor != this->blocks.end()) { + if (alloc_end_successor->virt - alloc_end_predecessor->virt < size || + alloc_end_predecessor->Mapped()) { + alloc_start = alloc_end_predecessor->virt; + break; + } + + alloc_end_predecessor = alloc_end_successor++; + + // Use the VA limit to calculate if we can fit in the final block since it has no + // successor + if (alloc_end_successor == this->blocks.end()) { + alloc_end = alloc_end_predecessor->virt + size; + + if (alloc_end >= alloc_end_predecessor->virt && alloc_end <= this->va_limit) { + alloc_start = alloc_end_predecessor->virt; + } + } + } + } + } + + if (alloc_start != UnmappedVa) { + current_linear_alloc_end = alloc_start + size; + } else { // If linear allocation overflows the AS then find a gap + if (this->blocks.size() <= 2) { + ASSERT_MSG(false, "Unexpected allocator state!"); + } + + auto search_predecessor{this->blocks.begin()}; + auto search_successor{std::next(search_predecessor)}; + + while (search_successor != this->blocks.end() && + (search_successor->virt - search_predecessor->virt < size || + search_predecessor->Mapped())) { + search_predecessor = search_successor++; + } + + if (search_successor != this->blocks.end()) { + alloc_start = search_predecessor->virt; + } else { + return {}; // AS is full + } + } + + this->MapLocked(alloc_start, true, size, {}); + return alloc_start; +} + +ALLOC_MEMBER(void)::AllocateFixed(VaType virt, VaType size) { + this->Map(virt, true, size); +} + +ALLOC_MEMBER(void)::Free(VaType virt, VaType size) { + this->Unmap(virt, size); +} +} // namespace Common diff --git a/src/common/algorithm.h b/src/common/algorithm.h index 9ddfd637b..c27c9241d 100644 --- a/src/common/algorithm.h +++ b/src/common/algorithm.h @@ -24,4 +24,12 @@ template <class ForwardIt, class T, class Compare = std::less<>> return first != last && !comp(value, *first) ? first : last; } +template <typename T, typename Func, typename... Args> +T FoldRight(T initial_value, Func&& func, Args&&... args) { + T value{initial_value}; + const auto high_func = [&value, &func]<typename U>(U x) { value = func(value, x); }; + (std::invoke(high_func, std::forward<Args>(args)), ...); + return value; +} + } // namespace Common diff --git a/src/common/hash.h b/src/common/hash.h index b6f3e6d6f..e8fe78b07 100644 --- a/src/common/hash.h +++ b/src/common/hash.h @@ -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 diff --git a/src/common/logging/backend.cpp b/src/common/logging/backend.cpp index 8ce1c2fd1..15d92505e 100644 --- a/src/common/logging/backend.cpp +++ b/src/common/logging/backend.cpp @@ -219,7 +219,7 @@ private: void StartBackendThread() { backend_thread = std::jthread([this](std::stop_token stop_token) { - Common::SetCurrentThreadName("yuzu:Log"); + Common::SetCurrentThreadName("Logger"); Entry entry; const auto write_logs = [this, &entry]() { ForEachBackend([&entry](Backend& backend) { backend.Write(entry); }); diff --git a/src/common/multi_level_page_table.cpp b/src/common/multi_level_page_table.cpp new file mode 100644 index 000000000..46e362f3b --- /dev/null +++ b/src/common/multi_level_page_table.cpp @@ -0,0 +1,9 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#include "common/multi_level_page_table.inc" + +namespace Common { +template class Common::MultiLevelPageTable<u64>; +template class Common::MultiLevelPageTable<u32>; +} // namespace Common diff --git a/src/common/multi_level_page_table.h b/src/common/multi_level_page_table.h new file mode 100644 index 000000000..31f6676a0 --- /dev/null +++ b/src/common/multi_level_page_table.h @@ -0,0 +1,78 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#pragma once + +#include <type_traits> +#include <utility> +#include <vector> + +#include "common/common_types.h" + +namespace Common { + +template <typename BaseAddr> +class MultiLevelPageTable final { +public: + constexpr MultiLevelPageTable() = default; + explicit MultiLevelPageTable(std::size_t address_space_bits, std::size_t first_level_bits, + std::size_t page_bits); + + ~MultiLevelPageTable() noexcept; + + MultiLevelPageTable(const MultiLevelPageTable&) = delete; + MultiLevelPageTable& operator=(const MultiLevelPageTable&) = delete; + + MultiLevelPageTable(MultiLevelPageTable&& other) noexcept + : address_space_bits{std::exchange(other.address_space_bits, 0)}, + first_level_bits{std::exchange(other.first_level_bits, 0)}, page_bits{std::exchange( + other.page_bits, 0)}, + first_level_shift{std::exchange(other.first_level_shift, 0)}, + first_level_chunk_size{std::exchange(other.first_level_chunk_size, 0)}, + first_level_map{std::move(other.first_level_map)}, base_ptr{std::exchange(other.base_ptr, + nullptr)} {} + + MultiLevelPageTable& operator=(MultiLevelPageTable&& other) noexcept { + address_space_bits = std::exchange(other.address_space_bits, 0); + first_level_bits = std::exchange(other.first_level_bits, 0); + page_bits = std::exchange(other.page_bits, 0); + first_level_shift = std::exchange(other.first_level_shift, 0); + first_level_chunk_size = std::exchange(other.first_level_chunk_size, 0); + alloc_size = std::exchange(other.alloc_size, 0); + first_level_map = std::move(other.first_level_map); + base_ptr = std::exchange(other.base_ptr, nullptr); + return *this; + } + + void ReserveRange(u64 start, std::size_t size); + + [[nodiscard]] const BaseAddr& operator[](std::size_t index) const { + return base_ptr[index]; + } + + [[nodiscard]] BaseAddr& operator[](std::size_t index) { + return base_ptr[index]; + } + + [[nodiscard]] BaseAddr* data() { + return base_ptr; + } + + [[nodiscard]] const BaseAddr* data() const { + return base_ptr; + } + +private: + void AllocateLevel(u64 level); + + std::size_t address_space_bits{}; + std::size_t first_level_bits{}; + std::size_t page_bits{}; + std::size_t first_level_shift{}; + std::size_t first_level_chunk_size{}; + std::size_t alloc_size{}; + std::vector<void*> first_level_map{}; + BaseAddr* base_ptr{}; +}; + +} // namespace Common diff --git a/src/common/multi_level_page_table.inc b/src/common/multi_level_page_table.inc new file mode 100644 index 000000000..8ac506fa0 --- /dev/null +++ b/src/common/multi_level_page_table.inc @@ -0,0 +1,84 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#ifdef _WIN32 +#include <windows.h> +#else +#include <sys/mman.h> +#endif + +#include "common/assert.h" +#include "common/multi_level_page_table.h" + +namespace Common { + +template <typename BaseAddr> +MultiLevelPageTable<BaseAddr>::MultiLevelPageTable(std::size_t address_space_bits_, + std::size_t first_level_bits_, + std::size_t page_bits_) + : address_space_bits{address_space_bits_}, + first_level_bits{first_level_bits_}, page_bits{page_bits_} { + if (page_bits == 0) { + return; + } + first_level_shift = address_space_bits - first_level_bits; + first_level_chunk_size = (1ULL << (first_level_shift - page_bits)) * sizeof(BaseAddr); + alloc_size = (1ULL << (address_space_bits - page_bits)) * sizeof(BaseAddr); + std::size_t first_level_size = 1ULL << first_level_bits; + first_level_map.resize(first_level_size, nullptr); +#ifdef _WIN32 + void* base{VirtualAlloc(nullptr, alloc_size, MEM_RESERVE, PAGE_READWRITE)}; +#else + void* base{mmap(nullptr, alloc_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0)}; + + if (base == MAP_FAILED) { + base = nullptr; + } +#endif + + ASSERT(base); + base_ptr = reinterpret_cast<BaseAddr*>(base); +} + +template <typename BaseAddr> +MultiLevelPageTable<BaseAddr>::~MultiLevelPageTable() noexcept { + if (!base_ptr) { + return; + } +#ifdef _WIN32 + ASSERT(VirtualFree(base_ptr, 0, MEM_RELEASE)); +#else + ASSERT(munmap(base_ptr, alloc_size) == 0); +#endif +} + +template <typename BaseAddr> +void MultiLevelPageTable<BaseAddr>::ReserveRange(u64 start, std::size_t size) { + const u64 new_start = start >> first_level_shift; + const u64 new_end = (start + size) >> first_level_shift; + for (u64 i = new_start; i <= new_end; i++) { + if (!first_level_map[i]) { + AllocateLevel(i); + } + } +} + +template <typename BaseAddr> +void MultiLevelPageTable<BaseAddr>::AllocateLevel(u64 level) { + void* ptr = reinterpret_cast<char *>(base_ptr) + level * first_level_chunk_size; +#ifdef _WIN32 + void* base{VirtualAlloc(ptr, first_level_chunk_size, MEM_COMMIT, PAGE_READWRITE)}; +#else + void* base{mmap(ptr, first_level_chunk_size, PROT_READ | PROT_WRITE, + MAP_ANONYMOUS | MAP_PRIVATE, -1, 0)}; + + if (base == MAP_FAILED) { + base = nullptr; + } +#endif + ASSERT(base); + + first_level_map[level] = base; +} + +} // namespace Common diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt index 8e3fd4505..95302c419 100644 --- a/src/core/CMakeLists.txt +++ b/src/core/CMakeLists.txt @@ -138,8 +138,6 @@ add_library(core STATIC frontend/emu_window.h frontend/framebuffer_layout.cpp frontend/framebuffer_layout.h - hardware_interrupt_manager.cpp - hardware_interrupt_manager.h hid/emulated_console.cpp hid/emulated_console.h hid/emulated_controller.cpp @@ -550,6 +548,12 @@ add_library(core STATIC hle/service/ns/ns.h hle/service/ns/pdm_qry.cpp hle/service/ns/pdm_qry.h + hle/service/nvdrv/core/container.cpp + hle/service/nvdrv/core/container.h + hle/service/nvdrv/core/nvmap.cpp + hle/service/nvdrv/core/nvmap.h + hle/service/nvdrv/core/syncpoint_manager.cpp + hle/service/nvdrv/core/syncpoint_manager.h hle/service/nvdrv/devices/nvdevice.h hle/service/nvdrv/devices/nvdisp_disp0.cpp hle/service/nvdrv/devices/nvdisp_disp0.h @@ -578,8 +582,6 @@ add_library(core STATIC hle/service/nvdrv/nvdrv_interface.h hle/service/nvdrv/nvmemp.cpp hle/service/nvdrv/nvmemp.h - hle/service/nvdrv/syncpoint_manager.cpp - hle/service/nvdrv/syncpoint_manager.h hle/service/nvflinger/binder.h hle/service/nvflinger/buffer_item.h hle/service/nvflinger/buffer_item_consumer.cpp diff --git a/src/core/core.cpp b/src/core/core.cpp index 121092868..1deeee154 100644 --- a/src/core/core.cpp +++ b/src/core/core.cpp @@ -27,7 +27,6 @@ #include "core/file_sys/savedata_factory.h" #include "core/file_sys/vfs_concat.h" #include "core/file_sys/vfs_real.h" -#include "core/hardware_interrupt_manager.h" #include "core/hid/hid_core.h" #include "core/hle/kernel/k_memory_manager.h" #include "core/hle/kernel/k_process.h" @@ -51,6 +50,7 @@ #include "core/telemetry_session.h" #include "core/tools/freezer.h" #include "network/network.h" +#include "video_core/host1x/host1x.h" #include "video_core/renderer_base.h" #include "video_core/video_core.h" @@ -215,6 +215,7 @@ struct System::Impl { telemetry_session = std::make_unique<Core::TelemetrySession>(); + host1x_core = std::make_unique<Tegra::Host1x::Host1x>(system); gpu_core = VideoCore::CreateGPU(emu_window, system); if (!gpu_core) { return SystemResultStatus::ErrorVideoCore; @@ -224,7 +225,6 @@ struct System::Impl { service_manager = std::make_shared<Service::SM::ServiceManager>(kernel); services = std::make_unique<Service::Services>(service_manager, system); - interrupt_manager = std::make_unique<Hardware::InterruptManager>(system); // Initialize time manager, which must happen after kernel is created time_manager.Initialize(); @@ -373,6 +373,7 @@ struct System::Impl { app_loader.reset(); audio_core.reset(); gpu_core.reset(); + host1x_core.reset(); perf_stats.reset(); kernel.Shutdown(); memory.Reset(); @@ -450,7 +451,7 @@ struct System::Impl { /// AppLoader used to load the current executing application std::unique_ptr<Loader::AppLoader> app_loader; std::unique_ptr<Tegra::GPU> gpu_core; - std::unique_ptr<Hardware::InterruptManager> interrupt_manager; + std::unique_ptr<Tegra::Host1x::Host1x> host1x_core; std::unique_ptr<Core::DeviceMemory> device_memory; std::unique_ptr<AudioCore::AudioCore> audio_core; Core::Memory::Memory memory; @@ -668,12 +669,12 @@ const Tegra::GPU& System::GPU() const { return *impl->gpu_core; } -Core::Hardware::InterruptManager& System::InterruptManager() { - return *impl->interrupt_manager; +Tegra::Host1x::Host1x& System::Host1x() { + return *impl->host1x_core; } -const Core::Hardware::InterruptManager& System::InterruptManager() const { - return *impl->interrupt_manager; +const Tegra::Host1x::Host1x& System::Host1x() const { + return *impl->host1x_core; } VideoCore::RendererBase& System::Renderer() { diff --git a/src/core/core.h b/src/core/core.h index 0ce3b1d60..7843cc8ad 100644 --- a/src/core/core.h +++ b/src/core/core.h @@ -74,6 +74,9 @@ class TimeManager; namespace Tegra { class DebugContext; class GPU; +namespace Host1x { +class Host1x; +} // namespace Host1x } // namespace Tegra namespace VideoCore { @@ -88,10 +91,6 @@ namespace Core::Timing { class CoreTiming; } -namespace Core::Hardware { -class InterruptManager; -} - namespace Core::HID { class HIDCore; } @@ -260,6 +259,12 @@ public: /// Gets an immutable reference to the GPU interface. [[nodiscard]] const Tegra::GPU& GPU() const; + /// Gets a mutable reference to the Host1x interface + [[nodiscard]] Tegra::Host1x::Host1x& Host1x(); + + /// Gets an immutable reference to the Host1x interface. + [[nodiscard]] const Tegra::Host1x::Host1x& Host1x() const; + /// Gets a mutable reference to the renderer. [[nodiscard]] VideoCore::RendererBase& Renderer(); @@ -296,12 +301,6 @@ public: /// Provides a constant reference to the core timing instance. [[nodiscard]] const Timing::CoreTiming& CoreTiming() const; - /// Provides a reference to the interrupt manager instance. - [[nodiscard]] Core::Hardware::InterruptManager& InterruptManager(); - - /// Provides a constant reference to the interrupt manager instance. - [[nodiscard]] const Core::Hardware::InterruptManager& InterruptManager() const; - /// Provides a reference to the kernel instance. [[nodiscard]] Kernel::KernelCore& Kernel(); diff --git a/src/core/core_timing.cpp b/src/core/core_timing.cpp index f6c4567ba..6c0fcb7b5 100644 --- a/src/core/core_timing.cpp +++ b/src/core/core_timing.cpp @@ -43,7 +43,7 @@ CoreTiming::CoreTiming() CoreTiming::~CoreTiming() = default; void CoreTiming::ThreadEntry(CoreTiming& instance) { - constexpr char name[] = "yuzu:HostTiming"; + constexpr char name[] = "HostTiming"; MicroProfileOnThreadCreate(name); Common::SetCurrentThreadName(name); Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical); diff --git a/src/core/cpu_manager.cpp b/src/core/cpu_manager.cpp index 9b1565ae1..0dd4c2196 100644 --- a/src/core/cpu_manager.cpp +++ b/src/core/cpu_manager.cpp @@ -189,9 +189,9 @@ void CpuManager::RunThread(std::size_t core) { system.RegisterCoreThread(core); std::string name; if (is_multicore) { - name = "yuzu:CPUCore_" + std::to_string(core); + name = "CPUCore_" + std::to_string(core); } else { - name = "yuzu:CPUThread"; + name = "CPUThread"; } MicroProfileOnThreadCreate(name.c_str()); Common::SetCurrentThreadName(name.c_str()); diff --git a/src/core/debugger/debugger.cpp b/src/core/debugger/debugger.cpp index e42bdd17d..339f971e6 100644 --- a/src/core/debugger/debugger.cpp +++ b/src/core/debugger/debugger.cpp @@ -140,7 +140,7 @@ private: } void ThreadLoop(std::stop_token stop_token) { - Common::SetCurrentThreadName("yuzu:Debugger"); + Common::SetCurrentThreadName("Debugger"); // Set up the client signals for new data. AsyncReceiveInto(signal_pipe, pipe_data, [&](auto d) { PipeData(d); }); diff --git a/src/core/file_sys/program_metadata.cpp b/src/core/file_sys/program_metadata.cpp index e0cdf3520..08d489eab 100644 --- a/src/core/file_sys/program_metadata.cpp +++ b/src/core/file_sys/program_metadata.cpp @@ -33,11 +33,55 @@ Loader::ResultStatus ProgramMetadata::Load(VirtualFile file) { return Loader::ResultStatus::ErrorBadACIHeader; } - if (sizeof(FileAccessControl) != file->ReadObject(&acid_file_access, acid_header.fac_offset)) { + // Load acid_file_access per-component instead of the entire struct, since this struct does not + // reflect the layout of the real data. + std::size_t current_offset = acid_header.fac_offset; + if (sizeof(FileAccessControl::version) != file->ReadBytes(&acid_file_access.version, + sizeof(FileAccessControl::version), + current_offset)) { + return Loader::ResultStatus::ErrorBadFileAccessControl; + } + if (sizeof(FileAccessControl::permissions) != + file->ReadBytes(&acid_file_access.permissions, sizeof(FileAccessControl::permissions), + current_offset += sizeof(FileAccessControl::version) + 3)) { + return Loader::ResultStatus::ErrorBadFileAccessControl; + } + if (sizeof(FileAccessControl::unknown) != + file->ReadBytes(&acid_file_access.unknown, sizeof(FileAccessControl::unknown), + current_offset + sizeof(FileAccessControl::permissions))) { return Loader::ResultStatus::ErrorBadFileAccessControl; } - if (sizeof(FileAccessHeader) != file->ReadObject(&aci_file_access, aci_header.fah_offset)) { + // Load aci_file_access per-component instead of the entire struct, same as acid_file_access + current_offset = aci_header.fah_offset; + if (sizeof(FileAccessHeader::version) != file->ReadBytes(&aci_file_access.version, + sizeof(FileAccessHeader::version), + current_offset)) { + return Loader::ResultStatus::ErrorBadFileAccessHeader; + } + if (sizeof(FileAccessHeader::permissions) != + file->ReadBytes(&aci_file_access.permissions, sizeof(FileAccessHeader::permissions), + current_offset += sizeof(FileAccessHeader::version) + 3)) { + return Loader::ResultStatus::ErrorBadFileAccessHeader; + } + if (sizeof(FileAccessHeader::unk_offset) != + file->ReadBytes(&aci_file_access.unk_offset, sizeof(FileAccessHeader::unk_offset), + current_offset += sizeof(FileAccessHeader::permissions))) { + return Loader::ResultStatus::ErrorBadFileAccessHeader; + } + if (sizeof(FileAccessHeader::unk_size) != + file->ReadBytes(&aci_file_access.unk_size, sizeof(FileAccessHeader::unk_size), + current_offset += sizeof(FileAccessHeader::unk_offset))) { + return Loader::ResultStatus::ErrorBadFileAccessHeader; + } + if (sizeof(FileAccessHeader::unk_offset_2) != + file->ReadBytes(&aci_file_access.unk_offset_2, sizeof(FileAccessHeader::unk_offset_2), + current_offset += sizeof(FileAccessHeader::unk_size))) { + return Loader::ResultStatus::ErrorBadFileAccessHeader; + } + if (sizeof(FileAccessHeader::unk_size_2) != + file->ReadBytes(&aci_file_access.unk_size_2, sizeof(FileAccessHeader::unk_size_2), + current_offset + sizeof(FileAccessHeader::unk_offset_2))) { return Loader::ResultStatus::ErrorBadFileAccessHeader; } @@ -152,9 +196,7 @@ void ProgramMetadata::Print() const { LOG_DEBUG(Service_FS, " > Is Retail: {}", acid_header.is_retail ? "YES" : "NO"); LOG_DEBUG(Service_FS, "Title ID Min: 0x{:016X}", acid_header.title_id_min); LOG_DEBUG(Service_FS, "Title ID Max: 0x{:016X}", acid_header.title_id_max); - u64_le permissions_l; // local copy to fix alignment error - std::memcpy(&permissions_l, &acid_file_access.permissions, sizeof(permissions_l)); - LOG_DEBUG(Service_FS, "Filesystem Access: 0x{:016X}\n", permissions_l); + LOG_DEBUG(Service_FS, "Filesystem Access: 0x{:016X}\n", acid_file_access.permissions); // Begin ACI0 printing (actual perms, unsigned) LOG_DEBUG(Service_FS, "Magic: {:.4}", aci_header.magic.data()); diff --git a/src/core/file_sys/program_metadata.h b/src/core/file_sys/program_metadata.h index e8fb4e27f..2e8960b07 100644 --- a/src/core/file_sys/program_metadata.h +++ b/src/core/file_sys/program_metadata.h @@ -144,20 +144,18 @@ private: static_assert(sizeof(AciHeader) == 0x40, "ACI0 header structure size is wrong"); -#pragma pack(push, 1) - + // FileAccessControl and FileAccessHeader need loaded per-component: this layout does not + // reflect the real layout to avoid reference binding to misaligned addresses struct FileAccessControl { u8 version; - INSERT_PADDING_BYTES(3); + // 3 padding bytes u64_le permissions; std::array<u8, 0x20> unknown; }; - static_assert(sizeof(FileAccessControl) == 0x2C, "FS access control structure size is wrong"); - struct FileAccessHeader { u8 version; - INSERT_PADDING_BYTES(3); + // 3 padding bytes u64_le permissions; u32_le unk_offset; u32_le unk_size; @@ -165,10 +163,6 @@ private: u32_le unk_size_2; }; - static_assert(sizeof(FileAccessHeader) == 0x1C, "FS access header structure size is wrong"); - -#pragma pack(pop) - Header npdm_header; AciHeader aci_header; AcidHeader acid_header; diff --git a/src/core/hardware_interrupt_manager.cpp b/src/core/hardware_interrupt_manager.cpp deleted file mode 100644 index d08cc3315..000000000 --- a/src/core/hardware_interrupt_manager.cpp +++ /dev/null @@ -1,32 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#include "core/core.h" -#include "core/core_timing.h" -#include "core/hardware_interrupt_manager.h" -#include "core/hle/service/nvdrv/nvdrv_interface.h" -#include "core/hle/service/sm/sm.h" - -namespace Core::Hardware { - -InterruptManager::InterruptManager(Core::System& system_in) : system(system_in) { - gpu_interrupt_event = Core::Timing::CreateEvent( - "GPUInterrupt", - [this](std::uintptr_t message, u64 time, - std::chrono::nanoseconds) -> std::optional<std::chrono::nanoseconds> { - auto nvdrv = system.ServiceManager().GetService<Service::Nvidia::NVDRV>("nvdrv"); - const u32 syncpt = static_cast<u32>(message >> 32); - const u32 value = static_cast<u32>(message); - nvdrv->SignalGPUInterruptSyncpt(syncpt, value); - return std::nullopt; - }); -} - -InterruptManager::~InterruptManager() = default; - -void InterruptManager::GPUInterruptSyncpt(const u32 syncpoint_id, const u32 value) { - const u64 msg = (static_cast<u64>(syncpoint_id) << 32ULL) | value; - system.CoreTiming().ScheduleEvent(std::chrono::nanoseconds{10}, gpu_interrupt_event, msg); -} - -} // namespace Core::Hardware diff --git a/src/core/hardware_interrupt_manager.h b/src/core/hardware_interrupt_manager.h deleted file mode 100644 index 5665c5918..000000000 --- a/src/core/hardware_interrupt_manager.h +++ /dev/null @@ -1,32 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#pragma once - -#include <memory> - -#include "common/common_types.h" - -namespace Core { -class System; -} - -namespace Core::Timing { -struct EventType; -} - -namespace Core::Hardware { - -class InterruptManager { -public: - explicit InterruptManager(Core::System& system); - ~InterruptManager(); - - void GPUInterruptSyncpt(u32 syncpoint_id, u32 value); - -private: - Core::System& system; - std::shared_ptr<Core::Timing::EventType> gpu_interrupt_event; -}; - -} // namespace Core::Hardware diff --git a/src/core/hid/emulated_controller.cpp b/src/core/hid/emulated_controller.cpp index e27d84734..025f1c78e 100644 --- a/src/core/hid/emulated_controller.cpp +++ b/src/core/hid/emulated_controller.cpp @@ -1017,9 +1017,11 @@ bool EmulatedController::SetPollingMode(Common::Input::PollingMode polling_mode) auto& output_device = output_devices[static_cast<std::size_t>(DeviceIndex::Right)]; auto& nfc_output_device = output_devices[3]; - nfc_output_device->SetPollingMode(polling_mode); + const auto virtual_nfc_result = nfc_output_device->SetPollingMode(polling_mode); + const auto mapped_nfc_result = output_device->SetPollingMode(polling_mode); - return output_device->SetPollingMode(polling_mode) == Common::Input::PollingError::None; + return virtual_nfc_result == Common::Input::PollingError::None || + mapped_nfc_result == Common::Input::PollingError::None; } bool EmulatedController::SetCameraFormat( diff --git a/src/core/hle/kernel/k_worker_task_manager.cpp b/src/core/hle/kernel/k_worker_task_manager.cpp index 221f341ee..04042bf8f 100644 --- a/src/core/hle/kernel/k_worker_task_manager.cpp +++ b/src/core/hle/kernel/k_worker_task_manager.cpp @@ -23,7 +23,7 @@ void KWorkerTask::DoWorkerTask() { } } -KWorkerTaskManager::KWorkerTaskManager() : m_waiting_thread(1, "yuzu:KWorkerTaskManager") {} +KWorkerTaskManager::KWorkerTaskManager() : m_waiting_thread(1, "KWorkerTaskManager") {} void KWorkerTaskManager::AddTask(KernelCore& kernel, WorkerType type, KWorkerTask* task) { ASSERT(type <= WorkerType::Count); diff --git a/src/core/hle/kernel/kernel.cpp b/src/core/hle/kernel/kernel.cpp index ce7fa8275..9251f29ad 100644 --- a/src/core/hle/kernel/kernel.cpp +++ b/src/core/hle/kernel/kernel.cpp @@ -48,7 +48,7 @@ namespace Kernel { struct KernelCore::Impl { explicit Impl(Core::System& system_, KernelCore& kernel_) : time_manager{system_}, - service_threads_manager{1, "yuzu:ServiceThreadsManager"}, system{system_} {} + service_threads_manager{1, "ServiceThreadsManager"}, system{system_} {} void SetMulticore(bool is_multi) { is_multicore = is_multi; diff --git a/src/core/hle/kernel/service_thread.cpp b/src/core/hle/kernel/service_thread.cpp index 2e87b4ea4..d23d76706 100644 --- a/src/core/hle/kernel/service_thread.cpp +++ b/src/core/hle/kernel/service_thread.cpp @@ -36,7 +36,7 @@ ServiceThread::Impl::Impl(KernelCore& kernel, std::size_t num_threads, const std : service_name{name} { for (std::size_t i = 0; i < num_threads; ++i) { threads.emplace_back([this, &kernel](std::stop_token stop_token) { - Common::SetCurrentThreadName(std::string{"yuzu:HleService:" + service_name}.c_str()); + Common::SetCurrentThreadName(std::string{service_name}.c_str()); // Wait for first request before trying to acquire a render context { diff --git a/src/core/hle/service/nfp/amiibo_crypto.cpp b/src/core/hle/service/nfp/amiibo_crypto.cpp index ce0bc3f75..c32a6816b 100644 --- a/src/core/hle/service/nfp/amiibo_crypto.cpp +++ b/src/core/hle/service/nfp/amiibo_crypto.cpp @@ -28,7 +28,7 @@ bool IsAmiiboValid(const EncryptedNTAG215File& ntag_file) { LOG_DEBUG(Service_NFP, "model_number=0x{0:x}", static_cast<u16>(amiibo_data.model_info.model_number)); LOG_DEBUG(Service_NFP, "series={}", amiibo_data.model_info.series); - LOG_DEBUG(Service_NFP, "fixed_value=0x{0:x}", amiibo_data.model_info.constant_value); + LOG_DEBUG(Service_NFP, "tag_type=0x{0:x}", amiibo_data.model_info.tag_type); LOG_DEBUG(Service_NFP, "tag_dynamic_lock=0x{0:x}", ntag_file.dynamic_lock); LOG_DEBUG(Service_NFP, "tag_CFG0=0x{0:x}", ntag_file.CFG0); @@ -55,7 +55,7 @@ bool IsAmiiboValid(const EncryptedNTAG215File& ntag_file) { if (amiibo_data.constant_value != 0xA5) { return false; } - if (amiibo_data.model_info.constant_value != 0x02) { + if (amiibo_data.model_info.tag_type != PackedTagType::Type2) { return false; } if ((ntag_file.dynamic_lock & 0xFFFFFF) != 0x0F0001U) { diff --git a/src/core/hle/service/nfp/nfp_device.cpp b/src/core/hle/service/nfp/nfp_device.cpp index 0d4ffd3a5..ec895ac01 100644 --- a/src/core/hle/service/nfp/nfp_device.cpp +++ b/src/core/hle/service/nfp/nfp_device.cpp @@ -98,11 +98,6 @@ bool NfpDevice::LoadAmiibo(std::span<const u8> data) { memcpy(&encrypted_tag_data, data.data(), sizeof(EncryptedNTAG215File)); - if (!AmiiboCrypto::IsAmiiboValid(encrypted_tag_data)) { - LOG_INFO(Service_NFP, "Invalid amiibo"); - return false; - } - device_state = DeviceState::TagFound; deactivate_event->GetReadableEvent().Clear(); activate_event->GetWritableEvent().Signal(); @@ -148,20 +143,28 @@ void NfpDevice::Finalize() { } Result NfpDevice::StartDetection(s32 protocol_) { - if (device_state == DeviceState::Initialized || device_state == DeviceState::TagRemoved) { - npad_device->SetPollingMode(Common::Input::PollingMode::NFC); - device_state = DeviceState::SearchingForTag; - protocol = protocol_; - return ResultSuccess; + if (device_state != DeviceState::Initialized && device_state != DeviceState::TagRemoved) { + LOG_ERROR(Service_NFP, "Wrong device state {}", device_state); + return WrongDeviceState; } - LOG_ERROR(Service_NFP, "Wrong device state {}", device_state); - return WrongDeviceState; + if (!npad_device->SetPollingMode(Common::Input::PollingMode::NFC)) { + LOG_ERROR(Service_NFP, "Nfc not supported"); + return NfcDisabled; + } + + device_state = DeviceState::SearchingForTag; + protocol = protocol_; + return ResultSuccess; } Result NfpDevice::StopDetection() { npad_device->SetPollingMode(Common::Input::PollingMode::Active); + if (device_state == DeviceState::Initialized) { + return ResultSuccess; + } + if (device_state == DeviceState::TagFound || device_state == DeviceState::TagMounted) { CloseAmiibo(); return ResultSuccess; @@ -225,6 +228,11 @@ Result NfpDevice::Mount(MountTarget mount_target_) { return WrongDeviceState; } + if (!AmiiboCrypto::IsAmiiboValid(encrypted_tag_data)) { + LOG_ERROR(Service_NFP, "Not an amiibo"); + return NotAnAmiibo; + } + if (!AmiiboCrypto::DecodeAmiibo(encrypted_tag_data, tag_data)) { LOG_ERROR(Service_NFP, "Can't decode amiibo {}", device_state); return CorruptedData; @@ -238,6 +246,9 @@ Result NfpDevice::Mount(MountTarget mount_target_) { Result NfpDevice::Unmount() { if (device_state != DeviceState::TagMounted) { LOG_ERROR(Service_NFP, "Wrong device state {}", device_state); + if (device_state == DeviceState::TagRemoved) { + return TagRemoved; + } return WrongDeviceState; } @@ -256,6 +267,9 @@ Result NfpDevice::Unmount() { Result NfpDevice::GetTagInfo(TagInfo& tag_info) const { if (device_state != DeviceState::TagFound && device_state != DeviceState::TagMounted) { LOG_ERROR(Service_NFP, "Wrong device state {}", device_state); + if (device_state == DeviceState::TagRemoved) { + return TagRemoved; + } return WrongDeviceState; } @@ -287,12 +301,7 @@ Result NfpDevice::GetCommonInfo(CommonInfo& common_info) const { // TODO: Validate this data common_info = { - .last_write_date = - { - settings.write_date.GetYear(), - settings.write_date.GetMonth(), - settings.write_date.GetDay(), - }, + .last_write_date = settings.write_date.GetWriteDate(), .write_counter = tag_data.write_counter, .version = 0, .application_area_size = sizeof(ApplicationArea), @@ -303,6 +312,9 @@ Result NfpDevice::GetCommonInfo(CommonInfo& common_info) const { Result NfpDevice::GetModelInfo(ModelInfo& model_info) const { if (device_state != DeviceState::TagMounted) { LOG_ERROR(Service_NFP, "Wrong device state {}", device_state); + if (device_state == DeviceState::TagRemoved) { + return TagRemoved; + } return WrongDeviceState; } @@ -341,12 +353,7 @@ Result NfpDevice::GetRegisterInfo(RegisterInfo& register_info) const { // TODO: Validate this data register_info = { .mii_char_info = manager.ConvertV3ToCharInfo(tag_data.owner_mii), - .creation_date = - { - settings.init_date.GetYear(), - settings.init_date.GetMonth(), - settings.init_date.GetDay(), - }, + .creation_date = settings.init_date.GetWriteDate(), .amiibo_name = GetAmiiboName(settings), .font_region = {}, }; @@ -478,8 +485,7 @@ Result NfpDevice::GetApplicationArea(std::vector<u8>& data) const { } if (data.size() > sizeof(ApplicationArea)) { - LOG_ERROR(Service_NFP, "Wrong data size {}", data.size()); - return ResultUnknown; + data.resize(sizeof(ApplicationArea)); } memcpy(data.data(), tag_data.application_area.data(), data.size()); @@ -518,7 +524,7 @@ Result NfpDevice::SetApplicationArea(std::span<const u8> data) { Common::TinyMT rng{}; std::memcpy(tag_data.application_area.data(), data.data(), data.size()); - // HW seems to fill excess data with garbage + // Fill remaining data with random numbers rng.GenerateRandomBytes(tag_data.application_area.data() + data.size(), sizeof(ApplicationArea) - data.size()); @@ -561,12 +567,12 @@ Result NfpDevice::RecreateApplicationArea(u32 access_id, std::span<const u8> dat if (data.size() > sizeof(ApplicationArea)) { LOG_ERROR(Service_NFP, "Wrong data size {}", data.size()); - return ResultUnknown; + return WrongApplicationAreaSize; } Common::TinyMT rng{}; std::memcpy(tag_data.application_area.data(), data.data(), data.size()); - // HW seems to fill excess data with garbage + // Fill remaining data with random numbers rng.GenerateRandomBytes(tag_data.application_area.data() + data.size(), sizeof(ApplicationArea) - data.size()); @@ -612,7 +618,6 @@ u64 NfpDevice::GetHandle() const { } u32 NfpDevice::GetApplicationAreaSize() const { - // Investigate if this value is really constant return sizeof(ApplicationArea); } diff --git a/src/core/hle/service/nfp/nfp_result.h b/src/core/hle/service/nfp/nfp_result.h index ac259e2ff..d8e4cf094 100644 --- a/src/core/hle/service/nfp/nfp_result.h +++ b/src/core/hle/service/nfp/nfp_result.h @@ -8,6 +8,8 @@ namespace Service::NFP { constexpr Result DeviceNotFound(ErrorModule::NFP, 64); +constexpr Result InvalidArgument(ErrorModule::NFP, 65); +constexpr Result WrongApplicationAreaSize(ErrorModule::NFP, 68); constexpr Result WrongDeviceState(ErrorModule::NFP, 73); constexpr Result NfcDisabled(ErrorModule::NFP, 80); constexpr Result WriteAmiiboFailed(ErrorModule::NFP, 88); diff --git a/src/core/hle/service/nfp/nfp_types.h b/src/core/hle/service/nfp/nfp_types.h index dd4525b61..c09f9ddb6 100644 --- a/src/core/hle/service/nfp/nfp_types.h +++ b/src/core/hle/service/nfp/nfp_types.h @@ -84,6 +84,15 @@ enum class TagType : u32 { Type5, // ISO15693 RW/RO 540 bytes 106kbit/s }; +enum class PackedTagType : u8 { + None, + Type1, // ISO14443A RW 96-2k bytes 106kbit/s + Type2, // ISO14443A RW/RO 540 bytes 106kbit/s + Type3, // Sony Felica RW/RO 2k bytes 212kbit/s + Type4, // ISO14443A RW/RO 4k-32k bytes 424kbit/s + Type5, // ISO15693 RW/RO 540 bytes 106kbit/s +}; + enum class TagProtocol : u32 { None, TypeA, // ISO14443A @@ -104,6 +113,13 @@ struct TagUuid { }; static_assert(sizeof(TagUuid) == 10, "TagUuid is an invalid size"); +struct WriteDate { + u16 year; + u8 month; + u8 day; +}; +static_assert(sizeof(WriteDate) == 0x4, "WriteDate is an invalid size"); + struct AmiiboDate { u16 raw_date{}; @@ -121,6 +137,21 @@ struct AmiiboDate { return static_cast<u8>(GetValue() & 0x001F); } + WriteDate GetWriteDate() const { + if (!IsValidDate()) { + return { + .year = 2000, + .month = 1, + .day = 1, + }; + } + return { + .year = GetYear(), + .month = GetMonth(), + .day = GetDay(), + }; + } + void SetYear(u16 year) { const u16 year_converted = static_cast<u16>((year - 2000) << 9); raw_date = Common::swap16((GetValue() & ~0xFE00) | year_converted); @@ -133,6 +164,13 @@ struct AmiiboDate { const u16 day_converted = static_cast<u16>(day); raw_date = Common::swap16((GetValue() & ~0x001F) | day_converted); } + + bool IsValidDate() const { + const bool is_day_valid = GetDay() > 0 && GetDay() < 32; + const bool is_month_valid = GetMonth() > 0 && GetMonth() < 13; + const bool is_year_valid = GetYear() >= 2000; + return is_year_valid && is_month_valid && is_day_valid; + } }; static_assert(sizeof(AmiiboDate) == 2, "AmiiboDate is an invalid size"); @@ -163,7 +201,7 @@ struct AmiiboModelInfo { AmiiboType amiibo_type; u16_be model_number; AmiiboSeries series; - u8 constant_value; // Must be 02 + PackedTagType tag_type; INSERT_PADDING_BYTES(0x4); // Unknown }; static_assert(sizeof(AmiiboModelInfo) == 0xC, "AmiiboModelInfo is an invalid size"); @@ -250,13 +288,6 @@ struct TagInfo { }; static_assert(sizeof(TagInfo) == 0x58, "TagInfo is an invalid size"); -struct WriteDate { - u16 year; - u8 month; - u8 day; -}; -static_assert(sizeof(WriteDate) == 0x4, "WriteDate is an invalid size"); - struct CommonInfo { WriteDate last_write_date; u16 write_counter; diff --git a/src/core/hle/service/nfp/nfp_user.cpp b/src/core/hle/service/nfp/nfp_user.cpp index c61df9401..4ed53b534 100644 --- a/src/core/hle/service/nfp/nfp_user.cpp +++ b/src/core/hle/service/nfp/nfp_user.cpp @@ -93,6 +93,18 @@ void IUser::ListDevices(Kernel::HLERequestContext& ctx) { return; } + if (!ctx.CanWriteBuffer()) { + IPC::ResponseBuilder rb{ctx, 2}; + rb.Push(InvalidArgument); + return; + } + + if (ctx.GetWriteBufferSize() == 0) { + IPC::ResponseBuilder rb{ctx, 2}; + rb.Push(InvalidArgument); + return; + } + std::vector<u64> nfp_devices; const std::size_t max_allowed_devices = ctx.GetWriteBufferSize() / sizeof(u64); @@ -255,6 +267,12 @@ void IUser::GetApplicationArea(Kernel::HLERequestContext& ctx) { return; } + if (!ctx.CanWriteBuffer()) { + IPC::ResponseBuilder rb{ctx, 2}; + rb.Push(InvalidArgument); + return; + } + auto device = GetNfpDevice(device_handle); if (!device.has_value()) { @@ -283,6 +301,12 @@ void IUser::SetApplicationArea(Kernel::HLERequestContext& ctx) { return; } + if (!ctx.CanReadBuffer()) { + IPC::ResponseBuilder rb{ctx, 2}; + rb.Push(InvalidArgument); + return; + } + auto device = GetNfpDevice(device_handle); if (!device.has_value()) { @@ -358,6 +382,12 @@ void IUser::CreateApplicationArea(Kernel::HLERequestContext& ctx) { return; } + if (!ctx.CanReadBuffer()) { + IPC::ResponseBuilder rb{ctx, 2}; + rb.Push(InvalidArgument); + return; + } + auto device = GetNfpDevice(device_handle); if (!device.has_value()) { diff --git a/src/core/hle/service/nvdrv/core/container.cpp b/src/core/hle/service/nvdrv/core/container.cpp new file mode 100644 index 000000000..37ca24f5d --- /dev/null +++ b/src/core/hle/service/nvdrv/core/container.cpp @@ -0,0 +1,50 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-FileCopyrightText: 2022 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "core/hle/service/nvdrv/core/container.h" +#include "core/hle/service/nvdrv/core/nvmap.h" +#include "core/hle/service/nvdrv/core/syncpoint_manager.h" +#include "video_core/host1x/host1x.h" + +namespace Service::Nvidia::NvCore { + +struct ContainerImpl { + explicit ContainerImpl(Tegra::Host1x::Host1x& host1x_) + : file{host1x_}, manager{host1x_}, device_file_data{} {} + NvMap file; + SyncpointManager manager; + Container::Host1xDeviceFileData device_file_data; +}; + +Container::Container(Tegra::Host1x::Host1x& host1x_) { + impl = std::make_unique<ContainerImpl>(host1x_); +} + +Container::~Container() = default; + +NvMap& Container::GetNvMapFile() { + return impl->file; +} + +const NvMap& Container::GetNvMapFile() const { + return impl->file; +} + +Container::Host1xDeviceFileData& Container::Host1xDeviceFile() { + return impl->device_file_data; +} + +const Container::Host1xDeviceFileData& Container::Host1xDeviceFile() const { + return impl->device_file_data; +} + +SyncpointManager& Container::GetSyncpointManager() { + return impl->manager; +} + +const SyncpointManager& Container::GetSyncpointManager() const { + return impl->manager; +} + +} // namespace Service::Nvidia::NvCore diff --git a/src/core/hle/service/nvdrv/core/container.h b/src/core/hle/service/nvdrv/core/container.h new file mode 100644 index 000000000..b4b63ac90 --- /dev/null +++ b/src/core/hle/service/nvdrv/core/container.h @@ -0,0 +1,52 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-FileCopyrightText: 2022 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <deque> +#include <memory> +#include <unordered_map> + +#include "core/hle/service/nvdrv/nvdata.h" + +namespace Tegra::Host1x { +class Host1x; +} // namespace Tegra::Host1x + +namespace Service::Nvidia::NvCore { + +class NvMap; +class SyncpointManager; + +struct ContainerImpl; + +class Container { +public: + explicit Container(Tegra::Host1x::Host1x& host1x); + ~Container(); + + NvMap& GetNvMapFile(); + + const NvMap& GetNvMapFile() const; + + SyncpointManager& GetSyncpointManager(); + + const SyncpointManager& GetSyncpointManager() const; + + struct Host1xDeviceFileData { + std::unordered_map<DeviceFD, u32> fd_to_id{}; + std::deque<u32> syncpts_accumulated{}; + u32 nvdec_next_id{}; + u32 vic_next_id{}; + }; + + Host1xDeviceFileData& Host1xDeviceFile(); + + const Host1xDeviceFileData& Host1xDeviceFile() const; + +private: + std::unique_ptr<ContainerImpl> impl; +}; + +} // namespace Service::Nvidia::NvCore diff --git a/src/core/hle/service/nvdrv/core/nvmap.cpp b/src/core/hle/service/nvdrv/core/nvmap.cpp new file mode 100644 index 000000000..fbd8a74a5 --- /dev/null +++ b/src/core/hle/service/nvdrv/core/nvmap.cpp @@ -0,0 +1,272 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-FileCopyrightText: 2022 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/alignment.h" +#include "common/assert.h" +#include "common/logging/log.h" +#include "core/hle/service/nvdrv/core/nvmap.h" +#include "core/memory.h" +#include "video_core/host1x/host1x.h" + +using Core::Memory::YUZU_PAGESIZE; + +namespace Service::Nvidia::NvCore { +NvMap::Handle::Handle(u64 size_, Id id_) + : size(size_), aligned_size(size), orig_size(size), id(id_) { + flags.raw = 0; +} + +NvResult NvMap::Handle::Alloc(Flags pFlags, u32 pAlign, u8 pKind, u64 pAddress) { + std::scoped_lock lock(mutex); + + // Handles cannot be allocated twice + if (allocated) { + return NvResult::AccessDenied; + } + + flags = pFlags; + kind = pKind; + align = pAlign < YUZU_PAGESIZE ? YUZU_PAGESIZE : pAlign; + + // This flag is only applicable for handles with an address passed + if (pAddress) { + flags.keep_uncached_after_free.Assign(0); + } else { + LOG_CRITICAL(Service_NVDRV, + "Mapping nvmap handles without a CPU side address is unimplemented!"); + } + + size = Common::AlignUp(size, YUZU_PAGESIZE); + aligned_size = Common::AlignUp(size, align); + address = pAddress; + allocated = true; + + return NvResult::Success; +} + +NvResult NvMap::Handle::Duplicate(bool internal_session) { + std::scoped_lock lock(mutex); + // Unallocated handles cannot be duplicated as duplication requires memory accounting (in HOS) + if (!allocated) [[unlikely]] { + return NvResult::BadValue; + } + + // If we internally use FromId the duplication tracking of handles won't work accurately due to + // us not implementing per-process handle refs. + if (internal_session) { + internal_dupes++; + } else { + dupes++; + } + + return NvResult::Success; +} + +NvMap::NvMap(Tegra::Host1x::Host1x& host1x_) : host1x{host1x_} {} + +void NvMap::AddHandle(std::shared_ptr<Handle> handle_description) { + std::scoped_lock lock(handles_lock); + + handles.emplace(handle_description->id, std::move(handle_description)); +} + +void NvMap::UnmapHandle(Handle& handle_description) { + // Remove pending unmap queue entry if needed + if (handle_description.unmap_queue_entry) { + unmap_queue.erase(*handle_description.unmap_queue_entry); + handle_description.unmap_queue_entry.reset(); + } + + // Free and unmap the handle from the SMMU + host1x.MemoryManager().Unmap(static_cast<GPUVAddr>(handle_description.pin_virt_address), + handle_description.aligned_size); + host1x.Allocator().Free(handle_description.pin_virt_address, + static_cast<u32>(handle_description.aligned_size)); + handle_description.pin_virt_address = 0; +} + +bool NvMap::TryRemoveHandle(const Handle& handle_description) { + // No dupes left, we can remove from handle map + if (handle_description.dupes == 0 && handle_description.internal_dupes == 0) { + std::scoped_lock lock(handles_lock); + + auto it{handles.find(handle_description.id)}; + if (it != handles.end()) { + handles.erase(it); + } + + return true; + } else { + return false; + } +} + +NvResult NvMap::CreateHandle(u64 size, std::shared_ptr<NvMap::Handle>& result_out) { + if (!size) [[unlikely]] { + return NvResult::BadValue; + } + + u32 id{next_handle_id.fetch_add(HandleIdIncrement, std::memory_order_relaxed)}; + auto handle_description{std::make_shared<Handle>(size, id)}; + AddHandle(handle_description); + + result_out = handle_description; + return NvResult::Success; +} + +std::shared_ptr<NvMap::Handle> NvMap::GetHandle(Handle::Id handle) { + std::scoped_lock lock(handles_lock); + try { + return handles.at(handle); + } catch (std::out_of_range&) { + return nullptr; + } +} + +VAddr NvMap::GetHandleAddress(Handle::Id handle) { + std::scoped_lock lock(handles_lock); + try { + return handles.at(handle)->address; + } catch (std::out_of_range&) { + return 0; + } +} + +u32 NvMap::PinHandle(NvMap::Handle::Id handle) { + auto handle_description{GetHandle(handle)}; + if (!handle_description) [[unlikely]] { + return 0; + } + + std::scoped_lock lock(handle_description->mutex); + if (!handle_description->pins) { + // If we're in the unmap queue we can just remove ourselves and return since we're already + // mapped + { + // Lock now to prevent our queue entry from being removed for allocation in-between the + // following check and erase + std::scoped_lock queueLock(unmap_queue_lock); + if (handle_description->unmap_queue_entry) { + unmap_queue.erase(*handle_description->unmap_queue_entry); + handle_description->unmap_queue_entry.reset(); + + handle_description->pins++; + return handle_description->pin_virt_address; + } + } + + // If not then allocate some space and map it + u32 address{}; + auto& smmu_allocator = host1x.Allocator(); + auto& smmu_memory_manager = host1x.MemoryManager(); + while (!(address = + smmu_allocator.Allocate(static_cast<u32>(handle_description->aligned_size)))) { + // Free handles until the allocation succeeds + std::scoped_lock queueLock(unmap_queue_lock); + if (auto freeHandleDesc{unmap_queue.front()}) { + // Handles in the unmap queue are guaranteed not to be pinned so don't bother + // checking if they are before unmapping + std::scoped_lock freeLock(freeHandleDesc->mutex); + if (handle_description->pin_virt_address) + UnmapHandle(*freeHandleDesc); + } else { + LOG_CRITICAL(Service_NVDRV, "Ran out of SMMU address space!"); + } + } + + smmu_memory_manager.Map(static_cast<GPUVAddr>(address), handle_description->address, + handle_description->aligned_size); + handle_description->pin_virt_address = address; + } + + handle_description->pins++; + return handle_description->pin_virt_address; +} + +void NvMap::UnpinHandle(Handle::Id handle) { + auto handle_description{GetHandle(handle)}; + if (!handle_description) { + return; + } + + std::scoped_lock lock(handle_description->mutex); + if (--handle_description->pins < 0) { + LOG_WARNING(Service_NVDRV, "Pin count imbalance detected!"); + } else if (!handle_description->pins) { + std::scoped_lock queueLock(unmap_queue_lock); + + // Add to the unmap queue allowing this handle's memory to be freed if needed + unmap_queue.push_back(handle_description); + handle_description->unmap_queue_entry = std::prev(unmap_queue.end()); + } +} + +void NvMap::DuplicateHandle(Handle::Id handle, bool internal_session) { + auto handle_description{GetHandle(handle)}; + if (!handle_description) { + LOG_CRITICAL(Service_NVDRV, "Unregistered handle!"); + return; + } + + auto result = handle_description->Duplicate(internal_session); + if (result != NvResult::Success) { + LOG_CRITICAL(Service_NVDRV, "Could not duplicate handle!"); + } +} + +std::optional<NvMap::FreeInfo> NvMap::FreeHandle(Handle::Id handle, bool internal_session) { + std::weak_ptr<Handle> hWeak{GetHandle(handle)}; + FreeInfo freeInfo; + + // We use a weak ptr here so we can tell when the handle has been freed and report that back to + // guest + if (auto handle_description = hWeak.lock()) { + std::scoped_lock lock(handle_description->mutex); + + if (internal_session) { + if (--handle_description->internal_dupes < 0) + LOG_WARNING(Service_NVDRV, "Internal duplicate count imbalance detected!"); + } else { + if (--handle_description->dupes < 0) { + LOG_WARNING(Service_NVDRV, "User duplicate count imbalance detected!"); + } else if (handle_description->dupes == 0) { + // Force unmap the handle + if (handle_description->pin_virt_address) { + std::scoped_lock queueLock(unmap_queue_lock); + UnmapHandle(*handle_description); + } + + handle_description->pins = 0; + } + } + + // Try to remove the shared ptr to the handle from the map, if nothing else is using the + // handle then it will now be freed when `handle_description` goes out of scope + if (TryRemoveHandle(*handle_description)) { + LOG_DEBUG(Service_NVDRV, "Removed nvmap handle: {}", handle); + } else { + LOG_DEBUG(Service_NVDRV, + "Tried to free nvmap handle: {} but didn't as it still has duplicates", + handle); + } + + freeInfo = { + .address = handle_description->address, + .size = handle_description->size, + .was_uncached = handle_description->flags.map_uncached.Value() != 0, + }; + } else { + return std::nullopt; + } + + // Handle hasn't been freed from memory, set address to 0 to mark that the handle wasn't freed + if (!hWeak.expired()) { + LOG_DEBUG(Service_NVDRV, "nvmap handle: {} wasn't freed as it is still in use", handle); + freeInfo.address = 0; + } + + return freeInfo; +} + +} // namespace Service::Nvidia::NvCore diff --git a/src/core/hle/service/nvdrv/core/nvmap.h b/src/core/hle/service/nvdrv/core/nvmap.h new file mode 100644 index 000000000..b9dd3801f --- /dev/null +++ b/src/core/hle/service/nvdrv/core/nvmap.h @@ -0,0 +1,175 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-FileCopyrightText: 2022 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <atomic> +#include <list> +#include <memory> +#include <mutex> +#include <optional> +#include <unordered_map> +#include <assert.h> + +#include "common/bit_field.h" +#include "common/common_types.h" +#include "core/hle/service/nvdrv/nvdata.h" + +namespace Tegra { + +namespace Host1x { +class Host1x; +} // namespace Host1x + +} // namespace Tegra + +namespace Service::Nvidia::NvCore { +/** + * @brief The nvmap core class holds the global state for nvmap and provides methods to manage + * handles + */ +class NvMap { +public: + /** + * @brief A handle to a contiguous block of memory in an application's address space + */ + struct Handle { + std::mutex mutex; + + u64 align{}; //!< The alignment to use when pinning the handle onto the SMMU + u64 size; //!< Page-aligned size of the memory the handle refers to + u64 aligned_size; //!< `align`-aligned size of the memory the handle refers to + u64 orig_size; //!< Original unaligned size of the memory this handle refers to + + s32 dupes{1}; //!< How many guest references there are to this handle + s32 internal_dupes{0}; //!< How many emulator-internal references there are to this handle + + using Id = u32; + Id id; //!< A globally unique identifier for this handle + + s32 pins{}; + u32 pin_virt_address{}; + std::optional<typename std::list<std::shared_ptr<Handle>>::iterator> unmap_queue_entry{}; + + union Flags { + u32 raw; + BitField<0, 1, u32> map_uncached; //!< If the handle should be mapped as uncached + BitField<2, 1, u32> keep_uncached_after_free; //!< Only applicable when the handle was + //!< allocated with a fixed address + BitField<4, 1, u32> _unk0_; //!< Passed to IOVMM for pins + } flags{}; + static_assert(sizeof(Flags) == sizeof(u32)); + + u64 address{}; //!< The memory location in the guest's AS that this handle corresponds to, + //!< this can also be in the nvdrv tmem + bool is_shared_mem_mapped{}; //!< If this nvmap has been mapped with the MapSharedMem IPC + //!< call + + u8 kind{}; //!< Used for memory compression + bool allocated{}; //!< If the handle has been allocated with `Alloc` + + u64 dma_map_addr{}; //! remove me after implementing pinning. + + Handle(u64 size, Id id); + + /** + * @brief Sets up the handle with the given memory config, can allocate memory from the tmem + * if a 0 address is passed + */ + [[nodiscard]] NvResult Alloc(Flags pFlags, u32 pAlign, u8 pKind, u64 pAddress); + + /** + * @brief Increases the dupe counter of the handle for the given session + */ + [[nodiscard]] NvResult Duplicate(bool internal_session); + + /** + * @brief Obtains a pointer to the handle's memory and marks the handle it as having been + * mapped + */ + u8* GetPointer() { + if (!address) { + return nullptr; + } + + is_shared_mem_mapped = true; + return reinterpret_cast<u8*>(address); + } + }; + + /** + * @brief Encapsulates the result of a FreeHandle operation + */ + struct FreeInfo { + u64 address; //!< Address the handle referred to before deletion + u64 size; //!< Page-aligned handle size + bool was_uncached; //!< If the handle was allocated as uncached + }; + + explicit NvMap(Tegra::Host1x::Host1x& host1x); + + /** + * @brief Creates an unallocated handle of the given size + */ + [[nodiscard]] NvResult CreateHandle(u64 size, std::shared_ptr<NvMap::Handle>& result_out); + + std::shared_ptr<Handle> GetHandle(Handle::Id handle); + + VAddr GetHandleAddress(Handle::Id handle); + + /** + * @brief Maps a handle into the SMMU address space + * @note This operation is refcounted, the number of calls to this must eventually match the + * number of calls to `UnpinHandle` + * @return The SMMU virtual address that the handle has been mapped to + */ + u32 PinHandle(Handle::Id handle); + + /** + * @brief When this has been called an equal number of times to `PinHandle` for the supplied + * handle it will be added to a list of handles to be freed when necessary + */ + void UnpinHandle(Handle::Id handle); + + /** + * @brief Tries to duplicate a handle + */ + void DuplicateHandle(Handle::Id handle, bool internal_session = false); + + /** + * @brief Tries to free a handle and remove a single dupe + * @note If a handle has no dupes left and has no other users a FreeInfo struct will be returned + * describing the prior state of the handle + */ + std::optional<FreeInfo> FreeHandle(Handle::Id handle, bool internal_session); + +private: + std::list<std::shared_ptr<Handle>> unmap_queue{}; + std::mutex unmap_queue_lock{}; //!< Protects access to `unmap_queue` + + std::unordered_map<Handle::Id, std::shared_ptr<Handle>> + handles{}; //!< Main owning map of handles + std::mutex handles_lock; //!< Protects access to `handles` + + static constexpr u32 HandleIdIncrement{ + 4}; //!< Each new handle ID is an increment of 4 from the previous + std::atomic<u32> next_handle_id{HandleIdIncrement}; + Tegra::Host1x::Host1x& host1x; + + void AddHandle(std::shared_ptr<Handle> handle); + + /** + * @brief Unmaps and frees the SMMU memory region a handle is mapped to + * @note Both `unmap_queue_lock` and `handle_description.mutex` MUST be locked when calling this + */ + void UnmapHandle(Handle& handle_description); + + /** + * @brief Removes a handle from the map taking its dupes into account + * @note handle_description.mutex MUST be locked when calling this + * @return If the handle was removed from the map + */ + bool TryRemoveHandle(const Handle& handle_description); +}; +} // namespace Service::Nvidia::NvCore diff --git a/src/core/hle/service/nvdrv/core/syncpoint_manager.cpp b/src/core/hle/service/nvdrv/core/syncpoint_manager.cpp new file mode 100644 index 000000000..eda2041a0 --- /dev/null +++ b/src/core/hle/service/nvdrv/core/syncpoint_manager.cpp @@ -0,0 +1,121 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-FileCopyrightText: 2022 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/assert.h" +#include "core/hle/service/nvdrv/core/syncpoint_manager.h" +#include "video_core/host1x/host1x.h" + +namespace Service::Nvidia::NvCore { + +SyncpointManager::SyncpointManager(Tegra::Host1x::Host1x& host1x_) : host1x{host1x_} { + constexpr u32 VBlank0SyncpointId{26}; + constexpr u32 VBlank1SyncpointId{27}; + + // Reserve both vblank syncpoints as client managed as they use Continuous Mode + // Refer to section 14.3.5.3 of the TRM for more information on Continuous Mode + // https://github.com/Jetson-TX1-AndroidTV/android_kernel_jetson_tx1_hdmi_primary/blob/8f74a72394efb871cb3f886a3de2998cd7ff2990/drivers/gpu/host1x/drm/dc.c#L660 + ReserveSyncpoint(VBlank0SyncpointId, true); + ReserveSyncpoint(VBlank1SyncpointId, true); + + for (u32 syncpoint_id : channel_syncpoints) { + if (syncpoint_id) { + ReserveSyncpoint(syncpoint_id, false); + } + } +} + +SyncpointManager::~SyncpointManager() = default; + +u32 SyncpointManager::ReserveSyncpoint(u32 id, bool client_managed) { + if (syncpoints.at(id).reserved) { + ASSERT_MSG(false, "Requested syncpoint is in use"); + return 0; + } + + syncpoints.at(id).reserved = true; + syncpoints.at(id).interface_managed = client_managed; + + return id; +} + +u32 SyncpointManager::FindFreeSyncpoint() { + for (u32 i{1}; i < syncpoints.size(); i++) { + if (!syncpoints[i].reserved) { + return i; + } + } + ASSERT_MSG(false, "Failed to find a free syncpoint!"); + return 0; +} + +u32 SyncpointManager::AllocateSyncpoint(bool client_managed) { + std::lock_guard lock(reservation_lock); + return ReserveSyncpoint(FindFreeSyncpoint(), client_managed); +} + +void SyncpointManager::FreeSyncpoint(u32 id) { + std::lock_guard lock(reservation_lock); + ASSERT(syncpoints.at(id).reserved); + syncpoints.at(id).reserved = false; +} + +bool SyncpointManager::IsSyncpointAllocated(u32 id) { + return (id <= SyncpointCount) && syncpoints[id].reserved; +} + +bool SyncpointManager::HasSyncpointExpired(u32 id, u32 threshold) const { + const SyncpointInfo& syncpoint{syncpoints.at(id)}; + + if (!syncpoint.reserved) { + ASSERT(false); + return 0; + } + + // If the interface manages counters then we don't keep track of the maximum value as it handles + // sanity checking the values then + if (syncpoint.interface_managed) { + return static_cast<s32>(syncpoint.counter_min - threshold) >= 0; + } else { + return (syncpoint.counter_max - threshold) >= (syncpoint.counter_min - threshold); + } +} + +u32 SyncpointManager::IncrementSyncpointMaxExt(u32 id, u32 amount) { + if (!syncpoints.at(id).reserved) { + ASSERT(false); + return 0; + } + + return syncpoints.at(id).counter_max += amount; +} + +u32 SyncpointManager::ReadSyncpointMinValue(u32 id) { + if (!syncpoints.at(id).reserved) { + ASSERT(false); + return 0; + } + + return syncpoints.at(id).counter_min; +} + +u32 SyncpointManager::UpdateMin(u32 id) { + if (!syncpoints.at(id).reserved) { + ASSERT(false); + return 0; + } + + syncpoints.at(id).counter_min = host1x.GetSyncpointManager().GetHostSyncpointValue(id); + return syncpoints.at(id).counter_min; +} + +NvFence SyncpointManager::GetSyncpointFence(u32 id) { + if (!syncpoints.at(id).reserved) { + ASSERT(false); + return NvFence{}; + } + + return {.id = static_cast<s32>(id), .value = syncpoints.at(id).counter_max}; +} + +} // namespace Service::Nvidia::NvCore diff --git a/src/core/hle/service/nvdrv/core/syncpoint_manager.h b/src/core/hle/service/nvdrv/core/syncpoint_manager.h new file mode 100644 index 000000000..b76ef9032 --- /dev/null +++ b/src/core/hle/service/nvdrv/core/syncpoint_manager.h @@ -0,0 +1,134 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-FileCopyrightText: 2022 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <array> +#include <atomic> +#include <mutex> + +#include "common/common_types.h" +#include "core/hle/service/nvdrv/nvdata.h" + +namespace Tegra::Host1x { +class Host1x; +} // namespace Tegra::Host1x + +namespace Service::Nvidia::NvCore { + +enum class ChannelType : u32 { + MsEnc = 0, + VIC = 1, + GPU = 2, + NvDec = 3, + Display = 4, + NvJpg = 5, + TSec = 6, + Max = 7 +}; + +/** + * @brief SyncpointManager handles allocating and accessing host1x syncpoints, these are cached + * versions of the HW syncpoints which are intermittently synced + * @note Refer to Chapter 14 of the Tegra X1 TRM for an exhaustive overview of them + * @url https://http.download.nvidia.com/tegra-public-appnotes/host1x.html + * @url + * https://github.com/Jetson-TX1-AndroidTV/android_kernel_jetson_tx1_hdmi_primary/blob/jetson-tx1/drivers/video/tegra/host/nvhost_syncpt.c + */ +class SyncpointManager final { +public: + explicit SyncpointManager(Tegra::Host1x::Host1x& host1x); + ~SyncpointManager(); + + /** + * @brief Checks if the given syncpoint is both allocated and below the number of HW syncpoints + */ + bool IsSyncpointAllocated(u32 id); + + /** + * @brief Finds a free syncpoint and reserves it + * @return The ID of the reserved syncpoint + */ + u32 AllocateSyncpoint(bool client_managed); + + /** + * @url + * https://github.com/Jetson-TX1-AndroidTV/android_kernel_jetson_tx1_hdmi_primary/blob/8f74a72394efb871cb3f886a3de2998cd7ff2990/drivers/gpu/host1x/syncpt.c#L259 + */ + bool HasSyncpointExpired(u32 id, u32 threshold) const; + + bool IsFenceSignalled(NvFence fence) const { + return HasSyncpointExpired(fence.id, fence.value); + } + + /** + * @brief Atomically increments the maximum value of a syncpoint by the given amount + * @return The new max value of the syncpoint + */ + u32 IncrementSyncpointMaxExt(u32 id, u32 amount); + + /** + * @return The minimum value of the syncpoint + */ + u32 ReadSyncpointMinValue(u32 id); + + /** + * @brief Synchronises the minimum value of the syncpoint to with the GPU + * @return The new minimum value of the syncpoint + */ + u32 UpdateMin(u32 id); + + /** + * @brief Frees the usage of a syncpoint. + */ + void FreeSyncpoint(u32 id); + + /** + * @return A fence that will be signalled once this syncpoint hits its maximum value + */ + NvFence GetSyncpointFence(u32 id); + + static constexpr std::array<u32, static_cast<u32>(ChannelType::Max)> channel_syncpoints{ + 0x0, // `MsEnc` is unimplemented + 0xC, // `VIC` + 0x0, // `GPU` syncpoints are allocated per-channel instead + 0x36, // `NvDec` + 0x0, // `Display` is unimplemented + 0x37, // `NvJpg` + 0x0, // `TSec` is unimplemented + }; //!< Maps each channel ID to a constant syncpoint + +private: + /** + * @note reservation_lock should be locked when calling this + */ + u32 ReserveSyncpoint(u32 id, bool client_managed); + + /** + * @return The ID of the first free syncpoint + */ + u32 FindFreeSyncpoint(); + + struct SyncpointInfo { + std::atomic<u32> counter_min; //!< The least value the syncpoint can be (The value it was + //!< when it was last synchronized with host1x) + std::atomic<u32> counter_max; //!< The maximum value the syncpoint can reach according to + //!< the current usage + bool interface_managed; //!< If the syncpoint is managed by a host1x client interface, a + //!< client interface is a HW block that can handle host1x + //!< transactions on behalf of a host1x client (Which would + //!< otherwise need to be manually synced using PIO which is + //!< synchronous and requires direct cooperation of the CPU) + bool reserved; //!< If the syncpoint is reserved or not, not to be confused with a reserved + //!< value + }; + + constexpr static std::size_t SyncpointCount{192}; + std::array<SyncpointInfo, SyncpointCount> syncpoints{}; + std::mutex reservation_lock; + + Tegra::Host1x::Host1x& host1x; +}; + +} // namespace Service::Nvidia::NvCore diff --git a/src/core/hle/service/nvdrv/devices/nvdevice.h b/src/core/hle/service/nvdrv/devices/nvdevice.h index 696e8121e..204b0e757 100644 --- a/src/core/hle/service/nvdrv/devices/nvdevice.h +++ b/src/core/hle/service/nvdrv/devices/nvdevice.h @@ -11,6 +11,10 @@ namespace Core { class System; } +namespace Kernel { +class KEvent; +} + namespace Service::Nvidia::Devices { /// Represents an abstract nvidia device node. It is to be subclassed by concrete device nodes to @@ -64,6 +68,10 @@ public: */ virtual void OnClose(DeviceFD fd) = 0; + virtual Kernel::KEvent* QueryEvent(u32 event_id) { + return nullptr; + } + protected: Core::System& system; }; diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp index 604711914..4122fc98d 100644 --- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp +++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp @@ -5,15 +5,16 @@ #include "common/logging/log.h" #include "core/core.h" #include "core/core_timing.h" +#include "core/hle/service/nvdrv/core/container.h" +#include "core/hle/service/nvdrv/core/nvmap.h" #include "core/hle/service/nvdrv/devices/nvdisp_disp0.h" -#include "core/hle/service/nvdrv/devices/nvmap.h" #include "core/perf_stats.h" #include "video_core/gpu.h" namespace Service::Nvidia::Devices { -nvdisp_disp0::nvdisp_disp0(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_) - : nvdevice{system_}, nvmap_dev{std::move(nvmap_dev_)} {} +nvdisp_disp0::nvdisp_disp0(Core::System& system_, NvCore::Container& core) + : nvdevice{system_}, container{core}, nvmap{core.GetNvMapFile()} {} nvdisp_disp0::~nvdisp_disp0() = default; NvResult nvdisp_disp0::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -39,8 +40,9 @@ void nvdisp_disp0::OnClose(DeviceFD fd) {} void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, android::PixelFormat format, u32 width, u32 height, u32 stride, android::BufferTransformFlags transform, - const Common::Rectangle<int>& crop_rect) { - const VAddr addr = nvmap_dev->GetObjectAddress(buffer_handle); + const Common::Rectangle<int>& crop_rect, + std::array<Service::Nvidia::NvFence, 4>& fences, u32 num_fences) { + const VAddr addr = nvmap.GetHandleAddress(buffer_handle); LOG_TRACE(Service, "Drawing from address {:X} offset {:08X} Width {} Height {} Stride {} Format {}", addr, offset, width, height, stride, format); @@ -48,10 +50,15 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, android::PixelFormat form const Tegra::FramebufferConfig framebuffer{addr, offset, width, height, stride, format, transform, crop_rect}; + system.GPU().RequestSwapBuffers(&framebuffer, fences, num_fences); system.GetPerfStats().EndSystemFrame(); - system.GPU().SwapBuffers(&framebuffer); system.SpeedLimiter().DoSpeedLimiting(system.CoreTiming().GetGlobalTimeUs()); system.GetPerfStats().BeginSystemFrame(); } +Kernel::KEvent* nvdisp_disp0::QueryEvent(u32 event_id) { + LOG_CRITICAL(Service_NVDRV, "Unknown DISP Event {}", event_id); + return nullptr; +} + } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h index 67b105e02..04217ab12 100644 --- a/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h +++ b/src/core/hle/service/nvdrv/devices/nvdisp_disp0.h @@ -11,13 +11,18 @@ #include "core/hle/service/nvflinger/buffer_transform_flags.h" #include "core/hle/service/nvflinger/pixel_format.h" +namespace Service::Nvidia::NvCore { +class Container; +class NvMap; +} // namespace Service::Nvidia::NvCore + namespace Service::Nvidia::Devices { class nvmap; class nvdisp_disp0 final : public nvdevice { public: - explicit nvdisp_disp0(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_); + explicit nvdisp_disp0(Core::System& system_, NvCore::Container& core); ~nvdisp_disp0() override; NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -33,10 +38,14 @@ public: /// Performs a screen flip, drawing the buffer pointed to by the handle. void flip(u32 buffer_handle, u32 offset, android::PixelFormat format, u32 width, u32 height, u32 stride, android::BufferTransformFlags transform, - const Common::Rectangle<int>& crop_rect); + const Common::Rectangle<int>& crop_rect, + std::array<Service::Nvidia::NvFence, 4>& fences, u32 num_fences); + + Kernel::KEvent* QueryEvent(u32 event_id) override; private: - std::shared_ptr<nvmap> nvmap_dev; + NvCore::Container& container; + NvCore::NvMap& nvmap; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp index 9867a648d..6411dbf43 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp @@ -1,21 +1,30 @@ -// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #include <cstring> #include <utility> +#include "common/alignment.h" #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" +#include "core/hle/service/nvdrv/core/container.h" +#include "core/hle/service/nvdrv/core/nvmap.h" #include "core/hle/service/nvdrv/devices/nvhost_as_gpu.h" -#include "core/hle/service/nvdrv/devices/nvmap.h" +#include "core/hle/service/nvdrv/devices/nvhost_gpu.h" +#include "core/hle/service/nvdrv/nvdrv.h" +#include "video_core/control/channel_state.h" +#include "video_core/gpu.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" namespace Service::Nvidia::Devices { -nvhost_as_gpu::nvhost_as_gpu(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_) - : nvdevice{system_}, nvmap_dev{std::move(nvmap_dev_)} {} +nvhost_as_gpu::nvhost_as_gpu(Core::System& system_, Module& module_, NvCore::Container& core) + : nvdevice{system_}, module{module_}, container{core}, nvmap{core.GetNvMapFile()}, vm{}, + gmmu{} {} + nvhost_as_gpu::~nvhost_as_gpu() = default; NvResult nvhost_as_gpu::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -82,12 +91,52 @@ NvResult nvhost_as_gpu::AllocAsEx(const std::vector<u8>& input, std::vector<u8>& IoctlAllocAsEx params{}; std::memcpy(¶ms, input.data(), input.size()); - LOG_WARNING(Service_NVDRV, "(STUBBED) called, big_page_size=0x{:X}", params.big_page_size); - if (params.big_page_size == 0) { - params.big_page_size = DEFAULT_BIG_PAGE_SIZE; + LOG_DEBUG(Service_NVDRV, "called, big_page_size=0x{:X}", params.big_page_size); + + std::scoped_lock lock(mutex); + + if (vm.initialised) { + ASSERT_MSG(false, "Cannot initialise an address space twice!"); + return NvResult::InvalidState; + } + + if (params.big_page_size) { + if (!std::has_single_bit(params.big_page_size)) { + LOG_ERROR(Service_NVDRV, "Non power-of-2 big page size: 0x{:X}!", params.big_page_size); + return NvResult::BadValue; + } + + if ((params.big_page_size & VM::SUPPORTED_BIG_PAGE_SIZES) == 0) { + LOG_ERROR(Service_NVDRV, "Unsupported big page size: 0x{:X}!", params.big_page_size); + return NvResult::BadValue; + } + + vm.big_page_size = params.big_page_size; + vm.big_page_size_bits = static_cast<u32>(std::countr_zero(params.big_page_size)); + + vm.va_range_start = params.big_page_size << VM::VA_START_SHIFT; + } + + // If this is unspecified then default values should be used + if (params.va_range_start) { + vm.va_range_start = params.va_range_start; + vm.va_range_split = params.va_range_split; + vm.va_range_end = params.va_range_end; } - big_page_size = params.big_page_size; + const auto start_pages{static_cast<u32>(vm.va_range_start >> VM::PAGE_SIZE_BITS)}; + const auto end_pages{static_cast<u32>(vm.va_range_split >> VM::PAGE_SIZE_BITS)}; + vm.small_page_allocator = std::make_shared<VM::Allocator>(start_pages, end_pages); + + const auto start_big_pages{static_cast<u32>(vm.va_range_split >> vm.big_page_size_bits)}; + const auto end_big_pages{ + static_cast<u32>((vm.va_range_end - vm.va_range_split) >> vm.big_page_size_bits)}; + vm.big_page_allocator = std::make_unique<VM::Allocator>(start_big_pages, end_big_pages); + + gmmu = std::make_shared<Tegra::MemoryManager>(system, 40, vm.big_page_size_bits, + VM::PAGE_SIZE_BITS); + system.GPU().InitAddressSpace(*gmmu); + vm.initialised = true; return NvResult::Success; } @@ -99,21 +148,76 @@ NvResult nvhost_as_gpu::AllocateSpace(const std::vector<u8>& input, std::vector< LOG_DEBUG(Service_NVDRV, "called, pages={:X}, page_size={:X}, flags={:X}", params.pages, params.page_size, params.flags); - const auto size{static_cast<u64>(params.pages) * static_cast<u64>(params.page_size)}; - if ((params.flags & AddressSpaceFlags::FixedOffset) != AddressSpaceFlags::None) { - params.offset = *system.GPU().MemoryManager().AllocateFixed(params.offset, size); + std::scoped_lock lock(mutex); + + if (!vm.initialised) { + return NvResult::BadValue; + } + + if (params.page_size != VM::YUZU_PAGESIZE && params.page_size != vm.big_page_size) { + return NvResult::BadValue; + } + + if (params.page_size != vm.big_page_size && + ((params.flags & MappingFlags::Sparse) != MappingFlags::None)) { + UNIMPLEMENTED_MSG("Sparse small pages are not implemented!"); + return NvResult::NotImplemented; + } + + const u32 page_size_bits{params.page_size == VM::YUZU_PAGESIZE ? VM::PAGE_SIZE_BITS + : vm.big_page_size_bits}; + + auto& allocator{params.page_size == VM::YUZU_PAGESIZE ? *vm.small_page_allocator + : *vm.big_page_allocator}; + + if ((params.flags & MappingFlags::Fixed) != MappingFlags::None) { + allocator.AllocateFixed(static_cast<u32>(params.offset >> page_size_bits), params.pages); } else { - params.offset = system.GPU().MemoryManager().Allocate(size, params.align); + params.offset = static_cast<u64>(allocator.Allocate(params.pages)) << page_size_bits; + if (!params.offset) { + ASSERT_MSG(false, "Failed to allocate free space in the GPU AS!"); + return NvResult::InsufficientMemory; + } } - auto result = NvResult::Success; - if (!params.offset) { - LOG_CRITICAL(Service_NVDRV, "allocation failed for size {}", size); - result = NvResult::InsufficientMemory; + u64 size{static_cast<u64>(params.pages) * params.page_size}; + + if ((params.flags & MappingFlags::Sparse) != MappingFlags::None) { + gmmu->MapSparse(params.offset, size); } + allocation_map[params.offset] = { + .size = size, + .mappings{}, + .page_size = params.page_size, + .sparse = (params.flags & MappingFlags::Sparse) != MappingFlags::None, + .big_pages = params.page_size != VM::YUZU_PAGESIZE, + }; + std::memcpy(output.data(), ¶ms, output.size()); - return result; + return NvResult::Success; +} + +void nvhost_as_gpu::FreeMappingLocked(u64 offset) { + auto mapping{mapping_map.at(offset)}; + + if (!mapping->fixed) { + auto& allocator{mapping->big_page ? *vm.big_page_allocator : *vm.small_page_allocator}; + u32 page_size_bits{mapping->big_page ? vm.big_page_size_bits : VM::PAGE_SIZE_BITS}; + + allocator.Free(static_cast<u32>(mapping->offset >> page_size_bits), + static_cast<u32>(mapping->size >> page_size_bits)); + } + + // Sparse mappings shouldn't be fully unmapped, just returned to their sparse state + // Only FreeSpace can unmap them fully + if (mapping->sparse_alloc) { + gmmu->MapSparse(offset, mapping->size, mapping->big_page); + } else { + gmmu->Unmap(offset, mapping->size); + } + + mapping_map.erase(offset); } NvResult nvhost_as_gpu::FreeSpace(const std::vector<u8>& input, std::vector<u8>& output) { @@ -123,8 +227,40 @@ NvResult nvhost_as_gpu::FreeSpace(const std::vector<u8>& input, std::vector<u8>& LOG_DEBUG(Service_NVDRV, "called, offset={:X}, pages={:X}, page_size={:X}", params.offset, params.pages, params.page_size); - system.GPU().MemoryManager().Unmap(params.offset, - static_cast<std::size_t>(params.pages) * params.page_size); + std::scoped_lock lock(mutex); + + if (!vm.initialised) { + return NvResult::BadValue; + } + + try { + auto allocation{allocation_map[params.offset]}; + + if (allocation.page_size != params.page_size || + allocation.size != (static_cast<u64>(params.pages) * params.page_size)) { + return NvResult::BadValue; + } + + for (const auto& mapping : allocation.mappings) { + FreeMappingLocked(mapping->offset); + } + + // Unset sparse flag if required + if (allocation.sparse) { + gmmu->Unmap(params.offset, allocation.size); + } + + auto& allocator{params.page_size == VM::YUZU_PAGESIZE ? *vm.small_page_allocator + : *vm.big_page_allocator}; + u32 page_size_bits{params.page_size == VM::YUZU_PAGESIZE ? VM::PAGE_SIZE_BITS + : vm.big_page_size_bits}; + + allocator.Free(static_cast<u32>(params.offset >> page_size_bits), + static_cast<u32>(allocation.size >> page_size_bits)); + allocation_map.erase(params.offset); + } catch (const std::out_of_range&) { + return NvResult::BadValue; + } std::memcpy(output.data(), ¶ms, output.size()); return NvResult::Success; @@ -135,35 +271,52 @@ NvResult nvhost_as_gpu::Remap(const std::vector<u8>& input, std::vector<u8>& out LOG_DEBUG(Service_NVDRV, "called, num_entries=0x{:X}", num_entries); - auto result = NvResult::Success; std::vector<IoctlRemapEntry> entries(num_entries); std::memcpy(entries.data(), input.data(), input.size()); + std::scoped_lock lock(mutex); + + if (!vm.initialised) { + return NvResult::BadValue; + } + for (const auto& entry : entries) { - LOG_DEBUG(Service_NVDRV, "remap entry, offset=0x{:X} handle=0x{:X} pages=0x{:X}", - entry.offset, entry.nvmap_handle, entry.pages); + GPUVAddr virtual_address{static_cast<u64>(entry.as_offset_big_pages) + << vm.big_page_size_bits}; + u64 size{static_cast<u64>(entry.big_pages) << vm.big_page_size_bits}; - const auto object{nvmap_dev->GetObject(entry.nvmap_handle)}; - if (!object) { - LOG_CRITICAL(Service_NVDRV, "invalid nvmap_handle={:X}", entry.nvmap_handle); - result = NvResult::InvalidState; - break; + auto alloc{allocation_map.upper_bound(virtual_address)}; + + if (alloc-- == allocation_map.begin() || + (virtual_address - alloc->first) + size > alloc->second.size) { + LOG_WARNING(Service_NVDRV, "Cannot remap into an unallocated region!"); + return NvResult::BadValue; } - const auto offset{static_cast<GPUVAddr>(entry.offset) << 0x10}; - const auto size{static_cast<u64>(entry.pages) << 0x10}; - const auto map_offset{static_cast<u64>(entry.map_offset) << 0x10}; - const auto addr{system.GPU().MemoryManager().Map(object->addr + map_offset, offset, size)}; + if (!alloc->second.sparse) { + LOG_WARNING(Service_NVDRV, "Cannot remap a non-sparse mapping!"); + return NvResult::BadValue; + } - if (!addr) { - LOG_CRITICAL(Service_NVDRV, "map returned an invalid address!"); - result = NvResult::InvalidState; - break; + const bool use_big_pages = alloc->second.big_pages; + if (!entry.handle) { + gmmu->MapSparse(virtual_address, size, use_big_pages); + } else { + auto handle{nvmap.GetHandle(entry.handle)}; + if (!handle) { + return NvResult::BadValue; + } + + VAddr cpu_address{static_cast<VAddr>( + handle->address + + (static_cast<u64>(entry.handle_offset_big_pages) << vm.big_page_size_bits))}; + + gmmu->Map(virtual_address, cpu_address, size, use_big_pages); } } std::memcpy(output.data(), entries.data(), output.size()); - return result; + return NvResult::Success; } NvResult nvhost_as_gpu::MapBufferEx(const std::vector<u8>& input, std::vector<u8>& output) { @@ -173,79 +326,98 @@ NvResult nvhost_as_gpu::MapBufferEx(const std::vector<u8>& input, std::vector<u8 LOG_DEBUG(Service_NVDRV, "called, flags={:X}, nvmap_handle={:X}, buffer_offset={}, mapping_size={}" ", offset={}", - params.flags, params.nvmap_handle, params.buffer_offset, params.mapping_size, + params.flags, params.handle, params.buffer_offset, params.mapping_size, params.offset); - const auto object{nvmap_dev->GetObject(params.nvmap_handle)}; - if (!object) { - LOG_CRITICAL(Service_NVDRV, "invalid nvmap_handle={:X}", params.nvmap_handle); - std::memcpy(output.data(), ¶ms, output.size()); - return NvResult::InvalidState; - } - - // The real nvservices doesn't make a distinction between handles and ids, and - // object can only have one handle and it will be the same as its id. Assert that this is the - // case to prevent unexpected behavior. - ASSERT(object->id == params.nvmap_handle); - auto& gpu = system.GPU(); + std::scoped_lock lock(mutex); - u64 page_size{params.page_size}; - if (!page_size) { - page_size = object->align; + if (!vm.initialised) { + return NvResult::BadValue; } - if ((params.flags & AddressSpaceFlags::Remap) != AddressSpaceFlags::None) { - if (const auto buffer_map{FindBufferMap(params.offset)}; buffer_map) { - const auto cpu_addr{static_cast<VAddr>(buffer_map->CpuAddr() + params.buffer_offset)}; - const auto gpu_addr{static_cast<GPUVAddr>(params.offset + params.buffer_offset)}; + // Remaps a subregion of an existing mapping to a different PA + if ((params.flags & MappingFlags::Remap) != MappingFlags::None) { + try { + auto mapping{mapping_map.at(params.offset)}; - if (!gpu.MemoryManager().Map(cpu_addr, gpu_addr, params.mapping_size)) { - LOG_CRITICAL(Service_NVDRV, - "remap failed, flags={:X}, nvmap_handle={:X}, buffer_offset={}, " - "mapping_size = {}, offset={}", - params.flags, params.nvmap_handle, params.buffer_offset, - params.mapping_size, params.offset); - - std::memcpy(output.data(), ¶ms, output.size()); - return NvResult::InvalidState; + if (mapping->size < params.mapping_size) { + LOG_WARNING(Service_NVDRV, + "Cannot remap a partially mapped GPU address space region: 0x{:X}", + params.offset); + return NvResult::BadValue; } - std::memcpy(output.data(), ¶ms, output.size()); - return NvResult::Success; - } else { - LOG_CRITICAL(Service_NVDRV, "address not mapped offset={}", params.offset); + u64 gpu_address{static_cast<u64>(params.offset + params.buffer_offset)}; + VAddr cpu_address{mapping->ptr + params.buffer_offset}; + + gmmu->Map(gpu_address, cpu_address, params.mapping_size, mapping->big_page); - std::memcpy(output.data(), ¶ms, output.size()); - return NvResult::InvalidState; + return NvResult::Success; + } catch (const std::out_of_range&) { + LOG_WARNING(Service_NVDRV, "Cannot remap an unmapped GPU address space region: 0x{:X}", + params.offset); + return NvResult::BadValue; } } - // We can only map objects that have already been assigned a CPU address. - ASSERT(object->status == nvmap::Object::Status::Allocated); - - const auto physical_address{object->addr + params.buffer_offset}; - u64 size{params.mapping_size}; - if (!size) { - size = object->size; + auto handle{nvmap.GetHandle(params.handle)}; + if (!handle) { + return NvResult::BadValue; } - const bool is_alloc{(params.flags & AddressSpaceFlags::FixedOffset) == AddressSpaceFlags::None}; - if (is_alloc) { - params.offset = gpu.MemoryManager().MapAllocate(physical_address, size, page_size); - } else { - params.offset = gpu.MemoryManager().Map(physical_address, params.offset, size); - } + VAddr cpu_address{static_cast<VAddr>(handle->address + params.buffer_offset)}; + u64 size{params.mapping_size ? params.mapping_size : handle->orig_size}; + + bool big_page{[&]() { + if (Common::IsAligned(handle->align, vm.big_page_size)) { + return true; + } else if (Common::IsAligned(handle->align, VM::YUZU_PAGESIZE)) { + return false; + } else { + ASSERT(false); + return false; + } + }()}; + + if ((params.flags & MappingFlags::Fixed) != MappingFlags::None) { + auto alloc{allocation_map.upper_bound(params.offset)}; - auto result = NvResult::Success; - if (!params.offset) { - LOG_CRITICAL(Service_NVDRV, "failed to map size={}", size); - result = NvResult::InvalidState; + if (alloc-- == allocation_map.begin() || + (params.offset - alloc->first) + size > alloc->second.size) { + ASSERT_MSG(false, "Cannot perform a fixed mapping into an unallocated region!"); + return NvResult::BadValue; + } + + const bool use_big_pages = alloc->second.big_pages && big_page; + gmmu->Map(params.offset, cpu_address, size, use_big_pages); + + auto mapping{std::make_shared<Mapping>(cpu_address, params.offset, size, true, + use_big_pages, alloc->second.sparse)}; + alloc->second.mappings.push_back(mapping); + mapping_map[params.offset] = mapping; } else { - AddBufferMap(params.offset, size, physical_address, is_alloc); + + auto& allocator{big_page ? *vm.big_page_allocator : *vm.small_page_allocator}; + u32 page_size{big_page ? vm.big_page_size : VM::YUZU_PAGESIZE}; + u32 page_size_bits{big_page ? vm.big_page_size_bits : VM::PAGE_SIZE_BITS}; + + params.offset = static_cast<u64>(allocator.Allocate( + static_cast<u32>(Common::AlignUp(size, page_size) >> page_size_bits))) + << page_size_bits; + if (!params.offset) { + ASSERT_MSG(false, "Failed to allocate free space in the GPU AS!"); + return NvResult::InsufficientMemory; + } + + gmmu->Map(params.offset, cpu_address, Common::AlignUp(size, page_size), big_page); + + auto mapping{ + std::make_shared<Mapping>(cpu_address, params.offset, size, false, big_page, false)}; + mapping_map[params.offset] = mapping; } std::memcpy(output.data(), ¶ms, output.size()); - return result; + return NvResult::Success; } NvResult nvhost_as_gpu::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output) { @@ -254,47 +426,82 @@ NvResult nvhost_as_gpu::UnmapBuffer(const std::vector<u8>& input, std::vector<u8 LOG_DEBUG(Service_NVDRV, "called, offset=0x{:X}", params.offset); - if (const auto size{RemoveBufferMap(params.offset)}; size) { - system.GPU().MemoryManager().Unmap(params.offset, *size); - } else { - LOG_ERROR(Service_NVDRV, "invalid offset=0x{:X}", params.offset); + std::scoped_lock lock(mutex); + + if (!vm.initialised) { + return NvResult::BadValue; + } + + try { + auto mapping{mapping_map.at(params.offset)}; + + if (!mapping->fixed) { + auto& allocator{mapping->big_page ? *vm.big_page_allocator : *vm.small_page_allocator}; + u32 page_size_bits{mapping->big_page ? vm.big_page_size_bits : VM::PAGE_SIZE_BITS}; + + allocator.Free(static_cast<u32>(mapping->offset >> page_size_bits), + static_cast<u32>(mapping->size >> page_size_bits)); + } + + // Sparse mappings shouldn't be fully unmapped, just returned to their sparse state + // Only FreeSpace can unmap them fully + if (mapping->sparse_alloc) { + gmmu->MapSparse(params.offset, mapping->size, mapping->big_page); + } else { + gmmu->Unmap(params.offset, mapping->size); + } + + mapping_map.erase(params.offset); + } catch (const std::out_of_range&) { + LOG_WARNING(Service_NVDRV, "Couldn't find region to unmap at 0x{:X}", params.offset); } - std::memcpy(output.data(), ¶ms, output.size()); return NvResult::Success; } NvResult nvhost_as_gpu::BindChannel(const std::vector<u8>& input, std::vector<u8>& output) { IoctlBindChannel params{}; std::memcpy(¶ms, input.data(), input.size()); - LOG_WARNING(Service_NVDRV, "(STUBBED) called, fd={:X}", params.fd); + LOG_DEBUG(Service_NVDRV, "called, fd={:X}", params.fd); - channel = params.fd; + auto gpu_channel_device = module.GetDevice<nvhost_gpu>(params.fd); + gpu_channel_device->channel_state->memory_manager = gmmu; return NvResult::Success; } +void nvhost_as_gpu::GetVARegionsImpl(IoctlGetVaRegions& params) { + params.buf_size = 2 * sizeof(VaRegion); + + params.regions = std::array<VaRegion, 2>{ + VaRegion{ + .offset = vm.small_page_allocator->GetVAStart() << VM::PAGE_SIZE_BITS, + .page_size = VM::YUZU_PAGESIZE, + ._pad0_{}, + .pages = vm.small_page_allocator->GetVALimit() - vm.small_page_allocator->GetVAStart(), + }, + VaRegion{ + .offset = vm.big_page_allocator->GetVAStart() << vm.big_page_size_bits, + .page_size = vm.big_page_size, + ._pad0_{}, + .pages = vm.big_page_allocator->GetVALimit() - vm.big_page_allocator->GetVAStart(), + }, + }; +} + NvResult nvhost_as_gpu::GetVARegions(const std::vector<u8>& input, std::vector<u8>& output) { IoctlGetVaRegions params{}; std::memcpy(¶ms, input.data(), input.size()); - LOG_WARNING(Service_NVDRV, "(STUBBED) called, buf_addr={:X}, buf_size={:X}", params.buf_addr, - params.buf_size); - - params.buf_size = 0x30; + LOG_DEBUG(Service_NVDRV, "called, buf_addr={:X}, buf_size={:X}", params.buf_addr, + params.buf_size); - params.small = IoctlVaRegion{ - .offset = 0x04000000, - .page_size = DEFAULT_SMALL_PAGE_SIZE, - .pages = 0x3fbfff, - }; + std::scoped_lock lock(mutex); - params.big = IoctlVaRegion{ - .offset = 0x04000000, - .page_size = big_page_size, - .pages = 0x1bffff, - }; + if (!vm.initialised) { + return NvResult::BadValue; + } - // TODO(ogniK): This probably can stay stubbed but should add support way way later + GetVARegionsImpl(params); std::memcpy(output.data(), ¶ms, output.size()); return NvResult::Success; @@ -305,62 +512,27 @@ NvResult nvhost_as_gpu::GetVARegions(const std::vector<u8>& input, std::vector<u IoctlGetVaRegions params{}; std::memcpy(¶ms, input.data(), input.size()); - LOG_WARNING(Service_NVDRV, "(STUBBED) called, buf_addr={:X}, buf_size={:X}", params.buf_addr, - params.buf_size); - - params.buf_size = 0x30; + LOG_DEBUG(Service_NVDRV, "called, buf_addr={:X}, buf_size={:X}", params.buf_addr, + params.buf_size); - params.small = IoctlVaRegion{ - .offset = 0x04000000, - .page_size = 0x1000, - .pages = 0x3fbfff, - }; + std::scoped_lock lock(mutex); - params.big = IoctlVaRegion{ - .offset = 0x04000000, - .page_size = big_page_size, - .pages = 0x1bffff, - }; + if (!vm.initialised) { + return NvResult::BadValue; + } - // TODO(ogniK): This probably can stay stubbed but should add support way way later + GetVARegionsImpl(params); std::memcpy(output.data(), ¶ms, output.size()); - std::memcpy(inline_output.data(), ¶ms.small, sizeof(IoctlVaRegion)); - std::memcpy(inline_output.data() + sizeof(IoctlVaRegion), ¶ms.big, sizeof(IoctlVaRegion)); + std::memcpy(inline_output.data(), ¶ms.regions[0], sizeof(VaRegion)); + std::memcpy(inline_output.data() + sizeof(VaRegion), ¶ms.regions[1], sizeof(VaRegion)); return NvResult::Success; } -std::optional<nvhost_as_gpu::BufferMap> nvhost_as_gpu::FindBufferMap(GPUVAddr gpu_addr) const { - const auto end{buffer_mappings.upper_bound(gpu_addr)}; - for (auto iter{buffer_mappings.begin()}; iter != end; ++iter) { - if (gpu_addr >= iter->second.StartAddr() && gpu_addr < iter->second.EndAddr()) { - return iter->second; - } - } - - return std::nullopt; -} - -void nvhost_as_gpu::AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr, - bool is_allocated) { - buffer_mappings[gpu_addr] = {gpu_addr, size, cpu_addr, is_allocated}; -} - -std::optional<std::size_t> nvhost_as_gpu::RemoveBufferMap(GPUVAddr gpu_addr) { - if (const auto iter{buffer_mappings.find(gpu_addr)}; iter != buffer_mappings.end()) { - std::size_t size{}; - - if (iter->second.IsAllocated()) { - size = iter->second.Size(); - } - - buffer_mappings.erase(iter); - - return size; - } - - return std::nullopt; +Kernel::KEvent* nvhost_as_gpu::QueryEvent(u32 event_id) { + LOG_CRITICAL(Service_NVDRV, "Unknown AS GPU Event {}", event_id); + return nullptr; } } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h index 555843a6f..86fe71c75 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h @@ -1,35 +1,50 @@ -// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once +#include <bit> +#include <list> #include <map> #include <memory> +#include <mutex> #include <optional> #include <vector> +#include "common/address_space.h" #include "common/common_funcs.h" #include "common/common_types.h" #include "common/swap.h" +#include "core/hle/service/nvdrv/core/nvmap.h" #include "core/hle/service/nvdrv/devices/nvdevice.h" -namespace Service::Nvidia::Devices { +namespace Tegra { +class MemoryManager; +} // namespace Tegra + +namespace Service::Nvidia { +class Module; +} -constexpr u32 DEFAULT_BIG_PAGE_SIZE = 1 << 16; -constexpr u32 DEFAULT_SMALL_PAGE_SIZE = 1 << 12; +namespace Service::Nvidia::NvCore { +class Container; +class NvMap; +} // namespace Service::Nvidia::NvCore -class nvmap; +namespace Service::Nvidia::Devices { -enum class AddressSpaceFlags : u32 { - None = 0x0, - FixedOffset = 0x1, - Remap = 0x100, +enum class MappingFlags : u32 { + None = 0, + Fixed = 1 << 0, + Sparse = 1 << 1, + Remap = 1 << 8, }; -DECLARE_ENUM_FLAG_OPERATORS(AddressSpaceFlags); +DECLARE_ENUM_FLAG_OPERATORS(MappingFlags); class nvhost_as_gpu final : public nvdevice { public: - explicit nvhost_as_gpu(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_); + explicit nvhost_as_gpu(Core::System& system_, Module& module, NvCore::Container& core); ~nvhost_as_gpu() override; NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -42,46 +57,17 @@ public: void OnOpen(DeviceFD fd) override; void OnClose(DeviceFD fd) override; -private: - class BufferMap final { - public: - constexpr BufferMap() = default; - - constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_) - : start_addr{start_addr_}, end_addr{start_addr_ + size_} {} - - constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_, VAddr cpu_addr_, - bool is_allocated_) - : start_addr{start_addr_}, end_addr{start_addr_ + size_}, cpu_addr{cpu_addr_}, - is_allocated{is_allocated_} {} - - constexpr VAddr StartAddr() const { - return start_addr; - } - - constexpr VAddr EndAddr() const { - return end_addr; - } - - constexpr std::size_t Size() const { - return end_addr - start_addr; - } - - constexpr VAddr CpuAddr() const { - return cpu_addr; - } - - constexpr bool IsAllocated() const { - return is_allocated; - } - - private: - GPUVAddr start_addr{}; - GPUVAddr end_addr{}; - VAddr cpu_addr{}; - bool is_allocated{}; + Kernel::KEvent* QueryEvent(u32 event_id) override; + + struct VaRegion { + u64 offset; + u32 page_size; + u32 _pad0_; + u64 pages; }; + static_assert(sizeof(VaRegion) == 0x18); +private: struct IoctlAllocAsEx { u32_le flags{}; // usually passes 1 s32_le as_fd{}; // ignored; passes 0 @@ -96,7 +82,7 @@ private: struct IoctlAllocSpace { u32_le pages{}; u32_le page_size{}; - AddressSpaceFlags flags{}; + MappingFlags flags{}; INSERT_PADDING_WORDS(1); union { u64_le offset; @@ -113,19 +99,19 @@ private: static_assert(sizeof(IoctlFreeSpace) == 16, "IoctlFreeSpace is incorrect size"); struct IoctlRemapEntry { - u16_le flags{}; - u16_le kind{}; - u32_le nvmap_handle{}; - u32_le map_offset{}; - u32_le offset{}; - u32_le pages{}; + u16 flags; + u16 kind; + NvCore::NvMap::Handle::Id handle; + u32 handle_offset_big_pages; + u32 as_offset_big_pages; + u32 big_pages; }; static_assert(sizeof(IoctlRemapEntry) == 20, "IoctlRemapEntry is incorrect size"); struct IoctlMapBufferEx { - AddressSpaceFlags flags{}; // bit0: fixed_offset, bit2: cacheable - u32_le kind{}; // -1 is default - u32_le nvmap_handle{}; + MappingFlags flags{}; // bit0: fixed_offset, bit2: cacheable + u32_le kind{}; // -1 is default + NvCore::NvMap::Handle::Id handle; u32_le page_size{}; // 0 means don't care s64_le buffer_offset{}; u64_le mapping_size{}; @@ -143,27 +129,15 @@ private: }; static_assert(sizeof(IoctlBindChannel) == 4, "IoctlBindChannel is incorrect size"); - struct IoctlVaRegion { - u64_le offset{}; - u32_le page_size{}; - INSERT_PADDING_WORDS(1); - u64_le pages{}; - }; - static_assert(sizeof(IoctlVaRegion) == 24, "IoctlVaRegion is incorrect size"); - struct IoctlGetVaRegions { u64_le buf_addr{}; // (contained output user ptr on linux, ignored) u32_le buf_size{}; // forced to 2*sizeof(struct va_region) u32_le reserved{}; - IoctlVaRegion small{}; - IoctlVaRegion big{}; + std::array<VaRegion, 2> regions{}; }; - static_assert(sizeof(IoctlGetVaRegions) == 16 + sizeof(IoctlVaRegion) * 2, + static_assert(sizeof(IoctlGetVaRegions) == 16 + sizeof(VaRegion) * 2, "IoctlGetVaRegions is incorrect size"); - s32 channel{}; - u32 big_page_size{DEFAULT_BIG_PAGE_SIZE}; - NvResult AllocAsEx(const std::vector<u8>& input, std::vector<u8>& output); NvResult AllocateSpace(const std::vector<u8>& input, std::vector<u8>& output); NvResult Remap(const std::vector<u8>& input, std::vector<u8>& output); @@ -172,18 +146,75 @@ private: NvResult FreeSpace(const std::vector<u8>& input, std::vector<u8>& output); NvResult BindChannel(const std::vector<u8>& input, std::vector<u8>& output); + void GetVARegionsImpl(IoctlGetVaRegions& params); NvResult GetVARegions(const std::vector<u8>& input, std::vector<u8>& output); NvResult GetVARegions(const std::vector<u8>& input, std::vector<u8>& output, std::vector<u8>& inline_output); - std::optional<BufferMap> FindBufferMap(GPUVAddr gpu_addr) const; - void AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr, bool is_allocated); - std::optional<std::size_t> RemoveBufferMap(GPUVAddr gpu_addr); + void FreeMappingLocked(u64 offset); + + Module& module; + + NvCore::Container& container; + NvCore::NvMap& nvmap; - std::shared_ptr<nvmap> nvmap_dev; + struct Mapping { + VAddr ptr; + u64 offset; + u64 size; + bool fixed; + bool big_page; // Only valid if fixed == false + bool sparse_alloc; + + Mapping(VAddr ptr_, u64 offset_, u64 size_, bool fixed_, bool big_page_, bool sparse_alloc_) + : ptr(ptr_), offset(offset_), size(size_), fixed(fixed_), big_page(big_page_), + sparse_alloc(sparse_alloc_) {} + }; + + struct Allocation { + u64 size; + std::list<std::shared_ptr<Mapping>> mappings; + u32 page_size; + bool sparse; + bool big_pages; + }; - // This is expected to be ordered, therefore we must use a map, not unordered_map - std::map<GPUVAddr, BufferMap> buffer_mappings; + std::map<u64, std::shared_ptr<Mapping>> + mapping_map; //!< This maps the base addresses of mapped buffers to their total sizes and + //!< mapping type, this is needed as what was originally a single buffer may + //!< have been split into multiple GPU side buffers with the remap flag. + std::map<u64, Allocation> allocation_map; //!< Holds allocations created by AllocSpace from + //!< which fixed buffers can be mapped into + std::mutex mutex; //!< Locks all AS operations + + struct VM { + static constexpr u32 YUZU_PAGESIZE{0x1000}; + static constexpr u32 PAGE_SIZE_BITS{std::countr_zero(YUZU_PAGESIZE)}; + + static constexpr u32 SUPPORTED_BIG_PAGE_SIZES{0x30000}; + static constexpr u32 DEFAULT_BIG_PAGE_SIZE{0x20000}; + u32 big_page_size{DEFAULT_BIG_PAGE_SIZE}; + u32 big_page_size_bits{std::countr_zero(DEFAULT_BIG_PAGE_SIZE)}; + + static constexpr u32 VA_START_SHIFT{10}; + static constexpr u64 DEFAULT_VA_SPLIT{1ULL << 34}; + static constexpr u64 DEFAULT_VA_RANGE{1ULL << 37}; + u64 va_range_start{DEFAULT_BIG_PAGE_SIZE << VA_START_SHIFT}; + u64 va_range_split{DEFAULT_VA_SPLIT}; + u64 va_range_end{DEFAULT_VA_RANGE}; + + using Allocator = Common::FlatAllocator<u32, 0, 32>; + + std::unique_ptr<Allocator> big_page_allocator; + std::shared_ptr<Allocator> + small_page_allocator; //! Shared as this is also used by nvhost::GpuChannel + + bool initialised{}; + } vm; + std::shared_ptr<Tegra::MemoryManager> gmmu; + + // s32 channel{}; + // u32 big_page_size{VM::DEFAULT_BIG_PAGE_SIZE}; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp index 527531f29..5bee4a3d3 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp @@ -1,24 +1,39 @@ -// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later +#include <bit> #include <cstdlib> #include <cstring> +#include <fmt/format.h> #include "common/assert.h" #include "common/logging/log.h" +#include "common/scope_exit.h" #include "core/core.h" #include "core/hle/kernel/k_event.h" #include "core/hle/kernel/k_writable_event.h" +#include "core/hle/service/nvdrv/core/container.h" +#include "core/hle/service/nvdrv/core/syncpoint_manager.h" #include "core/hle/service/nvdrv/devices/nvhost_ctrl.h" #include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" namespace Service::Nvidia::Devices { nvhost_ctrl::nvhost_ctrl(Core::System& system_, EventInterface& events_interface_, - SyncpointManager& syncpoint_manager_) - : nvdevice{system_}, events_interface{events_interface_}, syncpoint_manager{ - syncpoint_manager_} {} -nvhost_ctrl::~nvhost_ctrl() = default; + NvCore::Container& core_) + : nvdevice{system_}, events_interface{events_interface_}, core{core_}, + syncpoint_manager{core_.GetSyncpointManager()} {} + +nvhost_ctrl::~nvhost_ctrl() { + for (auto& event : events) { + if (!event.registered) { + continue; + } + events_interface.FreeEvent(event.kevent); + } +} NvResult nvhost_ctrl::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) { @@ -30,13 +45,15 @@ NvResult nvhost_ctrl::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& case 0x1c: return IocCtrlClearEventWait(input, output); case 0x1d: - return IocCtrlEventWait(input, output, false); - case 0x1e: return IocCtrlEventWait(input, output, true); + case 0x1e: + return IocCtrlEventWait(input, output, false); case 0x1f: return IocCtrlEventRegister(input, output); case 0x20: return IocCtrlEventUnregister(input, output); + case 0x21: + return IocCtrlEventUnregisterBatch(input, output); } break; default: @@ -60,6 +77,7 @@ NvResult nvhost_ctrl::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>& } void nvhost_ctrl::OnOpen(DeviceFD fd) {} + void nvhost_ctrl::OnClose(DeviceFD fd) {} NvResult nvhost_ctrl::NvOsGetConfigU32(const std::vector<u8>& input, std::vector<u8>& output) { @@ -71,116 +89,167 @@ NvResult nvhost_ctrl::NvOsGetConfigU32(const std::vector<u8>& input, std::vector } NvResult nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, - bool is_async) { + bool is_allocation) { IocCtrlEventWaitParams params{}; std::memcpy(¶ms, input.data(), sizeof(params)); - LOG_DEBUG(Service_NVDRV, "syncpt_id={}, threshold={}, timeout={}, is_async={}", - params.syncpt_id, params.threshold, params.timeout, is_async); + LOG_DEBUG(Service_NVDRV, "syncpt_id={}, threshold={}, timeout={}, is_allocation={}", + params.fence.id, params.fence.value, params.timeout, is_allocation); - if (params.syncpt_id >= MaxSyncPoints) { - return NvResult::BadParameter; - } + bool must_unmark_fail = !is_allocation; + const u32 event_id = params.value.raw; + SCOPE_EXIT({ + std::memcpy(output.data(), ¶ms, sizeof(params)); + if (must_unmark_fail) { + events[event_id].fails = 0; + } + }); - u32 event_id = params.value & 0x00FF; + const u32 fence_id = static_cast<u32>(params.fence.id); - if (event_id >= MaxNvEvents) { - std::memcpy(output.data(), ¶ms, sizeof(params)); + if (fence_id >= MaxSyncPoints) { return NvResult::BadParameter; } - if (syncpoint_manager.IsSyncpointExpired(params.syncpt_id, params.threshold)) { - params.value = syncpoint_manager.GetSyncpointMin(params.syncpt_id); - std::memcpy(output.data(), ¶ms, sizeof(params)); - events_interface.failed[event_id] = false; + if (params.fence.value == 0) { + if (!syncpoint_manager.IsSyncpointAllocated(params.fence.id)) { + LOG_WARNING(Service_NVDRV, + "Unallocated syncpt_id={}, threshold={}, timeout={}, is_allocation={}", + params.fence.id, params.fence.value, params.timeout, is_allocation); + } else { + params.value.raw = syncpoint_manager.ReadSyncpointMinValue(fence_id); + } return NvResult::Success; } - if (const auto new_value = syncpoint_manager.RefreshSyncpoint(params.syncpt_id); - syncpoint_manager.IsSyncpointExpired(params.syncpt_id, params.threshold)) { - params.value = new_value; - std::memcpy(output.data(), ¶ms, sizeof(params)); - events_interface.failed[event_id] = false; + if (syncpoint_manager.IsFenceSignalled(params.fence)) { + params.value.raw = syncpoint_manager.ReadSyncpointMinValue(fence_id); return NvResult::Success; } - auto& event = events_interface.events[event_id]; - auto& gpu = system.GPU(); - - // This is mostly to take into account unimplemented features. As synced - // gpu is always synced. - if (!gpu.IsAsync()) { - event.event->GetWritableEvent().Signal(); - return NvResult::Success; - } - const u32 current_syncpoint_value = event.fence.value; - const s32 diff = current_syncpoint_value - params.threshold; - if (diff >= 0) { - event.event->GetWritableEvent().Signal(); - params.value = current_syncpoint_value; - std::memcpy(output.data(), ¶ms, sizeof(params)); - events_interface.failed[event_id] = false; + if (const auto new_value = syncpoint_manager.UpdateMin(fence_id); + syncpoint_manager.IsFenceSignalled(params.fence)) { + params.value.raw = new_value; return NvResult::Success; } - const u32 target_value = current_syncpoint_value - diff; - if (!is_async) { - params.value = 0; + auto& host1x_syncpoint_manager = system.Host1x().GetSyncpointManager(); + const u32 target_value = params.fence.value; + + auto lock = NvEventsLock(); + + u32 slot = [&]() { + if (is_allocation) { + params.value.raw = 0; + return FindFreeNvEvent(fence_id); + } else { + return params.value.raw; + } + }(); + + must_unmark_fail = false; + + const auto check_failing = [&]() { + if (events[slot].fails > 2) { + { + auto lk = system.StallProcesses(); + host1x_syncpoint_manager.WaitHost(fence_id, target_value); + system.UnstallProcesses(); + } + params.value.raw = target_value; + return true; + } + return false; + }; + + if (slot >= MaxNvEvents) { + return NvResult::BadParameter; } if (params.timeout == 0) { - std::memcpy(output.data(), ¶ms, sizeof(params)); + if (check_failing()) { + events[slot].fails = 0; + return NvResult::Success; + } return NvResult::Timeout; } - EventState status = events_interface.status[event_id]; - const bool bad_parameter = status == EventState::Busy; - if (bad_parameter) { - std::memcpy(output.data(), ¶ms, sizeof(params)); + auto& event = events[slot]; + + if (!event.registered) { return NvResult::BadParameter; } - events_interface.SetEventStatus(event_id, EventState::Waiting); - events_interface.assigned_syncpt[event_id] = params.syncpt_id; - events_interface.assigned_value[event_id] = target_value; - if (is_async) { - params.value = params.syncpt_id << 4; - } else { - params.value = ((params.syncpt_id & 0xfff) << 16) | 0x10000000; - } - params.value |= event_id; - event.event->GetWritableEvent().Clear(); - if (events_interface.failed[event_id]) { - { - auto lk = system.StallProcesses(); - gpu.WaitFence(params.syncpt_id, target_value); - system.UnstallProcesses(); - } - std::memcpy(output.data(), ¶ms, sizeof(params)); - events_interface.failed[event_id] = false; + + if (event.IsBeingUsed()) { + return NvResult::BadParameter; + } + + if (check_failing()) { + event.fails = 0; return NvResult::Success; } - gpu.RegisterSyncptInterrupt(params.syncpt_id, target_value); - std::memcpy(output.data(), ¶ms, sizeof(params)); + + params.value.raw = 0; + + event.status.store(EventState::Waiting, std::memory_order_release); + event.assigned_syncpt = fence_id; + event.assigned_value = target_value; + if (is_allocation) { + params.value.syncpoint_id_for_allocation.Assign(static_cast<u16>(fence_id)); + params.value.event_allocated.Assign(1); + } else { + params.value.syncpoint_id.Assign(fence_id); + } + params.value.raw |= slot; + + event.wait_handle = + host1x_syncpoint_manager.RegisterHostAction(fence_id, target_value, [this, slot]() { + auto& event_ = events[slot]; + if (event_.status.exchange(EventState::Signalling, std::memory_order_acq_rel) == + EventState::Waiting) { + event_.kevent->GetWritableEvent().Signal(); + } + event_.status.store(EventState::Signalled, std::memory_order_release); + }); return NvResult::Timeout; } +NvResult nvhost_ctrl::FreeEvent(u32 slot) { + if (slot >= MaxNvEvents) { + return NvResult::BadParameter; + } + + auto& event = events[slot]; + + if (!event.registered) { + return NvResult::Success; + } + + if (event.IsBeingUsed()) { + return NvResult::Busy; + } + + FreeNvEvent(slot); + return NvResult::Success; +} + NvResult nvhost_ctrl::IocCtrlEventRegister(const std::vector<u8>& input, std::vector<u8>& output) { IocCtrlEventRegisterParams params{}; std::memcpy(¶ms, input.data(), sizeof(params)); - const u32 event_id = params.user_event_id & 0x00FF; + const u32 event_id = params.user_event_id; LOG_DEBUG(Service_NVDRV, " called, user_event_id: {:X}", event_id); if (event_id >= MaxNvEvents) { return NvResult::BadParameter; } - if (events_interface.registered[event_id]) { - const auto event_state = events_interface.status[event_id]; - if (event_state != EventState::Free) { - LOG_WARNING(Service_NVDRV, "Event already registered! Unregistering previous event"); - events_interface.UnregisterEvent(event_id); - } else { - return NvResult::BadParameter; + + auto lock = NvEventsLock(); + + if (events[event_id].registered) { + const auto result = FreeEvent(event_id); + if (result != NvResult::Success) { + return result; } } - events_interface.RegisterEvent(event_id); + CreateNvEvent(event_id); return NvResult::Success; } @@ -190,34 +259,142 @@ NvResult nvhost_ctrl::IocCtrlEventUnregister(const std::vector<u8>& input, std::memcpy(¶ms, input.data(), sizeof(params)); const u32 event_id = params.user_event_id & 0x00FF; LOG_DEBUG(Service_NVDRV, " called, user_event_id: {:X}", event_id); - if (event_id >= MaxNvEvents) { - return NvResult::BadParameter; - } - if (!events_interface.registered[event_id]) { - return NvResult::BadParameter; + + auto lock = NvEventsLock(); + return FreeEvent(event_id); +} + +NvResult nvhost_ctrl::IocCtrlEventUnregisterBatch(const std::vector<u8>& input, + std::vector<u8>& output) { + IocCtrlEventUnregisterBatchParams params{}; + std::memcpy(¶ms, input.data(), sizeof(params)); + u64 event_mask = params.user_events; + LOG_DEBUG(Service_NVDRV, " called, event_mask: {:X}", event_mask); + + auto lock = NvEventsLock(); + while (event_mask != 0) { + const u64 event_id = std::countr_zero(event_mask); + event_mask &= ~(1ULL << event_id); + const auto result = FreeEvent(static_cast<u32>(event_id)); + if (result != NvResult::Success) { + return result; + } } - events_interface.UnregisterEvent(event_id); return NvResult::Success; } NvResult nvhost_ctrl::IocCtrlClearEventWait(const std::vector<u8>& input, std::vector<u8>& output) { - IocCtrlEventSignalParams params{}; + IocCtrlEventClearParams params{}; std::memcpy(¶ms, input.data(), sizeof(params)); - u32 event_id = params.event_id & 0x00FF; - LOG_WARNING(Service_NVDRV, "cleared event wait on, event_id: {:X}", event_id); + u32 event_id = params.event_id.slot; + LOG_DEBUG(Service_NVDRV, "called, event_id: {:X}", event_id); if (event_id >= MaxNvEvents) { return NvResult::BadParameter; } - if (events_interface.status[event_id] == EventState::Waiting) { - events_interface.LiberateEvent(event_id); - } - events_interface.failed[event_id] = true; - syncpoint_manager.RefreshSyncpoint(events_interface.events[event_id].fence.id); + auto lock = NvEventsLock(); + + auto& event = events[event_id]; + if (event.status.exchange(EventState::Cancelling, std::memory_order_acq_rel) == + EventState::Waiting) { + auto& host1x_syncpoint_manager = system.Host1x().GetSyncpointManager(); + host1x_syncpoint_manager.DeregisterHostAction(event.assigned_syncpt, event.wait_handle); + syncpoint_manager.UpdateMin(event.assigned_syncpt); + event.wait_handle = {}; + } + event.fails++; + event.status.store(EventState::Cancelled, std::memory_order_release); + event.kevent->GetWritableEvent().Clear(); return NvResult::Success; } +Kernel::KEvent* nvhost_ctrl::QueryEvent(u32 event_id) { + const auto desired_event = SyncpointEventValue{.raw = event_id}; + + const bool allocated = desired_event.event_allocated.Value() != 0; + const u32 slot{allocated ? desired_event.partial_slot.Value() + : static_cast<u32>(desired_event.slot)}; + if (slot >= MaxNvEvents) { + ASSERT(false); + return nullptr; + } + + const u32 syncpoint_id{allocated ? desired_event.syncpoint_id_for_allocation.Value() + : desired_event.syncpoint_id.Value()}; + + auto lock = NvEventsLock(); + + auto& event = events[slot]; + if (event.registered && event.assigned_syncpt == syncpoint_id) { + ASSERT(event.kevent); + return event.kevent; + } + // Is this possible in hardware? + ASSERT_MSG(false, "Slot:{}, SyncpointID:{}, requested", slot, syncpoint_id); + return nullptr; +} + +std::unique_lock<std::mutex> nvhost_ctrl::NvEventsLock() { + return std::unique_lock<std::mutex>(events_mutex); +} + +void nvhost_ctrl::CreateNvEvent(u32 event_id) { + auto& event = events[event_id]; + ASSERT(!event.kevent); + ASSERT(!event.registered); + ASSERT(!event.IsBeingUsed()); + event.kevent = events_interface.CreateEvent(fmt::format("NVCTRL::NvEvent_{}", event_id)); + event.status = EventState::Available; + event.registered = true; + const u64 mask = 1ULL << event_id; + event.fails = 0; + events_mask |= mask; + event.assigned_syncpt = 0; +} + +void nvhost_ctrl::FreeNvEvent(u32 event_id) { + auto& event = events[event_id]; + ASSERT(event.kevent); + ASSERT(event.registered); + ASSERT(!event.IsBeingUsed()); + events_interface.FreeEvent(event.kevent); + event.kevent = nullptr; + event.status = EventState::Available; + event.registered = false; + const u64 mask = ~(1ULL << event_id); + events_mask &= mask; +} + +u32 nvhost_ctrl::FindFreeNvEvent(u32 syncpoint_id) { + u32 slot{MaxNvEvents}; + u32 free_slot{MaxNvEvents}; + for (u32 i = 0; i < MaxNvEvents; i++) { + auto& event = events[i]; + if (event.registered) { + if (!event.IsBeingUsed()) { + slot = i; + if (event.assigned_syncpt == syncpoint_id) { + return slot; + } + } + } else if (free_slot == MaxNvEvents) { + free_slot = i; + } + } + if (free_slot < MaxNvEvents) { + CreateNvEvent(free_slot); + return free_slot; + } + + if (slot < MaxNvEvents) { + return slot; + } + + LOG_CRITICAL(Service_NVDRV, "Failed to allocate an event"); + return 0; +} + } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h index 4fbb89b15..0b56d7070 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl.h @@ -1,20 +1,28 @@ -// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once #include <array> #include <vector> +#include "common/bit_field.h" #include "common/common_types.h" #include "core/hle/service/nvdrv/devices/nvdevice.h" #include "core/hle/service/nvdrv/nvdrv.h" +#include "video_core/host1x/syncpoint_manager.h" + +namespace Service::Nvidia::NvCore { +class Container; +class SyncpointManager; +} // namespace Service::Nvidia::NvCore namespace Service::Nvidia::Devices { class nvhost_ctrl final : public nvdevice { public: explicit nvhost_ctrl(Core::System& system_, EventInterface& events_interface_, - SyncpointManager& syncpoint_manager_); + NvCore::Container& core); ~nvhost_ctrl() override; NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -27,7 +35,70 @@ public: void OnOpen(DeviceFD fd) override; void OnClose(DeviceFD fd) override; + Kernel::KEvent* QueryEvent(u32 event_id) override; + + union SyncpointEventValue { + u32 raw; + + union { + BitField<0, 4, u32> partial_slot; + BitField<4, 28, u32> syncpoint_id; + }; + + struct { + u16 slot; + union { + BitField<0, 12, u16> syncpoint_id_for_allocation; + BitField<12, 1, u16> event_allocated; + }; + }; + }; + static_assert(sizeof(SyncpointEventValue) == sizeof(u32)); + private: + struct InternalEvent { + // Mask representing registered events + + // Each kernel event associated to an NV event + Kernel::KEvent* kevent{}; + // The status of the current NVEvent + std::atomic<EventState> status{}; + + // Tells the NVEvent that it has failed. + u32 fails{}; + // When an NVEvent is waiting on GPU interrupt, this is the sync_point + // associated with it. + u32 assigned_syncpt{}; + // This is the value of the GPU interrupt for which the NVEvent is waiting + // for. + u32 assigned_value{}; + + // Tells if an NVEvent is registered or not + bool registered{}; + + // Used for waiting on a syncpoint & canceling it. + Tegra::Host1x::SyncpointManager::ActionHandle wait_handle{}; + + bool IsBeingUsed() const { + const auto current_status = status.load(std::memory_order_acquire); + return current_status == EventState::Waiting || + current_status == EventState::Cancelling || + current_status == EventState::Signalling; + } + }; + + std::unique_lock<std::mutex> NvEventsLock(); + + void CreateNvEvent(u32 event_id); + + void FreeNvEvent(u32 event_id); + + u32 FindFreeNvEvent(u32 syncpoint_id); + + std::array<InternalEvent, MaxNvEvents> events{}; + std::mutex events_mutex; + u64 events_mask{}; + struct IocSyncptReadParams { u32_le id{}; u32_le value{}; @@ -83,27 +154,18 @@ private: }; static_assert(sizeof(IocGetConfigParams) == 387, "IocGetConfigParams is incorrect size"); - struct IocCtrlEventSignalParams { - u32_le event_id{}; + struct IocCtrlEventClearParams { + SyncpointEventValue event_id{}; }; - static_assert(sizeof(IocCtrlEventSignalParams) == 4, - "IocCtrlEventSignalParams is incorrect size"); + static_assert(sizeof(IocCtrlEventClearParams) == 4, + "IocCtrlEventClearParams is incorrect size"); struct IocCtrlEventWaitParams { - u32_le syncpt_id{}; - u32_le threshold{}; - s32_le timeout{}; - u32_le value{}; - }; - static_assert(sizeof(IocCtrlEventWaitParams) == 16, "IocCtrlEventWaitParams is incorrect size"); - - struct IocCtrlEventWaitAsyncParams { - u32_le syncpt_id{}; - u32_le threshold{}; + NvFence fence{}; u32_le timeout{}; - u32_le value{}; + SyncpointEventValue value{}; }; - static_assert(sizeof(IocCtrlEventWaitAsyncParams) == 16, + static_assert(sizeof(IocCtrlEventWaitParams) == 16, "IocCtrlEventWaitAsyncParams is incorrect size"); struct IocCtrlEventRegisterParams { @@ -118,19 +180,25 @@ private: static_assert(sizeof(IocCtrlEventUnregisterParams) == 4, "IocCtrlEventUnregisterParams is incorrect size"); - struct IocCtrlEventKill { + struct IocCtrlEventUnregisterBatchParams { u64_le user_events{}; }; - static_assert(sizeof(IocCtrlEventKill) == 8, "IocCtrlEventKill is incorrect size"); + static_assert(sizeof(IocCtrlEventUnregisterBatchParams) == 8, + "IocCtrlEventKill is incorrect size"); NvResult NvOsGetConfigU32(const std::vector<u8>& input, std::vector<u8>& output); - NvResult IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, bool is_async); + NvResult IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, + bool is_allocation); NvResult IocCtrlEventRegister(const std::vector<u8>& input, std::vector<u8>& output); NvResult IocCtrlEventUnregister(const std::vector<u8>& input, std::vector<u8>& output); + NvResult IocCtrlEventUnregisterBatch(const std::vector<u8>& input, std::vector<u8>& output); NvResult IocCtrlClearEventWait(const std::vector<u8>& input, std::vector<u8>& output); + NvResult FreeEvent(u32 slot); + EventInterface& events_interface; - SyncpointManager& syncpoint_manager; + NvCore::Container& core; + NvCore::SyncpointManager& syncpoint_manager; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp index 2b3b7efea..ced57dfe6 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp @@ -7,11 +7,19 @@ #include "core/core.h" #include "core/core_timing.h" #include "core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h" +#include "core/hle/service/nvdrv/nvdrv.h" namespace Service::Nvidia::Devices { -nvhost_ctrl_gpu::nvhost_ctrl_gpu(Core::System& system_) : nvdevice{system_} {} -nvhost_ctrl_gpu::~nvhost_ctrl_gpu() = default; +nvhost_ctrl_gpu::nvhost_ctrl_gpu(Core::System& system_, EventInterface& events_interface_) + : nvdevice{system_}, events_interface{events_interface_} { + error_notifier_event = events_interface.CreateEvent("CtrlGpuErrorNotifier"); + unknown_event = events_interface.CreateEvent("CtrlGpuUknownEvent"); +} +nvhost_ctrl_gpu::~nvhost_ctrl_gpu() { + events_interface.FreeEvent(error_notifier_event); + events_interface.FreeEvent(unknown_event); +} NvResult nvhost_ctrl_gpu::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) { @@ -286,4 +294,17 @@ NvResult nvhost_ctrl_gpu::GetGpuTime(const std::vector<u8>& input, std::vector<u return NvResult::Success; } +Kernel::KEvent* nvhost_ctrl_gpu::QueryEvent(u32 event_id) { + switch (event_id) { + case 1: + return error_notifier_event; + case 2: + return unknown_event; + default: { + LOG_CRITICAL(Service_NVDRV, "Unknown Ctrl GPU Event {}", event_id); + } + } + return nullptr; +} + } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h index 97e9a90cb..1e8f254e2 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h @@ -10,11 +10,15 @@ #include "common/swap.h" #include "core/hle/service/nvdrv/devices/nvdevice.h" +namespace Service::Nvidia { +class EventInterface; +} + namespace Service::Nvidia::Devices { class nvhost_ctrl_gpu final : public nvdevice { public: - explicit nvhost_ctrl_gpu(Core::System& system_); + explicit nvhost_ctrl_gpu(Core::System& system_, EventInterface& events_interface_); ~nvhost_ctrl_gpu() override; NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -27,6 +31,8 @@ public: void OnOpen(DeviceFD fd) override; void OnClose(DeviceFD fd) override; + Kernel::KEvent* QueryEvent(u32 event_id) override; + private: struct IoctlGpuCharacteristics { u32_le arch; // 0x120 (NVGPU_GPU_ARCH_GM200) @@ -160,6 +166,12 @@ private: NvResult ZBCQueryTable(const std::vector<u8>& input, std::vector<u8>& output); NvResult FlushL2(const std::vector<u8>& input, std::vector<u8>& output); NvResult GetGpuTime(const std::vector<u8>& input, std::vector<u8>& output); + + EventInterface& events_interface; + + // Events + Kernel::KEvent* error_notifier_event; + Kernel::KEvent* unknown_event; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp index b98e63011..45a759fa8 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp @@ -5,29 +5,46 @@ #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" +#include "core/hle/service/nvdrv/core/container.h" +#include "core/hle/service/nvdrv/core/nvmap.h" +#include "core/hle/service/nvdrv/core/syncpoint_manager.h" #include "core/hle/service/nvdrv/devices/nvhost_gpu.h" -#include "core/hle/service/nvdrv/syncpoint_manager.h" +#include "core/hle/service/nvdrv/nvdrv.h" #include "core/memory.h" +#include "video_core/control/channel_state.h" +#include "video_core/engines/puller.h" #include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" namespace Service::Nvidia::Devices { namespace { -Tegra::CommandHeader BuildFenceAction(Tegra::GPU::FenceOperation op, u32 syncpoint_id) { - Tegra::GPU::FenceAction result{}; +Tegra::CommandHeader BuildFenceAction(Tegra::Engines::Puller::FenceOperation op, u32 syncpoint_id) { + Tegra::Engines::Puller::FenceAction result{}; result.op.Assign(op); result.syncpoint_id.Assign(syncpoint_id); return {result.raw}; } } // namespace -nvhost_gpu::nvhost_gpu(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_) - : nvdevice{system_}, nvmap_dev{std::move(nvmap_dev_)}, syncpoint_manager{syncpoint_manager_} { - channel_fence.id = syncpoint_manager_.AllocateSyncpoint(); - channel_fence.value = system_.GPU().GetSyncpointValue(channel_fence.id); +nvhost_gpu::nvhost_gpu(Core::System& system_, EventInterface& events_interface_, + NvCore::Container& core_) + : nvdevice{system_}, events_interface{events_interface_}, core{core_}, + syncpoint_manager{core_.GetSyncpointManager()}, nvmap{core.GetNvMapFile()}, + channel_state{system.GPU().AllocateChannel()} { + channel_syncpoint = syncpoint_manager.AllocateSyncpoint(false); + sm_exception_breakpoint_int_report_event = + events_interface.CreateEvent("GpuChannelSMExceptionBreakpointInt"); + sm_exception_breakpoint_pause_report_event = + events_interface.CreateEvent("GpuChannelSMExceptionBreakpointPause"); + error_notifier_event = events_interface.CreateEvent("GpuChannelErrorNotifier"); } -nvhost_gpu::~nvhost_gpu() = default; +nvhost_gpu::~nvhost_gpu() { + events_interface.FreeEvent(sm_exception_breakpoint_int_report_event); + events_interface.FreeEvent(sm_exception_breakpoint_pause_report_event); + events_interface.FreeEvent(error_notifier_event); + syncpoint_manager.FreeSyncpoint(channel_syncpoint); +} NvResult nvhost_gpu::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) { @@ -167,9 +184,14 @@ NvResult nvhost_gpu::AllocGPFIFOEx2(const std::vector<u8>& input, std::vector<u8 params.num_entries, params.flags, params.unk0, params.unk1, params.unk2, params.unk3); - channel_fence.value = system.GPU().GetSyncpointValue(channel_fence.id); + if (channel_state->initialized) { + LOG_CRITICAL(Service_NVDRV, "Already allocated!"); + return NvResult::AlreadyAllocated; + } + + system.GPU().InitChannel(*channel_state); - params.fence_out = channel_fence; + params.fence_out = syncpoint_manager.GetSyncpointFence(channel_syncpoint); std::memcpy(output.data(), ¶ms, output.size()); return NvResult::Success; @@ -188,39 +210,37 @@ NvResult nvhost_gpu::AllocateObjectContext(const std::vector<u8>& input, std::ve static std::vector<Tegra::CommandHeader> BuildWaitCommandList(NvFence fence) { return { - Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1, + Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointPayload, 1, Tegra::SubmissionMode::Increasing), {fence.value}, - Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1, + Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointOperation, 1, Tegra::SubmissionMode::Increasing), - BuildFenceAction(Tegra::GPU::FenceOperation::Acquire, fence.id), + BuildFenceAction(Tegra::Engines::Puller::FenceOperation::Acquire, fence.id), }; } -static std::vector<Tegra::CommandHeader> BuildIncrementCommandList(NvFence fence, - u32 add_increment) { +static std::vector<Tegra::CommandHeader> BuildIncrementCommandList(NvFence fence) { std::vector<Tegra::CommandHeader> result{ - Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceValue, 1, + Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointPayload, 1, Tegra::SubmissionMode::Increasing), {}}; - for (u32 count = 0; count < add_increment; ++count) { - result.emplace_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::FenceAction, 1, + for (u32 count = 0; count < 2; ++count) { + result.emplace_back(Tegra::BuildCommandHeader(Tegra::BufferMethods::SyncpointOperation, 1, Tegra::SubmissionMode::Increasing)); - result.emplace_back(BuildFenceAction(Tegra::GPU::FenceOperation::Increment, fence.id)); + result.emplace_back( + BuildFenceAction(Tegra::Engines::Puller::FenceOperation::Increment, fence.id)); } return result; } -static std::vector<Tegra::CommandHeader> BuildIncrementWithWfiCommandList(NvFence fence, - u32 add_increment) { +static std::vector<Tegra::CommandHeader> BuildIncrementWithWfiCommandList(NvFence fence) { std::vector<Tegra::CommandHeader> result{ - Tegra::BuildCommandHeader(Tegra::BufferMethods::WaitForInterrupt, 1, + Tegra::BuildCommandHeader(Tegra::BufferMethods::WaitForIdle, 1, Tegra::SubmissionMode::Increasing), {}}; - const std::vector<Tegra::CommandHeader> increment{ - BuildIncrementCommandList(fence, add_increment)}; + const std::vector<Tegra::CommandHeader> increment{BuildIncrementCommandList(fence)}; result.insert(result.end(), increment.begin(), increment.end()); @@ -234,33 +254,41 @@ NvResult nvhost_gpu::SubmitGPFIFOImpl(IoctlSubmitGpfifo& params, std::vector<u8> auto& gpu = system.GPU(); - params.fence_out.id = channel_fence.id; + std::scoped_lock lock(channel_mutex); - if (params.flags.add_wait.Value() && - !syncpoint_manager.IsSyncpointExpired(params.fence_out.id, params.fence_out.value)) { - gpu.PushGPUEntries(Tegra::CommandList{BuildWaitCommandList(params.fence_out)}); - } + const auto bind_id = channel_state->bind_id; - if (params.flags.add_increment.Value() || params.flags.increment.Value()) { - const u32 increment_value = params.flags.increment.Value() ? params.fence_out.value : 0; - params.fence_out.value = syncpoint_manager.IncreaseSyncpoint( - params.fence_out.id, params.AddIncrementValue() + increment_value); - } else { - params.fence_out.value = syncpoint_manager.GetSyncpointMax(params.fence_out.id); + auto& flags = params.flags; + + if (flags.fence_wait.Value()) { + if (flags.increment_value.Value()) { + return NvResult::BadParameter; + } + + if (!syncpoint_manager.IsFenceSignalled(params.fence)) { + gpu.PushGPUEntries(bind_id, Tegra::CommandList{BuildWaitCommandList(params.fence)}); + } } - gpu.PushGPUEntries(std::move(entries)); + params.fence.id = channel_syncpoint; + + u32 increment{(flags.fence_increment.Value() != 0 ? 2 : 0) + + (flags.increment_value.Value() != 0 ? params.fence.value : 0)}; + params.fence.value = syncpoint_manager.IncrementSyncpointMaxExt(channel_syncpoint, increment); + gpu.PushGPUEntries(bind_id, std::move(entries)); - if (params.flags.add_increment.Value()) { - if (params.flags.suppress_wfi) { - gpu.PushGPUEntries(Tegra::CommandList{ - BuildIncrementCommandList(params.fence_out, params.AddIncrementValue())}); + if (flags.fence_increment.Value()) { + if (flags.suppress_wfi.Value()) { + gpu.PushGPUEntries(bind_id, + Tegra::CommandList{BuildIncrementCommandList(params.fence)}); } else { - gpu.PushGPUEntries(Tegra::CommandList{ - BuildIncrementWithWfiCommandList(params.fence_out, params.AddIncrementValue())}); + gpu.PushGPUEntries(bind_id, + Tegra::CommandList{BuildIncrementWithWfiCommandList(params.fence)}); } } + flags.raw = 0; + std::memcpy(output.data(), ¶ms, sizeof(IoctlSubmitGpfifo)); return NvResult::Success; } @@ -328,4 +356,19 @@ NvResult nvhost_gpu::ChannelSetTimeslice(const std::vector<u8>& input, std::vect return NvResult::Success; } +Kernel::KEvent* nvhost_gpu::QueryEvent(u32 event_id) { + switch (event_id) { + case 1: + return sm_exception_breakpoint_int_report_event; + case 2: + return sm_exception_breakpoint_pause_report_event; + case 3: + return error_notifier_event; + default: { + LOG_CRITICAL(Service_NVDRV, "Unknown Ctrl GPU Event {}", event_id); + } + } + return nullptr; +} + } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h index 8a9f7775a..1e4ecd55b 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_gpu.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_gpu.h @@ -13,17 +13,31 @@ #include "core/hle/service/nvdrv/nvdata.h" #include "video_core/dma_pusher.h" +namespace Tegra { +namespace Control { +struct ChannelState; +} +} // namespace Tegra + namespace Service::Nvidia { + +namespace NvCore { +class Container; +class NvMap; class SyncpointManager; -} +} // namespace NvCore + +class EventInterface; +} // namespace Service::Nvidia namespace Service::Nvidia::Devices { +class nvhost_as_gpu; class nvmap; class nvhost_gpu final : public nvdevice { public: - explicit nvhost_gpu(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_); + explicit nvhost_gpu(Core::System& system_, EventInterface& events_interface_, + NvCore::Container& core); ~nvhost_gpu() override; NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -36,7 +50,10 @@ public: void OnOpen(DeviceFD fd) override; void OnClose(DeviceFD fd) override; + Kernel::KEvent* QueryEvent(u32 event_id) override; + private: + friend class nvhost_as_gpu; enum class CtxObjects : u32_le { Ctx2D = 0x902D, Ctx3D = 0xB197, @@ -146,17 +163,13 @@ private: u32_le num_entries{}; // number of fence objects being submitted union { u32_le raw; - BitField<0, 1, u32_le> add_wait; // append a wait sync_point to the list - BitField<1, 1, u32_le> add_increment; // append an increment to the list - BitField<2, 1, u32_le> new_hw_format; // mostly ignored - BitField<4, 1, u32_le> suppress_wfi; // suppress wait for interrupt - BitField<8, 1, u32_le> increment; // increment the returned fence + BitField<0, 1, u32_le> fence_wait; // append a wait sync_point to the list + BitField<1, 1, u32_le> fence_increment; // append an increment to the list + BitField<2, 1, u32_le> new_hw_format; // mostly ignored + BitField<4, 1, u32_le> suppress_wfi; // suppress wait for interrupt + BitField<8, 1, u32_le> increment_value; // increment the returned fence } flags; - NvFence fence_out{}; // returned new fence object for others to wait on - - u32 AddIncrementValue() const { - return flags.add_increment.Value() << 1; - } + NvFence fence{}; // returned new fence object for others to wait on }; static_assert(sizeof(IoctlSubmitGpfifo) == 16 + sizeof(NvFence), "IoctlSubmitGpfifo is incorrect size"); @@ -191,9 +204,18 @@ private: NvResult ChannelSetTimeout(const std::vector<u8>& input, std::vector<u8>& output); NvResult ChannelSetTimeslice(const std::vector<u8>& input, std::vector<u8>& output); - std::shared_ptr<nvmap> nvmap_dev; - SyncpointManager& syncpoint_manager; - NvFence channel_fence; + EventInterface& events_interface; + NvCore::Container& core; + NvCore::SyncpointManager& syncpoint_manager; + NvCore::NvMap& nvmap; + std::shared_ptr<Tegra::Control::ChannelState> channel_state; + u32 channel_syncpoint; + std::mutex channel_mutex; + + // Events + Kernel::KEvent* sm_exception_breakpoint_int_report_event; + Kernel::KEvent* sm_exception_breakpoint_pause_report_event; + Kernel::KEvent* error_notifier_event; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp index a7385fce8..1703f9cc3 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp @@ -5,14 +5,14 @@ #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" +#include "core/hle/service/nvdrv/core/container.h" #include "core/hle/service/nvdrv/devices/nvhost_nvdec.h" #include "video_core/renderer_base.h" namespace Service::Nvidia::Devices { -nvhost_nvdec::nvhost_nvdec(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_) - : nvhost_nvdec_common{system_, std::move(nvmap_dev_), syncpoint_manager_} {} +nvhost_nvdec::nvhost_nvdec(Core::System& system_, NvCore::Container& core_) + : nvhost_nvdec_common{system_, core_, NvCore::ChannelType::NvDec} {} nvhost_nvdec::~nvhost_nvdec() = default; NvResult nvhost_nvdec::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -21,8 +21,9 @@ NvResult nvhost_nvdec::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& case 0x0: switch (command.cmd) { case 0x1: { - if (!fd_to_id.contains(fd)) { - fd_to_id[fd] = next_id++; + auto& host1x_file = core.Host1xDeviceFile(); + if (!host1x_file.fd_to_id.contains(fd)) { + host1x_file.fd_to_id[fd] = host1x_file.nvdec_next_id++; } return Submit(fd, input, output); } @@ -73,8 +74,9 @@ void nvhost_nvdec::OnOpen(DeviceFD fd) { void nvhost_nvdec::OnClose(DeviceFD fd) { LOG_INFO(Service_NVDRV, "NVDEC video stream ended"); - const auto iter = fd_to_id.find(fd); - if (iter != fd_to_id.end()) { + auto& host1x_file = core.Host1xDeviceFile(); + const auto iter = host1x_file.fd_to_id.find(fd); + if (iter != host1x_file.fd_to_id.end()) { system.GPU().ClearCdmaInstance(iter->second); } system.AudioCore().SetNVDECActive(false); diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h index 29b3e6a36..c1b4e53e8 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec.h @@ -10,8 +10,7 @@ namespace Service::Nvidia::Devices { class nvhost_nvdec final : public nvhost_nvdec_common { public: - explicit nvhost_nvdec(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_); + explicit nvhost_nvdec(Core::System& system_, NvCore::Container& core); ~nvhost_nvdec() override; NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -23,9 +22,6 @@ public: void OnOpen(DeviceFD fd) override; void OnClose(DeviceFD fd) override; - -private: - u32 next_id{}; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp index 8b2cd9bf1..99eede702 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.cpp @@ -8,10 +8,12 @@ #include "common/common_types.h" #include "common/logging/log.h" #include "core/core.h" +#include "core/hle/service/nvdrv/core/container.h" +#include "core/hle/service/nvdrv/core/nvmap.h" +#include "core/hle/service/nvdrv/core/syncpoint_manager.h" #include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h" -#include "core/hle/service/nvdrv/devices/nvmap.h" -#include "core/hle/service/nvdrv/syncpoint_manager.h" #include "core/memory.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" #include "video_core/renderer_base.h" @@ -44,10 +46,22 @@ std::size_t WriteVectors(std::vector<u8>& dst, const std::vector<T>& src, std::s } } // Anonymous namespace -nvhost_nvdec_common::nvhost_nvdec_common(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_) - : nvdevice{system_}, nvmap_dev{std::move(nvmap_dev_)}, syncpoint_manager{syncpoint_manager_} {} -nvhost_nvdec_common::~nvhost_nvdec_common() = default; +nvhost_nvdec_common::nvhost_nvdec_common(Core::System& system_, NvCore::Container& core_, + NvCore::ChannelType channel_type_) + : nvdevice{system_}, core{core_}, syncpoint_manager{core.GetSyncpointManager()}, + nvmap{core.GetNvMapFile()}, channel_type{channel_type_} { + auto& syncpts_accumulated = core.Host1xDeviceFile().syncpts_accumulated; + if (syncpts_accumulated.empty()) { + channel_syncpoint = syncpoint_manager.AllocateSyncpoint(false); + } else { + channel_syncpoint = syncpts_accumulated.front(); + syncpts_accumulated.pop_front(); + } +} + +nvhost_nvdec_common::~nvhost_nvdec_common() { + core.Host1xDeviceFile().syncpts_accumulated.push_back(channel_syncpoint); +} NvResult nvhost_nvdec_common::SetNVMAPfd(const std::vector<u8>& input) { IoctlSetNvmapFD params{}; @@ -84,16 +98,16 @@ NvResult nvhost_nvdec_common::Submit(DeviceFD fd, const std::vector<u8>& input, for (std::size_t i = 0; i < syncpt_increments.size(); i++) { const SyncptIncr& syncpt_incr = syncpt_increments[i]; fence_thresholds[i] = - syncpoint_manager.IncreaseSyncpoint(syncpt_incr.id, syncpt_incr.increments); + syncpoint_manager.IncrementSyncpointMaxExt(syncpt_incr.id, syncpt_incr.increments); } } for (const auto& cmd_buffer : command_buffers) { - const auto object = nvmap_dev->GetObject(cmd_buffer.memory_id); + const auto object = nvmap.GetHandle(cmd_buffer.memory_id); ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;); Tegra::ChCommandHeaderList cmdlist(cmd_buffer.word_count); - system.Memory().ReadBlock(object->addr + cmd_buffer.offset, cmdlist.data(), + system.Memory().ReadBlock(object->address + cmd_buffer.offset, cmdlist.data(), cmdlist.size() * sizeof(u32)); - gpu.PushCommandBuffer(fd_to_id[fd], cmdlist); + gpu.PushCommandBuffer(core.Host1xDeviceFile().fd_to_id[fd], cmdlist); } std::memcpy(output.data(), ¶ms, sizeof(IoctlSubmit)); // Some games expect command_buffers to be written back @@ -112,10 +126,8 @@ NvResult nvhost_nvdec_common::GetSyncpoint(const std::vector<u8>& input, std::ve std::memcpy(¶ms, input.data(), sizeof(IoctlGetSyncpoint)); LOG_DEBUG(Service_NVDRV, "called GetSyncpoint, id={}", params.param); - if (device_syncpoints[params.param] == 0 && system.GPU().UseNvdec()) { - device_syncpoints[params.param] = syncpoint_manager.AllocateSyncpoint(); - } - params.value = device_syncpoints[params.param]; + // const u32 id{NvCore::SyncpointManager::channel_syncpoints[static_cast<u32>(channel_type)]}; + params.value = channel_syncpoint; std::memcpy(output.data(), ¶ms, sizeof(IoctlGetSyncpoint)); return NvResult::Success; @@ -123,6 +135,7 @@ NvResult nvhost_nvdec_common::GetSyncpoint(const std::vector<u8>& input, std::ve NvResult nvhost_nvdec_common::GetWaitbase(const std::vector<u8>& input, std::vector<u8>& output) { IoctlGetWaitbase params{}; + LOG_CRITICAL(Service_NVDRV, "called WAITBASE"); std::memcpy(¶ms, input.data(), sizeof(IoctlGetWaitbase)); params.value = 0; // Seems to be hard coded at 0 std::memcpy(output.data(), ¶ms, sizeof(IoctlGetWaitbase)); @@ -136,28 +149,8 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer)); - auto& gpu = system.GPU(); - for (auto& cmd_buffer : cmd_buffer_handles) { - auto object{nvmap_dev->GetObject(cmd_buffer.map_handle)}; - if (!object) { - LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmd_buffer.map_handle); - std::memcpy(output.data(), ¶ms, output.size()); - return NvResult::InvalidState; - } - if (object->dma_map_addr == 0) { - // NVDEC and VIC memory is in the 32-bit address space - // MapAllocate32 will attempt to map a lower 32-bit value in the shared gpu memory space - const GPUVAddr low_addr = gpu.MemoryManager().MapAllocate32(object->addr, object->size); - object->dma_map_addr = static_cast<u32>(low_addr); - // Ensure that the dma_map_addr is indeed in the lower 32-bit address space. - ASSERT(object->dma_map_addr == low_addr); - } - if (!object->dma_map_addr) { - LOG_ERROR(Service_NVDRV, "failed to map size={}", object->size); - } else { - cmd_buffer.map_address = object->dma_map_addr; - } + cmd_buffer.map_address = nvmap.PinHandle(cmd_buffer.map_handle); } std::memcpy(output.data(), ¶ms, sizeof(IoctlMapBuffer)); std::memcpy(output.data() + sizeof(IoctlMapBuffer), cmd_buffer_handles.data(), @@ -167,11 +160,16 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto } NvResult nvhost_nvdec_common::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output) { - // This is intntionally stubbed. - // Skip unmapping buffers here, as to not break the continuity of the VP9 reference frame - // addresses, and risk invalidating data before the async GPU thread is done with it + IoctlMapBuffer params{}; + std::memcpy(¶ms, input.data(), sizeof(IoctlMapBuffer)); + std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries); + + SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer)); + for (auto& cmd_buffer : cmd_buffer_handles) { + nvmap.UnpinHandle(cmd_buffer.map_handle); + } + std::memset(output.data(), 0, output.size()); - LOG_DEBUG(Service_NVDRV, "(STUBBED) called"); return NvResult::Success; } @@ -182,4 +180,9 @@ NvResult nvhost_nvdec_common::SetSubmitTimeout(const std::vector<u8>& input, return NvResult::Success; } +Kernel::KEvent* nvhost_nvdec_common::QueryEvent(u32 event_id) { + LOG_CRITICAL(Service_NVDRV, "Unknown HOSTX1 Event {}", event_id); + return nullptr; +} + } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h index 12d39946d..fe76100c8 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h @@ -3,21 +3,26 @@ #pragma once +#include <deque> #include <vector> #include "common/common_types.h" #include "common/swap.h" +#include "core/hle/service/nvdrv/core/syncpoint_manager.h" #include "core/hle/service/nvdrv/devices/nvdevice.h" namespace Service::Nvidia { -class SyncpointManager; + +namespace NvCore { +class Container; +class NvMap; +} // namespace NvCore namespace Devices { -class nvmap; class nvhost_nvdec_common : public nvdevice { public: - explicit nvhost_nvdec_common(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_); + explicit nvhost_nvdec_common(Core::System& system_, NvCore::Container& core, + NvCore::ChannelType channel_type); ~nvhost_nvdec_common() override; protected: @@ -110,11 +115,15 @@ protected: NvResult UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output); NvResult SetSubmitTimeout(const std::vector<u8>& input, std::vector<u8>& output); - std::unordered_map<DeviceFD, u32> fd_to_id{}; + Kernel::KEvent* QueryEvent(u32 event_id) override; + + u32 channel_syncpoint; s32_le nvmap_fd{}; u32_le submit_timeout{}; - std::shared_ptr<nvmap> nvmap_dev; - SyncpointManager& syncpoint_manager; + NvCore::Container& core; + NvCore::SyncpointManager& syncpoint_manager; + NvCore::NvMap& nvmap; + NvCore::ChannelType channel_type; std::array<u32, MaxSyncPoints> device_syncpoints{}; }; }; // namespace Devices diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp index f58e8bada..73f97136e 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp +++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.cpp @@ -4,13 +4,14 @@ #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" +#include "core/hle/service/nvdrv/core/container.h" #include "core/hle/service/nvdrv/devices/nvhost_vic.h" #include "video_core/renderer_base.h" namespace Service::Nvidia::Devices { -nvhost_vic::nvhost_vic(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_) - : nvhost_nvdec_common{system_, std::move(nvmap_dev_), syncpoint_manager_} {} + +nvhost_vic::nvhost_vic(Core::System& system_, NvCore::Container& core_) + : nvhost_nvdec_common{system_, core_, NvCore::ChannelType::VIC} {} nvhost_vic::~nvhost_vic() = default; @@ -19,11 +20,13 @@ NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& i switch (command.group) { case 0x0: switch (command.cmd) { - case 0x1: - if (!fd_to_id.contains(fd)) { - fd_to_id[fd] = next_id++; + case 0x1: { + auto& host1x_file = core.Host1xDeviceFile(); + if (!host1x_file.fd_to_id.contains(fd)) { + host1x_file.fd_to_id[fd] = host1x_file.vic_next_id++; } return Submit(fd, input, output); + } case 0x2: return GetSyncpoint(input, output); case 0x3: @@ -67,8 +70,9 @@ NvResult nvhost_vic::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>& i void nvhost_vic::OnOpen(DeviceFD fd) {} void nvhost_vic::OnClose(DeviceFD fd) { - const auto iter = fd_to_id.find(fd); - if (iter != fd_to_id.end()) { + auto& host1x_file = core.Host1xDeviceFile(); + const auto iter = host1x_file.fd_to_id.find(fd); + if (iter != host1x_file.fd_to_id.end()) { system.GPU().ClearCdmaInstance(iter->second); } } diff --git a/src/core/hle/service/nvdrv/devices/nvhost_vic.h b/src/core/hle/service/nvdrv/devices/nvhost_vic.h index b41b195ae..f164caafb 100644 --- a/src/core/hle/service/nvdrv/devices/nvhost_vic.h +++ b/src/core/hle/service/nvdrv/devices/nvhost_vic.h @@ -9,8 +9,7 @@ namespace Service::Nvidia::Devices { class nvhost_vic final : public nvhost_nvdec_common { public: - explicit nvhost_vic(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_, - SyncpointManager& syncpoint_manager_); + explicit nvhost_vic(Core::System& system_, NvCore::Container& core); ~nvhost_vic(); NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -22,8 +21,5 @@ public: void OnOpen(DeviceFD fd) override; void OnClose(DeviceFD fd) override; - -private: - u32 next_id{}; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/devices/nvmap.cpp b/src/core/hle/service/nvdrv/devices/nvmap.cpp index d8518149d..ddf273b5e 100644 --- a/src/core/hle/service/nvdrv/devices/nvmap.cpp +++ b/src/core/hle/service/nvdrv/devices/nvmap.cpp @@ -2,19 +2,26 @@ // SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> +#include <bit> #include <cstring> +#include "common/alignment.h" #include "common/assert.h" #include "common/logging/log.h" +#include "core/core.h" +#include "core/hle/kernel/k_page_table.h" +#include "core/hle/kernel/k_process.h" +#include "core/hle/service/nvdrv/core/container.h" +#include "core/hle/service/nvdrv/core/nvmap.h" #include "core/hle/service/nvdrv/devices/nvmap.h" +#include "core/memory.h" + +using Core::Memory::YUZU_PAGESIZE; namespace Service::Nvidia::Devices { -nvmap::nvmap(Core::System& system_) : nvdevice{system_} { - // Handle 0 appears to be used when remapping, so we create a placeholder empty nvmap object to - // represent this. - CreateObject(0); -} +nvmap::nvmap(Core::System& system_, NvCore::Container& container_) + : nvdevice{system_}, container{container_}, file{container.GetNvMapFile()} {} nvmap::~nvmap() = default; @@ -62,39 +69,21 @@ NvResult nvmap::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>& input, void nvmap::OnOpen(DeviceFD fd) {} void nvmap::OnClose(DeviceFD fd) {} -VAddr nvmap::GetObjectAddress(u32 handle) const { - auto object = GetObject(handle); - ASSERT(object); - ASSERT(object->status == Object::Status::Allocated); - return object->addr; -} - -u32 nvmap::CreateObject(u32 size) { - // Create a new nvmap object and obtain a handle to it. - auto object = std::make_shared<Object>(); - object->id = next_id++; - object->size = size; - object->status = Object::Status::Created; - object->refcount = 1; - - const u32 handle = next_handle++; - - handles.insert_or_assign(handle, std::move(object)); - - return handle; -} - NvResult nvmap::IocCreate(const std::vector<u8>& input, std::vector<u8>& output) { IocCreateParams params; std::memcpy(¶ms, input.data(), sizeof(params)); - LOG_DEBUG(Service_NVDRV, "size=0x{:08X}", params.size); - - if (!params.size) { - LOG_ERROR(Service_NVDRV, "Size is 0"); - return NvResult::BadValue; + LOG_DEBUG(Service_NVDRV, "called, size=0x{:08X}", params.size); + + std::shared_ptr<NvCore::NvMap::Handle> handle_description{}; + auto result = + file.CreateHandle(Common::AlignUp(params.size, YUZU_PAGESIZE), handle_description); + if (result != NvResult::Success) { + LOG_CRITICAL(Service_NVDRV, "Failed to create Object"); + return result; } - - params.handle = CreateObject(params.size); + handle_description->orig_size = params.size; // Orig size is the unaligned size + params.handle = handle_description->id; + LOG_DEBUG(Service_NVDRV, "handle: {}, size: 0x{:X}", handle_description->id, params.size); std::memcpy(output.data(), ¶ms, sizeof(params)); return NvResult::Success; @@ -103,63 +92,68 @@ NvResult nvmap::IocCreate(const std::vector<u8>& input, std::vector<u8>& output) NvResult nvmap::IocAlloc(const std::vector<u8>& input, std::vector<u8>& output) { IocAllocParams params; std::memcpy(¶ms, input.data(), sizeof(params)); - LOG_DEBUG(Service_NVDRV, "called, addr={:X}", params.addr); + LOG_DEBUG(Service_NVDRV, "called, addr={:X}", params.address); if (!params.handle) { - LOG_ERROR(Service_NVDRV, "Handle is 0"); + LOG_CRITICAL(Service_NVDRV, "Handle is 0"); return NvResult::BadValue; } if ((params.align - 1) & params.align) { - LOG_ERROR(Service_NVDRV, "Incorrect alignment used, alignment={:08X}", params.align); + LOG_CRITICAL(Service_NVDRV, "Incorrect alignment used, alignment={:08X}", params.align); return NvResult::BadValue; } - const u32 min_alignment = 0x1000; - if (params.align < min_alignment) { - params.align = min_alignment; + // Force page size alignment at a minimum + if (params.align < YUZU_PAGESIZE) { + params.align = YUZU_PAGESIZE; } - auto object = GetObject(params.handle); - if (!object) { - LOG_ERROR(Service_NVDRV, "Object does not exist, handle={:08X}", params.handle); + auto handle_description{file.GetHandle(params.handle)}; + if (!handle_description) { + LOG_CRITICAL(Service_NVDRV, "Object does not exist, handle={:08X}", params.handle); return NvResult::BadValue; } - if (object->status == Object::Status::Allocated) { - LOG_ERROR(Service_NVDRV, "Object is already allocated, handle={:08X}", params.handle); + if (handle_description->allocated) { + LOG_CRITICAL(Service_NVDRV, "Object is already allocated, handle={:08X}", params.handle); return NvResult::InsufficientMemory; } - object->flags = params.flags; - object->align = params.align; - object->kind = params.kind; - object->addr = params.addr; - object->status = Object::Status::Allocated; - + const auto result = + handle_description->Alloc(params.flags, params.align, params.kind, params.address); + if (result != NvResult::Success) { + LOG_CRITICAL(Service_NVDRV, "Object failed to allocate, handle={:08X}", params.handle); + return result; + } + ASSERT(system.CurrentProcess() + ->PageTable() + .LockForDeviceAddressSpace(handle_description->address, handle_description->size) + .IsSuccess()); std::memcpy(output.data(), ¶ms, sizeof(params)); - return NvResult::Success; + return result; } NvResult nvmap::IocGetId(const std::vector<u8>& input, std::vector<u8>& output) { IocGetIdParams params; std::memcpy(¶ms, input.data(), sizeof(params)); - LOG_WARNING(Service_NVDRV, "called"); + LOG_DEBUG(Service_NVDRV, "called"); + // See the comment in FromId for extra info on this function if (!params.handle) { - LOG_ERROR(Service_NVDRV, "Handle is zero"); + LOG_CRITICAL(Service_NVDRV, "Error!"); return NvResult::BadValue; } - auto object = GetObject(params.handle); - if (!object) { - LOG_ERROR(Service_NVDRV, "Object does not exist, handle={:08X}", params.handle); - return NvResult::BadValue; + auto handle_description{file.GetHandle(params.handle)}; + if (!handle_description) { + LOG_CRITICAL(Service_NVDRV, "Error!"); + return NvResult::AccessDenied; // This will always return EPERM irrespective of if the + // handle exists or not } - params.id = object->id; - + params.id = handle_description->id; std::memcpy(output.data(), ¶ms, sizeof(params)); return NvResult::Success; } @@ -168,26 +162,29 @@ NvResult nvmap::IocFromId(const std::vector<u8>& input, std::vector<u8>& output) IocFromIdParams params; std::memcpy(¶ms, input.data(), sizeof(params)); - LOG_WARNING(Service_NVDRV, "(STUBBED) called"); + LOG_DEBUG(Service_NVDRV, "called, id:{}", params.id); - auto itr = std::find_if(handles.begin(), handles.end(), - [&](const auto& entry) { return entry.second->id == params.id; }); - if (itr == handles.end()) { - LOG_ERROR(Service_NVDRV, "Object does not exist, handle={:08X}", params.handle); + // Handles and IDs are always the same value in nvmap however IDs can be used globally given the + // right permissions. + // Since we don't plan on ever supporting multiprocess we can skip implementing handle refs and + // so this function just does simple validation and passes through the handle id. + if (!params.id) { + LOG_CRITICAL(Service_NVDRV, "Zero Id is invalid!"); return NvResult::BadValue; } - auto& object = itr->second; - if (object->status != Object::Status::Allocated) { - LOG_ERROR(Service_NVDRV, "Object is not allocated, handle={:08X}", params.handle); + auto handle_description{file.GetHandle(params.id)}; + if (!handle_description) { + LOG_CRITICAL(Service_NVDRV, "Unregistered handle!"); return NvResult::BadValue; } - itr->second->refcount++; - - // Return the existing handle instead of creating a new one. - params.handle = itr->first; - + auto result = handle_description->Duplicate(false); + if (result != NvResult::Success) { + LOG_CRITICAL(Service_NVDRV, "Could not duplicate handle!"); + return result; + } + params.handle = handle_description->id; std::memcpy(output.data(), ¶ms, sizeof(params)); return NvResult::Success; } @@ -198,35 +195,43 @@ NvResult nvmap::IocParam(const std::vector<u8>& input, std::vector<u8>& output) IocParamParams params; std::memcpy(¶ms, input.data(), sizeof(params)); - LOG_DEBUG(Service_NVDRV, "(STUBBED) called type={}", params.param); + LOG_DEBUG(Service_NVDRV, "called type={}", params.param); - auto object = GetObject(params.handle); - if (!object) { - LOG_ERROR(Service_NVDRV, "Object does not exist, handle={:08X}", params.handle); + if (!params.handle) { + LOG_CRITICAL(Service_NVDRV, "Invalid handle!"); return NvResult::BadValue; } - if (object->status != Object::Status::Allocated) { - LOG_ERROR(Service_NVDRV, "Object is not allocated, handle={:08X}", params.handle); + auto handle_description{file.GetHandle(params.handle)}; + if (!handle_description) { + LOG_CRITICAL(Service_NVDRV, "Not registered handle!"); return NvResult::BadValue; } - switch (static_cast<ParamTypes>(params.param)) { - case ParamTypes::Size: - params.result = object->size; + switch (params.param) { + case HandleParameterType::Size: + params.result = static_cast<u32_le>(handle_description->orig_size); + break; + case HandleParameterType::Alignment: + params.result = static_cast<u32_le>(handle_description->align); break; - case ParamTypes::Alignment: - params.result = object->align; + case HandleParameterType::Base: + params.result = static_cast<u32_le>(-22); // posix EINVAL break; - case ParamTypes::Heap: - // TODO(Subv): Seems to be a hardcoded value? - params.result = 0x40000000; + case HandleParameterType::Heap: + if (handle_description->allocated) + params.result = 0x40000000; + else + params.result = 0; break; - case ParamTypes::Kind: - params.result = object->kind; + case HandleParameterType::Kind: + params.result = handle_description->kind; + break; + case HandleParameterType::IsSharedMemMapped: + params.result = handle_description->is_shared_mem_mapped; break; default: - UNIMPLEMENTED(); + return NvResult::BadValue; } std::memcpy(output.data(), ¶ms, sizeof(params)); @@ -234,46 +239,29 @@ NvResult nvmap::IocParam(const std::vector<u8>& input, std::vector<u8>& output) } NvResult nvmap::IocFree(const std::vector<u8>& input, std::vector<u8>& output) { - // TODO(Subv): These flags are unconfirmed. - enum FreeFlags { - Freed = 0, - NotFreedYet = 1, - }; - IocFreeParams params; std::memcpy(¶ms, input.data(), sizeof(params)); - LOG_DEBUG(Service_NVDRV, "(STUBBED) called"); + LOG_DEBUG(Service_NVDRV, "called"); - auto itr = handles.find(params.handle); - if (itr == handles.end()) { - LOG_ERROR(Service_NVDRV, "Object does not exist, handle={:08X}", params.handle); - return NvResult::BadValue; - } - if (!itr->second->refcount) { - LOG_ERROR( - Service_NVDRV, - "There is no references to this object. The object is already freed. handle={:08X}", - params.handle); - return NvResult::BadValue; + if (!params.handle) { + LOG_CRITICAL(Service_NVDRV, "Handle null freed?"); + return NvResult::Success; } - itr->second->refcount--; - - params.size = itr->second->size; - - if (itr->second->refcount == 0) { - params.flags = Freed; - // The address of the nvmap is written to the output if we're finally freeing it, otherwise - // 0 is written. - params.address = itr->second->addr; + if (auto freeInfo{file.FreeHandle(params.handle, false)}) { + ASSERT(system.CurrentProcess() + ->PageTable() + .UnlockForDeviceAddressSpace(freeInfo->address, freeInfo->size) + .IsSuccess()); + params.address = freeInfo->address; + params.size = static_cast<u32>(freeInfo->size); + params.flags.raw = 0; + params.flags.map_uncached.Assign(freeInfo->was_uncached); } else { - params.flags = NotFreedYet; - params.address = 0; + // This is possible when there's internel dups or other duplicates. } - handles.erase(params.handle); - std::memcpy(output.data(), ¶ms, sizeof(params)); return NvResult::Success; } diff --git a/src/core/hle/service/nvdrv/devices/nvmap.h b/src/core/hle/service/nvdrv/devices/nvmap.h index d5360d6e5..e9bfd0358 100644 --- a/src/core/hle/service/nvdrv/devices/nvmap.h +++ b/src/core/hle/service/nvdrv/devices/nvmap.h @@ -9,15 +9,23 @@ #include "common/common_funcs.h" #include "common/common_types.h" #include "common/swap.h" +#include "core/hle/service/nvdrv/core/nvmap.h" #include "core/hle/service/nvdrv/devices/nvdevice.h" +namespace Service::Nvidia::NvCore { +class Container; +} // namespace Service::Nvidia::NvCore + namespace Service::Nvidia::Devices { class nvmap final : public nvdevice { public: - explicit nvmap(Core::System& system_); + explicit nvmap(Core::System& system_, NvCore::Container& container); ~nvmap() override; + nvmap(const nvmap&) = delete; + nvmap& operator=(const nvmap&) = delete; + NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override; NvResult Ioctl2(DeviceFD fd, Ioctl command, const std::vector<u8>& input, @@ -28,31 +36,15 @@ public: void OnOpen(DeviceFD fd) override; void OnClose(DeviceFD fd) override; - /// Returns the allocated address of an nvmap object given its handle. - VAddr GetObjectAddress(u32 handle) const; - - /// Represents an nvmap object. - struct Object { - enum class Status { Created, Allocated }; - u32 id; - u32 size; - u32 flags; - u32 align; - u8 kind; - VAddr addr; - Status status; - u32 refcount; - u32 dma_map_addr; + enum class HandleParameterType : u32_le { + Size = 1, + Alignment = 2, + Base = 3, + Heap = 4, + Kind = 5, + IsSharedMemMapped = 6 }; - std::shared_ptr<Object> GetObject(u32 handle) const { - auto itr = handles.find(handle); - if (itr != handles.end()) { - return itr->second; - } - return {}; - } - private: /// Id to use for the next handle that is created. u32 next_handle = 0; @@ -60,9 +52,6 @@ private: /// Id to use for the next object that is created. u32 next_id = 0; - /// Mapping of currently allocated handles to the objects they represent. - std::unordered_map<u32, std::shared_ptr<Object>> handles; - struct IocCreateParams { // Input u32_le size{}; @@ -83,11 +72,11 @@ private: // Input u32_le handle{}; u32_le heap_mask{}; - u32_le flags{}; + NvCore::NvMap::Handle::Flags flags{}; u32_le align{}; u8 kind{}; INSERT_PADDING_BYTES(7); - u64_le addr{}; + u64_le address{}; }; static_assert(sizeof(IocAllocParams) == 32, "IocAllocParams has wrong size"); @@ -96,14 +85,14 @@ private: INSERT_PADDING_BYTES(4); u64_le address{}; u32_le size{}; - u32_le flags{}; + NvCore::NvMap::Handle::Flags flags{}; }; static_assert(sizeof(IocFreeParams) == 24, "IocFreeParams has wrong size"); struct IocParamParams { // Input u32_le handle{}; - u32_le param{}; + HandleParameterType param{}; // Output u32_le result{}; }; @@ -117,14 +106,15 @@ private: }; static_assert(sizeof(IocGetIdParams) == 8, "IocGetIdParams has wrong size"); - u32 CreateObject(u32 size); - NvResult IocCreate(const std::vector<u8>& input, std::vector<u8>& output); NvResult IocAlloc(const std::vector<u8>& input, std::vector<u8>& output); NvResult IocGetId(const std::vector<u8>& input, std::vector<u8>& output); NvResult IocFromId(const std::vector<u8>& input, std::vector<u8>& output); NvResult IocParam(const std::vector<u8>& input, std::vector<u8>& output); NvResult IocFree(const std::vector<u8>& input, std::vector<u8>& output); + + NvCore::Container& container; + NvCore::NvMap& file; }; } // namespace Service::Nvidia::Devices diff --git a/src/core/hle/service/nvdrv/nvdata.h b/src/core/hle/service/nvdrv/nvdata.h index 1d00394c8..0e2f47075 100644 --- a/src/core/hle/service/nvdrv/nvdata.h +++ b/src/core/hle/service/nvdrv/nvdata.h @@ -1,5 +1,6 @@ -// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once @@ -78,11 +79,15 @@ enum class NvResult : u32 { ModuleNotPresent = 0xA000E, }; +// obtained from +// https://github.com/skyline-emu/skyline/blob/nvdec-dev/app/src/main/cpp/skyline/services/nvdrv/devices/nvhost/ctrl.h#L47 enum class EventState { - Free = 0, - Registered = 1, - Waiting = 2, - Busy = 3, + Available = 0, + Waiting = 1, + Cancelling = 2, + Signalling = 3, + Signalled = 4, + Cancelled = 5, }; union Ioctl { diff --git a/src/core/hle/service/nvdrv/nvdrv.cpp b/src/core/hle/service/nvdrv/nvdrv.cpp index 756eb7453..5e7b7468f 100644 --- a/src/core/hle/service/nvdrv/nvdrv.cpp +++ b/src/core/hle/service/nvdrv/nvdrv.cpp @@ -1,5 +1,6 @@ -// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #include <utility> @@ -8,6 +9,7 @@ #include "core/hle/ipc_helpers.h" #include "core/hle/kernel/k_event.h" #include "core/hle/kernel/k_writable_event.h" +#include "core/hle/service/nvdrv/core/container.h" #include "core/hle/service/nvdrv/devices/nvdevice.h" #include "core/hle/service/nvdrv/devices/nvdisp_disp0.h" #include "core/hle/service/nvdrv/devices/nvhost_as_gpu.h" @@ -15,17 +17,31 @@ #include "core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h" #include "core/hle/service/nvdrv/devices/nvhost_gpu.h" #include "core/hle/service/nvdrv/devices/nvhost_nvdec.h" +#include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h" #include "core/hle/service/nvdrv/devices/nvhost_nvjpg.h" #include "core/hle/service/nvdrv/devices/nvhost_vic.h" #include "core/hle/service/nvdrv/devices/nvmap.h" #include "core/hle/service/nvdrv/nvdrv.h" #include "core/hle/service/nvdrv/nvdrv_interface.h" #include "core/hle/service/nvdrv/nvmemp.h" -#include "core/hle/service/nvdrv/syncpoint_manager.h" #include "core/hle/service/nvflinger/nvflinger.h" +#include "video_core/gpu.h" namespace Service::Nvidia { +EventInterface::EventInterface(Module& module_) : module{module_}, guard{}, on_signal{} {} + +EventInterface::~EventInterface() = default; + +Kernel::KEvent* EventInterface::CreateEvent(std::string name) { + Kernel::KEvent* new_event = module.service_context.CreateEvent(std::move(name)); + return new_event; +} + +void EventInterface::FreeEvent(Kernel::KEvent* event) { + module.service_context.CloseEvent(event); +} + void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger& nvflinger, Core::System& system) { auto module_ = std::make_shared<Module>(system); @@ -38,34 +54,54 @@ void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger } Module::Module(Core::System& system) - : syncpoint_manager{system.GPU()}, service_context{system, "nvdrv"} { - for (u32 i = 0; i < MaxNvEvents; i++) { - events_interface.events[i].event = - service_context.CreateEvent(fmt::format("NVDRV::NvEvent_{}", i)); - events_interface.status[i] = EventState::Free; - events_interface.registered[i] = false; - } - auto nvmap_dev = std::make_shared<Devices::nvmap>(system); - devices["/dev/nvhost-as-gpu"] = std::make_shared<Devices::nvhost_as_gpu>(system, nvmap_dev); - devices["/dev/nvhost-gpu"] = - std::make_shared<Devices::nvhost_gpu>(system, nvmap_dev, syncpoint_manager); - devices["/dev/nvhost-ctrl-gpu"] = std::make_shared<Devices::nvhost_ctrl_gpu>(system); - devices["/dev/nvmap"] = nvmap_dev; - devices["/dev/nvdisp_disp0"] = std::make_shared<Devices::nvdisp_disp0>(system, nvmap_dev); - devices["/dev/nvhost-ctrl"] = - std::make_shared<Devices::nvhost_ctrl>(system, events_interface, syncpoint_manager); - devices["/dev/nvhost-nvdec"] = - std::make_shared<Devices::nvhost_nvdec>(system, nvmap_dev, syncpoint_manager); - devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>(system); - devices["/dev/nvhost-vic"] = - std::make_shared<Devices::nvhost_vic>(system, nvmap_dev, syncpoint_manager); + : service_context{system, "nvdrv"}, events_interface{*this}, container{system.Host1x()} { + builders["/dev/nvhost-as-gpu"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvhost_as_gpu>(system, *this, container); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvhost-gpu"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvhost_gpu>(system, events_interface, container); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvhost-ctrl-gpu"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvhost_ctrl_gpu>(system, events_interface); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvmap"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvmap>(system, container); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvdisp_disp0"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvdisp_disp0>(system, container); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvhost-ctrl"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvhost_ctrl>(system, events_interface, container); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvhost-nvdec"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvhost_nvdec>(system, container); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvhost-nvjpg"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = std::make_shared<Devices::nvhost_nvjpg>(system); + return open_files.emplace(fd, device).first; + }; + builders["/dev/nvhost-vic"] = [this, &system](DeviceFD fd) { + std::shared_ptr<Devices::nvdevice> device = + std::make_shared<Devices::nvhost_vic>(system, container); + return open_files.emplace(fd, device).first; + }; } -Module::~Module() { - for (u32 i = 0; i < MaxNvEvents; i++) { - service_context.CloseEvent(events_interface.events[i].event); - } -} +Module::~Module() {} NvResult Module::VerifyFD(DeviceFD fd) const { if (fd < 0) { @@ -82,18 +118,18 @@ NvResult Module::VerifyFD(DeviceFD fd) const { } DeviceFD Module::Open(const std::string& device_name) { - if (devices.find(device_name) == devices.end()) { + auto it = builders.find(device_name); + if (it == builders.end()) { LOG_ERROR(Service_NVDRV, "Trying to open unknown device {}", device_name); return INVALID_NVDRV_FD; } - auto device = devices[device_name]; const DeviceFD fd = next_fd++; + auto& builder = it->second; + auto device = builder(fd)->second; device->OnOpen(fd); - open_files[fd] = std::move(device); - return fd; } @@ -168,22 +204,24 @@ NvResult Module::Close(DeviceFD fd) { return NvResult::Success; } -void Module::SignalSyncpt(const u32 syncpoint_id, const u32 value) { - for (u32 i = 0; i < MaxNvEvents; i++) { - if (events_interface.assigned_syncpt[i] == syncpoint_id && - events_interface.assigned_value[i] == value) { - events_interface.LiberateEvent(i); - events_interface.events[i].event->GetWritableEvent().Signal(); - } +NvResult Module::QueryEvent(DeviceFD fd, u32 event_id, Kernel::KEvent*& event) { + if (fd < 0) { + LOG_ERROR(Service_NVDRV, "Invalid DeviceFD={}!", fd); + return NvResult::InvalidState; } -} -Kernel::KReadableEvent& Module::GetEvent(const u32 event_id) { - return events_interface.events[event_id].event->GetReadableEvent(); -} + const auto itr = open_files.find(fd); -Kernel::KWritableEvent& Module::GetEventWriteable(const u32 event_id) { - return events_interface.events[event_id].event->GetWritableEvent(); + if (itr == open_files.end()) { + LOG_ERROR(Service_NVDRV, "Could not find DeviceFD={}!", fd); + return NvResult::NotImplemented; + } + + event = itr->second->QueryEvent(event_id); + if (!event) { + return NvResult::BadParameter; + } + return NvResult::Success; } } // namespace Service::Nvidia diff --git a/src/core/hle/service/nvdrv/nvdrv.h b/src/core/hle/service/nvdrv/nvdrv.h index c929e5106..146d046a9 100644 --- a/src/core/hle/service/nvdrv/nvdrv.h +++ b/src/core/hle/service/nvdrv/nvdrv.h @@ -1,16 +1,20 @@ -// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once +#include <functional> +#include <list> #include <memory> +#include <string> #include <unordered_map> #include <vector> #include "common/common_types.h" #include "core/hle/service/kernel_helpers.h" +#include "core/hle/service/nvdrv/core/container.h" #include "core/hle/service/nvdrv/nvdata.h" -#include "core/hle/service/nvdrv/syncpoint_manager.h" #include "core/hle/service/nvflinger/ui/fence.h" #include "core/hle/service/service.h" @@ -28,81 +32,31 @@ class NVFlinger; namespace Service::Nvidia { +namespace NvCore { +class Container; class SyncpointManager; +} // namespace NvCore namespace Devices { class nvdevice; -} +class nvhost_ctrl; +} // namespace Devices -/// Represents an Nvidia event -struct NvEvent { - Kernel::KEvent* event{}; - NvFence fence{}; -}; +class Module; -struct EventInterface { - // Mask representing currently busy events - u64 events_mask{}; - // Each kernel event associated to an NV event - std::array<NvEvent, MaxNvEvents> events; - // The status of the current NVEvent - std::array<EventState, MaxNvEvents> status{}; - // Tells if an NVEvent is registered or not - std::array<bool, MaxNvEvents> registered{}; - // Tells the NVEvent that it has failed. - std::array<bool, MaxNvEvents> failed{}; - // When an NVEvent is waiting on GPU interrupt, this is the sync_point - // associated with it. - std::array<u32, MaxNvEvents> assigned_syncpt{}; - // This is the value of the GPU interrupt for which the NVEvent is waiting - // for. - std::array<u32, MaxNvEvents> assigned_value{}; - // Constant to denote an unasigned syncpoint. - static constexpr u32 unassigned_syncpt = 0xFFFFFFFF; - std::optional<u32> GetFreeEvent() const { - u64 mask = events_mask; - for (u32 i = 0; i < MaxNvEvents; i++) { - const bool is_free = (mask & 0x1) == 0; - if (is_free) { - if (status[i] == EventState::Registered || status[i] == EventState::Free) { - return {i}; - } - } - mask = mask >> 1; - } - return std::nullopt; - } - void SetEventStatus(const u32 event_id, EventState new_status) { - EventState old_status = status[event_id]; - if (old_status == new_status) { - return; - } - status[event_id] = new_status; - if (new_status == EventState::Registered) { - registered[event_id] = true; - } - if (new_status == EventState::Waiting || new_status == EventState::Busy) { - events_mask |= (1ULL << event_id); - } - } - void RegisterEvent(const u32 event_id) { - registered[event_id] = true; - if (status[event_id] == EventState::Free) { - status[event_id] = EventState::Registered; - } - } - void UnregisterEvent(const u32 event_id) { - registered[event_id] = false; - if (status[event_id] == EventState::Registered) { - status[event_id] = EventState::Free; - } - } - void LiberateEvent(const u32 event_id) { - status[event_id] = registered[event_id] ? EventState::Registered : EventState::Free; - events_mask &= ~(1ULL << event_id); - assigned_syncpt[event_id] = unassigned_syncpt; - assigned_value[event_id] = 0; - } +class EventInterface { +public: + explicit EventInterface(Module& module_); + ~EventInterface(); + + Kernel::KEvent* CreateEvent(std::string name); + + void FreeEvent(Kernel::KEvent* event); + +private: + Module& module; + std::mutex guard; + std::list<Devices::nvhost_ctrl*> on_signal; }; class Module final { @@ -112,9 +66,9 @@ public: /// Returns a pointer to one of the available devices, identified by its name. template <typename T> - std::shared_ptr<T> GetDevice(const std::string& name) { - auto itr = devices.find(name); - if (itr == devices.end()) + std::shared_ptr<T> GetDevice(DeviceFD fd) { + auto itr = open_files.find(fd); + if (itr == open_files.end()) return nullptr; return std::static_pointer_cast<T>(itr->second); } @@ -137,28 +91,27 @@ public: /// Closes a device file descriptor and returns operation success. NvResult Close(DeviceFD fd); - void SignalSyncpt(const u32 syncpoint_id, const u32 value); - - Kernel::KReadableEvent& GetEvent(u32 event_id); - - Kernel::KWritableEvent& GetEventWriteable(u32 event_id); + NvResult QueryEvent(DeviceFD fd, u32 event_id, Kernel::KEvent*& event); private: - /// Manages syncpoints on the host - SyncpointManager syncpoint_manager; + friend class EventInterface; + friend class Service::NVFlinger::NVFlinger; /// Id to use for the next open file descriptor. DeviceFD next_fd = 1; + using FilesContainerType = std::unordered_map<DeviceFD, std::shared_ptr<Devices::nvdevice>>; /// Mapping of file descriptors to the devices they reference. - std::unordered_map<DeviceFD, std::shared_ptr<Devices::nvdevice>> open_files; + FilesContainerType open_files; - /// Mapping of device node names to their implementation. - std::unordered_map<std::string, std::shared_ptr<Devices::nvdevice>> devices; + KernelHelpers::ServiceContext service_context; EventInterface events_interface; - KernelHelpers::ServiceContext service_context; + /// Manages syncpoints on the host + NvCore::Container container; + + std::unordered_map<std::string, std::function<FilesContainerType::iterator(DeviceFD)>> builders; }; /// Registers all NVDRV services with the specified service manager. diff --git a/src/core/hle/service/nvdrv/nvdrv_interface.cpp b/src/core/hle/service/nvdrv/nvdrv_interface.cpp index b5a980384..edbdfee43 100644 --- a/src/core/hle/service/nvdrv/nvdrv_interface.cpp +++ b/src/core/hle/service/nvdrv/nvdrv_interface.cpp @@ -1,10 +1,12 @@ -// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #include <cinttypes> #include "common/logging/log.h" #include "core/core.h" #include "core/hle/ipc_helpers.h" +#include "core/hle/kernel/k_event.h" #include "core/hle/kernel/k_readable_event.h" #include "core/hle/service/nvdrv/nvdata.h" #include "core/hle/service/nvdrv/nvdrv.h" @@ -12,10 +14,6 @@ namespace Service::Nvidia { -void NVDRV::SignalGPUInterruptSyncpt(const u32 syncpoint_id, const u32 value) { - nvdrv->SignalSyncpt(syncpoint_id, value); -} - void NVDRV::Open(Kernel::HLERequestContext& ctx) { LOG_DEBUG(Service_NVDRV, "called"); IPC::ResponseBuilder rb{ctx, 4}; @@ -164,8 +162,7 @@ void NVDRV::Initialize(Kernel::HLERequestContext& ctx) { void NVDRV::QueryEvent(Kernel::HLERequestContext& ctx) { IPC::RequestParser rp{ctx}; const auto fd = rp.Pop<DeviceFD>(); - const auto event_id = rp.Pop<u32>() & 0x00FF; - LOG_WARNING(Service_NVDRV, "(STUBBED) called, fd={:X}, event_id={:X}", fd, event_id); + const auto event_id = rp.Pop<u32>(); if (!is_initialized) { ServiceError(ctx, NvResult::NotInitialized); @@ -173,24 +170,20 @@ void NVDRV::QueryEvent(Kernel::HLERequestContext& ctx) { return; } - const auto nv_result = nvdrv->VerifyFD(fd); - if (nv_result != NvResult::Success) { - LOG_ERROR(Service_NVDRV, "Invalid FD specified DeviceFD={}!", fd); - ServiceError(ctx, nv_result); - return; - } + Kernel::KEvent* event = nullptr; + NvResult result = nvdrv->QueryEvent(fd, event_id, event); - if (event_id < MaxNvEvents) { + if (result == NvResult::Success) { IPC::ResponseBuilder rb{ctx, 3, 1}; rb.Push(ResultSuccess); - auto& event = nvdrv->GetEvent(event_id); - event.Clear(); - rb.PushCopyObjects(event); + auto& readable_event = event->GetReadableEvent(); + rb.PushCopyObjects(readable_event); rb.PushEnum(NvResult::Success); } else { + LOG_ERROR(Service_NVDRV, "Invalid event request!"); IPC::ResponseBuilder rb{ctx, 3}; rb.Push(ResultSuccess); - rb.PushEnum(NvResult::BadParameter); + rb.PushEnum(result); } } diff --git a/src/core/hle/service/nvdrv/nvdrv_interface.h b/src/core/hle/service/nvdrv/nvdrv_interface.h index cbd37b52b..cd58a4f35 100644 --- a/src/core/hle/service/nvdrv/nvdrv_interface.h +++ b/src/core/hle/service/nvdrv/nvdrv_interface.h @@ -18,8 +18,6 @@ public: explicit NVDRV(Core::System& system_, std::shared_ptr<Module> nvdrv_, const char* name); ~NVDRV() override; - void SignalGPUInterruptSyncpt(u32 syncpoint_id, u32 value); - private: void Open(Kernel::HLERequestContext& ctx); void Ioctl1(Kernel::HLERequestContext& ctx); diff --git a/src/core/hle/service/nvdrv/syncpoint_manager.cpp b/src/core/hle/service/nvdrv/syncpoint_manager.cpp deleted file mode 100644 index a6fa943e8..000000000 --- a/src/core/hle/service/nvdrv/syncpoint_manager.cpp +++ /dev/null @@ -1,38 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#include "common/assert.h" -#include "core/hle/service/nvdrv/syncpoint_manager.h" -#include "video_core/gpu.h" - -namespace Service::Nvidia { - -SyncpointManager::SyncpointManager(Tegra::GPU& gpu_) : gpu{gpu_} {} - -SyncpointManager::~SyncpointManager() = default; - -u32 SyncpointManager::RefreshSyncpoint(u32 syncpoint_id) { - syncpoints[syncpoint_id].min = gpu.GetSyncpointValue(syncpoint_id); - return GetSyncpointMin(syncpoint_id); -} - -u32 SyncpointManager::AllocateSyncpoint() { - for (u32 syncpoint_id = 1; syncpoint_id < MaxSyncPoints; syncpoint_id++) { - if (!syncpoints[syncpoint_id].is_allocated) { - syncpoints[syncpoint_id].is_allocated = true; - return syncpoint_id; - } - } - ASSERT_MSG(false, "No more available syncpoints!"); - return {}; -} - -u32 SyncpointManager::IncreaseSyncpoint(u32 syncpoint_id, u32 value) { - for (u32 index = 0; index < value; ++index) { - syncpoints[syncpoint_id].max.fetch_add(1, std::memory_order_relaxed); - } - - return GetSyncpointMax(syncpoint_id); -} - -} // namespace Service::Nvidia diff --git a/src/core/hle/service/nvdrv/syncpoint_manager.h b/src/core/hle/service/nvdrv/syncpoint_manager.h deleted file mode 100644 index 7f080f76e..000000000 --- a/src/core/hle/service/nvdrv/syncpoint_manager.h +++ /dev/null @@ -1,84 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#pragma once - -#include <array> -#include <atomic> - -#include "common/common_types.h" -#include "core/hle/service/nvdrv/nvdata.h" - -namespace Tegra { -class GPU; -} - -namespace Service::Nvidia { - -class SyncpointManager final { -public: - explicit SyncpointManager(Tegra::GPU& gpu_); - ~SyncpointManager(); - - /** - * Returns true if the specified syncpoint is expired for the given value. - * @param syncpoint_id Syncpoint ID to check. - * @param value Value to check against the specified syncpoint. - * @returns True if the specified syncpoint is expired for the given value, otherwise False. - */ - bool IsSyncpointExpired(u32 syncpoint_id, u32 value) const { - return (GetSyncpointMax(syncpoint_id) - value) >= (GetSyncpointMin(syncpoint_id) - value); - } - - /** - * Gets the lower bound for the specified syncpoint. - * @param syncpoint_id Syncpoint ID to get the lower bound for. - * @returns The lower bound for the specified syncpoint. - */ - u32 GetSyncpointMin(u32 syncpoint_id) const { - return syncpoints.at(syncpoint_id).min.load(std::memory_order_relaxed); - } - - /** - * Gets the uper bound for the specified syncpoint. - * @param syncpoint_id Syncpoint ID to get the upper bound for. - * @returns The upper bound for the specified syncpoint. - */ - u32 GetSyncpointMax(u32 syncpoint_id) const { - return syncpoints.at(syncpoint_id).max.load(std::memory_order_relaxed); - } - - /** - * Refreshes the minimum value for the specified syncpoint. - * @param syncpoint_id Syncpoint ID to be refreshed. - * @returns The new syncpoint minimum value. - */ - u32 RefreshSyncpoint(u32 syncpoint_id); - - /** - * Allocates a new syncoint. - * @returns The syncpoint ID for the newly allocated syncpoint. - */ - u32 AllocateSyncpoint(); - - /** - * Increases the maximum value for the specified syncpoint. - * @param syncpoint_id Syncpoint ID to be increased. - * @param value Value to increase the specified syncpoint by. - * @returns The new syncpoint maximum value. - */ - u32 IncreaseSyncpoint(u32 syncpoint_id, u32 value); - -private: - struct Syncpoint { - std::atomic<u32> min; - std::atomic<u32> max; - std::atomic<bool> is_allocated; - }; - - std::array<Syncpoint, MaxSyncPoints> syncpoints{}; - - Tegra::GPU& gpu; -}; - -} // namespace Service::Nvidia diff --git a/src/core/hle/service/nvflinger/buffer_queue_consumer.cpp b/src/core/hle/service/nvflinger/buffer_queue_consumer.cpp index 4b3d5efd6..1ce67c771 100644 --- a/src/core/hle/service/nvflinger/buffer_queue_consumer.cpp +++ b/src/core/hle/service/nvflinger/buffer_queue_consumer.cpp @@ -5,15 +5,18 @@ // https://cs.android.com/android/platform/superproject/+/android-5.1.1_r38:frameworks/native/libs/gui/BufferQueueConsumer.cpp #include "common/logging/log.h" +#include "core/hle/service/nvdrv/core/nvmap.h" #include "core/hle/service/nvflinger/buffer_item.h" #include "core/hle/service/nvflinger/buffer_queue_consumer.h" #include "core/hle/service/nvflinger/buffer_queue_core.h" #include "core/hle/service/nvflinger/producer_listener.h" +#include "core/hle/service/nvflinger/ui/graphic_buffer.h" namespace Service::android { -BufferQueueConsumer::BufferQueueConsumer(std::shared_ptr<BufferQueueCore> core_) - : core{std::move(core_)}, slots{core->slots} {} +BufferQueueConsumer::BufferQueueConsumer(std::shared_ptr<BufferQueueCore> core_, + Service::Nvidia::NvCore::NvMap& nvmap_) + : core{std::move(core_)}, slots{core->slots}, nvmap(nvmap_) {} BufferQueueConsumer::~BufferQueueConsumer() = default; @@ -133,6 +136,8 @@ Status BufferQueueConsumer::ReleaseBuffer(s32 slot, u64 frame_number, const Fenc slots[slot].buffer_state = BufferState::Free; + nvmap.FreeHandle(slots[slot].graphic_buffer->BufferId(), true); + listener = core->connected_producer_listener; LOG_DEBUG(Service_NVFlinger, "releasing slot {}", slot); diff --git a/src/core/hle/service/nvflinger/buffer_queue_consumer.h b/src/core/hle/service/nvflinger/buffer_queue_consumer.h index b598c314f..4ec06ca13 100644 --- a/src/core/hle/service/nvflinger/buffer_queue_consumer.h +++ b/src/core/hle/service/nvflinger/buffer_queue_consumer.h @@ -13,6 +13,10 @@ #include "core/hle/service/nvflinger/buffer_queue_defs.h" #include "core/hle/service/nvflinger/status.h" +namespace Service::Nvidia::NvCore { +class NvMap; +} // namespace Service::Nvidia::NvCore + namespace Service::android { class BufferItem; @@ -21,7 +25,8 @@ class IConsumerListener; class BufferQueueConsumer final { public: - explicit BufferQueueConsumer(std::shared_ptr<BufferQueueCore> core_); + explicit BufferQueueConsumer(std::shared_ptr<BufferQueueCore> core_, + Service::Nvidia::NvCore::NvMap& nvmap_); ~BufferQueueConsumer(); Status AcquireBuffer(BufferItem* out_buffer, std::chrono::nanoseconds expected_present); @@ -32,6 +37,7 @@ public: private: std::shared_ptr<BufferQueueCore> core; BufferQueueDefs::SlotsType& slots; + Service::Nvidia::NvCore::NvMap& nvmap; }; } // namespace Service::android diff --git a/src/core/hle/service/nvflinger/buffer_queue_producer.cpp b/src/core/hle/service/nvflinger/buffer_queue_producer.cpp index 337431488..d4ab23a10 100644 --- a/src/core/hle/service/nvflinger/buffer_queue_producer.cpp +++ b/src/core/hle/service/nvflinger/buffer_queue_producer.cpp @@ -14,7 +14,7 @@ #include "core/hle/kernel/k_writable_event.h" #include "core/hle/kernel/kernel.h" #include "core/hle/service/kernel_helpers.h" -#include "core/hle/service/nvdrv/nvdrv.h" +#include "core/hle/service/nvdrv/core/nvmap.h" #include "core/hle/service/nvflinger/buffer_queue_core.h" #include "core/hle/service/nvflinger/buffer_queue_producer.h" #include "core/hle/service/nvflinger/consumer_listener.h" @@ -26,8 +26,10 @@ namespace Service::android { BufferQueueProducer::BufferQueueProducer(Service::KernelHelpers::ServiceContext& service_context_, - std::shared_ptr<BufferQueueCore> buffer_queue_core_) - : service_context{service_context_}, core{std::move(buffer_queue_core_)}, slots(core->slots) { + std::shared_ptr<BufferQueueCore> buffer_queue_core_, + Service::Nvidia::NvCore::NvMap& nvmap_) + : service_context{service_context_}, core{std::move(buffer_queue_core_)}, slots(core->slots), + nvmap(nvmap_) { buffer_wait_event = service_context.CreateEvent("BufferQueue:WaitEvent"); } @@ -530,6 +532,8 @@ Status BufferQueueProducer::QueueBuffer(s32 slot, const QueueBufferInput& input, item.is_droppable = core->dequeue_buffer_cannot_block || async; item.swap_interval = swap_interval; + nvmap.DuplicateHandle(item.graphic_buffer->BufferId(), true); + sticky_transform = sticky_transform_; if (core->queue.empty()) { diff --git a/src/core/hle/service/nvflinger/buffer_queue_producer.h b/src/core/hle/service/nvflinger/buffer_queue_producer.h index 42d4722dc..0ba03a568 100644 --- a/src/core/hle/service/nvflinger/buffer_queue_producer.h +++ b/src/core/hle/service/nvflinger/buffer_queue_producer.h @@ -31,6 +31,10 @@ namespace Service::KernelHelpers { class ServiceContext; } // namespace Service::KernelHelpers +namespace Service::Nvidia::NvCore { +class NvMap; +} // namespace Service::Nvidia::NvCore + namespace Service::android { class BufferQueueCore; @@ -39,7 +43,8 @@ class IProducerListener; class BufferQueueProducer final : public IBinder { public: explicit BufferQueueProducer(Service::KernelHelpers::ServiceContext& service_context_, - std::shared_ptr<BufferQueueCore> buffer_queue_core_); + std::shared_ptr<BufferQueueCore> buffer_queue_core_, + Service::Nvidia::NvCore::NvMap& nvmap_); ~BufferQueueProducer(); void Transact(Kernel::HLERequestContext& ctx, android::TransactionId code, u32 flags) override; @@ -78,6 +83,8 @@ private: s32 next_callback_ticket{}; s32 current_callback_ticket{}; std::condition_variable_any callback_condition; + + Service::Nvidia::NvCore::NvMap& nvmap; }; } // namespace Service::android diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp index 93057e800..aa14d2cbc 100644 --- a/src/core/hle/service/nvflinger/nvflinger.cpp +++ b/src/core/hle/service/nvflinger/nvflinger.cpp @@ -24,6 +24,8 @@ #include "core/hle/service/vi/layer/vi_layer.h" #include "core/hle/service/vi/vi_results.h" #include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/syncpoint_manager.h" namespace Service::NVFlinger { @@ -31,7 +33,7 @@ constexpr auto frame_ns = std::chrono::nanoseconds{1000000000 / 60}; void NVFlinger::SplitVSync(std::stop_token stop_token) { system.RegisterHostThread(); - std::string name = "yuzu:VSyncThread"; + std::string name = "VSyncThread"; MicroProfileOnThreadCreate(name.c_str()); // Cleanup @@ -105,10 +107,15 @@ NVFlinger::~NVFlinger() { display.GetLayer(layer).Core().NotifyShutdown(); } } + + if (nvdrv) { + nvdrv->Close(disp_fd); + } } void NVFlinger::SetNVDrvInstance(std::shared_ptr<Nvidia::Module> instance) { nvdrv = std::move(instance); + disp_fd = nvdrv->Open("/dev/nvdisp_disp0"); } std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) { @@ -142,7 +149,7 @@ std::optional<u64> NVFlinger::CreateLayer(u64 display_id) { void NVFlinger::CreateLayerAtId(VI::Display& display, u64 layer_id) { const auto buffer_id = next_buffer_queue_id++; - display.CreateLayer(layer_id, buffer_id); + display.CreateLayer(layer_id, buffer_id, nvdrv->container); } void NVFlinger::CloseLayer(u64 layer_id) { @@ -262,30 +269,24 @@ void NVFlinger::Compose() { return; // We are likely shutting down } - auto& gpu = system.GPU(); - const auto& multi_fence = buffer.fence; - guard->unlock(); - for (u32 fence_id = 0; fence_id < multi_fence.num_fences; fence_id++) { - const auto& fence = multi_fence.fences[fence_id]; - gpu.WaitFence(fence.id, fence.value); - } - guard->lock(); - - MicroProfileFlip(); - // Now send the buffer to the GPU for drawing. // TODO(Subv): Support more than just disp0. The display device selection is probably based // on which display we're drawing (Default, Internal, External, etc) - auto nvdisp = nvdrv->GetDevice<Nvidia::Devices::nvdisp_disp0>("/dev/nvdisp_disp0"); + auto nvdisp = nvdrv->GetDevice<Nvidia::Devices::nvdisp_disp0>(disp_fd); ASSERT(nvdisp); + guard->unlock(); Common::Rectangle<int> crop_rect{ static_cast<int>(buffer.crop.Left()), static_cast<int>(buffer.crop.Top()), static_cast<int>(buffer.crop.Right()), static_cast<int>(buffer.crop.Bottom())}; nvdisp->flip(igbp_buffer.BufferId(), igbp_buffer.Offset(), igbp_buffer.ExternalFormat(), igbp_buffer.Width(), igbp_buffer.Height(), igbp_buffer.Stride(), - static_cast<android::BufferTransformFlags>(buffer.transform), crop_rect); + static_cast<android::BufferTransformFlags>(buffer.transform), crop_rect, + buffer.fence.fences, buffer.fence.num_fences); + + MicroProfileFlip(); + guard->lock(); swap_interval = buffer.swap_interval; diff --git a/src/core/hle/service/nvflinger/nvflinger.h b/src/core/hle/service/nvflinger/nvflinger.h index 3bbe5d92b..b62615de2 100644 --- a/src/core/hle/service/nvflinger/nvflinger.h +++ b/src/core/hle/service/nvflinger/nvflinger.h @@ -116,6 +116,7 @@ private: void SplitVSync(std::stop_token stop_token); std::shared_ptr<Nvidia::Module> nvdrv; + s32 disp_fd; std::list<VI::Display> displays; diff --git a/src/core/hle/service/vi/display/vi_display.cpp b/src/core/hle/service/vi/display/vi_display.cpp index aa49aa775..288aafaaf 100644 --- a/src/core/hle/service/vi/display/vi_display.cpp +++ b/src/core/hle/service/vi/display/vi_display.cpp @@ -12,6 +12,7 @@ #include "core/hle/kernel/k_readable_event.h" #include "core/hle/kernel/k_writable_event.h" #include "core/hle/service/kernel_helpers.h" +#include "core/hle/service/nvdrv/core/container.h" #include "core/hle/service/nvflinger/buffer_item_consumer.h" #include "core/hle/service/nvflinger/buffer_queue_consumer.h" #include "core/hle/service/nvflinger/buffer_queue_core.h" @@ -29,11 +30,13 @@ struct BufferQueue { std::unique_ptr<android::BufferQueueConsumer> consumer; }; -static BufferQueue CreateBufferQueue(KernelHelpers::ServiceContext& service_context) { +static BufferQueue CreateBufferQueue(KernelHelpers::ServiceContext& service_context, + Service::Nvidia::NvCore::NvMap& nvmap) { auto buffer_queue_core = std::make_shared<android::BufferQueueCore>(); - return {buffer_queue_core, - std::make_unique<android::BufferQueueProducer>(service_context, buffer_queue_core), - std::make_unique<android::BufferQueueConsumer>(buffer_queue_core)}; + return { + buffer_queue_core, + std::make_unique<android::BufferQueueProducer>(service_context, buffer_queue_core, nvmap), + std::make_unique<android::BufferQueueConsumer>(buffer_queue_core, nvmap)}; } Display::Display(u64 id, std::string name_, @@ -74,10 +77,11 @@ void Display::SignalVSyncEvent() { vsync_event->GetWritableEvent().Signal(); } -void Display::CreateLayer(u64 layer_id, u32 binder_id) { +void Display::CreateLayer(u64 layer_id, u32 binder_id, + Service::Nvidia::NvCore::Container& nv_core) { ASSERT_MSG(layers.empty(), "Only one layer is supported per display at the moment"); - auto [core, producer, consumer] = CreateBufferQueue(service_context); + auto [core, producer, consumer] = CreateBufferQueue(service_context, nv_core.GetNvMapFile()); auto buffer_item_consumer = std::make_shared<android::BufferItemConsumer>(std::move(consumer)); buffer_item_consumer->Connect(false); diff --git a/src/core/hle/service/vi/display/vi_display.h b/src/core/hle/service/vi/display/vi_display.h index 8dbb0ef80..33d5f398c 100644 --- a/src/core/hle/service/vi/display/vi_display.h +++ b/src/core/hle/service/vi/display/vi_display.h @@ -27,6 +27,11 @@ namespace Service::NVFlinger { class HosBinderDriverServer; } +namespace Service::Nvidia::NvCore { +class Container; +class NvMap; +} // namespace Service::Nvidia::NvCore + namespace Service::VI { class Layer; @@ -93,7 +98,7 @@ public: /// @param layer_id The ID to assign to the created layer. /// @param binder_id The ID assigned to the buffer queue. /// - void CreateLayer(u64 layer_id, u32 binder_id); + void CreateLayer(u64 layer_id, u32 binder_id, Service::Nvidia::NvCore::Container& core); /// Closes and removes a layer from this display with the given ID. /// diff --git a/src/core/hle/service/vi/vi.cpp b/src/core/hle/service/vi/vi.cpp index f083811ec..9c917cacf 100644 --- a/src/core/hle/service/vi/vi.cpp +++ b/src/core/hle/service/vi/vi.cpp @@ -58,6 +58,7 @@ static_assert(sizeof(DisplayInfo) == 0x60, "DisplayInfo has wrong size"); class NativeWindow final { public: constexpr explicit NativeWindow(u32 id_) : id{id_} {} + constexpr explicit NativeWindow(const NativeWindow& other) = default; private: const u32 magic = 2; diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 34ad7cadd..2ac792566 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -551,6 +551,11 @@ struct Memory::Impl { []() {}); } + [[nodiscard]] u8* GetPointerSilent(const VAddr vaddr) const { + return GetPointerImpl( + vaddr, []() {}, []() {}); + } + /** * Reads a particular data type out of memory at the given virtual address. * @@ -686,6 +691,10 @@ u8* Memory::GetPointer(VAddr vaddr) { return impl->GetPointer(vaddr); } +u8* Memory::GetPointerSilent(VAddr vaddr) { + return impl->GetPointerSilent(vaddr); +} + const u8* Memory::GetPointer(VAddr vaddr) const { return impl->GetPointer(vaddr); } diff --git a/src/core/memory.h b/src/core/memory.h index a11ff8766..81eac448b 100644 --- a/src/core/memory.h +++ b/src/core/memory.h @@ -114,6 +114,7 @@ public: * If the address is not valid, nullptr will be returned. */ u8* GetPointer(VAddr vaddr); + u8* GetPointerSilent(VAddr vaddr); template <typename T> T* GetPointer(VAddr vaddr) { diff --git a/src/dedicated_room/CMakeLists.txt b/src/dedicated_room/CMakeLists.txt index 1efdbc1f7..2d9731f19 100644 --- a/src/dedicated_room/CMakeLists.txt +++ b/src/dedicated_room/CMakeLists.txt @@ -23,5 +23,5 @@ endif() target_link_libraries(yuzu-room PRIVATE ${PLATFORM_LIBRARIES} Threads::Threads) if(UNIX AND NOT APPLE) - install(TARGETS yuzu-room RUNTIME DESTINATION "${CMAKE_INSTALL_PREFIX}/bin") + install(TARGETS yuzu-room) endif() diff --git a/src/input_common/drivers/gc_adapter.cpp b/src/input_common/drivers/gc_adapter.cpp index 27a0ffb0d..f4dd24e7d 100644 --- a/src/input_common/drivers/gc_adapter.cpp +++ b/src/input_common/drivers/gc_adapter.cpp @@ -90,7 +90,7 @@ GCAdapter::~GCAdapter() { void GCAdapter::AdapterInputThread(std::stop_token stop_token) { LOG_DEBUG(Input, "Input thread started"); - Common::SetCurrentThreadName("yuzu:input:GCAdapter"); + Common::SetCurrentThreadName("GCAdapter"); s32 payload_size{}; AdapterPayload adapter_payload{}; @@ -214,7 +214,7 @@ void GCAdapter::UpdateStateAxes(std::size_t port, const AdapterPayload& adapter_ } void GCAdapter::AdapterScanThread(std::stop_token stop_token) { - Common::SetCurrentThreadName("yuzu:input:ScanGCAdapter"); + Common::SetCurrentThreadName("ScanGCAdapter"); usb_adapter_handle = nullptr; pads = {}; while (!stop_token.stop_requested() && !Setup()) { diff --git a/src/input_common/drivers/mouse.cpp b/src/input_common/drivers/mouse.cpp index 4909fa8d7..98c3157a8 100644 --- a/src/input_common/drivers/mouse.cpp +++ b/src/input_common/drivers/mouse.cpp @@ -37,7 +37,7 @@ Mouse::Mouse(std::string input_engine_) : InputEngine(std::move(input_engine_)) } void Mouse::UpdateThread(std::stop_token stop_token) { - Common::SetCurrentThreadName("yuzu:input:Mouse"); + Common::SetCurrentThreadName("Mouse"); constexpr int update_time = 10; while (!stop_token.stop_requested()) { if (Settings::values.mouse_panning && !Settings::values.mouse_enabled) { diff --git a/src/input_common/drivers/sdl_driver.cpp b/src/input_common/drivers/sdl_driver.cpp index 5cc1ccbd9..b72e4b397 100644 --- a/src/input_common/drivers/sdl_driver.cpp +++ b/src/input_common/drivers/sdl_driver.cpp @@ -436,7 +436,7 @@ SDLDriver::SDLDriver(std::string input_engine_) : InputEngine(std::move(input_en initialized = true; if (start_thread) { poll_thread = std::thread([this] { - Common::SetCurrentThreadName("yuzu:input:SDL"); + Common::SetCurrentThreadName("SDL_MainLoop"); using namespace std::chrono_literals; while (initialized) { SDL_PumpEvents(); @@ -444,7 +444,7 @@ SDLDriver::SDLDriver(std::string input_engine_) : InputEngine(std::move(input_en } }); vibration_thread = std::thread([this] { - Common::SetCurrentThreadName("yuzu:input:SDL_Vibration"); + Common::SetCurrentThreadName("SDL_Vibration"); using namespace std::chrono_literals; while (initialized) { SendVibrations(); diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp index b5c08d611..7e8f37563 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp @@ -13,9 +13,6 @@ namespace Shader::Backend::GLASM { namespace { void GetCbuf(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, ScalarU32 offset, std::string_view size) { - if (!binding.IsImmediate()) { - throw NotImplementedException("Indirect constant buffer loading"); - } const Register ret{ctx.reg_alloc.Define(inst)}; if (offset.type == Type::U32) { // Avoid reading arrays out of bounds, matching hardware's behavior @@ -24,7 +21,27 @@ void GetCbuf(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, ScalarU return; } } - ctx.Add("LDC.{} {},c{}[{}];", size, ret, binding.U32(), offset); + + if (binding.IsImmediate()) { + ctx.Add("LDC.{} {},c{}[{}];", size, ret, binding.U32(), offset); + return; + } + + const ScalarU32 idx{ctx.reg_alloc.Consume(binding)}; + for (u32 i = 0; i < Info::MAX_INDIRECT_CBUFS; i++) { + ctx.Add("SEQ.S.CC RC.x,{},{};" + "IF NE.x;" + "LDC.{} {},c{}[{}];", + idx, i, size, ret, i, offset); + + if (i != Info::MAX_INDIRECT_CBUFS - 1) { + ctx.Add("ELSE;"); + } + } + + for (u32 i = 0; i < Info::MAX_INDIRECT_CBUFS; i++) { + ctx.Add("ENDIF;"); + } } bool IsInputArray(Stage stage) { diff --git a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp index 578bc8c1b..ce42475d4 100644 --- a/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp +++ b/src/shader_recompiler/frontend/maxwell/structured_control_flow.cpp @@ -964,9 +964,9 @@ private: demote_endif_node.type = Type::EndIf; demote_endif_node.data.end_if.merge = return_block_it->data.block; - asl.insert(return_block_it, demote_endif_node); - asl.insert(return_block_it, demote_node); - asl.insert(return_block_it, demote_if_node); + const auto next_it_1 = asl.insert(return_block_it, demote_endif_node); + const auto next_it_2 = asl.insert(next_it_1, demote_node); + asl.insert(next_it_2, demote_if_node); } ObjectPool<Statement>& stmt_pool; diff --git a/src/shader_recompiler/ir_opt/texture_pass.cpp b/src/shader_recompiler/ir_opt/texture_pass.cpp index 597112ba4..e8be58357 100644 --- a/src/shader_recompiler/ir_opt/texture_pass.cpp +++ b/src/shader_recompiler/ir_opt/texture_pass.cpp @@ -19,8 +19,10 @@ namespace { struct ConstBufferAddr { u32 index; u32 offset; + u32 shift_left; u32 secondary_index; u32 secondary_offset; + u32 secondary_shift_left; IR::U32 dynamic_offset; u32 count; bool has_secondary; @@ -172,19 +174,41 @@ bool IsTextureInstruction(const IR::Inst& inst) { return IndexedInstruction(inst) != IR::Opcode::Void; } -std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst); +std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env); -std::optional<ConstBufferAddr> Track(const IR::Value& value) { - return IR::BreadthFirstSearch(value, TryGetConstBuffer); +std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env) { + return IR::BreadthFirstSearch( + value, [&env](const IR::Inst* inst) { return TryGetConstBuffer(inst, env); }); } -std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { +std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) { + const IR::Inst* inst = value.InstRecursive(); + if (inst->GetOpcode() != IR::Opcode::GetCbufU32) { + return std::nullopt; + } + const IR::Value index{inst->Arg(0)}; + const IR::Value offset{inst->Arg(1)}; + if (!index.IsImmediate()) { + return std::nullopt; + } + if (!offset.IsImmediate()) { + return std::nullopt; + } + const auto index_number = index.U32(); + if (index_number != 1) { + return std::nullopt; + } + const auto offset_number = offset.U32(); + return env.ReadCbufValue(index_number, offset_number); +} + +std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env) { switch (inst->GetOpcode()) { default: return std::nullopt; case IR::Opcode::BitwiseOr32: { - std::optional lhs{Track(inst->Arg(0))}; - std::optional rhs{Track(inst->Arg(1))}; + std::optional lhs{Track(inst->Arg(0), env)}; + std::optional rhs{Track(inst->Arg(1), env)}; if (!lhs || !rhs) { return std::nullopt; } @@ -194,19 +218,62 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { if (lhs->count > 1 || rhs->count > 1) { return std::nullopt; } - if (lhs->index > rhs->index || lhs->offset > rhs->offset) { + if (lhs->shift_left > 0 || lhs->index > rhs->index || lhs->offset > rhs->offset) { std::swap(lhs, rhs); } return ConstBufferAddr{ .index = lhs->index, .offset = lhs->offset, + .shift_left = lhs->shift_left, .secondary_index = rhs->index, .secondary_offset = rhs->offset, + .secondary_shift_left = rhs->shift_left, .dynamic_offset = {}, .count = 1, .has_secondary = true, }; } + case IR::Opcode::ShiftLeftLogical32: { + const IR::Value shift{inst->Arg(1)}; + if (!shift.IsImmediate()) { + return std::nullopt; + } + std::optional lhs{Track(inst->Arg(0), env)}; + if (lhs) { + lhs->shift_left = shift.U32(); + } + return lhs; + break; + } + case IR::Opcode::BitwiseAnd32: { + IR::Value op1{inst->Arg(0)}; + IR::Value op2{inst->Arg(1)}; + if (op1.IsImmediate()) { + std::swap(op1, op2); + } + if (!op2.IsImmediate() && !op1.IsImmediate()) { + do { + auto try_index = TryGetConstant(op1, env); + if (try_index) { + op1 = op2; + op2 = IR::Value{*try_index}; + break; + } + auto try_index_2 = TryGetConstant(op2, env); + if (try_index_2) { + op2 = IR::Value{*try_index_2}; + break; + } + return std::nullopt; + } while (false); + } + std::optional lhs{Track(op1, env)}; + if (lhs) { + lhs->shift_left = static_cast<u32>(std::countr_zero(op2.U32())); + } + return lhs; + break; + } case IR::Opcode::GetCbufU32x2: case IR::Opcode::GetCbufU32: break; @@ -222,8 +289,10 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { return ConstBufferAddr{ .index = index.U32(), .offset = offset.U32(), + .shift_left = 0, .secondary_index = 0, .secondary_offset = 0, + .secondary_shift_left = 0, .dynamic_offset = {}, .count = 1, .has_secondary = false, @@ -247,8 +316,10 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { return ConstBufferAddr{ .index = index.U32(), .offset = base_offset, + .shift_left = 0, .secondary_index = 0, .secondary_offset = 0, + .secondary_shift_left = 0, .dynamic_offset = dynamic_offset, .count = 8, .has_secondary = false, @@ -258,7 +329,7 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) { ConstBufferAddr addr; if (IsBindless(inst)) { - const std::optional<ConstBufferAddr> track_addr{Track(inst.Arg(0))}; + const std::optional<ConstBufferAddr> track_addr{Track(inst.Arg(0), env)}; if (!track_addr) { throw NotImplementedException("Failed to track bindless texture constant buffer"); } @@ -267,8 +338,10 @@ TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) { addr = ConstBufferAddr{ .index = env.TextureBoundBuffer(), .offset = inst.Arg(0).U32(), + .shift_left = 0, .secondary_index = 0, .secondary_offset = 0, + .secondary_shift_left = 0, .dynamic_offset = {}, .count = 1, .has_secondary = false, @@ -284,8 +357,9 @@ TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) { TextureType ReadTextureType(Environment& env, const ConstBufferAddr& cbuf) { const u32 secondary_index{cbuf.has_secondary ? cbuf.secondary_index : cbuf.index}; const u32 secondary_offset{cbuf.has_secondary ? cbuf.secondary_offset : cbuf.offset}; - const u32 lhs_raw{env.ReadCbufValue(cbuf.index, cbuf.offset)}; - const u32 rhs_raw{env.ReadCbufValue(secondary_index, secondary_offset)}; + const u32 lhs_raw{env.ReadCbufValue(cbuf.index, cbuf.offset) << cbuf.shift_left}; + const u32 rhs_raw{env.ReadCbufValue(secondary_index, secondary_offset) + << cbuf.secondary_shift_left}; return env.ReadTextureType(lhs_raw | rhs_raw); } @@ -487,8 +561,10 @@ void TexturePass(Environment& env, IR::Program& program) { .has_secondary = cbuf.has_secondary, .cbuf_index = cbuf.index, .cbuf_offset = cbuf.offset, + .shift_left = cbuf.shift_left, .secondary_cbuf_index = cbuf.secondary_index, .secondary_cbuf_offset = cbuf.secondary_offset, + .secondary_shift_left = cbuf.secondary_shift_left, .count = cbuf.count, .size_shift = DESCRIPTOR_SIZE_SHIFT, }); @@ -499,8 +575,10 @@ void TexturePass(Environment& env, IR::Program& program) { .has_secondary = cbuf.has_secondary, .cbuf_index = cbuf.index, .cbuf_offset = cbuf.offset, + .shift_left = cbuf.shift_left, .secondary_cbuf_index = cbuf.secondary_index, .secondary_cbuf_offset = cbuf.secondary_offset, + .secondary_shift_left = cbuf.secondary_shift_left, .count = cbuf.count, .size_shift = DESCRIPTOR_SIZE_SHIFT, }); diff --git a/src/shader_recompiler/shader_info.h b/src/shader_recompiler/shader_info.h index f5690805c..cc596da4f 100644 --- a/src/shader_recompiler/shader_info.h +++ b/src/shader_recompiler/shader_info.h @@ -61,8 +61,10 @@ struct TextureBufferDescriptor { bool has_secondary; u32 cbuf_index; u32 cbuf_offset; + u32 shift_left; u32 secondary_cbuf_index; u32 secondary_cbuf_offset; + u32 secondary_shift_left; u32 count; u32 size_shift; }; @@ -85,8 +87,10 @@ struct TextureDescriptor { bool has_secondary; u32 cbuf_index; u32 cbuf_offset; + u32 shift_left; u32 secondary_cbuf_index; u32 secondary_cbuf_offset; + u32 secondary_shift_left; u32 count; u32 size_shift; }; diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index 5b3808351..40e6d1ec4 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -4,7 +4,7 @@ add_subdirectory(host_shaders) if(LIBVA_FOUND) - set_source_files_properties(command_classes/codecs/codec.cpp + set_source_files_properties(host1x/codecs/codec.cpp PROPERTIES COMPILE_DEFINITIONS LIBVA_FOUND=1) list(APPEND FFmpeg_LIBRARIES ${LIBVA_LIBRARIES}) endif() @@ -15,26 +15,14 @@ add_library(video_core STATIC buffer_cache/buffer_cache.h cdma_pusher.cpp cdma_pusher.h - command_classes/codecs/codec.cpp - command_classes/codecs/codec.h - command_classes/codecs/h264.cpp - command_classes/codecs/h264.h - command_classes/codecs/vp8.cpp - command_classes/codecs/vp8.h - command_classes/codecs/vp9.cpp - command_classes/codecs/vp9.h - command_classes/codecs/vp9_types.h - command_classes/host1x.cpp - command_classes/host1x.h - command_classes/nvdec.cpp - command_classes/nvdec.h - command_classes/nvdec_common.h - command_classes/sync_manager.cpp - command_classes/sync_manager.h - command_classes/vic.cpp - command_classes/vic.h compatible_formats.cpp compatible_formats.h + control/channel_state.cpp + control/channel_state.h + control/channel_state_cache.cpp + control/channel_state_cache.h + control/scheduler.cpp + control/scheduler.h delayed_destruction_ring.h dirty_flags.cpp dirty_flags.h @@ -54,7 +42,31 @@ add_library(video_core STATIC engines/maxwell_3d.h engines/maxwell_dma.cpp engines/maxwell_dma.h + engines/puller.cpp + engines/puller.h framebuffer_config.h + host1x/codecs/codec.cpp + host1x/codecs/codec.h + host1x/codecs/h264.cpp + host1x/codecs/h264.h + host1x/codecs/vp8.cpp + host1x/codecs/vp8.h + host1x/codecs/vp9.cpp + host1x/codecs/vp9.h + host1x/codecs/vp9_types.h + host1x/control.cpp + host1x/control.h + host1x/host1x.cpp + host1x/host1x.h + host1x/nvdec.cpp + host1x/nvdec.h + host1x/nvdec_common.h + host1x/sync_manager.cpp + host1x/sync_manager.h + host1x/syncpoint_manager.cpp + host1x/syncpoint_manager.h + host1x/vic.cpp + host1x/vic.h macro/macro.cpp macro/macro.h macro/macro_hle.cpp @@ -195,6 +207,7 @@ add_library(video_core STATIC texture_cache/render_targets.h texture_cache/samples_helper.h texture_cache/slot_vector.h + texture_cache/texture_cache.cpp texture_cache/texture_cache.h texture_cache/texture_cache_base.h texture_cache/types.h diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index f015dae56..8e26b3f95 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -5,7 +5,6 @@ #include <algorithm> #include <array> -#include <deque> #include <memory> #include <mutex> #include <numeric> @@ -23,6 +22,7 @@ #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" #include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" @@ -56,7 +56,7 @@ using UniformBufferSizes = std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFE using ComputeUniformBufferSizes = std::array<u32, NUM_COMPUTE_UNIFORM_BUFFERS>; template <typename P> -class BufferCache { +class BufferCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { // Page size for caching purposes. // This is unrelated to the CPU page size and it can be changed as it seems optimal. @@ -116,10 +116,7 @@ public: static constexpr u32 DEFAULT_SKIP_CACHE_SIZE = static_cast<u32>(4_KiB); explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - Runtime& runtime_); + Core::Memory::Memory& cpu_memory_, Runtime& runtime_); void TickFrame(); @@ -129,7 +126,7 @@ public: void DownloadMemory(VAddr cpu_addr, u64 size); - bool InlineMemory(VAddr dest_address, size_t copy_size, std::span<u8> inlined_buffer); + bool InlineMemory(VAddr dest_address, size_t copy_size, std::span<const u8> inlined_buffer); void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size); @@ -353,7 +350,7 @@ private: void NotifyBufferDeletion(); - [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr) const; + [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr, bool is_written = false) const; [[nodiscard]] TextureBufferBinding GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size, PixelFormat format); @@ -367,9 +364,6 @@ private: void ClearDownload(IntervalType subtract_interval); VideoCore::RasterizerInterface& rasterizer; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - Tegra::MemoryManager& gpu_memory; Core::Memory::Memory& cpu_memory; SlotVector<Buffer> slot_buffers; @@ -444,12 +438,8 @@ private: template <class P> BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - Runtime& runtime_) - : runtime{runtime_}, rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, - kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_} { + Core::Memory::Memory& cpu_memory_, Runtime& runtime_) + : runtime{runtime_}, rasterizer{rasterizer_}, cpu_memory{cpu_memory_} { // Ensure the first slot is used for the null buffer void(slot_buffers.insert(runtime, NullBufferParams{})); common_ranges.clear(); @@ -552,8 +542,8 @@ void BufferCache<P>::ClearDownload(IntervalType subtract_interval) { template <class P> bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { - const std::optional<VAddr> cpu_src_address = gpu_memory.GpuToCpuAddress(src_address); - const std::optional<VAddr> cpu_dest_address = gpu_memory.GpuToCpuAddress(dest_address); + const std::optional<VAddr> cpu_src_address = gpu_memory->GpuToCpuAddress(src_address); + const std::optional<VAddr> cpu_dest_address = gpu_memory->GpuToCpuAddress(dest_address); if (!cpu_src_address || !cpu_dest_address) { return false; } @@ -611,7 +601,7 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am template <class P> bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) { - const std::optional<VAddr> cpu_dst_address = gpu_memory.GpuToCpuAddress(dst_address); + const std::optional<VAddr> cpu_dst_address = gpu_memory->GpuToCpuAddress(dst_address); if (!cpu_dst_address) { return false; } @@ -635,7 +625,7 @@ bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) { template <class P> void BufferCache<P>::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) { - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); const Binding binding{ .cpu_addr = *cpu_addr, .size = size, @@ -673,7 +663,7 @@ void BufferCache<P>::BindHostGeometryBuffers(bool is_indexed) { if (is_indexed) { BindHostIndexBuffer(); } else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) { runtime.BindQuadArrayIndexBuffer(regs.vertex_buffer.first, regs.vertex_buffer.count); } @@ -733,9 +723,9 @@ void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, enabled_storage_buffers[stage] |= 1U << ssbo_index; written_storage_buffers[stage] |= (is_written ? 1U : 0U) << ssbo_index; - const auto& cbufs = maxwell3d.state.shader_stages[stage]; + const auto& cbufs = maxwell3d->state.shader_stages[stage]; const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset; - storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr); + storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr, is_written); } template <class P> @@ -770,12 +760,12 @@ void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, enabled_compute_storage_buffers |= 1U << ssbo_index; written_compute_storage_buffers |= (is_written ? 1U : 0U) << ssbo_index; - const auto& launch_desc = kepler_compute.launch_description; + const auto& launch_desc = kepler_compute->launch_description; ASSERT(((launch_desc.const_buffer_enable_mask >> cbuf_index) & 1) != 0); const auto& cbufs = launch_desc.const_buffer_config; const GPUVAddr ssbo_addr = cbufs[cbuf_index].Address() + cbuf_offset; - compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr); + compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr, is_written); } template <class P> @@ -836,6 +826,19 @@ void BufferCache<P>::CommitAsyncFlushesHigh() { const bool is_accuracy_normal = Settings::values.gpu_accuracy.GetValue() == Settings::GPUAccuracy::Normal; + auto it = committed_ranges.begin(); + while (it != committed_ranges.end()) { + auto& current_intervals = *it; + auto next_it = std::next(it); + while (next_it != committed_ranges.end()) { + for (auto& interval : *next_it) { + current_intervals.subtract(interval); + } + next_it++; + } + it++; + } + boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads; u64 total_size_bytes = 0; u64 largest_copy = 0; @@ -991,19 +994,19 @@ void BufferCache<P>::BindHostIndexBuffer() { const u32 size = index_buffer.size; SynchronizeBuffer(buffer, index_buffer.cpu_addr, size); if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { - const u32 new_offset = offset + maxwell3d.regs.index_array.first * - maxwell3d.regs.index_array.FormatSizeInBytes(); + const u32 new_offset = offset + maxwell3d->regs.index_array.first * + maxwell3d->regs.index_array.FormatSizeInBytes(); runtime.BindIndexBuffer(buffer, new_offset, size); } else { - runtime.BindIndexBuffer(maxwell3d.regs.draw.topology, maxwell3d.regs.index_array.format, - maxwell3d.regs.index_array.first, maxwell3d.regs.index_array.count, - buffer, offset, size); + runtime.BindIndexBuffer(maxwell3d->regs.draw.topology, maxwell3d->regs.index_array.format, + maxwell3d->regs.index_array.first, + maxwell3d->regs.index_array.count, buffer, offset, size); } } template <class P> void BufferCache<P>::BindHostVertexBuffers() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { const Binding& binding = vertex_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; @@ -1014,7 +1017,7 @@ void BufferCache<P>::BindHostVertexBuffers() { } flags[Dirty::VertexBuffer0 + index] = false; - const u32 stride = maxwell3d.regs.vertex_array[index].stride; + const u32 stride = maxwell3d->regs.vertex_array[index].stride; const u32 offset = buffer.Offset(binding.cpu_addr); runtime.BindVertexBuffer(index, buffer, offset, binding.size, stride); } @@ -1154,7 +1157,7 @@ void BufferCache<P>::BindHostGraphicsTextureBuffers(size_t stage) { template <class P> void BufferCache<P>::BindHostTransformFeedbackBuffers() { - if (maxwell3d.regs.tfb_enabled == 0) { + if (maxwell3d->regs.tfb_enabled == 0) { return; } for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) { @@ -1239,16 +1242,19 @@ void BufferCache<P>::BindHostComputeTextureBuffers() { template <class P> void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) { - if (is_indexed) { - UpdateIndexBuffer(); - } - UpdateVertexBuffers(); - UpdateTransformFeedbackBuffers(); - for (size_t stage = 0; stage < NUM_STAGES; ++stage) { - UpdateUniformBuffers(stage); - UpdateStorageBuffers(stage); - UpdateTextureBuffers(stage); - } + do { + has_deleted_buffers = false; + if (is_indexed) { + UpdateIndexBuffer(); + } + UpdateVertexBuffers(); + UpdateTransformFeedbackBuffers(); + for (size_t stage = 0; stage < NUM_STAGES; ++stage) { + UpdateUniformBuffers(stage); + UpdateStorageBuffers(stage); + UpdateTextureBuffers(stage); + } + } while (has_deleted_buffers); } template <class P> @@ -1262,8 +1268,8 @@ template <class P> void BufferCache<P>::UpdateIndexBuffer() { // We have to check for the dirty flags and index count // The index count is currently changed without updating the dirty flags - const auto& index_array = maxwell3d.regs.index_array; - auto& flags = maxwell3d.dirty.flags; + const auto& index_array = maxwell3d->regs.index_array; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::IndexBuffer] && last_index_count == index_array.count) { return; } @@ -1272,7 +1278,7 @@ void BufferCache<P>::UpdateIndexBuffer() { const GPUVAddr gpu_addr_begin = index_array.StartAddress(); const GPUVAddr gpu_addr_end = index_array.EndAddress(); - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr_begin); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin); const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin); const u32 draw_size = (index_array.count + index_array.first) * index_array.FormatSizeInBytes(); const u32 size = std::min(address_size, draw_size); @@ -1289,8 +1295,8 @@ void BufferCache<P>::UpdateIndexBuffer() { template <class P> void BufferCache<P>::UpdateVertexBuffers() { - auto& flags = maxwell3d.dirty.flags; - if (!maxwell3d.dirty.flags[Dirty::VertexBuffers]) { + auto& flags = maxwell3d->dirty.flags; + if (!maxwell3d->dirty.flags[Dirty::VertexBuffers]) { return; } flags[Dirty::VertexBuffers] = false; @@ -1302,33 +1308,25 @@ void BufferCache<P>::UpdateVertexBuffers() { template <class P> void BufferCache<P>::UpdateVertexBuffer(u32 index) { - if (!maxwell3d.dirty.flags[Dirty::VertexBuffer0 + index]) { + if (!maxwell3d->dirty.flags[Dirty::VertexBuffer0 + index]) { return; } - const auto& array = maxwell3d.regs.vertex_array[index]; - const auto& limit = maxwell3d.regs.vertex_array_limit[index]; + const auto& array = maxwell3d->regs.vertex_array[index]; + const auto& limit = maxwell3d->regs.vertex_array_limit[index]; const GPUVAddr gpu_addr_begin = array.StartAddress(); const GPUVAddr gpu_addr_end = limit.LimitAddress() + 1; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr_begin); - u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin); - if (address_size >= 64_MiB) { - // Reported vertex buffer size is very large, cap to mapped buffer size - GPUVAddr submapped_addr_end = gpu_addr_begin; - - const auto ranges{gpu_memory.GetSubmappedRange(gpu_addr_begin, address_size)}; - if (ranges.size() > 0) { - const auto& [addr, size] = *ranges.begin(); - submapped_addr_end = addr + size; - } - - address_size = - std::min(address_size, static_cast<u32>(submapped_addr_end - gpu_addr_begin)); - } - const u32 size = address_size; // TODO: Analyze stride and number of vertices - if (array.enable == 0 || size == 0 || !cpu_addr) { + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr_begin); + u32 address_size = static_cast<u32>( + std::min(gpu_addr_end - gpu_addr_begin, static_cast<u64>(std::numeric_limits<u32>::max()))); + if (array.enable == 0 || address_size == 0 || !cpu_addr) { vertex_buffers[index] = NULL_BINDING; return; } + if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end)) { + address_size = + static_cast<u32>(gpu_memory->MaxContinousRange(gpu_addr_begin, address_size)); + } + const u32 size = address_size; // TODO: Analyze stride and number of vertices vertex_buffers[index] = Binding{ .cpu_addr = *cpu_addr, .size = size, @@ -1382,7 +1380,7 @@ void BufferCache<P>::UpdateTextureBuffers(size_t stage) { template <class P> void BufferCache<P>::UpdateTransformFeedbackBuffers() { - if (maxwell3d.regs.tfb_enabled == 0) { + if (maxwell3d->regs.tfb_enabled == 0) { return; } for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) { @@ -1392,10 +1390,10 @@ void BufferCache<P>::UpdateTransformFeedbackBuffers() { template <class P> void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) { - const auto& binding = maxwell3d.regs.tfb_bindings[index]; + const auto& binding = maxwell3d->regs.tfb_bindings[index]; const GPUVAddr gpu_addr = binding.Address() + binding.buffer_offset; const u32 size = binding.buffer_size; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (binding.buffer_enable == 0 || size == 0 || !cpu_addr) { transform_feedback_buffers[index] = NULL_BINDING; return; @@ -1414,10 +1412,10 @@ void BufferCache<P>::UpdateComputeUniformBuffers() { ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) { Binding& binding = compute_uniform_buffers[index]; binding = NULL_BINDING; - const auto& launch_desc = kepler_compute.launch_description; + const auto& launch_desc = kepler_compute->launch_description; if (((launch_desc.const_buffer_enable_mask >> index) & 1) != 0) { const auto& cbuf = launch_desc.const_buffer_config[index]; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(cbuf.Address()); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(cbuf.Address()); if (cpu_addr) { binding.cpu_addr = *cpu_addr; binding.size = cbuf.size; @@ -1567,6 +1565,8 @@ BufferId BufferCache<P>::CreateBuffer(VAddr cpu_addr, u32 wanted_size) { const OverlapResult overlap = ResolveOverlaps(cpu_addr, wanted_size); const u32 size = static_cast<u32>(overlap.end - overlap.begin); const BufferId new_buffer_id = slot_buffers.insert(runtime, rasterizer, overlap.begin, size); + auto& new_buffer = slot_buffers[new_buffer_id]; + runtime.ClearBuffer(new_buffer, 0, new_buffer.SizeBytes(), 0); for (const BufferId overlap_id : overlap.ids) { JoinOverlap(new_buffer_id, overlap_id, !overlap.has_stream_leap); } @@ -1695,7 +1695,7 @@ void BufferCache<P>::MappedUploadMemory(Buffer& buffer, u64 total_size_bytes, template <class P> bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size, - std::span<u8> inlined_buffer) { + std::span<const u8> inlined_buffer) { const bool is_dirty = IsRegionRegistered(dest_address, copy_size); if (!is_dirty) { return false; @@ -1831,7 +1831,7 @@ void BufferCache<P>::NotifyBufferDeletion() { dirty_uniform_buffers.fill(~u32{0}); uniform_buffer_binding_sizes.fill({}); } - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; flags[Dirty::IndexBuffer] = true; flags[Dirty::VertexBuffers] = true; for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { @@ -1841,16 +1841,18 @@ void BufferCache<P>::NotifyBufferDeletion() { } template <class P> -typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr) const { - const GPUVAddr gpu_addr = gpu_memory.Read<u64>(ssbo_addr); - const u32 size = gpu_memory.Read<u32>(ssbo_addr + 8); - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); +typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr, + bool is_written) const { + const GPUVAddr gpu_addr = gpu_memory->Read<u64>(ssbo_addr); + const u32 size = gpu_memory->Read<u32>(ssbo_addr + 8); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (!cpu_addr || size == 0) { return NULL_BINDING; } + const VAddr cpu_end = Common::AlignUp(*cpu_addr + size, Core::Memory::YUZU_PAGESIZE); const Binding binding{ .cpu_addr = *cpu_addr, - .size = size, + .size = is_written ? size : static_cast<u32>(cpu_end - *cpu_addr), .buffer_id = BufferId{}, }; return binding; @@ -1859,7 +1861,7 @@ typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr s template <class P> typename BufferCache<P>::TextureBufferBinding BufferCache<P>::GetTextureBufferBinding( GPUVAddr gpu_addr, u32 size, PixelFormat format) { - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); TextureBufferBinding binding; if (!cpu_addr || size == 0) { binding.cpu_addr = 0; diff --git a/src/video_core/cdma_pusher.cpp b/src/video_core/cdma_pusher.cpp index 8e890a85e..28a2d2090 100644 --- a/src/video_core/cdma_pusher.cpp +++ b/src/video_core/cdma_pusher.cpp @@ -2,20 +2,22 @@ // SPDX-License-Identifier: MIT #include <bit> -#include "command_classes/host1x.h" -#include "command_classes/nvdec.h" -#include "command_classes/vic.h" #include "video_core/cdma_pusher.h" -#include "video_core/command_classes/sync_manager.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" +#include "video_core/host1x/control.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/nvdec.h" +#include "video_core/host1x/nvdec_common.h" +#include "video_core/host1x/sync_manager.h" +#include "video_core/host1x/vic.h" +#include "video_core/memory_manager.h" namespace Tegra { -CDmaPusher::CDmaPusher(GPU& gpu_) - : gpu{gpu_}, nvdec_processor(std::make_shared<Nvdec>(gpu)), - vic_processor(std::make_unique<Vic>(gpu, nvdec_processor)), - host1x_processor(std::make_unique<Host1x>(gpu)), - sync_manager(std::make_unique<SyncptIncrManager>(gpu)) {} +CDmaPusher::CDmaPusher(Host1x::Host1x& host1x_) + : host1x{host1x_}, nvdec_processor(std::make_shared<Host1x::Nvdec>(host1x)), + vic_processor(std::make_unique<Host1x::Vic>(host1x, nvdec_processor)), + host1x_processor(std::make_unique<Host1x::Control>(host1x)), + sync_manager(std::make_unique<Host1x::SyncptIncrManager>(host1x)) {} CDmaPusher::~CDmaPusher() = default; @@ -109,16 +111,17 @@ void CDmaPusher::ExecuteCommand(u32 state_offset, u32 data) { case ThiMethod::SetMethod1: LOG_DEBUG(Service_NVDRV, "VIC method 0x{:X}, Args=({})", static_cast<u32>(vic_thi_state.method_0), data); - vic_processor->ProcessMethod(static_cast<Vic::Method>(vic_thi_state.method_0), data); + vic_processor->ProcessMethod(static_cast<Host1x::Vic::Method>(vic_thi_state.method_0), + data); break; default: break; } break; - case ChClassId::Host1x: + case ChClassId::Control: // This device is mainly for syncpoint synchronization LOG_DEBUG(Service_NVDRV, "Host1X Class Method"); - host1x_processor->ProcessMethod(static_cast<Host1x::Method>(offset), data); + host1x_processor->ProcessMethod(static_cast<Host1x::Control::Method>(offset), data); break; default: UNIMPLEMENTED_MSG("Current class not implemented {:X}", static_cast<u32>(current_class)); diff --git a/src/video_core/cdma_pusher.h b/src/video_core/cdma_pusher.h index d6ffef95f..83112dfce 100644 --- a/src/video_core/cdma_pusher.h +++ b/src/video_core/cdma_pusher.h @@ -12,11 +12,13 @@ namespace Tegra { -class GPU; +namespace Host1x { +class Control; class Host1x; class Nvdec; class SyncptIncrManager; class Vic; +} // namespace Host1x enum class ChSubmissionMode : u32 { SetClass = 0, @@ -30,7 +32,7 @@ enum class ChSubmissionMode : u32 { enum class ChClassId : u32 { NoClass = 0x0, - Host1x = 0x1, + Control = 0x1, VideoEncodeMpeg = 0x20, VideoEncodeNvEnc = 0x21, VideoStreamingVi = 0x30, @@ -88,7 +90,7 @@ enum class ThiMethod : u32 { class CDmaPusher { public: - explicit CDmaPusher(GPU& gpu_); + explicit CDmaPusher(Host1x::Host1x& host1x); ~CDmaPusher(); /// Process the command entry @@ -101,11 +103,11 @@ private: /// Write arguments value to the ThiRegisters member at the specified offset void ThiStateWrite(ThiRegisters& state, u32 offset, u32 argument); - GPU& gpu; - std::shared_ptr<Tegra::Nvdec> nvdec_processor; - std::unique_ptr<Tegra::Vic> vic_processor; - std::unique_ptr<Tegra::Host1x> host1x_processor; - std::unique_ptr<SyncptIncrManager> sync_manager; + Host1x::Host1x& host1x; + std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor; + std::unique_ptr<Tegra::Host1x::Vic> vic_processor; + std::unique_ptr<Tegra::Host1x::Control> host1x_processor; + std::unique_ptr<Host1x::SyncptIncrManager> sync_manager; ChClassId current_class{}; ThiRegisters vic_thi_state{}; ThiRegisters nvdec_thi_state{}; diff --git a/src/video_core/command_classes/host1x.cpp b/src/video_core/command_classes/host1x.cpp deleted file mode 100644 index 11855fe10..000000000 --- a/src/video_core/command_classes/host1x.cpp +++ /dev/null @@ -1,29 +0,0 @@ -// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later - -#include "common/assert.h" -#include "video_core/command_classes/host1x.h" -#include "video_core/gpu.h" - -Tegra::Host1x::Host1x(GPU& gpu_) : gpu(gpu_) {} - -Tegra::Host1x::~Host1x() = default; - -void Tegra::Host1x::ProcessMethod(Method method, u32 argument) { - switch (method) { - case Method::LoadSyncptPayload32: - syncpoint_value = argument; - break; - case Method::WaitSyncpt: - case Method::WaitSyncpt32: - Execute(argument); - break; - default: - UNIMPLEMENTED_MSG("Host1x method 0x{:X}", static_cast<u32>(method)); - break; - } -} - -void Tegra::Host1x::Execute(u32 data) { - gpu.WaitFence(data, syncpoint_value); -} diff --git a/src/video_core/control/channel_state.cpp b/src/video_core/control/channel_state.cpp new file mode 100644 index 000000000..cdecc3a91 --- /dev/null +++ b/src/video_core/control/channel_state.cpp @@ -0,0 +1,40 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/assert.h" +#include "video_core/control/channel_state.h" +#include "video_core/dma_pusher.h" +#include "video_core/engines/fermi_2d.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/kepler_memory.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/engines/maxwell_dma.h" +#include "video_core/engines/puller.h" +#include "video_core/memory_manager.h" + +namespace Tegra::Control { + +ChannelState::ChannelState(s32 bind_id_) : bind_id{bind_id_}, initialized{} {} + +void ChannelState::Init(Core::System& system, GPU& gpu) { + ASSERT(memory_manager); + dma_pusher = std::make_unique<Tegra::DmaPusher>(system, gpu, *memory_manager, *this); + maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, *memory_manager); + fermi_2d = std::make_unique<Engines::Fermi2D>(); + kepler_compute = std::make_unique<Engines::KeplerCompute>(system, *memory_manager); + maxwell_dma = std::make_unique<Engines::MaxwellDMA>(system, *memory_manager); + kepler_memory = std::make_unique<Engines::KeplerMemory>(system, *memory_manager); + initialized = true; +} + +void ChannelState::BindRasterizer(VideoCore::RasterizerInterface* rasterizer) { + dma_pusher->BindRasterizer(rasterizer); + memory_manager->BindRasterizer(rasterizer); + maxwell_3d->BindRasterizer(rasterizer); + fermi_2d->BindRasterizer(rasterizer); + kepler_memory->BindRasterizer(rasterizer); + kepler_compute->BindRasterizer(rasterizer); + maxwell_dma->BindRasterizer(rasterizer); +} + +} // namespace Tegra::Control diff --git a/src/video_core/control/channel_state.h b/src/video_core/control/channel_state.h new file mode 100644 index 000000000..3a7b9872c --- /dev/null +++ b/src/video_core/control/channel_state.h @@ -0,0 +1,68 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <memory> + +#include "common/common_types.h" + +namespace Core { +class System; +} + +namespace VideoCore { +class RasterizerInterface; +} + +namespace Tegra { + +class GPU; + +namespace Engines { +class Puller; +class Fermi2D; +class Maxwell3D; +class MaxwellDMA; +class KeplerCompute; +class KeplerMemory; +} // namespace Engines + +class MemoryManager; +class DmaPusher; + +namespace Control { + +struct ChannelState { + explicit ChannelState(s32 bind_id); + ChannelState(const ChannelState& state) = delete; + ChannelState& operator=(const ChannelState&) = delete; + ChannelState(ChannelState&& other) noexcept = default; + ChannelState& operator=(ChannelState&& other) noexcept = default; + + void Init(Core::System& system, GPU& gpu); + + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + + s32 bind_id = -1; + /// 3D engine + std::unique_ptr<Engines::Maxwell3D> maxwell_3d; + /// 2D engine + std::unique_ptr<Engines::Fermi2D> fermi_2d; + /// Compute engine + std::unique_ptr<Engines::KeplerCompute> kepler_compute; + /// DMA engine + std::unique_ptr<Engines::MaxwellDMA> maxwell_dma; + /// Inline memory engine + std::unique_ptr<Engines::KeplerMemory> kepler_memory; + + std::shared_ptr<MemoryManager> memory_manager; + + std::unique_ptr<DmaPusher> dma_pusher; + + bool initialized{}; +}; + +} // namespace Control + +} // namespace Tegra diff --git a/src/video_core/control/channel_state_cache.cpp b/src/video_core/control/channel_state_cache.cpp new file mode 100644 index 000000000..4ebeb6356 --- /dev/null +++ b/src/video_core/control/channel_state_cache.cpp @@ -0,0 +1,14 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "video_core/control/channel_state_cache.inc" + +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>; + +} // namespace VideoCommon diff --git a/src/video_core/control/channel_state_cache.h b/src/video_core/control/channel_state_cache.h new file mode 100644 index 000000000..584a0c26c --- /dev/null +++ b/src/video_core/control/channel_state_cache.h @@ -0,0 +1,101 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <deque> +#include <limits> +#include <mutex> +#include <optional> +#include <unordered_map> +#include <vector> + +#include "common/common_types.h" + +namespace Tegra { + +namespace Engines { +class Maxwell3D; +class KeplerCompute; +} // namespace Engines + +class MemoryManager; + +namespace Control { +struct ChannelState; +} + +} // namespace Tegra + +namespace VideoCommon { + +class ChannelInfo { +public: + ChannelInfo() = delete; + explicit ChannelInfo(Tegra::Control::ChannelState& state); + ChannelInfo(const ChannelInfo& state) = delete; + ChannelInfo& operator=(const ChannelInfo&) = delete; + ChannelInfo(ChannelInfo&& other) = default; + ChannelInfo& operator=(ChannelInfo&& other) = default; + + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; + Tegra::MemoryManager& gpu_memory; +}; + +template <class P> +class ChannelSetupCaches { +public: + /// Operations for seting the channel of execution. + virtual ~ChannelSetupCaches(); + + /// Create channel state. + virtual void CreateChannel(Tegra::Control::ChannelState& channel); + + /// Bind a channel for execution. + void BindToChannel(s32 id); + + /// Erase channel's state. + 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; + } + + std::optional<size_t> getStorageID(size_t id) const { + std::unique_lock<std::mutex> lk(config_mutex); + const auto ref = address_spaces.find(id); + if (ref == address_spaces.end()) { + return std::nullopt; + } + return ref->second.storage_id; + } + +protected: + 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<size_t> free_channel_ids; + std::unordered_map<s32, size_t> channel_map; + std::vector<size_t> active_channel_ids; + struct AddresSpaceRef { + size_t ref_count; + size_t storage_id; + 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 diff --git a/src/video_core/control/channel_state_cache.inc b/src/video_core/control/channel_state_cache.inc new file mode 100644 index 000000000..460313893 --- /dev/null +++ b/src/video_core/control/channel_state_cache.inc @@ -0,0 +1,86 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <algorithm> + +#include "video_core/control/channel_state.h" +#include "video_core/control/channel_state_cache.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/memory_manager.h" + +namespace VideoCommon { + +template <class P> +ChannelSetupCaches<P>::~ChannelSetupCaches() = default; + +template <class P> +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); + auto new_id = [this, &channel]() { + if (!free_channel_ids.empty()) { + auto id = free_channel_ids.front(); + free_channel_ids.pop_front(); + new (&channel_storage[id]) P(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) { + channel_state = &channel_storage[current_channel_id]; + } + active_channel_ids.push_back(new_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. +template <class P> +void ChannelSetupCaches<P>::BindToChannel(s32 id) { + std::unique_lock<std::mutex> lk(config_mutex); + auto it = channel_map.find(id); + ASSERT(it != channel_map.end() && id >= 0); + current_channel_id = it->second; + channel_state = &channel_storage[current_channel_id]; + maxwell3d = &channel_state->maxwell3d; + kepler_compute = &channel_state->kepler_compute; + gpu_memory = &channel_state->gpu_memory; + current_address_space = gpu_memory->GetID(); +} + +/// Erase channel's channel_state. +template <class P> +void ChannelSetupCaches<P>::EraseChannel(s32 id) { + std::unique_lock<std::mutex> lk(config_mutex); + 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; + channel_state = nullptr; + maxwell3d = nullptr; + kepler_compute = nullptr; + gpu_memory = nullptr; + } else if (current_channel_id != UNSET_CHANNEL) { + channel_state = &channel_storage[current_channel_id]; + } + active_channel_ids.erase( + std::find(active_channel_ids.begin(), active_channel_ids.end(), this_id)); +} + +} // namespace VideoCommon diff --git a/src/video_core/control/scheduler.cpp b/src/video_core/control/scheduler.cpp new file mode 100644 index 000000000..f7cbe204e --- /dev/null +++ b/src/video_core/control/scheduler.cpp @@ -0,0 +1,32 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include <memory> + +#include "common/assert.h" +#include "video_core/control/channel_state.h" +#include "video_core/control/scheduler.h" +#include "video_core/gpu.h" + +namespace Tegra::Control { +Scheduler::Scheduler(GPU& gpu_) : gpu{gpu_} {} + +Scheduler::~Scheduler() = default; + +void Scheduler::Push(s32 channel, CommandList&& entries) { + std::unique_lock lk(scheduling_guard); + auto it = channels.find(channel); + ASSERT(it != channels.end()); + auto channel_state = it->second; + gpu.BindChannel(channel_state->bind_id); + channel_state->dma_pusher->Push(std::move(entries)); + channel_state->dma_pusher->DispatchCalls(); +} + +void Scheduler::DeclareChannel(std::shared_ptr<ChannelState> new_channel) { + s32 channel = new_channel->bind_id; + std::unique_lock lk(scheduling_guard); + channels.emplace(channel, new_channel); +} + +} // namespace Tegra::Control diff --git a/src/video_core/control/scheduler.h b/src/video_core/control/scheduler.h new file mode 100644 index 000000000..44addf61c --- /dev/null +++ b/src/video_core/control/scheduler.h @@ -0,0 +1,37 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <memory> +#include <mutex> +#include <unordered_map> + +#include "video_core/dma_pusher.h" + +namespace Tegra { + +class GPU; + +namespace Control { + +struct ChannelState; + +class Scheduler { +public: + explicit Scheduler(GPU& gpu_); + ~Scheduler(); + + void Push(s32 channel, CommandList&& entries); + + void DeclareChannel(std::shared_ptr<ChannelState> new_channel); + +private: + std::unordered_map<s32, std::shared_ptr<ChannelState>> channels; + std::mutex scheduling_guard; + GPU& gpu; +}; + +} // namespace Control + +} // namespace Tegra diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 29b8582ab..9835e3ac1 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -12,7 +12,10 @@ namespace Tegra { -DmaPusher::DmaPusher(Core::System& system_, GPU& gpu_) : gpu{gpu_}, system{system_} {} +DmaPusher::DmaPusher(Core::System& system_, GPU& gpu_, MemoryManager& memory_manager_, + Control::ChannelState& channel_state_) + : gpu{gpu_}, system{system_}, memory_manager{memory_manager_}, puller{gpu_, memory_manager_, + *this, channel_state_} {} DmaPusher::~DmaPusher() = default; @@ -21,8 +24,6 @@ MICROPROFILE_DEFINE(DispatchCalls, "GPU", "Execute command buffer", MP_RGB(128, void DmaPusher::DispatchCalls() { MICROPROFILE_SCOPE(DispatchCalls); - gpu.SyncGuestHost(); - dma_pushbuffer_subindex = 0; dma_state.is_last_call = true; @@ -33,7 +34,6 @@ void DmaPusher::DispatchCalls() { } } gpu.FlushCommands(); - gpu.SyncGuestHost(); gpu.OnCommandListEnd(); } @@ -76,11 +76,11 @@ bool DmaPusher::Step() { // Push buffer non-empty, read a word command_headers.resize(command_list_header.size); if (Settings::IsGPULevelHigh()) { - gpu.MemoryManager().ReadBlock(dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + memory_manager.ReadBlock(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); } else { - gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + memory_manager.ReadBlockUnsafe(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); } } for (std::size_t index = 0; index < command_headers.size();) { @@ -154,7 +154,7 @@ void DmaPusher::SetState(const CommandHeader& command_header) { void DmaPusher::CallMethod(u32 argument) const { if (dma_state.method < non_puller_methods) { - gpu.CallMethod(GPU::MethodCall{ + puller.CallPullerMethod(Engines::Puller::MethodCall{ dma_state.method, argument, dma_state.subchannel, @@ -168,12 +168,16 @@ void DmaPusher::CallMethod(u32 argument) const { void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const { if (dma_state.method < non_puller_methods) { - gpu.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods, - dma_state.method_count); + puller.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods, + dma_state.method_count); } else { subchannels[dma_state.subchannel]->CallMultiMethod(dma_state.method, base_start, num_methods, dma_state.method_count); } } +void DmaPusher::BindRasterizer(VideoCore::RasterizerInterface* rasterizer) { + puller.BindRasterizer(rasterizer); +} + } // namespace Tegra diff --git a/src/video_core/dma_pusher.h b/src/video_core/dma_pusher.h index 872fd146a..938f0f11c 100644 --- a/src/video_core/dma_pusher.h +++ b/src/video_core/dma_pusher.h @@ -10,6 +10,7 @@ #include "common/bit_field.h" #include "common/common_types.h" #include "video_core/engines/engine_interface.h" +#include "video_core/engines/puller.h" namespace Core { class System; @@ -17,7 +18,12 @@ class System; namespace Tegra { +namespace Control { +struct ChannelState; +} + class GPU; +class MemoryManager; enum class SubmissionMode : u32 { IncreasingOld = 0, @@ -31,24 +37,32 @@ enum class SubmissionMode : u32 { // Note that, traditionally, methods are treated as 4-byte addressable locations, and hence // their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4. // So the values you see in docs might be multiplied by 4. +// Register documentation: +// https://github.com/NVIDIA/open-gpu-doc/blob/ab27fc22db5de0d02a4cabe08e555663b62db4d4/classes/host/cla26f.h +// +// Register Description (approx): +// https://github.com/NVIDIA/open-gpu-doc/blob/ab27fc22db5de0d02a4cabe08e555663b62db4d4/manuals/volta/gv100/dev_pbdma.ref.txt enum class BufferMethods : u32 { BindObject = 0x0, + Illegal = 0x1, Nop = 0x2, SemaphoreAddressHigh = 0x4, SemaphoreAddressLow = 0x5, - SemaphoreSequence = 0x6, - SemaphoreTrigger = 0x7, - NotifyIntr = 0x8, + SemaphoreSequencePayload = 0x6, + SemaphoreOperation = 0x7, + NonStallInterrupt = 0x8, WrcacheFlush = 0x9, - Unk28 = 0xA, - UnkCacheFlush = 0xB, + MemOpA = 0xA, + MemOpB = 0xB, + MemOpC = 0xC, + MemOpD = 0xD, RefCnt = 0x14, SemaphoreAcquire = 0x1A, SemaphoreRelease = 0x1B, - FenceValue = 0x1C, - FenceAction = 0x1D, - WaitForInterrupt = 0x1E, - Unk7c = 0x1F, + SyncpointPayload = 0x1C, + SyncpointOperation = 0x1D, + WaitForIdle = 0x1E, + CRCCheck = 0x1F, Yield = 0x20, NonPullerMethods = 0x40, }; @@ -102,7 +116,8 @@ struct CommandList final { */ class DmaPusher final { public: - explicit DmaPusher(Core::System& system_, GPU& gpu_); + explicit DmaPusher(Core::System& system_, GPU& gpu_, MemoryManager& memory_manager_, + Control::ChannelState& channel_state_); ~DmaPusher(); void Push(CommandList&& entries) { @@ -115,6 +130,8 @@ public: subchannels[subchannel_id] = engine; } + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + private: static constexpr u32 non_puller_methods = 0x40; static constexpr u32 max_subchannels = 8; @@ -148,6 +165,8 @@ private: GPU& gpu; Core::System& system; + MemoryManager& memory_manager; + mutable Engines::Puller puller; }; } // namespace Tegra diff --git a/src/video_core/engines/engine_upload.cpp b/src/video_core/engines/engine_upload.cpp index 6ff5b1eca..a34819234 100644 --- a/src/video_core/engines/engine_upload.cpp +++ b/src/video_core/engines/engine_upload.cpp @@ -3,6 +3,7 @@ #include <cstring> +#include "common/algorithm.h" #include "common/assert.h" #include "video_core/engines/engine_upload.h" #include "video_core/memory_manager.h" @@ -34,21 +35,48 @@ void State::ProcessData(const u32 data, const bool is_last_call) { if (!is_last_call) { return; } + ProcessData(inner_buffer); +} + +void State::ProcessData(const u32* data, size_t num_data) { + std::span<const u8> read_buffer(reinterpret_cast<const u8*>(data), num_data * sizeof(u32)); + ProcessData(read_buffer); +} + +void State::ProcessData(std::span<const u8> read_buffer) { const GPUVAddr address{regs.dest.Address()}; if (is_linear) { - rasterizer->AccelerateInlineToMemory(address, copy_size, inner_buffer); + if (regs.line_count == 1) { + rasterizer->AccelerateInlineToMemory(address, copy_size, read_buffer); + } else { + for (u32 line = 0; line < regs.line_count; ++line) { + const GPUVAddr dest_line = address + static_cast<size_t>(line) * regs.dest.pitch; + memory_manager.WriteBlockUnsafe( + dest_line, read_buffer.data() + static_cast<size_t>(line) * regs.line_length_in, + regs.line_length_in); + } + memory_manager.InvalidateRegion(address, regs.dest.pitch * regs.line_count); + } } else { - UNIMPLEMENTED_IF(regs.dest.z != 0); - UNIMPLEMENTED_IF(regs.dest.depth != 1); - UNIMPLEMENTED_IF(regs.dest.BlockWidth() != 0); - UNIMPLEMENTED_IF(regs.dest.BlockDepth() != 0); + u32 width = regs.dest.width; + u32 x_elements = regs.line_length_in; + u32 x_offset = regs.dest.x; + const u32 bpp_shift = Common::FoldRight( + 4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); }, + width, x_elements, x_offset, static_cast<u32>(address)); + width >>= bpp_shift; + x_elements >>= bpp_shift; + x_offset >>= bpp_shift; + const u32 bytes_per_pixel = 1U << bpp_shift; const std::size_t dst_size = Tegra::Texture::CalculateSize( - true, 1, regs.dest.width, regs.dest.height, 1, regs.dest.BlockHeight(), 0); + true, bytes_per_pixel, width, regs.dest.height, regs.dest.depth, + regs.dest.BlockHeight(), regs.dest.BlockDepth()); tmp_buffer.resize(dst_size); memory_manager.ReadBlock(address, tmp_buffer.data(), dst_size); - Tegra::Texture::SwizzleKepler(regs.dest.width, regs.dest.height, regs.dest.x, regs.dest.y, - regs.dest.BlockHeight(), copy_size, inner_buffer.data(), - tmp_buffer.data()); + Tegra::Texture::SwizzleSubrect(tmp_buffer, read_buffer, bytes_per_pixel, width, + regs.dest.height, regs.dest.depth, x_offset, regs.dest.y, + x_elements, regs.line_count, regs.dest.BlockHeight(), + regs.dest.BlockDepth(), regs.line_length_in); memory_manager.WriteBlock(address, tmp_buffer.data(), dst_size); } } diff --git a/src/video_core/engines/engine_upload.h b/src/video_core/engines/engine_upload.h index 94ff3314a..f08f6e36a 100644 --- a/src/video_core/engines/engine_upload.h +++ b/src/video_core/engines/engine_upload.h @@ -3,6 +3,7 @@ #pragma once +#include <span> #include <vector> #include "common/bit_field.h" #include "common/common_types.h" @@ -33,7 +34,7 @@ struct Registers { u32 width; u32 height; u32 depth; - u32 z; + u32 layer; u32 x; u32 y; @@ -62,11 +63,14 @@ public: void ProcessExec(bool is_linear_); void ProcessData(u32 data, bool is_last_call); + void ProcessData(const u32* data, size_t num_data); /// Binds a rasterizer to this engine. void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); private: + void ProcessData(std::span<const u8> read_buffer); + u32 write_offset = 0; u32 copy_size = 0; std::vector<u8> inner_buffer; diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp index 5db254d94..7c50bdbe0 100644 --- a/src/video_core/engines/kepler_compute.cpp +++ b/src/video_core/engines/kepler_compute.cpp @@ -36,8 +36,6 @@ void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_cal } case KEPLER_COMPUTE_REG_INDEX(data_upload): { upload_state.ProcessData(method_argument, is_last_call); - if (is_last_call) { - } break; } case KEPLER_COMPUTE_REG_INDEX(launch): @@ -50,8 +48,15 @@ void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_cal void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) { - for (std::size_t i = 0; i < amount; i++) { - CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + switch (method) { + case KEPLER_COMPUTE_REG_INDEX(data_upload): + upload_state.ProcessData(base_start, static_cast<size_t>(amount)); + return; + default: + for (std::size_t i = 0; i < amount; i++) { + CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + } + break; } } diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp index e2b029542..a3fbab1e5 100644 --- a/src/video_core/engines/kepler_memory.cpp +++ b/src/video_core/engines/kepler_memory.cpp @@ -33,8 +33,6 @@ void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call } case KEPLERMEMORY_REG_INDEX(data): { upload_state.ProcessData(method_argument, is_last_call); - if (is_last_call) { - } break; } } @@ -42,8 +40,15 @@ void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call void KeplerMemory::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) { - for (std::size_t i = 0; i < amount; i++) { - CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + switch (method) { + case KEPLERMEMORY_REG_INDEX(data): + upload_state.ProcessData(base_start, static_cast<size_t>(amount)); + return; + default: + for (std::size_t i = 0; i < amount; i++) { + CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); + } + break; } } diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 3a4646289..3c6e44a25 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -219,6 +219,8 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume regs.index_array.count = regs.small_index_2.count; regs.index_array.first = regs.small_index_2.first; dirty.flags[VideoCommon::Dirty::IndexBuffer] = true; + // a macro calls this one over and over, should it increase instancing? + // Used by Hades and likely other Vulkan games. return DrawArrays(); case MAXWELL3D_REG_INDEX(topology_override): use_topology_override = true; @@ -237,11 +239,12 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume return upload_state.ProcessExec(regs.exec_upload.linear != 0); case MAXWELL3D_REG_INDEX(data_upload): upload_state.ProcessData(argument, is_last_call); - if (is_last_call) { - } return; case MAXWELL3D_REG_INDEX(fragment_barrier): return rasterizer->FragmentBarrier(); + case MAXWELL3D_REG_INDEX(invalidate_texture_data_cache): + rasterizer->InvalidateGPUCache(); + return rasterizer->WaitForIdle(); case MAXWELL3D_REG_INDEX(tiled_cache_barrier): return rasterizer->TiledCacheBarrier(); } @@ -311,6 +314,9 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, case MAXWELL3D_REG_INDEX(const_buffer.cb_data) + 15: ProcessCBMultiData(base_start, amount); break; + case MAXWELL3D_REG_INDEX(data_upload): + upload_state.ProcessData(base_start, static_cast<size_t>(amount)); + return; default: for (std::size_t i = 0; i < amount; i++) { CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1); @@ -447,18 +453,10 @@ void Maxwell3D::ProcessFirmwareCall4() { } void Maxwell3D::StampQueryResult(u64 payload, bool long_query) { - struct LongQueryResult { - u64_le value; - u64_le timestamp; - }; - static_assert(sizeof(LongQueryResult) == 16, "LongQueryResult has wrong size"); const GPUVAddr sequence_address{regs.query.QueryAddress()}; if (long_query) { - // Write the 128-bit result structure in long mode. Note: We emulate an infinitely fast - // GPU, this command may actually take a while to complete in real hardware due to GPU - // wait queues. - LongQueryResult query_result{payload, system.GPU().GetTicks()}; - memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result)); + memory_manager.Write<u64>(sequence_address + sizeof(u64), system.GPU().GetTicks()); + memory_manager.Write<u64>(sequence_address, payload); } else { memory_manager.Write<u32>(sequence_address, static_cast<u32>(payload)); } @@ -472,10 +470,25 @@ void Maxwell3D::ProcessQueryGet() { switch (regs.query.query_get.operation) { case Regs::QueryOperation::Release: - if (regs.query.query_get.fence == 1) { - rasterizer->SignalSemaphore(regs.query.QueryAddress(), regs.query.query_sequence); + if (regs.query.query_get.fence == 1 || regs.query.query_get.short_query != 0) { + const GPUVAddr sequence_address{regs.query.QueryAddress()}; + const u32 payload = regs.query.query_sequence; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u32>(sequence_address, payload); + }); + rasterizer->SignalFence(std::move(operation)); } else { - StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0); + struct LongQueryResult { + u64_le value; + u64_le timestamp; + }; + const GPUVAddr sequence_address{regs.query.QueryAddress()}; + const u32 payload = regs.query.query_sequence; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u64>(sequence_address + sizeof(u64), system.GPU().GetTicks()); + memory_manager.Write<u64>(sequence_address, payload); + }); + rasterizer->SyncOperation(std::move(operation)); } break; case Regs::QueryOperation::Acquire: diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 0efe58282..3909d36c1 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -1,6 +1,7 @@ // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include "common/algorithm.h" #include "common/assert.h" #include "common/logging/log.h" #include "common/microprofile.h" @@ -54,8 +55,6 @@ void MaxwellDMA::Launch() { const LaunchDMA& launch = regs.launch_dma; ASSERT(launch.interrupt_type == LaunchDMA::InterruptType::NONE); ASSERT(launch.data_transfer_type == LaunchDMA::DataTransferType::NON_PIPELINED); - ASSERT(regs.dst_params.origin.x == 0); - ASSERT(regs.dst_params.origin.y == 0); const bool is_src_pitch = launch.src_memory_layout == LaunchDMA::MemoryLayout::PITCH; const bool is_dst_pitch = launch.dst_memory_layout == LaunchDMA::MemoryLayout::PITCH; @@ -121,23 +120,40 @@ void MaxwellDMA::CopyPitchToPitch() { void MaxwellDMA::CopyBlockLinearToPitch() { UNIMPLEMENTED_IF(regs.src_params.block_size.width != 0); - UNIMPLEMENTED_IF(regs.src_params.block_size.depth != 0); UNIMPLEMENTED_IF(regs.src_params.layer != 0); + const bool is_remapping = regs.launch_dma.remap_enable != 0; + // Optimized path for micro copies. const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; - if (dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X && + if (!is_remapping && dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X && regs.src_params.height > GOB_SIZE_Y) { FastCopyBlockLinearToPitch(); return; } // Deswizzle the input and copy it over. - UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); - const u32 bytes_per_pixel = - regs.launch_dma.remap_enable ? regs.pitch_out / regs.line_length_in : 1; const Parameters& src_params = regs.src_params; - const u32 width = src_params.width; + + const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1; + const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1; + + const u32 base_bpp = !is_remapping ? 1U : num_remap_components * remap_components_size; + + u32 width = src_params.width; + u32 x_elements = regs.line_length_in; + u32 x_offset = src_params.origin.x; + u32 bpp_shift = 0U; + if (!is_remapping) { + bpp_shift = Common::FoldRight( + 4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); }, + width, x_elements, x_offset, static_cast<u32>(regs.offset_in)); + width >>= bpp_shift; + x_elements >>= bpp_shift; + x_offset >>= bpp_shift; + } + + const u32 bytes_per_pixel = base_bpp << bpp_shift; const u32 height = src_params.height; const u32 depth = src_params.depth; const u32 block_height = src_params.block_size.height; @@ -155,30 +171,45 @@ void MaxwellDMA::CopyBlockLinearToPitch() { memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); - UnswizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_out, width, bytes_per_pixel, - block_height, src_params.origin.x, src_params.origin.y, write_buffer.data(), - read_buffer.data()); + UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, + src_params.origin.y, x_elements, regs.line_count, block_height, block_depth, + regs.pitch_out); memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size); } void MaxwellDMA::CopyPitchToBlockLinear() { UNIMPLEMENTED_IF_MSG(regs.dst_params.block_size.width != 0, "Block width is not one"); - UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0); + UNIMPLEMENTED_IF(regs.dst_params.layer != 0); + + const bool is_remapping = regs.launch_dma.remap_enable != 0; + const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1; + const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1; const auto& dst_params = regs.dst_params; - const u32 bytes_per_pixel = - regs.launch_dma.remap_enable ? regs.pitch_in / regs.line_length_in : 1; - const u32 width = dst_params.width; + + const u32 base_bpp = !is_remapping ? 1U : num_remap_components * remap_components_size; + + u32 width = dst_params.width; + u32 x_elements = regs.line_length_in; + u32 x_offset = dst_params.origin.x; + u32 bpp_shift = 0U; + if (!is_remapping) { + bpp_shift = Common::FoldRight( + 4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); }, + width, x_elements, x_offset, static_cast<u32>(regs.offset_out)); + width >>= bpp_shift; + x_elements >>= bpp_shift; + x_offset >>= bpp_shift; + } + + const u32 bytes_per_pixel = base_bpp << bpp_shift; const u32 height = dst_params.height; const u32 depth = dst_params.depth; const u32 block_height = dst_params.block_size.height; const u32 block_depth = dst_params.block_size.depth; const size_t dst_size = CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth); - const size_t dst_layer_size = - CalculateSize(true, bytes_per_pixel, width, height, 1, block_height, block_depth); - const size_t src_size = static_cast<size_t>(regs.pitch_in) * regs.line_count; if (read_buffer.size() < src_size) { @@ -188,32 +219,23 @@ void MaxwellDMA::CopyPitchToBlockLinear() { write_buffer.resize(dst_size); } + memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); if (Settings::IsGPULevelExtreme()) { - memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size); } else { - memory_manager.ReadBlockUnsafe(regs.offset_in, read_buffer.data(), src_size); memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); } // If the input is linear and the output is tiled, swizzle the input and copy it over. - if (regs.dst_params.block_size.depth > 0) { - ASSERT(dst_params.layer == 0); - SwizzleSliceToVoxel(regs.line_length_in, regs.line_count, regs.pitch_in, width, height, - bytes_per_pixel, block_height, block_depth, dst_params.origin.x, - dst_params.origin.y, write_buffer.data(), read_buffer.data()); - } else { - SwizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_in, width, bytes_per_pixel, - write_buffer.data() + dst_layer_size * dst_params.layer, read_buffer.data(), - block_height, dst_params.origin.x, dst_params.origin.y); - } + SwizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset, + dst_params.origin.y, x_elements, regs.line_count, block_height, block_depth, + regs.pitch_in); memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size); } void MaxwellDMA::FastCopyBlockLinearToPitch() { - const u32 bytes_per_pixel = - regs.launch_dma.remap_enable ? regs.pitch_out / regs.line_length_in : 1; + const u32 bytes_per_pixel = 1U; const size_t src_size = GOB_SIZE; const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count; u32 pos_x = regs.src_params.origin.x; @@ -239,9 +261,10 @@ void MaxwellDMA::FastCopyBlockLinearToPitch() { memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size); } - UnswizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_out, regs.src_params.width, - bytes_per_pixel, regs.src_params.block_size.height, pos_x, pos_y, - write_buffer.data(), read_buffer.data()); + UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, regs.src_params.width, + regs.src_params.height, 1, pos_x, pos_y, regs.line_length_in, regs.line_count, + regs.src_params.block_size.height, regs.src_params.block_size.depth, + regs.pitch_out); memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size); } @@ -249,16 +272,24 @@ void MaxwellDMA::FastCopyBlockLinearToPitch() { void MaxwellDMA::ReleaseSemaphore() { const auto type = regs.launch_dma.semaphore_type; const GPUVAddr address = regs.semaphore.address; + const u32 payload = regs.semaphore.payload; switch (type) { case LaunchDMA::SemaphoreType::NONE: break; - case LaunchDMA::SemaphoreType::RELEASE_ONE_WORD_SEMAPHORE: - memory_manager.Write<u32>(address, regs.semaphore.payload); + case LaunchDMA::SemaphoreType::RELEASE_ONE_WORD_SEMAPHORE: { + std::function<void()> operation( + [this, address, payload] { memory_manager.Write<u32>(address, payload); }); + rasterizer->SignalFence(std::move(operation)); break; - case LaunchDMA::SemaphoreType::RELEASE_FOUR_WORD_SEMAPHORE: - memory_manager.Write<u64>(address, static_cast<u64>(regs.semaphore.payload)); - memory_manager.Write<u64>(address + 8, system.GPU().GetTicks()); + } + case LaunchDMA::SemaphoreType::RELEASE_FOUR_WORD_SEMAPHORE: { + std::function<void()> operation([this, address, payload] { + memory_manager.Write<u64>(address + sizeof(u64), system.GPU().GetTicks()); + memory_manager.Write<u64>(address, payload); + }); + rasterizer->SignalFence(std::move(operation)); break; + } default: ASSERT_MSG(false, "Unknown semaphore type: {}", static_cast<u32>(type.Value())); } diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 074bac92c..bc48320ce 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -189,10 +189,16 @@ public: BitField<4, 3, Swizzle> dst_y; BitField<8, 3, Swizzle> dst_z; BitField<12, 3, Swizzle> dst_w; + BitField<0, 12, u32> dst_components_raw; BitField<16, 2, u32> component_size_minus_one; BitField<20, 2, u32> num_src_components_minus_one; BitField<24, 2, u32> num_dst_components_minus_one; }; + + Swizzle GetComponent(size_t i) const { + const u32 raw = dst_components_raw; + return static_cast<Swizzle>((raw >> (i * 3)) & 0x7); + } }; static_assert(sizeof(RemapConst) == 12); diff --git a/src/video_core/engines/puller.cpp b/src/video_core/engines/puller.cpp new file mode 100644 index 000000000..cca890792 --- /dev/null +++ b/src/video_core/engines/puller.cpp @@ -0,0 +1,306 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/assert.h" +#include "common/logging/log.h" +#include "common/settings.h" +#include "core/core.h" +#include "video_core/control/channel_state.h" +#include "video_core/dma_pusher.h" +#include "video_core/engines/fermi_2d.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/kepler_memory.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/engines/maxwell_dma.h" +#include "video_core/engines/puller.h" +#include "video_core/gpu.h" +#include "video_core/memory_manager.h" +#include "video_core/rasterizer_interface.h" + +namespace Tegra::Engines { + +Puller::Puller(GPU& gpu_, MemoryManager& memory_manager_, DmaPusher& dma_pusher_, + Control::ChannelState& channel_state_) + : gpu{gpu_}, memory_manager{memory_manager_}, dma_pusher{dma_pusher_}, channel_state{ + channel_state_} {} + +Puller::~Puller() = default; + +void Puller::ProcessBindMethod(const MethodCall& method_call) { + // Bind the current subchannel to the desired engine id. + LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel, + method_call.argument); + const auto engine_id = static_cast<EngineID>(method_call.argument); + bound_engines[method_call.subchannel] = static_cast<EngineID>(engine_id); + switch (engine_id) { + case EngineID::FERMI_TWOD_A: + dma_pusher.BindSubchannel(channel_state.fermi_2d.get(), method_call.subchannel); + break; + case EngineID::MAXWELL_B: + dma_pusher.BindSubchannel(channel_state.maxwell_3d.get(), method_call.subchannel); + break; + case EngineID::KEPLER_COMPUTE_B: + dma_pusher.BindSubchannel(channel_state.kepler_compute.get(), method_call.subchannel); + break; + case EngineID::MAXWELL_DMA_COPY_A: + dma_pusher.BindSubchannel(channel_state.maxwell_dma.get(), method_call.subchannel); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + dma_pusher.BindSubchannel(channel_state.kepler_memory.get(), method_call.subchannel); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine {:04X}", engine_id); + } +} + +void Puller::ProcessFenceActionMethod() { + switch (regs.fence_action.op) { + case Puller::FenceOperation::Acquire: + // UNIMPLEMENTED_MSG("Channel Scheduling pending."); + // WaitFence(regs.fence_action.syncpoint_id, regs.fence_value); + rasterizer->ReleaseFences(); + break; + case Puller::FenceOperation::Increment: + rasterizer->SignalSyncPoint(regs.fence_action.syncpoint_id); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented operation {}", regs.fence_action.op.Value()); + } +} + +void Puller::ProcessSemaphoreTriggerMethod() { + const auto semaphoreOperationMask = 0xF; + const auto op = + static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask); + if (op == GpuSemaphoreOperation::WriteLong) { + const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()}; + const u32 payload = regs.semaphore_sequence; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u64>(sequence_address + sizeof(u64), gpu.GetTicks()); + memory_manager.Write<u64>(sequence_address, payload); + }); + rasterizer->SignalFence(std::move(operation)); + } else { + do { + const u32 word{memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress())}; + regs.acquire_source = true; + regs.acquire_value = regs.semaphore_sequence; + if (op == GpuSemaphoreOperation::AcquireEqual) { + regs.acquire_active = true; + regs.acquire_mode = false; + if (word != regs.acquire_value) { + rasterizer->ReleaseFences(); + continue; + } + } else if (op == GpuSemaphoreOperation::AcquireGequal) { + regs.acquire_active = true; + regs.acquire_mode = true; + if (word < regs.acquire_value) { + rasterizer->ReleaseFences(); + continue; + } + } else if (op == GpuSemaphoreOperation::AcquireMask) { + if (word && regs.semaphore_sequence == 0) { + rasterizer->ReleaseFences(); + continue; + } + } else { + LOG_ERROR(HW_GPU, "Invalid semaphore operation"); + } + } while (false); + } +} + +void Puller::ProcessSemaphoreRelease() { + const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()}; + const u32 payload = regs.semaphore_release; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u32>(sequence_address, payload); + }); + rasterizer->SyncOperation(std::move(operation)); +} + +void Puller::ProcessSemaphoreAcquire() { + u32 word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress()); + const auto value = regs.semaphore_acquire; + while (word != value) { + regs.acquire_active = true; + regs.acquire_value = value; + std::this_thread::sleep_for(std::chrono::milliseconds(1)); + rasterizer->ReleaseFences(); + word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress()); + // TODO(kemathe73) figure out how to do the acquire_timeout + regs.acquire_mode = false; + regs.acquire_source = false; + } +} + +/// Calls a GPU puller method. +void Puller::CallPullerMethod(const MethodCall& method_call) { + regs.reg_array[method_call.method] = method_call.argument; + const auto method = static_cast<BufferMethods>(method_call.method); + + switch (method) { + case BufferMethods::BindObject: { + ProcessBindMethod(method_call); + break; + } + case BufferMethods::Nop: + case BufferMethods::SemaphoreAddressHigh: + case BufferMethods::SemaphoreAddressLow: + case BufferMethods::SemaphoreSequencePayload: + case BufferMethods::SyncpointPayload: + break; + case BufferMethods::WrcacheFlush: + case BufferMethods::RefCnt: + rasterizer->SignalReference(); + break; + case BufferMethods::SyncpointOperation: + ProcessFenceActionMethod(); + break; + case BufferMethods::WaitForIdle: + rasterizer->WaitForIdle(); + break; + case BufferMethods::SemaphoreOperation: { + ProcessSemaphoreTriggerMethod(); + break; + } + case BufferMethods::NonStallInterrupt: { + LOG_ERROR(HW_GPU, "Special puller engine method NonStallInterrupt not implemented"); + break; + } + case BufferMethods::MemOpA: { + LOG_ERROR(HW_GPU, "Memory Operation A"); + break; + } + case BufferMethods::MemOpB: { + // Implement this better. + rasterizer->InvalidateGPUCache(); + break; + } + case BufferMethods::MemOpC: + case BufferMethods::MemOpD: { + LOG_ERROR(HW_GPU, "Memory Operation C,D"); + break; + } + case BufferMethods::SemaphoreAcquire: { + ProcessSemaphoreAcquire(); + break; + } + case BufferMethods::SemaphoreRelease: { + ProcessSemaphoreRelease(); + break; + } + case BufferMethods::Yield: { + // TODO(Kmather73): Research and implement this method. + LOG_ERROR(HW_GPU, "Special puller engine method Yield not implemented"); + break; + } + default: + LOG_ERROR(HW_GPU, "Special puller engine method {:X} not implemented", method); + break; + } +} + +/// Calls a GPU engine method. +void Puller::CallEngineMethod(const MethodCall& method_call) { + const EngineID engine = bound_engines[method_call.subchannel]; + + switch (engine) { + case EngineID::FERMI_TWOD_A: + channel_state.fermi_2d->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::MAXWELL_B: + channel_state.maxwell_3d->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::KEPLER_COMPUTE_B: + channel_state.kepler_compute->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::MAXWELL_DMA_COPY_A: + channel_state.maxwell_dma->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + channel_state.kepler_memory->CallMethod(method_call.method, method_call.argument, + method_call.IsLastCall()); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine"); + } +} + +/// Calls a GPU engine multivalue method. +void Puller::CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending) { + const EngineID engine = bound_engines[subchannel]; + + switch (engine) { + case EngineID::FERMI_TWOD_A: + channel_state.fermi_2d->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::MAXWELL_B: + channel_state.maxwell_3d->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::KEPLER_COMPUTE_B: + channel_state.kepler_compute->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::MAXWELL_DMA_COPY_A: + channel_state.maxwell_dma->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + channel_state.kepler_memory->CallMultiMethod(method, base_start, amount, methods_pending); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine"); + } +} + +/// Calls a GPU method. +void Puller::CallMethod(const MethodCall& method_call) { + LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method_call.method, + method_call.subchannel); + + ASSERT(method_call.subchannel < bound_engines.size()); + + if (ExecuteMethodOnEngine(method_call.method)) { + CallEngineMethod(method_call); + } else { + CallPullerMethod(method_call); + } +} + +/// Calls a GPU multivalue method. +void Puller::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending) { + LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, subchannel); + + ASSERT(subchannel < bound_engines.size()); + + if (ExecuteMethodOnEngine(method)) { + CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending); + } else { + for (std::size_t i = 0; i < amount; i++) { + CallPullerMethod(MethodCall{ + method, + base_start[i], + subchannel, + methods_pending - static_cast<u32>(i), + }); + } + } +} + +void Puller::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; +} + +/// Determines where the method should be executed. +[[nodiscard]] bool Puller::ExecuteMethodOnEngine(u32 method) { + const auto buffer_method = static_cast<BufferMethods>(method); + return buffer_method >= BufferMethods::NonPullerMethods; +} + +} // namespace Tegra::Engines diff --git a/src/video_core/engines/puller.h b/src/video_core/engines/puller.h new file mode 100644 index 000000000..d4175ee94 --- /dev/null +++ b/src/video_core/engines/puller.h @@ -0,0 +1,177 @@ +// SPDX-FileCopyrightText: 2022 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <array> +#include <cstddef> +#include <vector> +#include "common/bit_field.h" +#include "common/common_funcs.h" +#include "common/common_types.h" +#include "video_core/engines/engine_interface.h" + +namespace Core { +class System; +} + +namespace Tegra { +class MemoryManager; +class DmaPusher; + +enum class EngineID { + FERMI_TWOD_A = 0x902D, // 2D Engine + MAXWELL_B = 0xB197, // 3D Engine + KEPLER_COMPUTE_B = 0xB1C0, + KEPLER_INLINE_TO_MEMORY_B = 0xA140, + MAXWELL_DMA_COPY_A = 0xB0B5, +}; + +namespace Control { +struct ChannelState; +} +} // namespace Tegra + +namespace VideoCore { +class RasterizerInterface; +} + +namespace Tegra::Engines { + +class Puller final { +public: + struct MethodCall { + u32 method{}; + u32 argument{}; + u32 subchannel{}; + u32 method_count{}; + + explicit MethodCall(u32 method_, u32 argument_, u32 subchannel_ = 0, u32 method_count_ = 0) + : method(method_), argument(argument_), subchannel(subchannel_), + method_count(method_count_) {} + + [[nodiscard]] bool IsLastCall() const { + return method_count <= 1; + } + }; + + enum class FenceOperation : u32 { + Acquire = 0, + Increment = 1, + }; + + union FenceAction { + u32 raw; + BitField<0, 1, FenceOperation> op; + BitField<8, 24, u32> syncpoint_id; + }; + + explicit Puller(GPU& gpu_, MemoryManager& memory_manager_, DmaPusher& dma_pusher, + Control::ChannelState& channel_state); + ~Puller(); + + void CallMethod(const MethodCall& method_call); + + void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending); + + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + + void CallPullerMethod(const MethodCall& method_call); + + void CallEngineMethod(const MethodCall& method_call); + + void CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending); + +private: + Tegra::GPU& gpu; + + MemoryManager& memory_manager; + DmaPusher& dma_pusher; + Control::ChannelState& channel_state; + VideoCore::RasterizerInterface* rasterizer = nullptr; + + static constexpr std::size_t NUM_REGS = 0x800; + struct Regs { + static constexpr size_t NUM_REGS = 0x40; + + union { + struct { + INSERT_PADDING_WORDS_NOINIT(0x4); + struct { + u32 address_high; + u32 address_low; + + [[nodiscard]] GPUVAddr SemaphoreAddress() const { + return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | + address_low); + } + } semaphore_address; + + u32 semaphore_sequence; + u32 semaphore_trigger; + INSERT_PADDING_WORDS_NOINIT(0xC); + + // The pusher and the puller share the reference counter, the pusher only has read + // access + u32 reference_count; + INSERT_PADDING_WORDS_NOINIT(0x5); + + u32 semaphore_acquire; + u32 semaphore_release; + u32 fence_value; + FenceAction fence_action; + INSERT_PADDING_WORDS_NOINIT(0xE2); + + // Puller state + u32 acquire_mode; + u32 acquire_source; + u32 acquire_active; + u32 acquire_timeout; + u32 acquire_value; + }; + std::array<u32, NUM_REGS> reg_array; + }; + } regs{}; + + void ProcessBindMethod(const MethodCall& method_call); + void ProcessFenceActionMethod(); + void ProcessSemaphoreAcquire(); + void ProcessSemaphoreRelease(); + void ProcessSemaphoreTriggerMethod(); + [[nodiscard]] bool ExecuteMethodOnEngine(u32 method); + + /// Mapping of command subchannels to their bound engine ids + std::array<EngineID, 8> bound_engines{}; + + enum class GpuSemaphoreOperation { + AcquireEqual = 0x1, + WriteLong = 0x2, + AcquireGequal = 0x4, + AcquireMask = 0x8, + }; + +#define ASSERT_REG_POSITION(field_name, position) \ + static_assert(offsetof(Regs, field_name) == position * 4, \ + "Field " #field_name " has invalid position") + + ASSERT_REG_POSITION(semaphore_address, 0x4); + ASSERT_REG_POSITION(semaphore_sequence, 0x6); + ASSERT_REG_POSITION(semaphore_trigger, 0x7); + ASSERT_REG_POSITION(reference_count, 0x14); + ASSERT_REG_POSITION(semaphore_acquire, 0x1A); + ASSERT_REG_POSITION(semaphore_release, 0x1B); + ASSERT_REG_POSITION(fence_value, 0x1C); + ASSERT_REG_POSITION(fence_action, 0x1D); + + ASSERT_REG_POSITION(acquire_mode, 0x100); + ASSERT_REG_POSITION(acquire_source, 0x101); + ASSERT_REG_POSITION(acquire_active, 0x102); + ASSERT_REG_POSITION(acquire_timeout, 0x103); + ASSERT_REG_POSITION(acquire_value, 0x104); + +#undef ASSERT_REG_POSITION +}; + +} // namespace Tegra::Engines diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h index 1e9832ddd..c390ac91b 100644 --- a/src/video_core/fence_manager.h +++ b/src/video_core/fence_manager.h @@ -4,40 +4,24 @@ #pragma once #include <algorithm> +#include <cstring> +#include <deque> +#include <functional> +#include <memory> #include <queue> #include "common/common_types.h" #include "video_core/delayed_destruction_ring.h" #include "video_core/gpu.h" -#include "video_core/memory_manager.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/syncpoint_manager.h" #include "video_core/rasterizer_interface.h" namespace VideoCommon { class FenceBase { public: - explicit FenceBase(u32 payload_, bool is_stubbed_) - : address{}, payload{payload_}, is_semaphore{false}, is_stubbed{is_stubbed_} {} - - explicit FenceBase(GPUVAddr address_, u32 payload_, bool is_stubbed_) - : address{address_}, payload{payload_}, is_semaphore{true}, is_stubbed{is_stubbed_} {} - - GPUVAddr GetAddress() const { - return address; - } - - u32 GetPayload() const { - return payload; - } - - bool IsSemaphore() const { - return is_semaphore; - } - -private: - GPUVAddr address; - u32 payload; - bool is_semaphore; + explicit FenceBase(bool is_stubbed_) : is_stubbed{is_stubbed_} {} protected: bool is_stubbed; @@ -57,30 +41,28 @@ public: buffer_cache.AccumulateFlushes(); } - void SignalSemaphore(GPUVAddr addr, u32 value) { + void SyncOperation(std::function<void()>&& func) { + uncommitted_operations.emplace_back(std::move(func)); + } + + void SignalFence(std::function<void()>&& func) { TryReleasePendingFences(); const bool should_flush = ShouldFlush(); CommitAsyncFlushes(); - TFence new_fence = CreateFence(addr, value, !should_flush); + uncommitted_operations.emplace_back(std::move(func)); + CommitOperations(); + TFence new_fence = CreateFence(!should_flush); fences.push(new_fence); QueueFence(new_fence); if (should_flush) { rasterizer.FlushCommands(); } - rasterizer.SyncGuestHost(); } void SignalSyncPoint(u32 value) { - TryReleasePendingFences(); - const bool should_flush = ShouldFlush(); - CommitAsyncFlushes(); - TFence new_fence = CreateFence(value, !should_flush); - fences.push(new_fence); - QueueFence(new_fence); - if (should_flush) { - rasterizer.FlushCommands(); - } - rasterizer.SyncGuestHost(); + syncpoint_manager.IncrementGuest(value); + std::function<void()> func([this, value] { syncpoint_manager.IncrementHost(value); }); + SignalFence(std::move(func)); } void WaitPendingFences() { @@ -90,11 +72,10 @@ public: WaitFence(current_fence); } PopAsyncFlushes(); - if (current_fence->IsSemaphore()) { - gpu_memory.template Write<u32>(current_fence->GetAddress(), - current_fence->GetPayload()); - } else { - gpu.IncrementSyncPoint(current_fence->GetPayload()); + auto operations = std::move(pending_operations.front()); + pending_operations.pop_front(); + for (auto& operation : operations) { + operation(); } PopFence(); } @@ -104,16 +85,14 @@ protected: explicit FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, TTextureCache& texture_cache_, TTBufferCache& buffer_cache_, TQueryCache& query_cache_) - : rasterizer{rasterizer_}, gpu{gpu_}, gpu_memory{gpu.MemoryManager()}, + : rasterizer{rasterizer_}, gpu{gpu_}, syncpoint_manager{gpu.Host1x().GetSyncpointManager()}, texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, query_cache{query_cache_} {} virtual ~FenceManager() = default; - /// Creates a Sync Point Fence Interface, does not create a backend fence if 'is_stubbed' is + /// Creates a Fence Interface, does not create a backend fence if 'is_stubbed' is /// true - virtual TFence CreateFence(u32 value, bool is_stubbed) = 0; - /// Creates a Semaphore Fence Interface, does not create a backend fence if 'is_stubbed' is true - virtual TFence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) = 0; + virtual TFence CreateFence(bool is_stubbed) = 0; /// Queues a fence into the backend if the fence isn't stubbed. virtual void QueueFence(TFence& fence) = 0; /// Notifies that the backend fence has been signaled/reached in host GPU. @@ -123,7 +102,7 @@ protected: VideoCore::RasterizerInterface& rasterizer; Tegra::GPU& gpu; - Tegra::MemoryManager& gpu_memory; + Tegra::Host1x::SyncpointManager& syncpoint_manager; TTextureCache& texture_cache; TTBufferCache& buffer_cache; TQueryCache& query_cache; @@ -136,11 +115,10 @@ private: return; } PopAsyncFlushes(); - if (current_fence->IsSemaphore()) { - gpu_memory.template Write<u32>(current_fence->GetAddress(), - current_fence->GetPayload()); - } else { - gpu.IncrementSyncPoint(current_fence->GetPayload()); + auto operations = std::move(pending_operations.front()); + pending_operations.pop_front(); + for (auto& operation : operations) { + operation(); } PopFence(); } @@ -159,16 +137,20 @@ private: } void PopAsyncFlushes() { - std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - texture_cache.PopAsyncFlushes(); - buffer_cache.PopAsyncFlushes(); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.PopAsyncFlushes(); + buffer_cache.PopAsyncFlushes(); + } query_cache.PopAsyncFlushes(); } void CommitAsyncFlushes() { - std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - texture_cache.CommitAsyncFlushes(); - buffer_cache.CommitAsyncFlushes(); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.CommitAsyncFlushes(); + buffer_cache.CommitAsyncFlushes(); + } query_cache.CommitAsyncFlushes(); } @@ -177,7 +159,13 @@ private: fences.pop(); } + void CommitOperations() { + pending_operations.emplace_back(std::move(uncommitted_operations)); + } + std::queue<TFence> fences; + std::deque<std::function<void()>> uncommitted_operations; + std::deque<std::deque<std::function<void()>>> pending_operations; DelayedDestructionRing<TFence, 6> delayed_destruction_ring; }; diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 33431f2a0..28b38273e 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -14,10 +14,11 @@ #include "core/core.h" #include "core/core_timing.h" #include "core/frontend/emu_window.h" -#include "core/hardware_interrupt_manager.h" #include "core/hle/service/nvdrv/nvdata.h" #include "core/perf_stats.h" #include "video_core/cdma_pusher.h" +#include "video_core/control/channel_state.h" +#include "video_core/control/scheduler.h" #include "video_core/dma_pusher.h" #include "video_core/engines/fermi_2d.h" #include "video_core/engines/kepler_compute.h" @@ -26,75 +27,64 @@ #include "video_core/engines/maxwell_dma.h" #include "video_core/gpu.h" #include "video_core/gpu_thread.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/syncpoint_manager.h" #include "video_core/memory_manager.h" #include "video_core/renderer_base.h" #include "video_core/shader_notify.h" namespace Tegra { -MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192)); - struct GPU::Impl { explicit Impl(GPU& gpu_, Core::System& system_, bool is_async_, bool use_nvdec_) - : gpu{gpu_}, system{system_}, memory_manager{std::make_unique<Tegra::MemoryManager>( - system)}, - dma_pusher{std::make_unique<Tegra::DmaPusher>(system, gpu)}, use_nvdec{use_nvdec_}, - maxwell_3d{std::make_unique<Engines::Maxwell3D>(system, *memory_manager)}, - fermi_2d{std::make_unique<Engines::Fermi2D>()}, - kepler_compute{std::make_unique<Engines::KeplerCompute>(system, *memory_manager)}, - maxwell_dma{std::make_unique<Engines::MaxwellDMA>(system, *memory_manager)}, - kepler_memory{std::make_unique<Engines::KeplerMemory>(system, *memory_manager)}, + : gpu{gpu_}, system{system_}, host1x{system.Host1x()}, use_nvdec{use_nvdec_}, shader_notify{std::make_unique<VideoCore::ShaderNotify>()}, is_async{is_async_}, - gpu_thread{system_, is_async_} {} + gpu_thread{system_, is_async_}, scheduler{std::make_unique<Control::Scheduler>(gpu)} {} ~Impl() = default; - /// Binds a renderer to the GPU. - void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) { - renderer = std::move(renderer_); - rasterizer = renderer->ReadRasterizer(); - - memory_manager->BindRasterizer(rasterizer); - maxwell_3d->BindRasterizer(rasterizer); - fermi_2d->BindRasterizer(rasterizer); - kepler_compute->BindRasterizer(rasterizer); - kepler_memory->BindRasterizer(rasterizer); - maxwell_dma->BindRasterizer(rasterizer); + std::shared_ptr<Control::ChannelState> CreateChannel(s32 channel_id) { + auto channel_state = std::make_shared<Tegra::Control::ChannelState>(channel_id); + channels.emplace(channel_id, channel_state); + scheduler->DeclareChannel(channel_state); + return channel_state; } - /// Calls a GPU method. - void CallMethod(const GPU::MethodCall& method_call) { - LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method_call.method, - method_call.subchannel); + void BindChannel(s32 channel_id) { + if (bound_channel == channel_id) { + return; + } + auto it = channels.find(channel_id); + ASSERT(it != channels.end()); + bound_channel = channel_id; + current_channel = it->second.get(); - ASSERT(method_call.subchannel < bound_engines.size()); + rasterizer->BindChannel(*current_channel); + } - if (ExecuteMethodOnEngine(method_call.method)) { - CallEngineMethod(method_call); - } else { - CallPullerMethod(method_call); - } + std::shared_ptr<Control::ChannelState> AllocateChannel() { + return CreateChannel(new_channel_id++); } - /// Calls a GPU multivalue method. - void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending) { - LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, subchannel); + void InitChannel(Control::ChannelState& to_init) { + to_init.Init(system, gpu); + to_init.BindRasterizer(rasterizer); + rasterizer->InitializeChannel(to_init); + } - ASSERT(subchannel < bound_engines.size()); + void InitAddressSpace(Tegra::MemoryManager& memory_manager) { + memory_manager.BindRasterizer(rasterizer); + } - if (ExecuteMethodOnEngine(method)) { - CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending); - } else { - for (std::size_t i = 0; i < amount; i++) { - CallPullerMethod(GPU::MethodCall{ - method, - base_start[i], - subchannel, - methods_pending - static_cast<u32>(i), - }); - } - } + void ReleaseChannel(Control::ChannelState& to_release) { + UNIMPLEMENTED(); + } + + /// Binds a renderer to the GPU. + void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) { + renderer = std::move(renderer_); + rasterizer = renderer->ReadRasterizer(); + host1x.MemoryManager().BindRasterizer(rasterizer); } /// Flush all current written commands into the host GPU for execution. @@ -103,85 +93,82 @@ struct GPU::Impl { } /// Synchronizes CPU writes with Host GPU memory. - void SyncGuestHost() { - rasterizer->SyncGuestHost(); + void InvalidateGPUCache() { + rasterizer->InvalidateGPUCache(); } /// Signal the ending of command list. void OnCommandListEnd() { - if (is_async) { - // This command only applies to asynchronous GPU mode - gpu_thread.OnCommandListEnd(); - } + gpu_thread.OnCommandListEnd(); } /// Request a host GPU memory flush from the CPU. - [[nodiscard]] u64 RequestFlush(VAddr addr, std::size_t size) { - std::unique_lock lck{flush_request_mutex}; - const u64 fence = ++last_flush_fence; - flush_requests.emplace_back(fence, addr, size); + template <typename Func> + [[nodiscard]] u64 RequestSyncOperation(Func&& action) { + std::unique_lock lck{sync_request_mutex}; + const u64 fence = ++last_sync_fence; + sync_requests.emplace_back(action); return fence; } /// Obtains current flush request fence id. - [[nodiscard]] u64 CurrentFlushRequestFence() const { - return current_flush_fence.load(std::memory_order_relaxed); + [[nodiscard]] u64 CurrentSyncRequestFence() const { + return current_sync_fence.load(std::memory_order_relaxed); + } + + void WaitForSyncOperation(const u64 fence) { + std::unique_lock lck{sync_request_mutex}; + sync_request_cv.wait(lck, [this, fence] { return CurrentSyncRequestFence() >= fence; }); } /// Tick pending requests within the GPU. void TickWork() { - std::unique_lock lck{flush_request_mutex}; - while (!flush_requests.empty()) { - auto& request = flush_requests.front(); - const u64 fence = request.fence; - const VAddr addr = request.addr; - const std::size_t size = request.size; - flush_requests.pop_front(); - flush_request_mutex.unlock(); - rasterizer->FlushRegion(addr, size); - current_flush_fence.store(fence); - flush_request_mutex.lock(); + std::unique_lock lck{sync_request_mutex}; + while (!sync_requests.empty()) { + auto request = std::move(sync_requests.front()); + sync_requests.pop_front(); + sync_request_mutex.unlock(); + request(); + current_sync_fence.fetch_add(1, std::memory_order_release); + sync_request_mutex.lock(); + sync_request_cv.notify_all(); } } /// Returns a reference to the Maxwell3D GPU engine. [[nodiscard]] Engines::Maxwell3D& Maxwell3D() { - return *maxwell_3d; + ASSERT(current_channel); + return *current_channel->maxwell_3d; } /// Returns a const reference to the Maxwell3D GPU engine. [[nodiscard]] const Engines::Maxwell3D& Maxwell3D() const { - return *maxwell_3d; + ASSERT(current_channel); + return *current_channel->maxwell_3d; } /// Returns a reference to the KeplerCompute GPU engine. [[nodiscard]] Engines::KeplerCompute& KeplerCompute() { - return *kepler_compute; + ASSERT(current_channel); + return *current_channel->kepler_compute; } /// Returns a reference to the KeplerCompute GPU engine. [[nodiscard]] const Engines::KeplerCompute& KeplerCompute() const { - return *kepler_compute; - } - - /// Returns a reference to the GPU memory manager. - [[nodiscard]] Tegra::MemoryManager& MemoryManager() { - return *memory_manager; - } - - /// Returns a const reference to the GPU memory manager. - [[nodiscard]] const Tegra::MemoryManager& MemoryManager() const { - return *memory_manager; + ASSERT(current_channel); + return *current_channel->kepler_compute; } /// Returns a reference to the GPU DMA pusher. [[nodiscard]] Tegra::DmaPusher& DmaPusher() { - return *dma_pusher; + ASSERT(current_channel); + return *current_channel->dma_pusher; } /// Returns a const reference to the GPU DMA pusher. [[nodiscard]] const Tegra::DmaPusher& DmaPusher() const { - return *dma_pusher; + ASSERT(current_channel); + return *current_channel->dma_pusher; } /// Returns a reference to the underlying renderer. @@ -204,77 +191,6 @@ struct GPU::Impl { return *shader_notify; } - /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame. - void WaitFence(u32 syncpoint_id, u32 value) { - // Synced GPU, is always in sync - if (!is_async) { - return; - } - if (syncpoint_id == UINT32_MAX) { - // TODO: Research what this does. - LOG_ERROR(HW_GPU, "Waiting for syncpoint -1 not implemented"); - return; - } - MICROPROFILE_SCOPE(GPU_wait); - std::unique_lock lock{sync_mutex}; - sync_cv.wait(lock, [=, this] { - if (shutting_down.load(std::memory_order_relaxed)) { - // We're shutting down, ensure no threads continue to wait for the next syncpoint - return true; - } - return syncpoints.at(syncpoint_id).load() >= value; - }); - } - - void IncrementSyncPoint(u32 syncpoint_id) { - auto& syncpoint = syncpoints.at(syncpoint_id); - syncpoint++; - std::scoped_lock lock{sync_mutex}; - sync_cv.notify_all(); - auto& interrupt = syncpt_interrupts.at(syncpoint_id); - if (!interrupt.empty()) { - u32 value = syncpoint.load(); - auto it = interrupt.begin(); - while (it != interrupt.end()) { - if (value >= *it) { - TriggerCpuInterrupt(syncpoint_id, *it); - it = interrupt.erase(it); - continue; - } - it++; - } - } - } - - [[nodiscard]] u32 GetSyncpointValue(u32 syncpoint_id) const { - return syncpoints.at(syncpoint_id).load(); - } - - void RegisterSyncptInterrupt(u32 syncpoint_id, u32 value) { - std::scoped_lock lock{sync_mutex}; - auto& interrupt = syncpt_interrupts.at(syncpoint_id); - bool contains = std::any_of(interrupt.begin(), interrupt.end(), - [value](u32 in_value) { return in_value == value; }); - if (contains) { - return; - } - interrupt.emplace_back(value); - } - - [[nodiscard]] bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value) { - std::scoped_lock lock{sync_mutex}; - auto& interrupt = syncpt_interrupts.at(syncpoint_id); - const auto iter = - std::find_if(interrupt.begin(), interrupt.end(), - [value](u32 interrupt_value) { return value == interrupt_value; }); - - if (iter == interrupt.end()) { - return false; - } - interrupt.erase(iter); - return true; - } - [[nodiscard]] u64 GetTicks() const { // This values were reversed engineered by fincs from NVN // The gpu clock is reported in units of 385/625 nanoseconds @@ -306,7 +222,7 @@ struct GPU::Impl { /// This can be used to launch any necessary threads and register any necessary /// core timing events. void Start() { - gpu_thread.StartThread(*renderer, renderer->Context(), *dma_pusher); + gpu_thread.StartThread(*renderer, renderer->Context(), *scheduler); cpu_context = renderer->GetRenderWindow().CreateSharedContext(); cpu_context->MakeCurrent(); } @@ -328,8 +244,8 @@ struct GPU::Impl { } /// Push GPU command entries to be processed - void PushGPUEntries(Tegra::CommandList&& entries) { - gpu_thread.SubmitList(std::move(entries)); + void PushGPUEntries(s32 channel, Tegra::CommandList&& entries) { + gpu_thread.SubmitList(channel, std::move(entries)); } /// Push GPU command buffer entries to be processed @@ -339,7 +255,7 @@ struct GPU::Impl { } if (!cdma_pushers.contains(id)) { - cdma_pushers.insert_or_assign(id, std::make_unique<Tegra::CDmaPusher>(gpu)); + cdma_pushers.insert_or_assign(id, std::make_unique<Tegra::CDmaPusher>(host1x)); } // SubmitCommandBuffer would make the nvdec operations async, this is not currently working @@ -376,308 +292,55 @@ struct GPU::Impl { gpu_thread.FlushAndInvalidateRegion(addr, size); } - void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const { - auto& interrupt_manager = system.InterruptManager(); - interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value); - } - - void ProcessBindMethod(const GPU::MethodCall& method_call) { - // Bind the current subchannel to the desired engine id. - LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel, - method_call.argument); - const auto engine_id = static_cast<EngineID>(method_call.argument); - bound_engines[method_call.subchannel] = static_cast<EngineID>(engine_id); - switch (engine_id) { - case EngineID::FERMI_TWOD_A: - dma_pusher->BindSubchannel(fermi_2d.get(), method_call.subchannel); - break; - case EngineID::MAXWELL_B: - dma_pusher->BindSubchannel(maxwell_3d.get(), method_call.subchannel); - break; - case EngineID::KEPLER_COMPUTE_B: - dma_pusher->BindSubchannel(kepler_compute.get(), method_call.subchannel); - break; - case EngineID::MAXWELL_DMA_COPY_A: - dma_pusher->BindSubchannel(maxwell_dma.get(), method_call.subchannel); - break; - case EngineID::KEPLER_INLINE_TO_MEMORY_B: - dma_pusher->BindSubchannel(kepler_memory.get(), method_call.subchannel); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented engine {:04X}", engine_id); - } - } - - void ProcessFenceActionMethod() { - switch (regs.fence_action.op) { - case GPU::FenceOperation::Acquire: - WaitFence(regs.fence_action.syncpoint_id, regs.fence_value); - break; - case GPU::FenceOperation::Increment: - IncrementSyncPoint(regs.fence_action.syncpoint_id); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented operation {}", regs.fence_action.op.Value()); - } - } - - void ProcessWaitForInterruptMethod() { - // TODO(bunnei) ImplementMe - LOG_WARNING(HW_GPU, "(STUBBED) called"); - } - - void ProcessSemaphoreTriggerMethod() { - const auto semaphoreOperationMask = 0xF; - const auto op = - static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask); - if (op == GpuSemaphoreOperation::WriteLong) { - struct Block { - u32 sequence; - u32 zeros = 0; - u64 timestamp; - }; - - Block block{}; - block.sequence = regs.semaphore_sequence; - // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of - // CoreTiming - block.timestamp = GetTicks(); - memory_manager->WriteBlock(regs.semaphore_address.SemaphoreAddress(), &block, - sizeof(block)); - } else { - const u32 word{memory_manager->Read<u32>(regs.semaphore_address.SemaphoreAddress())}; - if ((op == GpuSemaphoreOperation::AcquireEqual && word == regs.semaphore_sequence) || - (op == GpuSemaphoreOperation::AcquireGequal && - static_cast<s32>(word - regs.semaphore_sequence) > 0) || - (op == GpuSemaphoreOperation::AcquireMask && (word & regs.semaphore_sequence))) { - // Nothing to do in this case + void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer, + std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) { + size_t current_request_counter{}; + { + std::unique_lock<std::mutex> lk(request_swap_mutex); + if (free_swap_counters.empty()) { + current_request_counter = request_swap_counters.size(); + request_swap_counters.emplace_back(num_fences); } else { - regs.acquire_source = true; - regs.acquire_value = regs.semaphore_sequence; - if (op == GpuSemaphoreOperation::AcquireEqual) { - regs.acquire_active = true; - regs.acquire_mode = false; - } else if (op == GpuSemaphoreOperation::AcquireGequal) { - regs.acquire_active = true; - regs.acquire_mode = true; - } else if (op == GpuSemaphoreOperation::AcquireMask) { - // TODO(kemathe) The acquire mask operation waits for a value that, ANDed with - // semaphore_sequence, gives a non-0 result - LOG_ERROR(HW_GPU, "Invalid semaphore operation AcquireMask not implemented"); - } else { - LOG_ERROR(HW_GPU, "Invalid semaphore operation"); - } + current_request_counter = free_swap_counters.front(); + request_swap_counters[current_request_counter] = num_fences; + free_swap_counters.pop_front(); } } - } - - void ProcessSemaphoreRelease() { - memory_manager->Write<u32>(regs.semaphore_address.SemaphoreAddress(), - regs.semaphore_release); - } - - void ProcessSemaphoreAcquire() { - const u32 word = memory_manager->Read<u32>(regs.semaphore_address.SemaphoreAddress()); - const auto value = regs.semaphore_acquire; - if (word != value) { - regs.acquire_active = true; - regs.acquire_value = value; - // TODO(kemathe73) figure out how to do the acquire_timeout - regs.acquire_mode = false; - regs.acquire_source = false; - } - } - - /// Calls a GPU puller method. - void CallPullerMethod(const GPU::MethodCall& method_call) { - regs.reg_array[method_call.method] = method_call.argument; - const auto method = static_cast<BufferMethods>(method_call.method); - - switch (method) { - case BufferMethods::BindObject: { - ProcessBindMethod(method_call); - break; - } - case BufferMethods::Nop: - case BufferMethods::SemaphoreAddressHigh: - case BufferMethods::SemaphoreAddressLow: - case BufferMethods::SemaphoreSequence: - break; - case BufferMethods::UnkCacheFlush: - rasterizer->SyncGuestHost(); - break; - case BufferMethods::WrcacheFlush: - rasterizer->SignalReference(); - break; - case BufferMethods::FenceValue: - break; - case BufferMethods::RefCnt: - rasterizer->SignalReference(); - break; - case BufferMethods::FenceAction: - ProcessFenceActionMethod(); - break; - case BufferMethods::WaitForInterrupt: - rasterizer->WaitForIdle(); - break; - case BufferMethods::SemaphoreTrigger: { - ProcessSemaphoreTriggerMethod(); - break; - } - case BufferMethods::NotifyIntr: { - // TODO(Kmather73): Research and implement this method. - LOG_ERROR(HW_GPU, "Special puller engine method NotifyIntr not implemented"); - break; - } - case BufferMethods::Unk28: { - // TODO(Kmather73): Research and implement this method. - LOG_ERROR(HW_GPU, "Special puller engine method Unk28 not implemented"); - break; - } - case BufferMethods::SemaphoreAcquire: { - ProcessSemaphoreAcquire(); - break; - } - case BufferMethods::SemaphoreRelease: { - ProcessSemaphoreRelease(); - break; - } - case BufferMethods::Yield: { - // TODO(Kmather73): Research and implement this method. - LOG_ERROR(HW_GPU, "Special puller engine method Yield not implemented"); - break; - } - default: - LOG_ERROR(HW_GPU, "Special puller engine method {:X} not implemented", method); - break; - } - } - - /// Calls a GPU engine method. - void CallEngineMethod(const GPU::MethodCall& method_call) { - const EngineID engine = bound_engines[method_call.subchannel]; - - switch (engine) { - case EngineID::FERMI_TWOD_A: - fermi_2d->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::MAXWELL_B: - maxwell_3d->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::KEPLER_COMPUTE_B: - kepler_compute->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::MAXWELL_DMA_COPY_A: - maxwell_dma->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - case EngineID::KEPLER_INLINE_TO_MEMORY_B: - kepler_memory->CallMethod(method_call.method, method_call.argument, - method_call.IsLastCall()); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented engine"); - } - } - - /// Calls a GPU engine multivalue method. - void CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending) { - const EngineID engine = bound_engines[subchannel]; - - switch (engine) { - case EngineID::FERMI_TWOD_A: - fermi_2d->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::MAXWELL_B: - maxwell_3d->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::KEPLER_COMPUTE_B: - kepler_compute->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::MAXWELL_DMA_COPY_A: - maxwell_dma->CallMultiMethod(method, base_start, amount, methods_pending); - break; - case EngineID::KEPLER_INLINE_TO_MEMORY_B: - kepler_memory->CallMultiMethod(method, base_start, amount, methods_pending); - break; - default: - UNIMPLEMENTED_MSG("Unimplemented engine"); - } - } - - /// Determines where the method should be executed. - [[nodiscard]] bool ExecuteMethodOnEngine(u32 method) { - const auto buffer_method = static_cast<BufferMethods>(method); - return buffer_method >= BufferMethods::NonPullerMethods; - } - - struct Regs { - static constexpr size_t NUM_REGS = 0x40; - - union { - struct { - INSERT_PADDING_WORDS_NOINIT(0x4); - struct { - u32 address_high; - u32 address_low; - - [[nodiscard]] GPUVAddr SemaphoreAddress() const { - return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | - address_low); + const auto wait_fence = + RequestSyncOperation([this, current_request_counter, framebuffer, fences, num_fences] { + auto& syncpoint_manager = host1x.GetSyncpointManager(); + if (num_fences == 0) { + renderer->SwapBuffers(framebuffer); + } + const auto executer = [this, current_request_counter, + framebuffer_copy = *framebuffer]() { + { + std::unique_lock<std::mutex> lk(request_swap_mutex); + if (--request_swap_counters[current_request_counter] != 0) { + return; + } + free_swap_counters.push_back(current_request_counter); } - } semaphore_address; - - u32 semaphore_sequence; - u32 semaphore_trigger; - INSERT_PADDING_WORDS_NOINIT(0xC); - - // The pusher and the puller share the reference counter, the pusher only has read - // access - u32 reference_count; - INSERT_PADDING_WORDS_NOINIT(0x5); - - u32 semaphore_acquire; - u32 semaphore_release; - u32 fence_value; - GPU::FenceAction fence_action; - INSERT_PADDING_WORDS_NOINIT(0xE2); - - // Puller state - u32 acquire_mode; - u32 acquire_source; - u32 acquire_active; - u32 acquire_timeout; - u32 acquire_value; - }; - std::array<u32, NUM_REGS> reg_array; - }; - } regs{}; + renderer->SwapBuffers(&framebuffer_copy); + }; + for (size_t i = 0; i < num_fences; i++) { + syncpoint_manager.RegisterGuestAction(fences[i].id, fences[i].value, executer); + } + }); + gpu_thread.TickGPU(); + WaitForSyncOperation(wait_fence); + } GPU& gpu; Core::System& system; - std::unique_ptr<Tegra::MemoryManager> memory_manager; - std::unique_ptr<Tegra::DmaPusher> dma_pusher; + Host1x::Host1x& host1x; + std::map<u32, std::unique_ptr<Tegra::CDmaPusher>> cdma_pushers; std::unique_ptr<VideoCore::RendererBase> renderer; VideoCore::RasterizerInterface* rasterizer = nullptr; const bool use_nvdec; - /// Mapping of command subchannels to their bound engine ids - std::array<EngineID, 8> bound_engines{}; - /// 3D engine - std::unique_ptr<Engines::Maxwell3D> maxwell_3d; - /// 2D engine - std::unique_ptr<Engines::Fermi2D> fermi_2d; - /// Compute engine - std::unique_ptr<Engines::KeplerCompute> kepler_compute; - /// DMA engine - std::unique_ptr<Engines::MaxwellDMA> maxwell_dma; - /// Inline memory engine - std::unique_ptr<Engines::KeplerMemory> kepler_memory; + s32 new_channel_id{1}; /// Shader build notifier std::unique_ptr<VideoCore::ShaderNotify> shader_notify; /// When true, we are about to shut down emulation session, so terminate outstanding tasks @@ -692,51 +355,25 @@ struct GPU::Impl { std::condition_variable sync_cv; - struct FlushRequest { - explicit FlushRequest(u64 fence_, VAddr addr_, std::size_t size_) - : fence{fence_}, addr{addr_}, size{size_} {} - u64 fence; - VAddr addr; - std::size_t size; - }; - - std::list<FlushRequest> flush_requests; - std::atomic<u64> current_flush_fence{}; - u64 last_flush_fence{}; - std::mutex flush_request_mutex; + std::list<std::function<void()>> sync_requests; + std::atomic<u64> current_sync_fence{}; + u64 last_sync_fence{}; + std::mutex sync_request_mutex; + std::condition_variable sync_request_cv; const bool is_async; VideoCommon::GPUThread::ThreadManager gpu_thread; std::unique_ptr<Core::Frontend::GraphicsContext> cpu_context; -#define ASSERT_REG_POSITION(field_name, position) \ - static_assert(offsetof(Regs, field_name) == position * 4, \ - "Field " #field_name " has invalid position") - - ASSERT_REG_POSITION(semaphore_address, 0x4); - ASSERT_REG_POSITION(semaphore_sequence, 0x6); - ASSERT_REG_POSITION(semaphore_trigger, 0x7); - ASSERT_REG_POSITION(reference_count, 0x14); - ASSERT_REG_POSITION(semaphore_acquire, 0x1A); - ASSERT_REG_POSITION(semaphore_release, 0x1B); - ASSERT_REG_POSITION(fence_value, 0x1C); - ASSERT_REG_POSITION(fence_action, 0x1D); - - ASSERT_REG_POSITION(acquire_mode, 0x100); - ASSERT_REG_POSITION(acquire_source, 0x101); - ASSERT_REG_POSITION(acquire_active, 0x102); - ASSERT_REG_POSITION(acquire_timeout, 0x103); - ASSERT_REG_POSITION(acquire_value, 0x104); - -#undef ASSERT_REG_POSITION - - enum class GpuSemaphoreOperation { - AcquireEqual = 0x1, - WriteLong = 0x2, - AcquireGequal = 0x4, - AcquireMask = 0x8, - }; + std::unique_ptr<Tegra::Control::Scheduler> scheduler; + std::unordered_map<s32, std::shared_ptr<Tegra::Control::ChannelState>> channels; + Tegra::Control::ChannelState* current_channel; + s32 bound_channel{-1}; + + std::deque<size_t> free_swap_counters; + std::deque<size_t> request_swap_counters; + std::mutex request_swap_mutex; }; GPU::GPU(Core::System& system, bool is_async, bool use_nvdec) @@ -744,25 +381,36 @@ GPU::GPU(Core::System& system, bool is_async, bool use_nvdec) GPU::~GPU() = default; -void GPU::BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer) { - impl->BindRenderer(std::move(renderer)); +std::shared_ptr<Control::ChannelState> GPU::AllocateChannel() { + return impl->AllocateChannel(); +} + +void GPU::InitChannel(Control::ChannelState& to_init) { + impl->InitChannel(to_init); +} + +void GPU::BindChannel(s32 channel_id) { + impl->BindChannel(channel_id); } -void GPU::CallMethod(const MethodCall& method_call) { - impl->CallMethod(method_call); +void GPU::ReleaseChannel(Control::ChannelState& to_release) { + impl->ReleaseChannel(to_release); } -void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending) { - impl->CallMultiMethod(method, subchannel, base_start, amount, methods_pending); +void GPU::InitAddressSpace(Tegra::MemoryManager& memory_manager) { + impl->InitAddressSpace(memory_manager); +} + +void GPU::BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer) { + impl->BindRenderer(std::move(renderer)); } void GPU::FlushCommands() { impl->FlushCommands(); } -void GPU::SyncGuestHost() { - impl->SyncGuestHost(); +void GPU::InvalidateGPUCache() { + impl->InvalidateGPUCache(); } void GPU::OnCommandListEnd() { @@ -770,17 +418,32 @@ void GPU::OnCommandListEnd() { } u64 GPU::RequestFlush(VAddr addr, std::size_t size) { - return impl->RequestFlush(addr, size); + return impl->RequestSyncOperation( + [this, addr, size]() { impl->rasterizer->FlushRegion(addr, size); }); } -u64 GPU::CurrentFlushRequestFence() const { - return impl->CurrentFlushRequestFence(); +u64 GPU::CurrentSyncRequestFence() const { + return impl->CurrentSyncRequestFence(); +} + +void GPU::WaitForSyncOperation(u64 fence) { + return impl->WaitForSyncOperation(fence); } void GPU::TickWork() { impl->TickWork(); } +/// Gets a mutable reference to the Host1x interface +Host1x::Host1x& GPU::Host1x() { + return impl->host1x; +} + +/// Gets an immutable reference to the Host1x interface. +const Host1x::Host1x& GPU::Host1x() const { + return impl->host1x; +} + Engines::Maxwell3D& GPU::Maxwell3D() { return impl->Maxwell3D(); } @@ -797,14 +460,6 @@ const Engines::KeplerCompute& GPU::KeplerCompute() const { return impl->KeplerCompute(); } -Tegra::MemoryManager& GPU::MemoryManager() { - return impl->MemoryManager(); -} - -const Tegra::MemoryManager& GPU::MemoryManager() const { - return impl->MemoryManager(); -} - Tegra::DmaPusher& GPU::DmaPusher() { return impl->DmaPusher(); } @@ -829,24 +484,9 @@ const VideoCore::ShaderNotify& GPU::ShaderNotify() const { return impl->ShaderNotify(); } -void GPU::WaitFence(u32 syncpoint_id, u32 value) { - impl->WaitFence(syncpoint_id, value); -} - -void GPU::IncrementSyncPoint(u32 syncpoint_id) { - impl->IncrementSyncPoint(syncpoint_id); -} - -u32 GPU::GetSyncpointValue(u32 syncpoint_id) const { - return impl->GetSyncpointValue(syncpoint_id); -} - -void GPU::RegisterSyncptInterrupt(u32 syncpoint_id, u32 value) { - impl->RegisterSyncptInterrupt(syncpoint_id, value); -} - -bool GPU::CancelSyncptInterrupt(u32 syncpoint_id, u32 value) { - return impl->CancelSyncptInterrupt(syncpoint_id, value); +void GPU::RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer, + std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences) { + impl->RequestSwapBuffers(framebuffer, fences, num_fences); } u64 GPU::GetTicks() const { @@ -881,8 +521,8 @@ void GPU::ReleaseContext() { impl->ReleaseContext(); } -void GPU::PushGPUEntries(Tegra::CommandList&& entries) { - impl->PushGPUEntries(std::move(entries)); +void GPU::PushGPUEntries(s32 channel, Tegra::CommandList&& entries) { + impl->PushGPUEntries(channel, std::move(entries)); } void GPU::PushCommandBuffer(u32 id, Tegra::ChCommandHeaderList& entries) { diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index b939ba315..0a4a8b14f 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -89,73 +89,58 @@ class Maxwell3D; class KeplerCompute; } // namespace Engines -enum class EngineID { - FERMI_TWOD_A = 0x902D, // 2D Engine - MAXWELL_B = 0xB197, // 3D Engine - KEPLER_COMPUTE_B = 0xB1C0, - KEPLER_INLINE_TO_MEMORY_B = 0xA140, - MAXWELL_DMA_COPY_A = 0xB0B5, -}; +namespace Control { +struct ChannelState; +} + +namespace Host1x { +class Host1x; +} // namespace Host1x class MemoryManager; class GPU final { public: - struct MethodCall { - u32 method{}; - u32 argument{}; - u32 subchannel{}; - u32 method_count{}; - - explicit MethodCall(u32 method_, u32 argument_, u32 subchannel_ = 0, u32 method_count_ = 0) - : method(method_), argument(argument_), subchannel(subchannel_), - method_count(method_count_) {} - - [[nodiscard]] bool IsLastCall() const { - return method_count <= 1; - } - }; - - enum class FenceOperation : u32 { - Acquire = 0, - Increment = 1, - }; - - union FenceAction { - u32 raw; - BitField<0, 1, FenceOperation> op; - BitField<8, 24, u32> syncpoint_id; - }; - explicit GPU(Core::System& system, bool is_async, bool use_nvdec); ~GPU(); /// Binds a renderer to the GPU. void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer); - /// Calls a GPU method. - void CallMethod(const MethodCall& method_call); - - /// Calls a GPU multivalue method. - void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, - u32 methods_pending); - /// Flush all current written commands into the host GPU for execution. void FlushCommands(); /// Synchronizes CPU writes with Host GPU memory. - void SyncGuestHost(); + void InvalidateGPUCache(); /// Signal the ending of command list. void OnCommandListEnd(); + std::shared_ptr<Control::ChannelState> AllocateChannel(); + + void InitChannel(Control::ChannelState& to_init); + + void BindChannel(s32 channel_id); + + void ReleaseChannel(Control::ChannelState& to_release); + + void InitAddressSpace(Tegra::MemoryManager& memory_manager); + /// Request a host GPU memory flush from the CPU. [[nodiscard]] u64 RequestFlush(VAddr addr, std::size_t size); /// Obtains current flush request fence id. - [[nodiscard]] u64 CurrentFlushRequestFence() const; + [[nodiscard]] u64 CurrentSyncRequestFence() const; + + void WaitForSyncOperation(u64 fence); /// Tick pending requests within the GPU. void TickWork(); + /// Gets a mutable reference to the Host1x interface + [[nodiscard]] Host1x::Host1x& Host1x(); + + /// Gets an immutable reference to the Host1x interface. + [[nodiscard]] const Host1x::Host1x& Host1x() const; + /// Returns a reference to the Maxwell3D GPU engine. [[nodiscard]] Engines::Maxwell3D& Maxwell3D(); @@ -168,12 +153,6 @@ public: /// Returns a reference to the KeplerCompute GPU engine. [[nodiscard]] const Engines::KeplerCompute& KeplerCompute() const; - /// Returns a reference to the GPU memory manager. - [[nodiscard]] Tegra::MemoryManager& MemoryManager(); - - /// Returns a const reference to the GPU memory manager. - [[nodiscard]] const Tegra::MemoryManager& MemoryManager() const; - /// Returns a reference to the GPU DMA pusher. [[nodiscard]] Tegra::DmaPusher& DmaPusher(); @@ -192,17 +171,6 @@ public: /// Returns a const reference to the shader notifier. [[nodiscard]] const VideoCore::ShaderNotify& ShaderNotify() const; - /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame. - void WaitFence(u32 syncpoint_id, u32 value); - - void IncrementSyncPoint(u32 syncpoint_id); - - [[nodiscard]] u32 GetSyncpointValue(u32 syncpoint_id) const; - - void RegisterSyncptInterrupt(u32 syncpoint_id, u32 value); - - [[nodiscard]] bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value); - [[nodiscard]] u64 GetTicks() const; [[nodiscard]] bool IsAsync() const; @@ -211,6 +179,9 @@ public: void RendererFrameEndNotify(); + void RequestSwapBuffers(const Tegra::FramebufferConfig* framebuffer, + std::array<Service::Nvidia::NvFence, 4>& fences, size_t num_fences); + /// Performs any additional setup necessary in order to begin GPU emulation. /// This can be used to launch any necessary threads and register any necessary /// core timing events. @@ -226,7 +197,7 @@ public: void ReleaseContext(); /// Push GPU command entries to be processed - void PushGPUEntries(Tegra::CommandList&& entries); + void PushGPUEntries(s32 channel, Tegra::CommandList&& entries); /// Push GPU command buffer entries to be processed void PushCommandBuffer(u32 id, Tegra::ChCommandHeaderList& entries); @@ -248,7 +219,7 @@ public: private: struct Impl; - std::unique_ptr<Impl> impl; + mutable std::unique_ptr<Impl> impl; }; } // namespace Tegra diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index d43f7175a..1bd477011 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -8,6 +8,7 @@ #include "common/thread.h" #include "core/core.h" #include "core/frontend/emu_window.h" +#include "video_core/control/scheduler.h" #include "video_core/dma_pusher.h" #include "video_core/gpu.h" #include "video_core/gpu_thread.h" @@ -18,8 +19,8 @@ namespace VideoCommon::GPUThread { /// Runs the GPU thread static void RunThread(std::stop_token stop_token, Core::System& system, VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, - Tegra::DmaPusher& dma_pusher, SynchState& state) { - std::string name = "yuzu:GPU"; + Tegra::Control::Scheduler& scheduler, SynchState& state) { + std::string name = "GPU"; MicroProfileOnThreadCreate(name.c_str()); SCOPE_EXIT({ MicroProfileOnThreadExit(); }); @@ -36,8 +37,7 @@ static void RunThread(std::stop_token stop_token, Core::System& system, break; } if (auto* submit_list = std::get_if<SubmitListCommand>(&next.data)) { - dma_pusher.Push(std::move(submit_list->entries)); - dma_pusher.DispatchCalls(); + scheduler.Push(submit_list->channel, std::move(submit_list->entries)); } else if (const auto* data = std::get_if<SwapBuffersCommand>(&next.data)) { renderer.SwapBuffers(data->framebuffer ? &*data->framebuffer : nullptr); } else if (std::holds_alternative<OnCommandListEndCommand>(next.data)) { @@ -68,14 +68,14 @@ ThreadManager::~ThreadManager() = default; void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, - Tegra::DmaPusher& dma_pusher) { + Tegra::Control::Scheduler& scheduler) { rasterizer = renderer.ReadRasterizer(); thread = std::jthread(RunThread, std::ref(system), std::ref(renderer), std::ref(context), - std::ref(dma_pusher), std::ref(state)); + std::ref(scheduler), std::ref(state)); } -void ThreadManager::SubmitList(Tegra::CommandList&& entries) { - PushCommand(SubmitListCommand(std::move(entries))); +void ThreadManager::SubmitList(s32 channel, Tegra::CommandList&& entries) { + PushCommand(SubmitListCommand(channel, std::move(entries))); } void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { @@ -93,8 +93,12 @@ void ThreadManager::FlushRegion(VAddr addr, u64 size) { } auto& gpu = system.GPU(); u64 fence = gpu.RequestFlush(addr, size); - PushCommand(GPUTickCommand(), true); - ASSERT(fence <= gpu.CurrentFlushRequestFence()); + TickGPU(); + gpu.WaitForSyncOperation(fence); +} + +void ThreadManager::TickGPU() { + PushCommand(GPUTickCommand()); } void ThreadManager::InvalidateRegion(VAddr addr, u64 size) { diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h index 2f8210cb9..64628d3e3 100644 --- a/src/video_core/gpu_thread.h +++ b/src/video_core/gpu_thread.h @@ -15,7 +15,9 @@ namespace Tegra { struct FramebufferConfig; -class DmaPusher; +namespace Control { +class Scheduler; +} } // namespace Tegra namespace Core { @@ -34,8 +36,10 @@ namespace VideoCommon::GPUThread { /// Command to signal to the GPU thread that a command list is ready for processing struct SubmitListCommand final { - explicit SubmitListCommand(Tegra::CommandList&& entries_) : entries{std::move(entries_)} {} + explicit SubmitListCommand(s32 channel_, Tegra::CommandList&& entries_) + : channel{channel_}, entries{std::move(entries_)} {} + s32 channel; Tegra::CommandList entries; }; @@ -112,10 +116,10 @@ public: /// Creates and starts the GPU thread. void StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, - Tegra::DmaPusher& dma_pusher); + Tegra::Control::Scheduler& scheduler); /// Push GPU command entries to be processed - void SubmitList(Tegra::CommandList&& entries); + void SubmitList(s32 channel, Tegra::CommandList&& entries); /// Swap buffers (render frame) void SwapBuffers(const Tegra::FramebufferConfig* framebuffer); @@ -131,6 +135,8 @@ public: void OnCommandListEnd(); + void TickGPU(); + private: /// Pushes a command to be executed by the GPU thread u64 PushCommand(CommandData&& command_data, bool block = false); diff --git a/src/video_core/command_classes/codecs/codec.cpp b/src/video_core/host1x/codecs/codec.cpp index a5eb97b7f..42e7d6e4f 100644 --- a/src/video_core/command_classes/codecs/codec.cpp +++ b/src/video_core/host1x/codecs/codec.cpp @@ -6,11 +6,11 @@ #include <vector> #include "common/assert.h" #include "common/settings.h" -#include "video_core/command_classes/codecs/codec.h" -#include "video_core/command_classes/codecs/h264.h" -#include "video_core/command_classes/codecs/vp8.h" -#include "video_core/command_classes/codecs/vp9.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/codec.h" +#include "video_core/host1x/codecs/h264.h" +#include "video_core/host1x/codecs/vp8.h" +#include "video_core/host1x/codecs/vp9.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" extern "C" { @@ -73,10 +73,10 @@ void AVFrameDeleter(AVFrame* ptr) { av_frame_free(&ptr); } -Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs) - : gpu(gpu_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(gpu)), - vp8_decoder(std::make_unique<Decoder::VP8>(gpu)), - vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {} +Codec::Codec(Host1x::Host1x& host1x_, const Host1x::NvdecCommon::NvdecRegisters& regs) + : host1x(host1x_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(host1x)), + vp8_decoder(std::make_unique<Decoder::VP8>(host1x)), + vp9_decoder(std::make_unique<Decoder::VP9>(host1x)) {} Codec::~Codec() { if (!initialized) { @@ -168,11 +168,11 @@ void Codec::InitializeGpuDecoder() { void Codec::Initialize() { const AVCodecID codec = [&] { switch (current_codec) { - case NvdecCommon::VideoCodec::H264: + case Host1x::NvdecCommon::VideoCodec::H264: return AV_CODEC_ID_H264; - case NvdecCommon::VideoCodec::VP8: + case Host1x::NvdecCommon::VideoCodec::VP8: return AV_CODEC_ID_VP8; - case NvdecCommon::VideoCodec::VP9: + case Host1x::NvdecCommon::VideoCodec::VP9: return AV_CODEC_ID_VP9; default: UNIMPLEMENTED_MSG("Unknown codec {}", current_codec); @@ -197,7 +197,7 @@ void Codec::Initialize() { initialized = true; } -void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) { +void Codec::SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec) { if (current_codec != codec) { current_codec = codec; LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", GetCurrentCodecName()); @@ -215,11 +215,11 @@ void Codec::Decode() { bool vp9_hidden_frame = false; const auto& frame_data = [&]() { switch (current_codec) { - case Tegra::NvdecCommon::VideoCodec::H264: + case Tegra::Host1x::NvdecCommon::VideoCodec::H264: return h264_decoder->ComposeFrame(state, is_first_frame); - case Tegra::NvdecCommon::VideoCodec::VP8: + case Tegra::Host1x::NvdecCommon::VideoCodec::VP8: return vp8_decoder->ComposeFrame(state); - case Tegra::NvdecCommon::VideoCodec::VP9: + case Tegra::Host1x::NvdecCommon::VideoCodec::VP9: vp9_decoder->ComposeFrame(state); vp9_hidden_frame = vp9_decoder->WasFrameHidden(); return vp9_decoder->GetFrameBytes(); @@ -287,21 +287,21 @@ AVFramePtr Codec::GetCurrentFrame() { return frame; } -NvdecCommon::VideoCodec Codec::GetCurrentCodec() const { +Host1x::NvdecCommon::VideoCodec Codec::GetCurrentCodec() const { return current_codec; } std::string_view Codec::GetCurrentCodecName() const { switch (current_codec) { - case NvdecCommon::VideoCodec::None: + case Host1x::NvdecCommon::VideoCodec::None: return "None"; - case NvdecCommon::VideoCodec::H264: + case Host1x::NvdecCommon::VideoCodec::H264: return "H264"; - case NvdecCommon::VideoCodec::VP8: + case Host1x::NvdecCommon::VideoCodec::VP8: return "VP8"; - case NvdecCommon::VideoCodec::H265: + case Host1x::NvdecCommon::VideoCodec::H265: return "H265"; - case NvdecCommon::VideoCodec::VP9: + case Host1x::NvdecCommon::VideoCodec::VP9: return "VP9"; default: return "Unknown"; diff --git a/src/video_core/command_classes/codecs/codec.h b/src/video_core/host1x/codecs/codec.h index 0c2405465..0d45fb7fe 100644 --- a/src/video_core/command_classes/codecs/codec.h +++ b/src/video_core/host1x/codecs/codec.h @@ -6,8 +6,8 @@ #include <memory> #include <string_view> #include <queue> - -#include "video_core/command_classes/nvdec_common.h" +#include "common/common_types.h" +#include "video_core/host1x/nvdec_common.h" extern "C" { #if defined(__GNUC__) || defined(__clang__) @@ -21,7 +21,6 @@ extern "C" { } namespace Tegra { -class GPU; void AVFrameDeleter(AVFrame* ptr); using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>; @@ -32,16 +31,20 @@ class VP8; class VP9; } // namespace Decoder +namespace Host1x { +class Host1x; +} // namespace Host1x + class Codec { public: - explicit Codec(GPU& gpu, const NvdecCommon::NvdecRegisters& regs); + explicit Codec(Host1x::Host1x& host1x, const Host1x::NvdecCommon::NvdecRegisters& regs); ~Codec(); /// Initialize the codec, returning success or failure void Initialize(); /// Sets NVDEC video stream codec - void SetTargetCodec(NvdecCommon::VideoCodec codec); + void SetTargetCodec(Host1x::NvdecCommon::VideoCodec codec); /// Call decoders to construct headers, decode AVFrame with ffmpeg void Decode(); @@ -50,7 +53,7 @@ public: [[nodiscard]] AVFramePtr GetCurrentFrame(); /// Returns the value of current_codec - [[nodiscard]] NvdecCommon::VideoCodec GetCurrentCodec() const; + [[nodiscard]] Host1x::NvdecCommon::VideoCodec GetCurrentCodec() const; /// Return name of the current codec [[nodiscard]] std::string_view GetCurrentCodecName() const; @@ -63,14 +66,14 @@ private: bool CreateGpuAvDevice(); bool initialized{}; - NvdecCommon::VideoCodec current_codec{NvdecCommon::VideoCodec::None}; + Host1x::NvdecCommon::VideoCodec current_codec{Host1x::NvdecCommon::VideoCodec::None}; const AVCodec* av_codec{nullptr}; AVCodecContext* av_codec_ctx{nullptr}; AVBufferRef* av_gpu_decoder{nullptr}; - GPU& gpu; - const NvdecCommon::NvdecRegisters& state; + Host1x::Host1x& host1x; + const Host1x::NvdecCommon::NvdecRegisters& state; std::unique_ptr<Decoder::H264> h264_decoder; std::unique_ptr<Decoder::VP8> vp8_decoder; std::unique_ptr<Decoder::VP9> vp9_decoder; diff --git a/src/video_core/command_classes/codecs/h264.cpp b/src/video_core/host1x/codecs/h264.cpp index e2acd54d4..e87bd65fa 100644 --- a/src/video_core/command_classes/codecs/h264.cpp +++ b/src/video_core/host1x/codecs/h264.cpp @@ -5,8 +5,8 @@ #include <bit> #include "common/settings.h" -#include "video_core/command_classes/codecs/h264.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/h264.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" namespace Tegra::Decoder { @@ -24,19 +24,20 @@ constexpr std::array<u8, 16> zig_zag_scan{ }; } // Anonymous namespace -H264::H264(GPU& gpu_) : gpu(gpu_) {} +H264::H264(Host1x::Host1x& host1x_) : host1x{host1x_} {} H264::~H264() = default; -const std::vector<u8>& H264::ComposeFrame(const NvdecCommon::NvdecRegisters& state, +const std::vector<u8>& H264::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state, bool is_first_frame) { H264DecoderContext context; - gpu.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext)); + host1x.MemoryManager().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(context.stream_len); - gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size()); + host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size()); return frame; } @@ -155,8 +156,8 @@ const std::vector<u8>& H264::ComposeFrame(const NvdecCommon::NvdecRegisters& sta frame.resize(encoded_header.size() + context.stream_len); std::memcpy(frame.data(), encoded_header.data(), encoded_header.size()); - gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, - frame.data() + encoded_header.size(), context.stream_len); + host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, + frame.data() + encoded_header.size(), context.stream_len); return frame; } diff --git a/src/video_core/command_classes/codecs/h264.h b/src/video_core/host1x/codecs/h264.h index 261574364..5cc86454e 100644 --- a/src/video_core/command_classes/codecs/h264.h +++ b/src/video_core/host1x/codecs/h264.h @@ -8,10 +8,14 @@ #include "common/bit_field.h" #include "common/common_funcs.h" #include "common/common_types.h" -#include "video_core/command_classes/nvdec_common.h" +#include "video_core/host1x/nvdec_common.h" namespace Tegra { -class GPU; + +namespace Host1x { +class Host1x; +} // namespace Host1x + namespace Decoder { class H264BitWriter { @@ -55,16 +59,16 @@ private: class H264 { public: - explicit H264(GPU& gpu); + explicit H264(Host1x::Host1x& host1x); ~H264(); /// Compose the H264 frame for FFmpeg decoding - [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state, - bool is_first_frame = false); + [[nodiscard]] const std::vector<u8>& ComposeFrame( + const Host1x::NvdecCommon::NvdecRegisters& state, bool is_first_frame = false); private: std::vector<u8> frame; - GPU& gpu; + Host1x::Host1x& host1x; struct H264ParameterSet { s32 log2_max_pic_order_cnt_lsb_minus4; ///< 0x00 diff --git a/src/video_core/command_classes/codecs/vp8.cpp b/src/video_core/host1x/codecs/vp8.cpp index c83b9bbc2..28fb12cb8 100644 --- a/src/video_core/command_classes/codecs/vp8.cpp +++ b/src/video_core/host1x/codecs/vp8.cpp @@ -3,18 +3,18 @@ #include <vector> -#include "video_core/command_classes/codecs/vp8.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/vp8.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" namespace Tegra::Decoder { -VP8::VP8(GPU& gpu_) : gpu(gpu_) {} +VP8::VP8(Host1x::Host1x& host1x_) : host1x{host1x_} {} VP8::~VP8() = default; -const std::vector<u8>& VP8::ComposeFrame(const NvdecCommon::NvdecRegisters& state) { +const std::vector<u8>& VP8::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { VP8PictureInfo info; - gpu.MemoryManager().ReadBlock(state.picture_info_offset, &info, sizeof(VP8PictureInfo)); + host1x.MemoryManager().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 @@ const std::vector<u8>& VP8::ComposeFrame(const NvdecCommon::NvdecRegisters& stat frame[9] = static_cast<u8>(((info.frame_height >> 8) & 0x3f)); } const u64 bitstream_offset = state.frame_bitstream_offset; - gpu.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size); + host1x.MemoryManager().ReadBlock(bitstream_offset, frame.data() + header_size, bitstream_size); return frame; } diff --git a/src/video_core/command_classes/codecs/vp8.h b/src/video_core/host1x/codecs/vp8.h index 3357667b0..5bf07ecab 100644 --- a/src/video_core/command_classes/codecs/vp8.h +++ b/src/video_core/host1x/codecs/vp8.h @@ -8,23 +8,28 @@ #include "common/common_funcs.h" #include "common/common_types.h" -#include "video_core/command_classes/nvdec_common.h" +#include "video_core/host1x/nvdec_common.h" namespace Tegra { -class GPU; + +namespace Host1x { +class Host1x; +} // namespace Host1x + namespace Decoder { class VP8 { public: - explicit VP8(GPU& gpu); + explicit VP8(Host1x::Host1x& host1x); ~VP8(); /// Compose the VP8 frame for FFmpeg decoding - [[nodiscard]] const std::vector<u8>& ComposeFrame(const NvdecCommon::NvdecRegisters& state); + [[nodiscard]] const std::vector<u8>& ComposeFrame( + const Host1x::NvdecCommon::NvdecRegisters& state); private: std::vector<u8> frame; - GPU& gpu; + Host1x::Host1x& host1x; struct VP8PictureInfo { INSERT_PADDING_WORDS_NOINIT(14); diff --git a/src/video_core/command_classes/codecs/vp9.cpp b/src/video_core/host1x/codecs/vp9.cpp index c01431441..cf40c9012 100644 --- a/src/video_core/command_classes/codecs/vp9.cpp +++ b/src/video_core/host1x/codecs/vp9.cpp @@ -4,8 +4,8 @@ #include <algorithm> // for std::copy #include <numeric> #include "common/assert.h" -#include "video_core/command_classes/codecs/vp9.h" -#include "video_core/gpu.h" +#include "video_core/host1x/codecs/vp9.h" +#include "video_core/host1x/host1x.h" #include "video_core/memory_manager.h" namespace Tegra::Decoder { @@ -236,7 +236,7 @@ constexpr std::array<u8, 254> map_lut{ } } // Anonymous namespace -VP9::VP9(GPU& gpu_) : gpu{gpu_} {} +VP9::VP9(Host1x::Host1x& host1x_) : host1x{host1x_} {} VP9::~VP9() = default; @@ -355,9 +355,9 @@ void VP9::WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_ } } -Vp9PictureInfo VP9::GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state) { +Vp9PictureInfo VP9::GetVp9PictureInfo(const Host1x::NvdecCommon::NvdecRegisters& state) { PictureInfo picture_info; - gpu.MemoryManager().ReadBlock(state.picture_info_offset, &picture_info, sizeof(PictureInfo)); + host1x.MemoryManager().ReadBlock(state.picture_info_offset, &picture_info, sizeof(PictureInfo)); Vp9PictureInfo vp9_info = picture_info.Convert(); InsertEntropy(state.vp9_entropy_probs_offset, vp9_info.entropy); @@ -372,18 +372,19 @@ Vp9PictureInfo VP9::GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state) void VP9::InsertEntropy(u64 offset, Vp9EntropyProbs& dst) { EntropyProbs entropy; - gpu.MemoryManager().ReadBlock(offset, &entropy, sizeof(EntropyProbs)); + host1x.MemoryManager().ReadBlock(offset, &entropy, sizeof(EntropyProbs)); entropy.Convert(dst); } -Vp9FrameContainer VP9::GetCurrentFrame(const NvdecCommon::NvdecRegisters& state) { +Vp9FrameContainer VP9::GetCurrentFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { Vp9FrameContainer current_frame{}; { - gpu.SyncGuestHost(); + // gpu.SyncGuestHost(); epic, why? current_frame.info = GetVp9PictureInfo(state); current_frame.bit_stream.resize(current_frame.info.bitstream_size); - gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, current_frame.bit_stream.data(), - current_frame.info.bitstream_size); + host1x.MemoryManager().ReadBlock(state.frame_bitstream_offset, + current_frame.bit_stream.data(), + current_frame.info.bitstream_size); } if (!next_frame.bit_stream.empty()) { Vp9FrameContainer temp{ @@ -769,7 +770,7 @@ VpxBitStreamWriter VP9::ComposeUncompressedHeader() { return uncomp_writer; } -void VP9::ComposeFrame(const NvdecCommon::NvdecRegisters& state) { +void VP9::ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state) { std::vector<u8> bitstream; { Vp9FrameContainer curr_frame = GetCurrentFrame(state); diff --git a/src/video_core/command_classes/codecs/vp9.h b/src/video_core/host1x/codecs/vp9.h index ecc40e8b1..d4083e8d3 100644 --- a/src/video_core/command_classes/codecs/vp9.h +++ b/src/video_core/host1x/codecs/vp9.h @@ -8,11 +8,15 @@ #include "common/common_types.h" #include "common/stream.h" -#include "video_core/command_classes/codecs/vp9_types.h" -#include "video_core/command_classes/nvdec_common.h" +#include "video_core/host1x/codecs/vp9_types.h" +#include "video_core/host1x/nvdec_common.h" namespace Tegra { -class GPU; + +namespace Host1x { +class Host1x; +} // namespace Host1x + namespace Decoder { /// The VpxRangeEncoder, and VpxBitStreamWriter classes are used to compose the @@ -106,7 +110,7 @@ private: class VP9 { public: - explicit VP9(GPU& gpu_); + explicit VP9(Host1x::Host1x& host1x); ~VP9(); VP9(const VP9&) = delete; @@ -117,7 +121,7 @@ public: /// Composes the VP9 frame from the GPU state information. /// Based on the official VP9 spec documentation - void ComposeFrame(const NvdecCommon::NvdecRegisters& state); + void ComposeFrame(const Host1x::NvdecCommon::NvdecRegisters& state); /// Returns true if the most recent frame was a hidden frame. [[nodiscard]] bool WasFrameHidden() const { @@ -162,19 +166,21 @@ private: void WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_prob); /// Returns VP9 information from NVDEC provided offset and size - [[nodiscard]] Vp9PictureInfo GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state); + [[nodiscard]] Vp9PictureInfo GetVp9PictureInfo( + const Host1x::NvdecCommon::NvdecRegisters& state); /// Read and convert NVDEC provided entropy probs to Vp9EntropyProbs struct void InsertEntropy(u64 offset, Vp9EntropyProbs& dst); /// Returns frame to be decoded after buffering - [[nodiscard]] Vp9FrameContainer GetCurrentFrame(const NvdecCommon::NvdecRegisters& state); + [[nodiscard]] Vp9FrameContainer GetCurrentFrame( + const Host1x::NvdecCommon::NvdecRegisters& state); /// Use NVDEC providied information to compose the headers for the current frame [[nodiscard]] std::vector<u8> ComposeCompressedHeader(); [[nodiscard]] VpxBitStreamWriter ComposeUncompressedHeader(); - GPU& gpu; + Host1x::Host1x& host1x; std::vector<u8> frame; std::array<s8, 4> loop_filter_ref_deltas{}; diff --git a/src/video_core/command_classes/codecs/vp9_types.h b/src/video_core/host1x/codecs/vp9_types.h index bb3d8df6e..adad8ed7e 100644 --- a/src/video_core/command_classes/codecs/vp9_types.h +++ b/src/video_core/host1x/codecs/vp9_types.h @@ -9,7 +9,6 @@ #include "common/common_types.h" namespace Tegra { -class GPU; namespace Decoder { struct Vp9FrameDimensions { diff --git a/src/video_core/host1x/control.cpp b/src/video_core/host1x/control.cpp new file mode 100644 index 000000000..dceefdb7f --- /dev/null +++ b/src/video_core/host1x/control.cpp @@ -0,0 +1,33 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/assert.h" +#include "video_core/host1x/control.h" +#include "video_core/host1x/host1x.h" + +namespace Tegra::Host1x { + +Control::Control(Host1x& host1x_) : host1x(host1x_) {} + +Control::~Control() = default; + +void Control::ProcessMethod(Method method, u32 argument) { + switch (method) { + case Method::LoadSyncptPayload32: + syncpoint_value = argument; + break; + case Method::WaitSyncpt: + case Method::WaitSyncpt32: + Execute(argument); + break; + default: + UNIMPLEMENTED_MSG("Control method 0x{:X}", static_cast<u32>(method)); + break; + } +} + +void Control::Execute(u32 data) { + host1x.GetSyncpointManager().WaitHost(data, syncpoint_value); +} + +} // namespace Tegra::Host1x diff --git a/src/video_core/command_classes/host1x.h b/src/video_core/host1x/control.h index bb48a4381..e117888a3 100644 --- a/src/video_core/command_classes/host1x.h +++ b/src/video_core/host1x/control.h @@ -1,15 +1,19 @@ -// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-FileCopyrightText: 2021 Skyline Team and Contributors +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once #include "common/common_types.h" namespace Tegra { -class GPU; + +namespace Host1x { + +class Host1x; class Nvdec; -class Host1x { +class Control { public: enum class Method : u32 { WaitSyncpt = 0x8, @@ -17,8 +21,8 @@ public: WaitSyncpt32 = 0x50, }; - explicit Host1x(GPU& gpu); - ~Host1x(); + explicit Control(Host1x& host1x); + ~Control(); /// Writes the method into the state, Invoke Execute() if encountered void ProcessMethod(Method method, u32 argument); @@ -28,7 +32,9 @@ private: void Execute(u32 data); u32 syncpoint_value{}; - GPU& gpu; + Host1x& host1x; }; +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/host1x/host1x.cpp b/src/video_core/host1x/host1x.cpp new file mode 100644 index 000000000..7c317a85d --- /dev/null +++ b/src/video_core/host1x/host1x.cpp @@ -0,0 +1,17 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "core/core.h" +#include "video_core/host1x/host1x.h" + +namespace Tegra { + +namespace Host1x { + +Host1x::Host1x(Core::System& system_) + : system{system_}, syncpoint_manager{}, memory_manager{system, 32, 12}, + allocator{std::make_unique<Common::FlatAllocator<u32, 0, 32>>(1 << 12)} {} + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/host1x/host1x.h b/src/video_core/host1x/host1x.h new file mode 100644 index 000000000..57082ae54 --- /dev/null +++ b/src/video_core/host1x/host1x.h @@ -0,0 +1,57 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include "common/common_types.h" + +#include "common/address_space.h" +#include "video_core/host1x/syncpoint_manager.h" +#include "video_core/memory_manager.h" + +namespace Core { +class System; +} // namespace Core + +namespace Tegra { + +namespace Host1x { + +class Host1x { +public: + explicit Host1x(Core::System& system); + + SyncpointManager& GetSyncpointManager() { + return syncpoint_manager; + } + + const SyncpointManager& GetSyncpointManager() const { + return syncpoint_manager; + } + + Tegra::MemoryManager& MemoryManager() { + return memory_manager; + } + + const Tegra::MemoryManager& MemoryManager() const { + return memory_manager; + } + + Common::FlatAllocator<u32, 0, 32>& Allocator() { + return *allocator; + } + + const Common::FlatAllocator<u32, 0, 32>& Allocator() const { + return *allocator; + } + +private: + Core::System& system; + SyncpointManager syncpoint_manager; + Tegra::MemoryManager memory_manager; + std::unique_ptr<Common::FlatAllocator<u32, 0, 32>> allocator; +}; + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/command_classes/nvdec.cpp b/src/video_core/host1x/nvdec.cpp index 4fbbe3da6..a4bd5b79f 100644 --- a/src/video_core/command_classes/nvdec.cpp +++ b/src/video_core/host1x/nvdec.cpp @@ -2,15 +2,16 @@ // SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" -#include "video_core/command_classes/nvdec.h" -#include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/nvdec.h" -namespace Tegra { +namespace Tegra::Host1x { #define NVDEC_REG_INDEX(field_name) \ (offsetof(NvdecCommon::NvdecRegisters, field_name) / sizeof(u64)) -Nvdec::Nvdec(GPU& gpu_) : gpu(gpu_), state{}, codec(std::make_unique<Codec>(gpu, state)) {} +Nvdec::Nvdec(Host1x& host1x_) + : host1x(host1x_), state{}, codec(std::make_unique<Codec>(host1x, state)) {} Nvdec::~Nvdec() = default; @@ -44,4 +45,4 @@ void Nvdec::Execute() { } } -} // namespace Tegra +} // namespace Tegra::Host1x diff --git a/src/video_core/command_classes/nvdec.h b/src/video_core/host1x/nvdec.h index 488531fc6..3949d5181 100644 --- a/src/video_core/command_classes/nvdec.h +++ b/src/video_core/host1x/nvdec.h @@ -6,14 +6,17 @@ #include <memory> #include <vector> #include "common/common_types.h" -#include "video_core/command_classes/codecs/codec.h" +#include "video_core/host1x/codecs/codec.h" namespace Tegra { -class GPU; + +namespace Host1x { + +class Host1x; class Nvdec { public: - explicit Nvdec(GPU& gpu); + explicit Nvdec(Host1x& host1x); ~Nvdec(); /// Writes the method into the state, Invoke Execute() if encountered @@ -26,8 +29,11 @@ private: /// Invoke codec to decode a frame void Execute(); - GPU& gpu; + Host1x& host1x; NvdecCommon::NvdecRegisters state; std::unique_ptr<Codec> codec; }; + +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/command_classes/nvdec_common.h b/src/video_core/host1x/nvdec_common.h index 521e5b52b..49d67ebbe 100644 --- a/src/video_core/command_classes/nvdec_common.h +++ b/src/video_core/host1x/nvdec_common.h @@ -7,7 +7,7 @@ #include "common/common_funcs.h" #include "common/common_types.h" -namespace Tegra::NvdecCommon { +namespace Tegra::Host1x::NvdecCommon { enum class VideoCodec : u64 { None = 0x0, @@ -94,4 +94,4 @@ ASSERT_REG_POSITION(vp9_curr_frame_mvs_offset, 0x176); #undef ASSERT_REG_POSITION -} // namespace Tegra::NvdecCommon +} // namespace Tegra::Host1x::NvdecCommon diff --git a/src/video_core/command_classes/sync_manager.cpp b/src/video_core/host1x/sync_manager.cpp index 67e58046f..5ef9ea217 100644 --- a/src/video_core/command_classes/sync_manager.cpp +++ b/src/video_core/host1x/sync_manager.cpp @@ -3,10 +3,13 @@ #include <algorithm> #include "sync_manager.h" -#include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/syncpoint_manager.h" namespace Tegra { -SyncptIncrManager::SyncptIncrManager(GPU& gpu_) : gpu(gpu_) {} +namespace Host1x { + +SyncptIncrManager::SyncptIncrManager(Host1x& host1x_) : host1x(host1x_) {} SyncptIncrManager::~SyncptIncrManager() = default; void SyncptIncrManager::Increment(u32 id) { @@ -36,8 +39,12 @@ void SyncptIncrManager::IncrementAllDone() { if (!increments[done_count].complete) { break; } - gpu.IncrementSyncPoint(increments[done_count].syncpt_id); + auto& syncpoint_manager = host1x.GetSyncpointManager(); + syncpoint_manager.IncrementGuest(increments[done_count].syncpt_id); + syncpoint_manager.IncrementHost(increments[done_count].syncpt_id); } increments.erase(increments.begin(), increments.begin() + done_count); } + +} // namespace Host1x } // namespace Tegra diff --git a/src/video_core/command_classes/sync_manager.h b/src/video_core/host1x/sync_manager.h index 6dfaae080..7bb77fa27 100644 --- a/src/video_core/command_classes/sync_manager.h +++ b/src/video_core/host1x/sync_manager.h @@ -8,7 +8,11 @@ #include "common/common_types.h" namespace Tegra { -class GPU; + +namespace Host1x { + +class Host1x; + struct SyncptIncr { u32 id; u32 class_id; @@ -21,7 +25,7 @@ struct SyncptIncr { class SyncptIncrManager { public: - explicit SyncptIncrManager(GPU& gpu); + explicit SyncptIncrManager(Host1x& host1x); ~SyncptIncrManager(); /// Add syncpoint id and increment all @@ -41,7 +45,9 @@ private: std::mutex increment_lock; u32 current_id{}; - GPU& gpu; + Host1x& host1x; }; +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/host1x/syncpoint_manager.cpp b/src/video_core/host1x/syncpoint_manager.cpp new file mode 100644 index 000000000..326e8355a --- /dev/null +++ b/src/video_core/host1x/syncpoint_manager.cpp @@ -0,0 +1,96 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#include "common/microprofile.h" +#include "video_core/host1x/syncpoint_manager.h" + +namespace Tegra { + +namespace Host1x { + +MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192)); + +SyncpointManager::ActionHandle SyncpointManager::RegisterAction( + std::atomic<u32>& syncpoint, std::list<RegisteredAction>& action_storage, u32 expected_value, + std::function<void()>&& action) { + if (syncpoint.load(std::memory_order_acquire) >= expected_value) { + action(); + return {}; + } + + std::unique_lock lk(guard); + if (syncpoint.load(std::memory_order_relaxed) >= expected_value) { + action(); + return {}; + } + auto it = action_storage.begin(); + while (it != action_storage.end()) { + if (it->expected_value >= expected_value) { + break; + } + ++it; + } + return action_storage.emplace(it, expected_value, std::move(action)); +} + +void SyncpointManager::DeregisterAction(std::list<RegisteredAction>& action_storage, + ActionHandle& handle) { + std::unique_lock lk(guard); + action_storage.erase(handle); +} + +void SyncpointManager::DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle) { + DeregisterAction(guest_action_storage[syncpoint_id], handle); +} + +void SyncpointManager::DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle) { + DeregisterAction(host_action_storage[syncpoint_id], handle); +} + +void SyncpointManager::IncrementGuest(u32 syncpoint_id) { + Increment(syncpoints_guest[syncpoint_id], wait_guest_cv, guest_action_storage[syncpoint_id]); +} + +void SyncpointManager::IncrementHost(u32 syncpoint_id) { + Increment(syncpoints_host[syncpoint_id], wait_host_cv, host_action_storage[syncpoint_id]); +} + +void SyncpointManager::WaitGuest(u32 syncpoint_id, u32 expected_value) { + Wait(syncpoints_guest[syncpoint_id], wait_guest_cv, expected_value); +} + +void SyncpointManager::WaitHost(u32 syncpoint_id, u32 expected_value) { + MICROPROFILE_SCOPE(GPU_wait); + Wait(syncpoints_host[syncpoint_id], wait_host_cv, expected_value); +} + +void SyncpointManager::Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, + std::list<RegisteredAction>& action_storage) { + auto new_value{syncpoint.fetch_add(1, std::memory_order_acq_rel) + 1}; + + std::unique_lock lk(guard); + auto it = action_storage.begin(); + while (it != action_storage.end()) { + if (it->expected_value > new_value) { + break; + } + it->action(); + it = action_storage.erase(it); + } + wait_cv.notify_all(); +} + +void SyncpointManager::Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, + u32 expected_value) { + const auto pred = [&]() { return syncpoint.load(std::memory_order_acquire) >= expected_value; }; + if (pred()) { + return; + } + + std::unique_lock lk(guard); + wait_cv.wait(lk, pred); +} + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/host1x/syncpoint_manager.h b/src/video_core/host1x/syncpoint_manager.h new file mode 100644 index 000000000..50a264e23 --- /dev/null +++ b/src/video_core/host1x/syncpoint_manager.h @@ -0,0 +1,98 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#pragma once + +#include <array> +#include <atomic> +#include <condition_variable> +#include <functional> +#include <list> +#include <mutex> + +#include "common/common_types.h" + +namespace Tegra { + +namespace Host1x { + +class SyncpointManager { +public: + u32 GetGuestSyncpointValue(u32 id) const { + return syncpoints_guest[id].load(std::memory_order_acquire); + } + + u32 GetHostSyncpointValue(u32 id) const { + return syncpoints_host[id].load(std::memory_order_acquire); + } + + struct RegisteredAction { + explicit RegisteredAction(u32 expected_value_, std::function<void()>&& action_) + : expected_value{expected_value_}, action{std::move(action_)} {} + u32 expected_value; + std::function<void()> action; + }; + using ActionHandle = std::list<RegisteredAction>::iterator; + + template <typename Func> + ActionHandle RegisterGuestAction(u32 syncpoint_id, u32 expected_value, Func&& action) { + std::function<void()> func(action); + return RegisterAction(syncpoints_guest[syncpoint_id], guest_action_storage[syncpoint_id], + expected_value, std::move(func)); + } + + template <typename Func> + ActionHandle RegisterHostAction(u32 syncpoint_id, u32 expected_value, Func&& action) { + std::function<void()> func(action); + return RegisterAction(syncpoints_host[syncpoint_id], host_action_storage[syncpoint_id], + expected_value, std::move(func)); + } + + void DeregisterGuestAction(u32 syncpoint_id, ActionHandle& handle); + + void DeregisterHostAction(u32 syncpoint_id, ActionHandle& handle); + + void IncrementGuest(u32 syncpoint_id); + + void IncrementHost(u32 syncpoint_id); + + void WaitGuest(u32 syncpoint_id, u32 expected_value); + + void WaitHost(u32 syncpoint_id, u32 expected_value); + + bool IsReadyGuest(u32 syncpoint_id, u32 expected_value) const { + return syncpoints_guest[syncpoint_id].load(std::memory_order_acquire) >= expected_value; + } + + bool IsReadyHost(u32 syncpoint_id, u32 expected_value) const { + return syncpoints_host[syncpoint_id].load(std::memory_order_acquire) >= expected_value; + } + +private: + void Increment(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, + std::list<RegisteredAction>& action_storage); + + ActionHandle RegisterAction(std::atomic<u32>& syncpoint, + std::list<RegisteredAction>& action_storage, u32 expected_value, + std::function<void()>&& action); + + void DeregisterAction(std::list<RegisteredAction>& action_storage, ActionHandle& handle); + + void Wait(std::atomic<u32>& syncpoint, std::condition_variable& wait_cv, u32 expected_value); + + static constexpr size_t NUM_MAX_SYNCPOINTS = 192; + + std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_guest{}; + std::array<std::atomic<u32>, NUM_MAX_SYNCPOINTS> syncpoints_host{}; + + std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> guest_action_storage; + std::array<std::list<RegisteredAction>, NUM_MAX_SYNCPOINTS> host_action_storage; + + std::mutex guard; + std::condition_variable wait_guest_cv; + std::condition_variable wait_host_cv; +}; + +} // namespace Host1x + +} // namespace Tegra diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/host1x/vic.cpp index 7c17df353..ac0b7d20e 100644 --- a/src/video_core/command_classes/vic.cpp +++ b/src/video_core/host1x/vic.cpp @@ -18,14 +18,17 @@ extern "C" { #include "common/bit_field.h" #include "common/logging/log.h" -#include "video_core/command_classes/nvdec.h" -#include "video_core/command_classes/vic.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" +#include "video_core/host1x/host1x.h" +#include "video_core/host1x/nvdec.h" +#include "video_core/host1x/vic.h" #include "video_core/memory_manager.h" #include "video_core/textures/decoders.h" namespace Tegra { + +namespace Host1x { + namespace { enum class VideoPixelFormat : u64_le { RGBA8 = 0x1f, @@ -46,8 +49,8 @@ union VicConfig { BitField<46, 14, u64_le> surface_height_minus1; }; -Vic::Vic(GPU& gpu_, std::shared_ptr<Nvdec> nvdec_processor_) - : gpu(gpu_), +Vic::Vic(Host1x& host1x_, std::shared_ptr<Nvdec> nvdec_processor_) + : host1x(host1x_), nvdec_processor(std::move(nvdec_processor_)), converted_frame_buffer{nullptr, av_free} {} Vic::~Vic() = default; @@ -78,7 +81,7 @@ void Vic::Execute() { LOG_ERROR(Service_NVDRV, "VIC Luma address not set."); return; } - const VicConfig config{gpu.MemoryManager().Read<u64>(config_struct_address + 0x20)}; + const VicConfig config{host1x.MemoryManager().Read<u64>(config_struct_address + 0x20)}; const AVFramePtr frame_ptr = nvdec_processor->GetFrame(); const auto* frame = frame_ptr.get(); if (!frame) { @@ -153,15 +156,16 @@ void Vic::WriteRGBFrame(const AVFrame* frame, const VicConfig& config) { const u32 block_height = static_cast<u32>(config.block_linear_height_log2); const auto size = Texture::CalculateSize(true, 4, width, height, 1, block_height, 0); luma_buffer.resize(size); - Texture::SwizzleSubrect(width, height, width * 4, width, 4, luma_buffer.data(), - converted_frame_buf_addr, block_height, 0, 0); + std::span<const u8> frame_buff(converted_frame_buf_addr, 4 * width * height); + Texture::SwizzleSubrect(luma_buffer, frame_buff, 4, width, height, 1, 0, 0, width, height, + block_height, 0, width * 4); - gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size); + host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size); } else { // send pitch linear frame const size_t linear_size = width * height * 4; - gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr, - linear_size); + host1x.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr, + linear_size); } } @@ -189,8 +193,8 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) { luma_buffer[dst + x] = luma_src[src + x]; } } - gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), - luma_buffer.size()); + host1x.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), + luma_buffer.size()); // Chroma const std::size_t half_height = frame_height / 2; @@ -231,8 +235,10 @@ void Vic::WriteYUVFrame(const AVFrame* frame, const VicConfig& config) { ASSERT(false); break; } - gpu.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(), - chroma_buffer.size()); + host1x.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(), + chroma_buffer.size()); } +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/command_classes/vic.h b/src/video_core/host1x/vic.h index 010daa6b6..2b78786e8 100644 --- a/src/video_core/command_classes/vic.h +++ b/src/video_core/host1x/vic.h @@ -10,7 +10,10 @@ struct SwsContext; namespace Tegra { -class GPU; + +namespace Host1x { + +class Host1x; class Nvdec; union VicConfig; @@ -25,7 +28,7 @@ public: SetOutputSurfaceChromaUnusedOffset = 0x1ca }; - explicit Vic(GPU& gpu, std::shared_ptr<Nvdec> nvdec_processor); + explicit Vic(Host1x& host1x, std::shared_ptr<Nvdec> nvdec_processor); ~Vic(); @@ -39,8 +42,8 @@ private: void WriteYUVFrame(const AVFrame* frame, const VicConfig& config); - GPU& gpu; - std::shared_ptr<Tegra::Nvdec> nvdec_processor; + Host1x& host1x; + std::shared_ptr<Tegra::Host1x::Nvdec> nvdec_processor; /// Avoid reallocation of the following buffers every frame, as their /// size does not change during a stream @@ -58,4 +61,6 @@ private: s32 scaler_height{}; }; +} // namespace Host1x + } // namespace Tegra diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp index 43f8b5904..f61d5998e 100644 --- a/src/video_core/macro/macro.cpp +++ b/src/video_core/macro/macro.cpp @@ -8,6 +8,7 @@ #include <boost/container_hash/hash.hpp> +#include <fstream> #include "common/assert.h" #include "common/fs/fs.h" #include "common/fs/path_util.h" diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp index aca25d902..a302a9603 100644 --- a/src/video_core/macro/macro_jit_x64.cpp +++ b/src/video_core/macro/macro_jit_x64.cpp @@ -279,28 +279,13 @@ void MacroJITx64Impl::Compile_ExtractInsert(Macro::Opcode opcode) { auto dst = Compile_GetRegister(opcode.src_a, RESULT); auto src = Compile_GetRegister(opcode.src_b, eax); - if (opcode.bf_src_bit != 0 && opcode.bf_src_bit != 31) { - shr(src, opcode.bf_src_bit); - } else if (opcode.bf_src_bit == 31) { - xor_(src, src); - } - // Don't bother masking the whole register since we're using a 32 bit register - if (opcode.bf_size != 31 && opcode.bf_size != 0) { - and_(src, opcode.GetBitfieldMask()); - } else if (opcode.bf_size == 0) { - xor_(src, src); - } - if (opcode.bf_dst_bit != 31 && opcode.bf_dst_bit != 0) { - shl(src, opcode.bf_dst_bit); - } else if (opcode.bf_dst_bit == 31) { - xor_(src, src); - } - const u32 mask = ~(opcode.GetBitfieldMask() << opcode.bf_dst_bit); - if (mask != 0xffffffff) { - and_(dst, mask); - } + and_(dst, mask); + shr(src, opcode.bf_src_bit); + and_(src, opcode.GetBitfieldMask()); + shl(src, opcode.bf_dst_bit); or_(dst, src); + Compile_ProcessResult(opcode.result_operation, opcode.dst); } @@ -309,17 +294,9 @@ void MacroJITx64Impl::Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode) { const auto src = Compile_GetRegister(opcode.src_b, RESULT); shr(src, dst.cvt8()); - if (opcode.bf_size != 0 && opcode.bf_size != 31) { - and_(src, opcode.GetBitfieldMask()); - } else if (opcode.bf_size == 0) { - xor_(src, src); - } + and_(src, opcode.GetBitfieldMask()); + shl(src, opcode.bf_dst_bit); - if (opcode.bf_dst_bit != 0 && opcode.bf_dst_bit != 31) { - shl(src, opcode.bf_dst_bit); - } else if (opcode.bf_dst_bit == 31) { - xor_(src, src); - } Compile_ProcessResult(opcode.result_operation, opcode.dst); } @@ -327,13 +304,8 @@ void MacroJITx64Impl::Compile_ExtractShiftLeftRegister(Macro::Opcode opcode) { const auto dst = Compile_GetRegister(opcode.src_a, ecx); const auto src = Compile_GetRegister(opcode.src_b, RESULT); - if (opcode.bf_src_bit != 0) { - shr(src, opcode.bf_src_bit); - } - - if (opcode.bf_size != 31) { - and_(src, opcode.GetBitfieldMask()); - } + shr(src, opcode.bf_src_bit); + and_(src, opcode.GetBitfieldMask()); shl(src, dst.cvt8()); Compile_ProcessResult(opcode.result_operation, opcode.dst); @@ -429,17 +401,11 @@ void MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) { Xbyak::Label handle_post_exit{}; Xbyak::Label skip{}; jmp(skip, T_NEAR); - if (opcode.is_exit) { - L(handle_post_exit); - // Execute 1 instruction - mov(BRANCH_HOLDER, end_of_code); - // Jump to next instruction to skip delay slot check - jmp(labels[jump_address], T_NEAR); - } else { - L(handle_post_exit); - xor_(BRANCH_HOLDER, BRANCH_HOLDER); - jmp(labels[jump_address], T_NEAR); - } + + L(handle_post_exit); + xor_(BRANCH_HOLDER, BRANCH_HOLDER); + jmp(labels[jump_address], T_NEAR); + L(skip); mov(BRANCH_HOLDER, handle_post_exit); jmp(delay_skip[pc], T_NEAR); diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index bf9eb735d..cca401c74 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -7,6 +7,7 @@ #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 "core/memory.h" @@ -16,172 +17,198 @@ namespace Tegra { -MemoryManager::MemoryManager(Core::System& system_) - : system{system_}, page_table(page_table_size) {} +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.Memory()}, 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}, + unique_identifier{unique_identifier_generator.fetch_add(1, std::memory_order_acq_rel)} { + address_space_size = 1ULL << address_space_bits; + page_size = 1ULL << page_bits; + page_mask = page_size - 1ULL; + big_page_size = 1ULL << big_page_bits; + big_page_mask = big_page_size - 1ULL; + const u64 page_table_bits = address_space_bits - page_bits; + const u64 big_page_table_bits = address_space_bits - big_page_bits; + const u64 page_table_size = 1ULL << page_table_bits; + const u64 big_page_table_size = 1ULL << big_page_table_bits; + page_table_mask = page_table_size - 1; + 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_continous.resize(big_page_table_size / continous_bits, 0); + entries.resize(page_table_size / 32, 0); +} MemoryManager::~MemoryManager() = default; -void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { - rasterizer = rasterizer_; -} - -GPUVAddr MemoryManager::UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size) { - u64 remaining_size{size}; - for (u64 offset{}; offset < size; offset += page_size) { - if (remaining_size < page_size) { - SetPageEntry(gpu_addr + offset, page_entry + offset, remaining_size); - } else { - SetPageEntry(gpu_addr + offset, page_entry + offset); - } - remaining_size -= page_size; +template <bool is_big_page> +MemoryManager::EntryType MemoryManager::GetEntry(size_t position) const { + if constexpr (is_big_page) { + position = position >> big_page_bits; + const u64 entry_mask = big_entries[position / 32]; + const size_t sub_index = position % 32; + return static_cast<EntryType>((entry_mask >> (2 * sub_index)) & 0x03ULL); + } else { + position = position >> page_bits; + const u64 entry_mask = entries[position / 32]; + const size_t sub_index = position % 32; + return static_cast<EntryType>((entry_mask >> (2 * sub_index)) & 0x03ULL); } - return gpu_addr; } -GPUVAddr MemoryManager::Map(VAddr cpu_addr, GPUVAddr gpu_addr, std::size_t size) { - const auto it = std::ranges::lower_bound(map_ranges, gpu_addr, {}, &MapRange::first); - if (it != map_ranges.end() && it->first == gpu_addr) { - it->second = size; +template <bool is_big_page> +void MemoryManager::SetEntry(size_t position, MemoryManager::EntryType entry) { + if constexpr (is_big_page) { + position = position >> big_page_bits; + const u64 entry_mask = big_entries[position / 32]; + const size_t sub_index = position % 32; + big_entries[position / 32] = + (~(3ULL << sub_index * 2) & entry_mask) | (static_cast<u64>(entry) << sub_index * 2); } else { - map_ranges.insert(it, MapRange{gpu_addr, size}); + position = position >> page_bits; + const u64 entry_mask = entries[position / 32]; + const size_t sub_index = position % 32; + entries[position / 32] = + (~(3ULL << sub_index * 2) & entry_mask) | (static_cast<u64>(entry) << sub_index * 2); } - return UpdateRange(gpu_addr, cpu_addr, size); } -GPUVAddr MemoryManager::MapAllocate(VAddr cpu_addr, std::size_t size, std::size_t align) { - return Map(cpu_addr, *FindFreeRange(size, align), size); +inline bool MemoryManager::IsBigPageContinous(size_t big_page_index) const { + const u64 entry_mask = big_page_continous[big_page_index / continous_bits]; + const size_t sub_index = big_page_index % continous_bits; + return ((entry_mask >> sub_index) & 0x1ULL) != 0; } -GPUVAddr MemoryManager::MapAllocate32(VAddr cpu_addr, std::size_t size) { - const std::optional<GPUVAddr> gpu_addr = FindFreeRange(size, 1, true); - ASSERT(gpu_addr); - return Map(cpu_addr, *gpu_addr, size); +inline void MemoryManager::SetBigPageContinous(size_t big_page_index, bool value) { + const u64 continous_mask = big_page_continous[big_page_index / continous_bits]; + const size_t sub_index = big_page_index % continous_bits; + big_page_continous[big_page_index / continous_bits] = + (~(1ULL << sub_index) & continous_mask) | (value ? 1ULL << sub_index : 0); } -void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) { - if (size == 0) { - return; - } - const auto it = std::ranges::lower_bound(map_ranges, gpu_addr, {}, &MapRange::first); - if (it != map_ranges.end()) { - ASSERT(it->first == gpu_addr); - map_ranges.erase(it); - } else { - ASSERT_MSG(false, "Unmapping non-existent GPU address=0x{:x}", gpu_addr); - } - const auto submapped_ranges = GetSubmappedRange(gpu_addr, size); - - for (const auto& [map_addr, map_size] : submapped_ranges) { - // Flush and invalidate through the GPU interface, to be asynchronous if possible. - const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map_addr); - ASSERT(cpu_addr); - - rasterizer->UnmapMemory(*cpu_addr, map_size); +template <MemoryManager::EntryType entry_type> +GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, + size_t size) { + u64 remaining_size{size}; + if constexpr (entry_type == EntryType::Mapped) { + page_table.ReserveRange(gpu_addr, size); } - - UpdateRange(gpu_addr, PageEntry::State::Unmapped, size); -} - -std::optional<GPUVAddr> MemoryManager::AllocateFixed(GPUVAddr gpu_addr, std::size_t size) { for (u64 offset{}; offset < size; offset += page_size) { - if (!GetPageEntry(gpu_addr + offset).IsUnmapped()) { - return std::nullopt; + const GPUVAddr current_gpu_addr = gpu_addr + offset; + [[maybe_unused]] const auto current_entry_type = GetEntry<false>(current_gpu_addr); + SetEntry<false>(current_gpu_addr, entry_type); + if (current_entry_type != entry_type) { + rasterizer->ModifyGPUMemory(unique_identifier, gpu_addr, page_size); + } + if constexpr (entry_type == EntryType::Mapped) { + const VAddr current_cpu_addr = cpu_addr + offset; + const auto index = PageEntryIndex<false>(current_gpu_addr); + const u32 sub_value = static_cast<u32>(current_cpu_addr >> cpu_page_bits); + page_table[index] = sub_value; } + remaining_size -= page_size; } - - return UpdateRange(gpu_addr, PageEntry::State::Allocated, size); -} - -GPUVAddr MemoryManager::Allocate(std::size_t size, std::size_t align) { - return *AllocateFixed(*FindFreeRange(size, align), size); + return gpu_addr; } -void MemoryManager::TryLockPage(PageEntry page_entry, std::size_t size) { - if (!page_entry.IsValid()) { - return; +template <MemoryManager::EntryType entry_type> +GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, + size_t size) { + u64 remaining_size{size}; + for (u64 offset{}; offset < size; offset += big_page_size) { + const GPUVAddr current_gpu_addr = gpu_addr + offset; + [[maybe_unused]] const auto current_entry_type = GetEntry<true>(current_gpu_addr); + SetEntry<true>(current_gpu_addr, entry_type); + if (current_entry_type != entry_type) { + rasterizer->ModifyGPUMemory(unique_identifier, gpu_addr, big_page_size); + } + if constexpr (entry_type == EntryType::Mapped) { + const VAddr current_cpu_addr = cpu_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 bool is_continous = ([&] { + uintptr_t base_ptr{ + reinterpret_cast<uintptr_t>(memory.GetPointerSilent(current_cpu_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) { + base_ptr += page_size; + auto next_ptr = reinterpret_cast<uintptr_t>(memory.GetPointerSilent(start_cpu)); + if (next_ptr == 0 || base_ptr != next_ptr) { + return false; + } + } + return true; + })(); + SetBigPageContinous(index, is_continous); + } + remaining_size -= big_page_size; } - - ASSERT(system.CurrentProcess() - ->PageTable() - .LockForDeviceAddressSpace(page_entry.ToAddress(), size) - .IsSuccess()); + return gpu_addr; } -void MemoryManager::TryUnlockPage(PageEntry page_entry, std::size_t size) { - if (!page_entry.IsValid()) { - return; - } - - ASSERT(system.CurrentProcess() - ->PageTable() - .UnlockForDeviceAddressSpace(page_entry.ToAddress(), size) - .IsSuccess()); +void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; } -PageEntry MemoryManager::GetPageEntry(GPUVAddr gpu_addr) const { - return page_table[PageEntryIndex(gpu_addr)]; +GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size, + bool is_big_pages) { + if (is_big_pages) [[likely]] { + return BigPageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size); + } + return PageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size); } -void MemoryManager::SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size) { - // TODO(bunnei): We should lock/unlock device regions. This currently causes issues due to - // improper tracking, but should be fixed in the future. - - //// Unlock the old page - // TryUnlockPage(page_table[PageEntryIndex(gpu_addr)], size); - - //// Lock the new page - // TryLockPage(page_entry, size); - auto& current_page = page_table[PageEntryIndex(gpu_addr)]; - - if ((!current_page.IsValid() && page_entry.IsValid()) || - current_page.ToAddress() != page_entry.ToAddress()) { - rasterizer->ModifyGPUMemory(gpu_addr, size); +GPUVAddr MemoryManager::MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages) { + if (is_big_pages) [[likely]] { + return BigPageTableOp<EntryType::Reserved>(gpu_addr, 0, size); } - - current_page = page_entry; + return PageTableOp<EntryType::Reserved>(gpu_addr, 0, size); } -std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size_t align, - bool start_32bit_address) const { - if (!align) { - align = page_size; - } else { - align = Common::AlignUp(align, page_size); +void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) { + if (size == 0) { + return; } + const auto submapped_ranges = GetSubmappedRange(gpu_addr, size); - u64 available_size{}; - GPUVAddr gpu_addr{start_32bit_address ? address_space_start_low : address_space_start}; - while (gpu_addr + available_size < address_space_size) { - if (GetPageEntry(gpu_addr + available_size).IsUnmapped()) { - available_size += page_size; - - if (available_size >= size) { - return gpu_addr; - } - } else { - gpu_addr += available_size + page_size; - available_size = 0; + for (const auto& [map_addr, map_size] : submapped_ranges) { + // Flush and invalidate through the GPU interface, to be asynchronous if possible. + const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map_addr); + ASSERT(cpu_addr); - const auto remainder{gpu_addr % align}; - if (remainder) { - gpu_addr = (gpu_addr - remainder) + align; - } - } + rasterizer->UnmapMemory(*cpu_addr, map_size); } - return std::nullopt; + BigPageTableOp<EntryType::Free>(gpu_addr, 0, size); + PageTableOp<EntryType::Free>(gpu_addr, 0, size); } std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) const { - if (gpu_addr == 0) { + if (!IsWithinGPUAddressRange(gpu_addr)) [[unlikely]] { return std::nullopt; } - const auto page_entry{GetPageEntry(gpu_addr)}; - if (!page_entry.IsValid()) { - return std::nullopt; + if (GetEntry<true>(gpu_addr) != EntryType::Mapped) [[unlikely]] { + if (GetEntry<false>(gpu_addr) != EntryType::Mapped) { + return std::nullopt; + } + + const VAddr cpu_addr_base = static_cast<VAddr>(page_table[PageEntryIndex<false>(gpu_addr)]) + << cpu_page_bits; + return cpu_addr_base + (gpu_addr & page_mask); } - return page_entry.ToAddress() + (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); } std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr, std::size_t size) const { @@ -189,7 +216,7 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr, std::size_t s const size_t page_last{(addr + size + page_size - 1) >> page_bits}; while (page_index < page_last) { const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (page_addr && *page_addr != 0) { + if (page_addr) { return page_addr; } ++page_index; @@ -232,126 +259,298 @@ template void MemoryManager::Write<u32>(GPUVAddr addr, u32 data); template void MemoryManager::Write<u64>(GPUVAddr addr, u64 data); u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) { - if (!GetPageEntry(gpu_addr).IsValid()) { - return {}; - } - const auto address{GpuToCpuAddress(gpu_addr)}; if (!address) { return {}; } - return system.Memory().GetPointer(*address); + return memory.GetPointer(*address); } const u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) const { - if (!GetPageEntry(gpu_addr).IsValid()) { - return {}; - } - const auto address{GpuToCpuAddress(gpu_addr)}; if (!address) { return {}; } - return system.Memory().GetPointer(*address); -} - -size_t MemoryManager::BytesToMapEnd(GPUVAddr gpu_addr) const noexcept { - auto it = std::ranges::upper_bound(map_ranges, gpu_addr, {}, &MapRange::first); - --it; - return it->second - (gpu_addr - it->first); -} - -void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, - bool is_safe) const { + return memory.GetPointer(*address); +} + +#ifdef _MSC_VER // no need for gcc / clang but msvc's compiler is more conservative with inlining. +#pragma inline_recursion(on) +#endif + +template <bool is_big_pages, typename FuncMapped, typename FuncReserved, typename FuncUnmapped> +inline void MemoryManager::MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, + FuncMapped&& func_mapped, FuncReserved&& func_reserved, + FuncUnmapped&& func_unmapped) const { + static constexpr bool BOOL_BREAK_MAPPED = std::is_same_v<FuncMapped, bool>; + static constexpr bool BOOL_BREAK_RESERVED = std::is_same_v<FuncReserved, bool>; + static constexpr bool BOOL_BREAK_UNMAPPED = std::is_same_v<FuncUnmapped, bool>; + u64 used_page_size; + u64 used_page_mask; + u64 used_page_bits; + if constexpr (is_big_pages) { + used_page_size = big_page_size; + used_page_mask = big_page_mask; + used_page_bits = big_page_bits; + } else { + used_page_size = page_size; + used_page_mask = page_mask; + used_page_bits = page_bits; + } std::size_t remaining_size{size}; - std::size_t page_index{gpu_src_addr >> page_bits}; - std::size_t page_offset{gpu_src_addr & page_mask}; + std::size_t page_index{gpu_src_addr >> used_page_bits}; + std::size_t page_offset{gpu_src_addr & used_page_mask}; + GPUVAddr current_address = gpu_src_addr; while (remaining_size > 0) { const std::size_t copy_amount{ - std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (page_addr && *page_addr != 0) { - const auto src_addr{*page_addr + page_offset}; - if (is_safe) { - // Flush must happen on the rasterizer interface, such that memory is always - // synchronous when it is read (even when in asynchronous GPU mode). - // Fixes Dead Cells title menu. - rasterizer->FlushRegion(src_addr, copy_amount); + std::min(static_cast<std::size_t>(used_page_size) - page_offset, remaining_size)}; + auto entry = GetEntry<is_big_pages>(current_address); + if (entry == EntryType::Mapped) [[likely]] { + if constexpr (BOOL_BREAK_MAPPED) { + if (func_mapped(page_index, page_offset, copy_amount)) { + return; + } + } else { + func_mapped(page_index, page_offset, copy_amount); } - system.Memory().ReadBlockUnsafe(src_addr, dest_buffer, copy_amount); - } else { - std::memset(dest_buffer, 0, copy_amount); - } + } else if (entry == EntryType::Reserved) { + if constexpr (BOOL_BREAK_RESERVED) { + if (func_reserved(page_index, page_offset, copy_amount)) { + return; + } + } else { + func_reserved(page_index, page_offset, copy_amount); + } + + } else [[unlikely]] { + if constexpr (BOOL_BREAK_UNMAPPED) { + if (func_unmapped(page_index, page_offset, copy_amount)) { + return; + } + } else { + func_unmapped(page_index, page_offset, copy_amount); + } + } page_index++; page_offset = 0; - dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount; remaining_size -= copy_amount; + current_address += copy_amount; } } +template <bool is_safe> +void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, + std::size_t size) const { + auto set_to_zero = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, std::size_t copy_amount) { + std::memset(dest_buffer, 0, copy_amount); + 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; + if constexpr (is_safe) { + rasterizer->FlushRegion(cpu_addr_base, copy_amount); + } + u8* physical = memory.GetPointer(cpu_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; + if constexpr (is_safe) { + rasterizer->FlushRegion(cpu_addr_base, copy_amount); + } + if (!IsBigPageContinous(page_index)) [[unlikely]] { + memory.ReadBlockUnsafe(cpu_addr_base, dest_buffer, copy_amount); + } else { + u8* physical = memory.GetPointer(cpu_addr_base); + std::memcpy(dest_buffer, physical, copy_amount); + } + dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount; + }; + auto read_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, set_to_zero, set_to_zero); + }; + MemoryOperation<true>(gpu_src_addr, size, mapped_big, set_to_zero, read_short_pages); +} + void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const { - ReadBlockImpl(gpu_src_addr, dest_buffer, size, true); + ReadBlockImpl<true>(gpu_src_addr, dest_buffer, size); } void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer, const std::size_t size) const { - ReadBlockImpl(gpu_src_addr, dest_buffer, size, false); + ReadBlockImpl<false>(gpu_src_addr, dest_buffer, size); } -void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, - bool is_safe) { - std::size_t remaining_size{size}; - std::size_t page_index{gpu_dest_addr >> page_bits}; - std::size_t page_offset{gpu_dest_addr & page_mask}; - - while (remaining_size > 0) { - const std::size_t copy_amount{ - std::min(static_cast<std::size_t>(page_size) - page_offset, remaining_size)}; - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (page_addr && *page_addr != 0) { - const auto dest_addr{*page_addr + page_offset}; - - if (is_safe) { - // Invalidate must happen on the rasterizer interface, such that memory is always - // synchronous when it is written (even when in asynchronous GPU mode). - rasterizer->InvalidateRegion(dest_addr, copy_amount); - } - system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount); +template <bool is_safe> +void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, + std::size_t size) { + auto just_advance = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, std::size_t copy_amount) { + 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; + if constexpr (is_safe) { + rasterizer->InvalidateRegion(cpu_addr_base, copy_amount); } - - page_index++; - page_offset = 0; + u8* physical = memory.GetPointer(cpu_addr_base); + std::memcpy(physical, src_buffer, copy_amount); src_buffer = static_cast<const u8*>(src_buffer) + copy_amount; - remaining_size -= 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; + if constexpr (is_safe) { + rasterizer->InvalidateRegion(cpu_addr_base, copy_amount); + } + if (!IsBigPageContinous(page_index)) [[unlikely]] { + memory.WriteBlockUnsafe(cpu_addr_base, src_buffer, copy_amount); + } else { + u8* physical = memory.GetPointer(cpu_addr_base); + std::memcpy(physical, src_buffer, copy_amount); + } + src_buffer = static_cast<const u8*>(src_buffer) + copy_amount; + }; + auto write_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, just_advance, just_advance); + }; + MemoryOperation<true>(gpu_dest_addr, size, mapped_big, just_advance, write_short_pages); } void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) { - WriteBlockImpl(gpu_dest_addr, src_buffer, size, true); + WriteBlockImpl<true>(gpu_dest_addr, src_buffer, size); } void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) { - WriteBlockImpl(gpu_dest_addr, src_buffer, size, false); + WriteBlockImpl<false>(gpu_dest_addr, src_buffer, size); } void MemoryManager::FlushRegion(GPUVAddr gpu_addr, size_t size) const { - size_t remaining_size{size}; - size_t page_index{gpu_addr >> page_bits}; - size_t page_offset{gpu_addr & page_mask}; - while (remaining_size > 0) { - const size_t num_bytes{std::min(page_size - page_offset, remaining_size)}; - if (const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; page_addr) { - rasterizer->FlushRegion(*page_addr + page_offset, num_bytes); + auto do_nothing = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, + [[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); + }; + 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); + }; + auto flush_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing); + }; + MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, flush_short_pages); +} + +bool MemoryManager::IsMemoryDirty(GPUVAddr gpu_addr, size_t size) const { + bool result = false; + auto do_nothing = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, + [[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); + 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); + return result; + }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing); + return result; + }; + MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, check_short_pages); + return result; +} + +size_t MemoryManager::MaxContinousRange(GPUVAddr gpu_addr, size_t size) const { + std::optional<VAddr> 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, + std::size_t copy_amount) { + result = true; + 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) { + result = true; + return true; } - ++page_index; - page_offset = 0; - remaining_size -= num_bytes; - } + range_so_far += copy_amount; + old_page_addr = {cpu_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) { + return true; + } + range_so_far += copy_amount; + old_page_addr = {cpu_addr_base + copy_amount}; + return false; + }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, short_check, fail, fail); + return result; + }; + MemoryOperation<true>(gpu_addr, size, big_check, fail, check_short_pages); + return range_so_far; +} + +void MemoryManager::InvalidateRegion(GPUVAddr gpu_addr, size_t size) const { + auto do_nothing = [&]([[maybe_unused]] std::size_t page_index, + [[maybe_unused]] std::size_t offset, + [[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); + }; + 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); + }; + auto invalidate_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing); + }; + MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, invalidate_short_pages); } void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size) { @@ -365,87 +564,134 @@ void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std } bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const { - const auto cpu_addr{GpuToCpuAddress(gpu_addr)}; - if (!cpu_addr) { + if (GetEntry<true>(gpu_addr) == EntryType::Mapped) [[likely]] { + size_t page_index = gpu_addr >> big_page_bits; + if (IsBigPageContinous(page_index)) [[likely]] { + 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; + } + if (GetEntry<false>(gpu_addr) != EntryType::Mapped) { return false; } - const std::size_t page{(*cpu_addr & Core::Memory::YUZU_PAGEMASK) + size}; + const std::size_t page{(gpu_addr & Core::Memory::YUZU_PAGEMASK) + size}; return page <= Core::Memory::YUZU_PAGESIZE; } bool MemoryManager::IsContinousRange(GPUVAddr gpu_addr, std::size_t size) const { - size_t page_index{gpu_addr >> page_bits}; - const size_t page_last{(gpu_addr + size + page_size - 1) >> page_bits}; std::optional<VAddr> old_page_addr{}; - while (page_index != page_last) { - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (!page_addr || *page_addr == 0) { - return false; + bool result{true}; + auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, + std::size_t copy_amount) { + result = false; + 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) { + result = false; + return true; } - if (old_page_addr) { - if (*old_page_addr + page_size != *page_addr) { - return false; - } + old_page_addr = {cpu_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) { + result = false; + return true; } - old_page_addr = page_addr; - ++page_index; - } - return true; + old_page_addr = {cpu_addr_base + copy_amount}; + return false; + }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, short_check, fail, fail); + return !result; + }; + MemoryOperation<true>(gpu_addr, size, big_check, fail, check_short_pages); + return result; } bool MemoryManager::IsFullyMappedRange(GPUVAddr gpu_addr, std::size_t size) const { - size_t page_index{gpu_addr >> page_bits}; - const size_t page_last{(gpu_addr + size + page_size - 1) >> page_bits}; - while (page_index < page_last) { - if (!page_table[page_index].IsValid() || page_table[page_index].ToAddress() == 0) { - return false; - } - ++page_index; - } - return true; + bool result{true}; + auto fail = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) { + result = false; + return true; + }; + auto pass = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, + [[maybe_unused]] std::size_t copy_amount) { return false; }; + auto check_short_pages = [&](std::size_t page_index, std::size_t offset, + std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, pass, pass, fail); + return !result; + }; + MemoryOperation<true>(gpu_addr, size, pass, fail, check_short_pages); + return result; } std::vector<std::pair<GPUVAddr, std::size_t>> MemoryManager::GetSubmappedRange( GPUVAddr gpu_addr, std::size_t size) const { std::vector<std::pair<GPUVAddr, std::size_t>> result{}; - size_t page_index{gpu_addr >> page_bits}; - size_t remaining_size{size}; - size_t page_offset{gpu_addr & page_mask}; std::optional<std::pair<GPUVAddr, std::size_t>> last_segment{}; std::optional<VAddr> old_page_addr{}; - const auto extend_size = [&last_segment, &page_index, &page_offset](std::size_t bytes) { - if (!last_segment) { - const GPUVAddr new_base_addr = (page_index << page_bits) + page_offset; - last_segment = {new_base_addr, bytes}; - } else { - last_segment->second += bytes; - } - }; - const auto split = [&last_segment, &result] { + 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) { if (last_segment) { result.push_back(*last_segment); last_segment = std::nullopt; } }; - while (remaining_size > 0) { - const size_t num_bytes{std::min(page_size - page_offset, remaining_size)}; - const auto page_addr{GpuToCpuAddress(page_index << page_bits)}; - if (!page_addr || *page_addr == 0) { - split(); - } else if (old_page_addr) { - if (*old_page_addr + page_size != *page_addr) { - split(); + 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; + if (old_page_addr) { + if (*old_page_addr != cpu_addr_base) { + split(0, 0, 0); + } + } + old_page_addr = {cpu_addr_base + copy_amount}; + if (!last_segment) { + const GPUVAddr new_base_addr = (page_index << big_page_bits) + offset; + last_segment = {new_base_addr, copy_amount}; + } else { + last_segment->second += copy_amount; + } + }; + 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; + if (old_page_addr) { + if (*old_page_addr != cpu_addr_base) { + split(0, 0, 0); } - extend_size(num_bytes); + } + old_page_addr = {cpu_addr_base + copy_amount}; + if (!last_segment) { + const GPUVAddr new_base_addr = (page_index << page_bits) + offset; + last_segment = {new_base_addr, copy_amount}; } else { - extend_size(num_bytes); + last_segment->second += copy_amount; } - ++page_index; - page_offset = 0; - remaining_size -= num_bytes; - old_page_addr = page_addr; - } - split(); + }; + auto do_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) { + GPUVAddr base = (page_index << big_page_bits) + offset; + MemoryOperation<false>(base, copy_amount, extend_size_short, split, split); + }; + MemoryOperation<true>(gpu_addr, size, extend_size_big, split, do_short_pages); + split(0, 0, 0); return result; } diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index 74f9ce175..f992e29f3 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -3,73 +3,39 @@ #pragma once +#include <atomic> #include <map> #include <optional> #include <vector> #include "common/common_types.h" +#include "common/multi_level_page_table.h" +#include "common/virtual_buffer.h" namespace VideoCore { class RasterizerInterface; } namespace Core { +class DeviceMemory; +namespace Memory { +class Memory; +} // namespace Memory class System; -} +} // namespace Core namespace Tegra { -class PageEntry final { -public: - enum class State : u32 { - Unmapped = static_cast<u32>(-1), - Allocated = static_cast<u32>(-2), - }; - - constexpr PageEntry() = default; - constexpr PageEntry(State state_) : state{state_} {} - constexpr PageEntry(VAddr addr) : state{static_cast<State>(addr >> ShiftBits)} {} - - [[nodiscard]] constexpr bool IsUnmapped() const { - return state == State::Unmapped; - } - - [[nodiscard]] constexpr bool IsAllocated() const { - return state == State::Allocated; - } - - [[nodiscard]] constexpr bool IsValid() const { - return !IsUnmapped() && !IsAllocated(); - } - - [[nodiscard]] constexpr VAddr ToAddress() const { - if (!IsValid()) { - return {}; - } - - return static_cast<VAddr>(state) << ShiftBits; - } - - [[nodiscard]] constexpr PageEntry operator+(u64 offset) const { - // If this is a reserved value, offsets do not apply - if (!IsValid()) { - return *this; - } - return PageEntry{(static_cast<VAddr>(state) << ShiftBits) + offset}; - } - -private: - static constexpr std::size_t ShiftBits{12}; - - State state{State::Unmapped}; -}; -static_assert(sizeof(PageEntry) == 4, "PageEntry is too large"); - class MemoryManager final { public: - explicit MemoryManager(Core::System& system_); + explicit MemoryManager(Core::System& system_, u64 address_space_bits_ = 40, + u64 big_page_bits_ = 16, u64 page_bits_ = 12); ~MemoryManager(); + size_t GetID() const { + return unique_identifier; + } + /// Binds a renderer to the memory manager. void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); @@ -86,9 +52,6 @@ public: [[nodiscard]] u8* GetPointer(GPUVAddr addr); [[nodiscard]] const u8* GetPointer(GPUVAddr addr) const; - /// Returns the number of bytes until the end of the memory map containing the given GPU address - [[nodiscard]] size_t BytesToMapEnd(GPUVAddr gpu_addr) const noexcept; - /** * ReadBlock and WriteBlock are full read and write operations over virtual * GPU Memory. It's important to use these when GPU memory may not be continuous @@ -135,54 +98,95 @@ public: std::vector<std::pair<GPUVAddr, std::size_t>> GetSubmappedRange(GPUVAddr gpu_addr, std::size_t size) const; - [[nodiscard]] GPUVAddr Map(VAddr cpu_addr, GPUVAddr gpu_addr, std::size_t size); - [[nodiscard]] GPUVAddr MapAllocate(VAddr cpu_addr, std::size_t size, std::size_t align); - [[nodiscard]] GPUVAddr MapAllocate32(VAddr cpu_addr, std::size_t size); - [[nodiscard]] std::optional<GPUVAddr> AllocateFixed(GPUVAddr gpu_addr, std::size_t size); - [[nodiscard]] GPUVAddr Allocate(std::size_t size, std::size_t align); + GPUVAddr Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size, 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); void FlushRegion(GPUVAddr gpu_addr, size_t size) const; + void InvalidateRegion(GPUVAddr gpu_addr, size_t size) const; + + bool IsMemoryDirty(GPUVAddr gpu_addr, size_t size) const; + + size_t MaxContinousRange(GPUVAddr gpu_addr, size_t size) const; + + bool IsWithinGPUAddressRange(GPUVAddr gpu_addr) const { + return gpu_addr < address_space_size; + } + private: - [[nodiscard]] PageEntry GetPageEntry(GPUVAddr gpu_addr) const; - void SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size = page_size); - GPUVAddr UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size); - [[nodiscard]] std::optional<GPUVAddr> FindFreeRange(std::size_t size, std::size_t align, - bool start_32bit_address = false) const; - - void TryLockPage(PageEntry page_entry, std::size_t size); - void TryUnlockPage(PageEntry page_entry, std::size_t size); - - void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, - bool is_safe) const; - void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, - bool is_safe); - - [[nodiscard]] static constexpr std::size_t PageEntryIndex(GPUVAddr gpu_addr) { - return (gpu_addr >> page_bits) & page_table_mask; + template <bool is_big_pages, typename FuncMapped, typename FuncReserved, typename FuncUnmapped> + inline void MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, FuncMapped&& func_mapped, + FuncReserved&& func_reserved, FuncUnmapped&& func_unmapped) const; + + template <bool is_safe> + void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size) const; + + template <bool is_safe> + void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size); + + template <bool is_big_page> + [[nodiscard]] std::size_t PageEntryIndex(GPUVAddr gpu_addr) const { + if constexpr (is_big_page) { + return (gpu_addr >> big_page_bits) & big_page_table_mask; + } else { + return (gpu_addr >> page_bits) & page_table_mask; + } } - static constexpr u64 address_space_size = 1ULL << 40; - static constexpr u64 address_space_start = 1ULL << 32; - static constexpr u64 address_space_start_low = 1ULL << 16; - static constexpr u64 page_bits{16}; - static constexpr u64 page_size{1 << page_bits}; - static constexpr u64 page_mask{page_size - 1}; - static constexpr u64 page_table_bits{24}; - static constexpr u64 page_table_size{1 << page_table_bits}; - static constexpr u64 page_table_mask{page_table_size - 1}; + inline bool IsBigPageContinous(size_t big_page_index) const; + inline void SetBigPageContinous(size_t big_page_index, bool value); Core::System& system; + Core::Memory::Memory& memory; + Core::DeviceMemory& device_memory; + + const u64 address_space_bits; + const u64 page_bits; + u64 address_space_size; + u64 page_size; + u64 page_mask; + u64 page_table_mask; + static constexpr u64 cpu_page_bits{12}; + + const u64 big_page_bits; + u64 big_page_size; + u64 big_page_mask; + u64 big_page_table_mask; VideoCore::RasterizerInterface* rasterizer = nullptr; - std::vector<PageEntry> page_table; + enum class EntryType : u64 { + Free = 0, + Reserved = 1, + Mapped = 2, + }; + + std::vector<u64> entries; + std::vector<u64> big_entries; + + template <EntryType entry_type> + GPUVAddr PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size); + + template <EntryType entry_type> + GPUVAddr BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] VAddr cpu_addr, size_t size); + + template <bool is_big_page> + inline EntryType GetEntry(size_t position) const; + + template <bool is_big_page> + inline void SetEntry(size_t position, EntryType entry); + + Common::MultiLevelPageTable<u32> page_table; + Common::VirtualBuffer<u32> big_page_table_cpu; + + std::vector<u64> big_page_continous; + + constexpr static size_t continous_bits = 64; - using MapRange = std::pair<GPUVAddr, size_t>; - std::vector<MapRange> map_ranges; + const size_t unique_identifier; - std::vector<std::pair<VAddr, std::size_t>> cache_invalidate_queue; + static std::atomic<size_t> unique_identifier_generator; }; } // namespace Tegra diff --git a/src/video_core/query_cache.h b/src/video_core/query_cache.h index 889b606b3..b0ebe71b7 100644 --- a/src/video_core/query_cache.h +++ b/src/video_core/query_cache.h @@ -17,6 +17,7 @@ #include "common/assert.h" #include "common/settings.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" @@ -90,13 +91,10 @@ private: }; template <class QueryCache, class CachedQuery, class CounterStream, class HostCounter> -class QueryCacheBase { +class QueryCacheBase : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { public: - explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::MemoryManager& gpu_memory_) - : rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, - gpu_memory{gpu_memory_}, streams{{CounterStream{static_cast<QueryCache&>(*this), + explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_) + : rasterizer{rasterizer_}, streams{{CounterStream{static_cast<QueryCache&>(*this), VideoCore::QueryType::SamplesPassed}}} {} void InvalidateRegion(VAddr addr, std::size_t size) { @@ -117,13 +115,13 @@ public: */ void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) { std::unique_lock lock{mutex}; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); + const std::optional<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); ASSERT(cpu_addr); CachedQuery* query = TryGet(*cpu_addr); if (!query) { ASSERT_OR_EXECUTE(cpu_addr, return;); - u8* const host_ptr = gpu_memory.GetPointer(gpu_addr); + u8* const host_ptr = gpu_memory->GetPointer(gpu_addr); query = Register(type, *cpu_addr, host_ptr, timestamp.has_value()); } @@ -137,8 +135,10 @@ public: /// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch. void UpdateCounters() { std::unique_lock lock{mutex}; - const auto& regs = maxwell3d.regs; - Stream(VideoCore::QueryType::SamplesPassed).Update(regs.samplecnt_enable); + if (maxwell3d) { + const auto& regs = maxwell3d->regs; + Stream(VideoCore::QueryType::SamplesPassed).Update(regs.samplecnt_enable); + } } /// Resets a counter to zero. It doesn't disable the query after resetting. @@ -264,8 +264,6 @@ private: static constexpr unsigned YUZU_PAGEBITS = 12; VideoCore::RasterizerInterface& rasterizer; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::MemoryManager& gpu_memory; std::recursive_mutex mutex; diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index a04a76481..d2d40884c 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -16,6 +16,9 @@ class MemoryManager; namespace Engines { class AccelerateDMAInterface; } +namespace Control { +struct ChannelState; +} } // namespace Tegra namespace VideoCore { @@ -59,7 +62,10 @@ public: virtual void DisableGraphicsUniformBuffer(size_t stage, u32 index) = 0; /// Signal a GPU based semaphore as a fence - virtual void SignalSemaphore(GPUVAddr addr, u32 value) = 0; + virtual void SignalFence(std::function<void()>&& func) = 0; + + /// Send an operation to be done after a certain amount of flushes. + virtual void SyncOperation(std::function<void()>&& func) = 0; /// Signal a GPU based syncpoint as a fence virtual void SignalSyncPoint(u32 value) = 0; @@ -86,13 +92,13 @@ public: virtual void OnCPUWrite(VAddr addr, u64 size) = 0; /// Sync memory between guest and host. - virtual void SyncGuestHost() = 0; + virtual void InvalidateGPUCache() = 0; /// Unmap memory range virtual void UnmapMemory(VAddr addr, u64 size) = 0; /// Remap GPU memory range. This means underneath backing memory changed - virtual void ModifyGPUMemory(GPUVAddr addr, u64 size) = 0; + virtual void ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) = 0; /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory /// and invalidated @@ -123,7 +129,7 @@ public: [[nodiscard]] virtual Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() = 0; virtual void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) = 0; + std::span<const u8> memory) = 0; /// Attempt to use a faster method to display the framebuffer to screen [[nodiscard]] virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config, @@ -137,5 +143,11 @@ public: /// Initialize disk cached resources for the game being emulated virtual void LoadDiskResources(u64 title_id, std::stop_token stop_loading, const DiskResourceLoadCallback& callback) {} + + virtual void InitializeChannel(Tegra::Control::ChannelState& channel) {} + + virtual void BindChannel(Tegra::Control::ChannelState& channel) {} + + virtual void ReleaseChannel(s32 channel_id) {} }; } // namespace VideoCore diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp index 1f0f156ed..26d066004 100644 --- a/src/video_core/renderer_opengl/gl_compute_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.cpp @@ -28,12 +28,11 @@ bool ComputePipelineKey::operator==(const ComputePipelineKey& rhs) const noexcep } ComputePipeline::ComputePipeline(const Device& device, TextureCache& texture_cache_, - BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, - Tegra::Engines::KeplerCompute& kepler_compute_, - ProgramManager& program_manager_, const Shader::Info& info_, - std::string code, std::vector<u32> code_v) - : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, gpu_memory{gpu_memory_}, - kepler_compute{kepler_compute_}, program_manager{program_manager_}, info{info_} { + BufferCache& buffer_cache_, ProgramManager& program_manager_, + const Shader::Info& info_, std::string code, + std::vector<u32> code_v) + : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, + program_manager{program_manager_}, info{info_} { switch (device.GetShaderBackend()) { case Settings::ShaderBackend::GLSL: source_program = CreateProgram(code, GL_COMPUTE_SHADER); @@ -86,7 +85,7 @@ void ComputePipeline::Configure() { GLsizei texture_binding{}; GLsizei image_binding{}; - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const auto& cbufs{qmd.const_buffer_config}; const bool via_header_index{qmd.linked_tsc != 0}; const auto read_handle{[&](const auto& desc, u32 index) { @@ -101,12 +100,13 @@ void ComputePipeline::Configure() { const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() + secondary_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory->Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory->Read<u32>(separate_addr) + << desc.secondary_shift_left}; return TexturePair(lhs_raw | rhs_raw, via_header_index); } } - return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + return TexturePair(gpu_memory->Read<u32>(addr), via_header_index); }}; const auto add_image{[&](const auto& desc, bool blacklist) { for (u32 index = 0; index < desc.count; ++index) { diff --git a/src/video_core/renderer_opengl/gl_compute_pipeline.h b/src/video_core/renderer_opengl/gl_compute_pipeline.h index 723f27f11..6534dec32 100644 --- a/src/video_core/renderer_opengl/gl_compute_pipeline.h +++ b/src/video_core/renderer_opengl/gl_compute_pipeline.h @@ -49,10 +49,8 @@ static_assert(std::is_trivially_constructible_v<ComputePipelineKey>); class ComputePipeline { public: explicit ComputePipeline(const Device& device, TextureCache& texture_cache_, - BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, - Tegra::Engines::KeplerCompute& kepler_compute_, - ProgramManager& program_manager_, const Shader::Info& info_, - std::string code, std::vector<u32> code_v); + BufferCache& buffer_cache_, ProgramManager& program_manager_, + const Shader::Info& info_, std::string code, std::vector<u32> code_v); void Configure(); @@ -60,11 +58,17 @@ public: return writes_global_memory; } + void SetEngine(Tegra::Engines::KeplerCompute* kepler_compute_, + Tegra::MemoryManager* gpu_memory_) { + kepler_compute = kepler_compute_; + gpu_memory = gpu_memory_; + } + private: TextureCache& texture_cache; BufferCache& buffer_cache; - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::KeplerCompute& kepler_compute; + Tegra::MemoryManager* gpu_memory; + Tegra::Engines::KeplerCompute* kepler_compute; ProgramManager& program_manager; Shader::Info info; diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp index 6e82c2e28..91463f854 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.cpp +++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp @@ -10,10 +10,7 @@ namespace OpenGL { -GLInnerFence::GLInnerFence(u32 payload_, bool is_stubbed_) : FenceBase{payload_, is_stubbed_} {} - -GLInnerFence::GLInnerFence(GPUVAddr address_, u32 payload_, bool is_stubbed_) - : FenceBase{address_, payload_, is_stubbed_} {} +GLInnerFence::GLInnerFence(bool is_stubbed_) : FenceBase{is_stubbed_} {} GLInnerFence::~GLInnerFence() = default; @@ -48,12 +45,8 @@ FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterize BufferCache& buffer_cache_, QueryCache& query_cache_) : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_} {} -Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) { - return std::make_shared<GLInnerFence>(value, is_stubbed); -} - -Fence FenceManagerOpenGL::CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) { - return std::make_shared<GLInnerFence>(addr, value, is_stubbed); +Fence FenceManagerOpenGL::CreateFence(bool is_stubbed) { + return std::make_shared<GLInnerFence>(is_stubbed); } void FenceManagerOpenGL::QueueFence(Fence& fence) { diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h index 14ff00db2..f1446e732 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.h +++ b/src/video_core/renderer_opengl/gl_fence_manager.h @@ -16,8 +16,7 @@ namespace OpenGL { class GLInnerFence : public VideoCommon::FenceBase { public: - explicit GLInnerFence(u32 payload_, bool is_stubbed_); - explicit GLInnerFence(GPUVAddr address_, u32 payload_, bool is_stubbed_); + explicit GLInnerFence(bool is_stubbed_); ~GLInnerFence(); void Queue(); @@ -40,8 +39,7 @@ public: QueryCache& query_cache); protected: - Fence CreateFence(u32 value, bool is_stubbed) override; - Fence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) override; + Fence CreateFence(bool is_stubbed) override; void QueueFence(Fence& fence) override; bool IsFenceSignaled(Fence& fence) const override; void WaitFence(Fence& fence) override; diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp index 67eae369d..41493a7da 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp @@ -169,15 +169,15 @@ ConfigureFuncPtr ConfigureFunc(const std::array<Shader::Info, 5>& infos, u32 ena } } // Anonymous namespace -GraphicsPipeline::GraphicsPipeline( - const Device& device, TextureCache& texture_cache_, BufferCache& buffer_cache_, - Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, - ProgramManager& program_manager_, StateTracker& state_tracker_, ShaderWorker* thread_worker, - VideoCore::ShaderNotify* shader_notify, std::array<std::string, 5> sources, - std::array<std::vector<u32>, 5> sources_spirv, const std::array<const Shader::Info*, 5>& infos, - const GraphicsPipelineKey& key_) - : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, - gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, program_manager{program_manager_}, +GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_cache_, + BufferCache& buffer_cache_, ProgramManager& program_manager_, + StateTracker& state_tracker_, ShaderWorker* thread_worker, + VideoCore::ShaderNotify* shader_notify, + std::array<std::string, 5> sources, + std::array<std::vector<u32>, 5> sources_spirv, + const std::array<const Shader::Info*, 5>& infos, + const GraphicsPipelineKey& key_) + : texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, program_manager{program_manager_}, state_tracker{state_tracker_}, key{key_} { if (shader_notify) { shader_notify->MarkShaderBuilding(); @@ -285,7 +285,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { buffer_cache.runtime.SetBaseStorageBindings(base_storage_bindings); buffer_cache.runtime.SetEnableStorageBuffers(use_storage_buffers); - const auto& regs{maxwell3d.regs}; + const auto& regs{maxwell3d->regs}; const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex}; const auto config_stage{[&](size_t stage) LAMBDA_FORCEINLINE { const Shader::Info& info{stage_infos[stage]}; @@ -299,7 +299,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { ++ssbo_index; } } - const auto& cbufs{maxwell3d.state.shader_stages[stage].const_buffers}; + const auto& cbufs{maxwell3d->state.shader_stages[stage].const_buffers}; const auto read_handle{[&](const auto& desc, u32 index) { ASSERT(cbufs[desc.cbuf_index].enabled); const u32 index_offset{index << desc.size_shift}; @@ -312,13 +312,14 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { const u32 second_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].address + second_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory->Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory->Read<u32>(separate_addr) + << desc.secondary_shift_left}; const u32 raw{lhs_raw | rhs_raw}; return TexturePair(raw, via_header_index); } } - return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + return TexturePair(gpu_memory->Read<u32>(addr), via_header_index); }}; const auto add_image{[&](const auto& desc, bool blacklist) LAMBDA_FORCEINLINE { for (u32 index = 0; index < desc.count; ++index) { diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.h b/src/video_core/renderer_opengl/gl_graphics_pipeline.h index 4ec15b966..a0f0e63cb 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.h +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.h @@ -71,10 +71,9 @@ static_assert(std::is_trivially_constructible_v<GraphicsPipelineKey>); class GraphicsPipeline { public: explicit GraphicsPipeline(const Device& device, TextureCache& texture_cache_, - BufferCache& buffer_cache_, Tegra::MemoryManager& gpu_memory_, - Tegra::Engines::Maxwell3D& maxwell3d_, - ProgramManager& program_manager_, StateTracker& state_tracker_, - ShaderWorker* thread_worker, VideoCore::ShaderNotify* shader_notify, + BufferCache& buffer_cache_, ProgramManager& program_manager_, + StateTracker& state_tracker_, ShaderWorker* thread_worker, + VideoCore::ShaderNotify* shader_notify, std::array<std::string, 5> sources, std::array<std::vector<u32>, 5> sources_spirv, const std::array<const Shader::Info*, 5>& infos, @@ -107,6 +106,11 @@ public: }; } + void SetEngine(Tegra::Engines::Maxwell3D* maxwell3d_, Tegra::MemoryManager* gpu_memory_) { + maxwell3d = maxwell3d_; + gpu_memory = gpu_memory_; + } + private: template <typename Spec> void ConfigureImpl(bool is_indexed); @@ -119,8 +123,8 @@ private: TextureCache& texture_cache; BufferCache& buffer_cache; - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::MemoryManager* gpu_memory; + Tegra::Engines::Maxwell3D* maxwell3d; ProgramManager& program_manager; StateTracker& state_tracker; const GraphicsPipelineKey key; diff --git a/src/video_core/renderer_opengl/gl_query_cache.cpp b/src/video_core/renderer_opengl/gl_query_cache.cpp index ed40f5791..5070db441 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.cpp +++ b/src/video_core/renderer_opengl/gl_query_cache.cpp @@ -26,9 +26,8 @@ constexpr GLenum GetTarget(VideoCore::QueryType type) { } // Anonymous namespace -QueryCache::QueryCache(RasterizerOpenGL& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::MemoryManager& gpu_memory_) - : QueryCacheBase(rasterizer_, maxwell3d_, gpu_memory_), gl_rasterizer{rasterizer_} {} +QueryCache::QueryCache(RasterizerOpenGL& rasterizer_) + : QueryCacheBase(rasterizer_), gl_rasterizer{rasterizer_} {} QueryCache::~QueryCache() = default; diff --git a/src/video_core/renderer_opengl/gl_query_cache.h b/src/video_core/renderer_opengl/gl_query_cache.h index 8a49f1ef0..14ce59990 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.h +++ b/src/video_core/renderer_opengl/gl_query_cache.h @@ -28,8 +28,7 @@ using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>; class QueryCache final : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> { public: - explicit QueryCache(RasterizerOpenGL& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::MemoryManager& gpu_memory_); + explicit QueryCache(RasterizerOpenGL& rasterizer_); ~QueryCache(); OGLQuery AllocateQuery(VideoCore::QueryType type); diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index a0d048b0b..c2d80605d 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -16,7 +16,7 @@ #include "common/microprofile.h" #include "common/scope_exit.h" #include "common/settings.h" - +#include "video_core/control/channel_state.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" @@ -56,22 +56,20 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra Core::Memory::Memory& cpu_memory_, const Device& device_, ScreenInfo& screen_info_, ProgramManager& program_manager_, StateTracker& state_tracker_) - : RasterizerAccelerated(cpu_memory_), gpu(gpu_), maxwell3d(gpu.Maxwell3D()), - kepler_compute(gpu.KeplerCompute()), gpu_memory(gpu.MemoryManager()), device(device_), - screen_info(screen_info_), program_manager(program_manager_), state_tracker(state_tracker_), + : RasterizerAccelerated(cpu_memory_), gpu(gpu_), device(device_), screen_info(screen_info_), + program_manager(program_manager_), state_tracker(state_tracker_), texture_cache_runtime(device, program_manager, state_tracker), - texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), - buffer_cache_runtime(device), - buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime), - shader_cache(*this, emu_window_, maxwell3d, kepler_compute, gpu_memory, device, texture_cache, - buffer_cache, program_manager, state_tracker, gpu.ShaderNotify()), - query_cache(*this, maxwell3d, gpu_memory), accelerate_dma(buffer_cache), + texture_cache(texture_cache_runtime, *this), buffer_cache_runtime(device), + 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), accelerate_dma(buffer_cache), fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache) {} RasterizerOpenGL::~RasterizerOpenGL() = default; void RasterizerOpenGL::SyncVertexFormats() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::VertexFormats]) { return; } @@ -89,7 +87,7 @@ void RasterizerOpenGL::SyncVertexFormats() { } flags[Dirty::VertexFormat0 + index] = false; - const auto attrib = maxwell3d.regs.vertex_attrib_format[index]; + const auto attrib = maxwell3d->regs.vertex_attrib_format[index]; const auto gl_index = static_cast<GLuint>(index); // Disable constant attributes. @@ -113,13 +111,13 @@ void RasterizerOpenGL::SyncVertexFormats() { } void RasterizerOpenGL::SyncVertexInstances() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::VertexInstances]) { return; } flags[Dirty::VertexInstances] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_ATTRIBUTES; ++index) { if (!flags[Dirty::VertexInstance0 + index]) { continue; @@ -140,11 +138,11 @@ void RasterizerOpenGL::LoadDiskResources(u64 title_id, std::stop_token stop_load void RasterizerOpenGL::Clear() { MICROPROFILE_SCOPE(OpenGL_Clears); - if (!maxwell3d.ShouldExecute()) { + if (!maxwell3d->ShouldExecute()) { return; } - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; bool use_color{}; bool use_depth{}; bool use_stencil{}; @@ -217,22 +215,26 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { if (!pipeline) { return; } + + gpu.TickWork(); + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + pipeline->SetEngine(maxwell3d, gpu_memory); pipeline->Configure(is_indexed); SyncState(); - const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology); + const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d->regs.draw.topology); BeginTransformFeedback(pipeline, primitive_mode); - const GLuint base_instance = static_cast<GLuint>(maxwell3d.regs.vb_base_instance); + const GLuint base_instance = static_cast<GLuint>(maxwell3d->regs.vb_base_instance); const GLsizei num_instances = - static_cast<GLsizei>(is_instanced ? maxwell3d.mme_draw.instance_count : 1); + static_cast<GLsizei>(is_instanced ? maxwell3d->mme_draw.instance_count : 1); if (is_indexed) { - const GLint base_vertex = static_cast<GLint>(maxwell3d.regs.vb_element_base); - const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d.regs.index_array.count); + const GLint base_vertex = static_cast<GLint>(maxwell3d->regs.vb_element_base); + const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d->regs.index_array.count); const GLvoid* const offset = buffer_cache_runtime.IndexOffset(); - const GLenum format = MaxwellToGL::IndexFormat(maxwell3d.regs.index_array.format); + const GLenum format = MaxwellToGL::IndexFormat(maxwell3d->regs.index_array.format); if (num_instances == 1 && base_instance == 0 && base_vertex == 0) { glDrawElements(primitive_mode, num_vertices, format, offset); } else if (num_instances == 1 && base_instance == 0) { @@ -251,8 +253,8 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { base_instance); } } else { - const GLint base_vertex = static_cast<GLint>(maxwell3d.regs.vertex_buffer.first); - const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d.regs.vertex_buffer.count); + const GLint base_vertex = static_cast<GLint>(maxwell3d->regs.vertex_buffer.first); + const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d->regs.vertex_buffer.count); if (num_instances == 1 && base_instance == 0) { glDrawArrays(primitive_mode, base_vertex, num_vertices); } else if (base_instance == 0) { @@ -273,8 +275,9 @@ void RasterizerOpenGL::DispatchCompute() { if (!pipeline) { return; } + pipeline->SetEngine(kepler_compute, gpu_memory); pipeline->Configure(); - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; glDispatchCompute(qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z); ++num_queued_commands; has_written_global_memory |= pipeline->WritesGlobalMemory(); @@ -359,7 +362,7 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { } } -void RasterizerOpenGL::SyncGuestHost() { +void RasterizerOpenGL::InvalidateGPUCache() { MICROPROFILE_SCOPE(OpenGL_CacheManagement); shader_cache.SyncGuestHost(); { @@ -380,40 +383,30 @@ void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { shader_cache.OnCPUWrite(addr, size); } -void RasterizerOpenGL::ModifyGPUMemory(GPUVAddr addr, u64 size) { +void RasterizerOpenGL::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { { std::scoped_lock lock{texture_cache.mutex}; - texture_cache.UnmapGPUMemory(addr, size); + texture_cache.UnmapGPUMemory(as_id, addr, size); } } -void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) { - if (!gpu.IsAsync()) { - gpu_memory.Write<u32>(addr, value); - return; - } - fence_manager.SignalSemaphore(addr, value); +void RasterizerOpenGL::SignalFence(std::function<void()>&& func) { + fence_manager.SignalFence(std::move(func)); +} + +void RasterizerOpenGL::SyncOperation(std::function<void()>&& func) { + fence_manager.SyncOperation(std::move(func)); } void RasterizerOpenGL::SignalSyncPoint(u32 value) { - if (!gpu.IsAsync()) { - gpu.IncrementSyncPoint(value); - return; - } fence_manager.SignalSyncPoint(value); } void RasterizerOpenGL::SignalReference() { - if (!gpu.IsAsync()) { - return; - } fence_manager.SignalOrdering(); } void RasterizerOpenGL::ReleaseFences() { - if (!gpu.IsAsync()) { - return; - } fence_manager.WaitPendingFences(); } @@ -430,6 +423,7 @@ void RasterizerOpenGL::WaitForIdle() { } void RasterizerOpenGL::FragmentBarrier() { + glTextureBarrier(); glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT | GL_TEXTURE_FETCH_BARRIER_BIT); } @@ -482,13 +476,13 @@ Tegra::Engines::AccelerateDMAInterface& RasterizerOpenGL::AccessAccelerateDMA() } void RasterizerOpenGL::AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) { - auto cpu_addr = gpu_memory.GpuToCpuAddress(address); + std::span<const u8> memory) { + auto cpu_addr = gpu_memory->GpuToCpuAddress(address); if (!cpu_addr) [[unlikely]] { - gpu_memory.WriteBlock(address, memory.data(), copy_size); + gpu_memory->WriteBlock(address, memory.data(), copy_size); return; } - gpu_memory.WriteBlockUnsafe(address, memory.data(), copy_size); + gpu_memory->WriteBlockUnsafe(address, memory.data(), copy_size); { std::unique_lock<std::mutex> lock{buffer_cache.mutex}; if (!buffer_cache.InlineMemory(*cpu_addr, copy_size, memory)) { @@ -551,8 +545,8 @@ void RasterizerOpenGL::SyncState() { } void RasterizerOpenGL::SyncViewport() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; const bool rescale_viewports = flags[VideoCommon::Dirty::RescaleViewports]; const bool dirty_viewport = flags[Dirty::Viewports] || rescale_viewports; @@ -657,23 +651,23 @@ void RasterizerOpenGL::SyncViewport() { } void RasterizerOpenGL::SyncDepthClamp() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::DepthClampEnabled]) { return; } flags[Dirty::DepthClampEnabled] = false; - oglEnable(GL_DEPTH_CLAMP, maxwell3d.regs.view_volume_clip_control.depth_clamp_disabled == 0); + oglEnable(GL_DEPTH_CLAMP, maxwell3d->regs.view_volume_clip_control.depth_clamp_disabled == 0); } void RasterizerOpenGL::SyncClipEnabled(u32 clip_mask) { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::ClipDistances] && !flags[VideoCommon::Dirty::Shaders]) { return; } flags[Dirty::ClipDistances] = false; - clip_mask &= maxwell3d.regs.clip_distance_enabled; + clip_mask &= maxwell3d->regs.clip_distance_enabled; if (clip_mask == last_clip_distance_mask) { return; } @@ -689,8 +683,8 @@ void RasterizerOpenGL::SyncClipCoef() { } void RasterizerOpenGL::SyncCullMode() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; if (flags[Dirty::CullTest]) { flags[Dirty::CullTest] = false; @@ -705,23 +699,23 @@ void RasterizerOpenGL::SyncCullMode() { } void RasterizerOpenGL::SyncPrimitiveRestart() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PrimitiveRestart]) { return; } flags[Dirty::PrimitiveRestart] = false; - if (maxwell3d.regs.primitive_restart.enabled) { + if (maxwell3d->regs.primitive_restart.enabled) { glEnable(GL_PRIMITIVE_RESTART); - glPrimitiveRestartIndex(maxwell3d.regs.primitive_restart.index); + glPrimitiveRestartIndex(maxwell3d->regs.primitive_restart.index); } else { glDisable(GL_PRIMITIVE_RESTART); } } void RasterizerOpenGL::SyncDepthTestState() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; if (flags[Dirty::DepthMask]) { flags[Dirty::DepthMask] = false; @@ -740,13 +734,13 @@ void RasterizerOpenGL::SyncDepthTestState() { } void RasterizerOpenGL::SyncStencilTestState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::StencilTest]) { return; } flags[Dirty::StencilTest] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_STENCIL_TEST, regs.stencil_enable); glStencilFuncSeparate(GL_FRONT, MaxwellToGL::ComparisonOp(regs.stencil_front_func_func), @@ -771,23 +765,23 @@ void RasterizerOpenGL::SyncStencilTestState() { } void RasterizerOpenGL::SyncRasterizeEnable() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::RasterizeEnable]) { return; } flags[Dirty::RasterizeEnable] = false; - oglEnable(GL_RASTERIZER_DISCARD, maxwell3d.regs.rasterize_enable == 0); + oglEnable(GL_RASTERIZER_DISCARD, maxwell3d->regs.rasterize_enable == 0); } void RasterizerOpenGL::SyncPolygonModes() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PolygonModes]) { return; } flags[Dirty::PolygonModes] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.fill_rectangle) { if (!GLAD_GL_NV_fill_rectangle) { LOG_ERROR(Render_OpenGL, "GL_NV_fill_rectangle used and not supported"); @@ -820,7 +814,7 @@ void RasterizerOpenGL::SyncPolygonModes() { } void RasterizerOpenGL::SyncColorMask() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::ColorMasks]) { return; } @@ -829,7 +823,7 @@ void RasterizerOpenGL::SyncColorMask() { const bool force = flags[Dirty::ColorMaskCommon]; flags[Dirty::ColorMaskCommon] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.color_mask_common) { if (!force && !flags[Dirty::ColorMask0]) { return; @@ -854,30 +848,30 @@ void RasterizerOpenGL::SyncColorMask() { } void RasterizerOpenGL::SyncMultiSampleState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::MultisampleControl]) { return; } flags[Dirty::MultisampleControl] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_SAMPLE_ALPHA_TO_COVERAGE, regs.multisample_control.alpha_to_coverage); oglEnable(GL_SAMPLE_ALPHA_TO_ONE, regs.multisample_control.alpha_to_one); } void RasterizerOpenGL::SyncFragmentColorClampState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::FragmentClampColor]) { return; } flags[Dirty::FragmentClampColor] = false; - glClampColor(GL_CLAMP_FRAGMENT_COLOR, maxwell3d.regs.frag_color_clamp ? GL_TRUE : GL_FALSE); + glClampColor(GL_CLAMP_FRAGMENT_COLOR, maxwell3d->regs.frag_color_clamp ? GL_TRUE : GL_FALSE); } void RasterizerOpenGL::SyncBlendState() { - auto& flags = maxwell3d.dirty.flags; - const auto& regs = maxwell3d.regs; + auto& flags = maxwell3d->dirty.flags; + const auto& regs = maxwell3d->regs; if (flags[Dirty::BlendColor]) { flags[Dirty::BlendColor] = false; @@ -934,13 +928,13 @@ void RasterizerOpenGL::SyncBlendState() { } void RasterizerOpenGL::SyncLogicOpState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::LogicOp]) { return; } flags[Dirty::LogicOp] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.logic_op.enable) { glEnable(GL_COLOR_LOGIC_OP); glLogicOp(MaxwellToGL::LogicOp(regs.logic_op.operation)); @@ -950,7 +944,7 @@ void RasterizerOpenGL::SyncLogicOpState() { } void RasterizerOpenGL::SyncScissorTest() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::Scissors] && !flags[VideoCommon::Dirty::RescaleScissors]) { return; } @@ -959,7 +953,7 @@ void RasterizerOpenGL::SyncScissorTest() { const bool force = flags[VideoCommon::Dirty::RescaleScissors]; flags[VideoCommon::Dirty::RescaleScissors] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; const auto& resolution = Settings::values.resolution_info; const bool is_rescaling{texture_cache.IsRescaling()}; @@ -995,39 +989,39 @@ void RasterizerOpenGL::SyncScissorTest() { } void RasterizerOpenGL::SyncPointState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PointSize]) { return; } flags[Dirty::PointSize] = false; - oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable); - oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d.regs.vp_point_size.enable); + oglEnable(GL_POINT_SPRITE, maxwell3d->regs.point_sprite_enable); + oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d->regs.vp_point_size.enable); const bool is_rescaling{texture_cache.IsRescaling()}; const float scale = is_rescaling ? Settings::values.resolution_info.up_factor : 1.0f; - glPointSize(std::max(1.0f, maxwell3d.regs.point_size * scale)); + glPointSize(std::max(1.0f, maxwell3d->regs.point_size * scale)); } void RasterizerOpenGL::SyncLineState() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::LineWidth]) { return; } flags[Dirty::LineWidth] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_LINE_SMOOTH, regs.line_smooth_enable); glLineWidth(regs.line_smooth_enable ? regs.line_width_smooth : regs.line_width_aliased); } void RasterizerOpenGL::SyncPolygonOffset() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::PolygonOffset]) { return; } flags[Dirty::PolygonOffset] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; oglEnable(GL_POLYGON_OFFSET_FILL, regs.polygon_offset_fill_enable); oglEnable(GL_POLYGON_OFFSET_LINE, regs.polygon_offset_line_enable); oglEnable(GL_POLYGON_OFFSET_POINT, regs.polygon_offset_point_enable); @@ -1041,13 +1035,13 @@ void RasterizerOpenGL::SyncPolygonOffset() { } void RasterizerOpenGL::SyncAlphaTest() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::AlphaTest]) { return; } flags[Dirty::AlphaTest] = false; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.alpha_test_enabled) { glEnable(GL_ALPHA_TEST); glAlphaFunc(MaxwellToGL::ComparisonOp(regs.alpha_test_func), regs.alpha_test_ref); @@ -1057,17 +1051,17 @@ void RasterizerOpenGL::SyncAlphaTest() { } void RasterizerOpenGL::SyncFramebufferSRGB() { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::FramebufferSRGB]) { return; } flags[Dirty::FramebufferSRGB] = false; - oglEnable(GL_FRAMEBUFFER_SRGB, maxwell3d.regs.framebuffer_srgb); + oglEnable(GL_FRAMEBUFFER_SRGB, maxwell3d->regs.framebuffer_srgb); } void RasterizerOpenGL::BeginTransformFeedback(GraphicsPipeline* program, GLenum primitive_mode) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.tfb_enabled == 0) { return; } @@ -1086,11 +1080,48 @@ void RasterizerOpenGL::BeginTransformFeedback(GraphicsPipeline* program, GLenum } void RasterizerOpenGL::EndTransformFeedback() { - if (maxwell3d.regs.tfb_enabled != 0) { + if (maxwell3d->regs.tfb_enabled != 0) { glEndTransformFeedback(); } } +void RasterizerOpenGL::InitializeChannel(Tegra::Control::ChannelState& channel) { + CreateChannel(channel); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.CreateChannel(channel); + buffer_cache.CreateChannel(channel); + } + shader_cache.CreateChannel(channel); + query_cache.CreateChannel(channel); + state_tracker.SetupTables(channel); +} + +void RasterizerOpenGL::BindChannel(Tegra::Control::ChannelState& channel) { + const s32 channel_id = channel.bind_id; + BindToChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.BindToChannel(channel_id); + buffer_cache.BindToChannel(channel_id); + } + shader_cache.BindToChannel(channel_id); + query_cache.BindToChannel(channel_id); + state_tracker.ChangeChannel(channel); + state_tracker.InvalidateState(); +} + +void RasterizerOpenGL::ReleaseChannel(s32 channel_id) { + EraseChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.EraseChannel(channel_id); + buffer_cache.EraseChannel(channel_id); + } + shader_cache.EraseChannel(channel_id); + query_cache.EraseChannel(channel_id); +} + AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {} bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) { diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 31a16fcba..45131b785 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -12,6 +12,7 @@ #include <glad/glad.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/rasterizer_interface.h" @@ -58,7 +59,8 @@ private: BufferCache& buffer_cache; }; -class RasterizerOpenGL : public VideoCore::RasterizerAccelerated { +class RasterizerOpenGL : public VideoCore::RasterizerAccelerated, + protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { public: explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, Core::Memory::Memory& cpu_memory_, const Device& device_, @@ -78,10 +80,11 @@ public: bool MustFlushRegion(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size) override; void OnCPUWrite(VAddr addr, u64 size) override; - void SyncGuestHost() override; + void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; - void ModifyGPUMemory(GPUVAddr addr, u64 size) override; - void SignalSemaphore(GPUVAddr addr, u32 value) 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; void SignalSyncPoint(u32 value) override; void SignalReference() override; void ReleaseFences() override; @@ -96,7 +99,7 @@ public: const Tegra::Engines::Fermi2D::Config& copy_config) override; Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override; void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) override; + std::span<const u8> memory) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; void LoadDiskResources(u64 title_id, std::stop_token stop_loading, @@ -107,6 +110,12 @@ public: return num_queued_commands > 0; } + void InitializeChannel(Tegra::Control::ChannelState& channel) override; + + void BindChannel(Tegra::Control::ChannelState& channel) override; + + void ReleaseChannel(s32 channel_id) override; + private: static constexpr size_t MAX_TEXTURES = 192; static constexpr size_t MAX_IMAGES = 48; @@ -191,9 +200,6 @@ private: void EndTransformFeedback(); Tegra::GPU& gpu; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - Tegra::MemoryManager& gpu_memory; const Device& device; ScreenInfo& screen_info; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index ddb70934c..5a29a41d2 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -151,16 +151,13 @@ void SetXfbState(VideoCommon::TransformFeedbackState& state, const Maxwell& regs } // Anonymous namespace ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_, - TextureCache& texture_cache_, BufferCache& buffer_cache_, - ProgramManager& program_manager_, StateTracker& state_tracker_, - VideoCore::ShaderNotify& shader_notify_) - : VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_}, - 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()}, + 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_}, + 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()}, profile{ .supported_spirv = 0x00010000, @@ -310,7 +307,7 @@ GraphicsPipeline* ShaderCache::CurrentGraphicsPipeline() { current_pipeline = nullptr; return nullptr; } - const auto& regs{maxwell3d.regs}; + const auto& regs{maxwell3d->regs}; graphics_key.raw = 0; graphics_key.early_z.Assign(regs.force_early_fragment_tests != 0 ? 1 : 0); graphics_key.gs_input_topology.Assign(graphics_key.unique_hashes[4] != 0 @@ -351,13 +348,13 @@ GraphicsPipeline* ShaderCache::BuiltPipeline(GraphicsPipeline* pipeline) const n } // If something is using depth, we can assume that games are not rendering anything which // will be used one time. - if (maxwell3d.regs.zeta_enable) { + if (maxwell3d->regs.zeta_enable) { return nullptr; } // If games are using a small index count, we can assume these are full screen quads. // Usually these shaders are only used once for building textures so we can assume they // can't be built async - if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) { + if (maxwell3d->regs.index_array.count <= 6 || maxwell3d->regs.vertex_buffer.count <= 6) { return pipeline; } return nullptr; @@ -368,7 +365,7 @@ ComputePipeline* ShaderCache::CurrentComputePipeline() { if (!shader) { return nullptr; } - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const ComputePipelineKey key{ .unique_hash = shader->unique_hash, .shared_memory_size = qmd.shared_alloc, @@ -480,9 +477,9 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline( previous_program = &program; } auto* const thread_worker{build_in_parallel ? workers.get() : nullptr}; - return std::make_unique<GraphicsPipeline>( - device, texture_cache, buffer_cache, gpu_memory, maxwell3d, program_manager, state_tracker, - thread_worker, &shader_notify, sources, sources_spirv, infos, key); + return std::make_unique<GraphicsPipeline>(device, texture_cache, buffer_cache, program_manager, + state_tracker, thread_worker, &shader_notify, sources, + sources_spirv, infos, key); } catch (Shader::Exception& exception) { LOG_ERROR(Render_OpenGL, "{}", exception.what()); @@ -491,9 +488,9 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline( std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( const ComputePipelineKey& key, const VideoCommon::ShaderInfo* shader) { - const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; - const auto& qmd{kepler_compute.launch_description}; - ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + const GPUVAddr program_base{kepler_compute->regs.code_loc.Address()}; + const auto& qmd{kepler_compute->launch_description}; + ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start}; env.SetCachedSize(shader->size_bytes); main_pools.ReleaseContents(); @@ -536,9 +533,8 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( break; } - return std::make_unique<ComputePipeline>(device, texture_cache, buffer_cache, gpu_memory, - kepler_compute, program_manager, program.info, code, - code_spirv); + return std::make_unique<ComputePipeline>(device, texture_cache, buffer_cache, program_manager, + program.info, code, code_spirv); } catch (Shader::Exception& exception) { LOG_ERROR(Render_OpenGL, "{}", exception.what()); return nullptr; @@ -546,7 +542,7 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline( std::unique_ptr<ShaderWorker> ShaderCache::CreateWorkers() const { return std::make_unique<ShaderWorker>(std::max(std::thread::hardware_concurrency(), 2U) - 1, - "yuzu:ShaderBuilder", + "GlShaderBuilder", [this] { return Context{emu_window}; }); } diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index a14269dea..89f181fe3 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -30,12 +30,9 @@ using ShaderWorker = Common::StatefulThreadWorker<ShaderContext::Context>; class ShaderCache : public VideoCommon::ShaderCache { public: explicit ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_, - TextureCache& texture_cache_, BufferCache& buffer_cache_, - ProgramManager& program_manager_, StateTracker& state_tracker_, - VideoCore::ShaderNotify& shader_notify_); + 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, diff --git a/src/video_core/renderer_opengl/gl_state_tracker.cpp b/src/video_core/renderer_opengl/gl_state_tracker.cpp index 912725ef7..a8f3a0f57 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.cpp +++ b/src/video_core/renderer_opengl/gl_state_tracker.cpp @@ -7,8 +7,8 @@ #include "common/common_types.h" #include "core/core.h" +#include "video_core/control/channel_state.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" #include "video_core/renderer_opengl/gl_state_tracker.h" #define OFF(field_name) MAXWELL3D_REG_INDEX(field_name) @@ -202,9 +202,8 @@ void SetupDirtyMisc(Tables& tables) { } // Anonymous namespace -StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} { - auto& dirty = gpu.Maxwell3D().dirty; - auto& tables = dirty.tables; +void StateTracker::SetupTables(Tegra::Control::ChannelState& channel_state) { + auto& tables{channel_state.maxwell_3d->dirty.tables}; SetupDirtyFlags(tables); SetupDirtyColorMasks(tables); SetupDirtyViewports(tables); @@ -230,4 +229,14 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} SetupDirtyMisc(tables); } +void StateTracker::ChangeChannel(Tegra::Control::ChannelState& channel_state) { + flags = &channel_state.maxwell_3d->dirty.flags; +} + +void StateTracker::InvalidateState() { + flags->set(); +} + +StateTracker::StateTracker() : flags{&default_flags} {} + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_state_tracker.h b/src/video_core/renderer_opengl/gl_state_tracker.h index 04e024f08..19bcf3f35 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.h +++ b/src/video_core/renderer_opengl/gl_state_tracker.h @@ -12,8 +12,10 @@ #include "video_core/engines/maxwell_3d.h" namespace Tegra { -class GPU; +namespace Control { +struct ChannelState; } +} // namespace Tegra namespace OpenGL { @@ -83,7 +85,7 @@ static_assert(Last <= std::numeric_limits<u8>::max()); class StateTracker { public: - explicit StateTracker(Tegra::GPU& gpu); + explicit StateTracker(); void BindIndexBuffer(GLuint new_index_buffer) { if (index_buffer == new_index_buffer) { @@ -121,94 +123,107 @@ public: } void NotifyScreenDrawVertexArray() { - flags[OpenGL::Dirty::VertexFormats] = true; - flags[OpenGL::Dirty::VertexFormat0 + 0] = true; - flags[OpenGL::Dirty::VertexFormat0 + 1] = true; + (*flags)[OpenGL::Dirty::VertexFormats] = true; + (*flags)[OpenGL::Dirty::VertexFormat0 + 0] = true; + (*flags)[OpenGL::Dirty::VertexFormat0 + 1] = true; - flags[VideoCommon::Dirty::VertexBuffers] = true; - flags[VideoCommon::Dirty::VertexBuffer0] = true; + (*flags)[VideoCommon::Dirty::VertexBuffers] = true; + (*flags)[VideoCommon::Dirty::VertexBuffer0] = true; - flags[OpenGL::Dirty::VertexInstances] = true; - flags[OpenGL::Dirty::VertexInstance0 + 0] = true; - flags[OpenGL::Dirty::VertexInstance0 + 1] = true; + (*flags)[OpenGL::Dirty::VertexInstances] = true; + (*flags)[OpenGL::Dirty::VertexInstance0 + 0] = true; + (*flags)[OpenGL::Dirty::VertexInstance0 + 1] = true; } void NotifyPolygonModes() { - flags[OpenGL::Dirty::PolygonModes] = true; - flags[OpenGL::Dirty::PolygonModeFront] = true; - flags[OpenGL::Dirty::PolygonModeBack] = true; + (*flags)[OpenGL::Dirty::PolygonModes] = true; + (*flags)[OpenGL::Dirty::PolygonModeFront] = true; + (*flags)[OpenGL::Dirty::PolygonModeBack] = true; } void NotifyViewport0() { - flags[OpenGL::Dirty::Viewports] = true; - flags[OpenGL::Dirty::Viewport0] = true; + (*flags)[OpenGL::Dirty::Viewports] = true; + (*flags)[OpenGL::Dirty::Viewport0] = true; } void NotifyScissor0() { - flags[OpenGL::Dirty::Scissors] = true; - flags[OpenGL::Dirty::Scissor0] = true; + (*flags)[OpenGL::Dirty::Scissors] = true; + (*flags)[OpenGL::Dirty::Scissor0] = true; } void NotifyColorMask(size_t index) { - flags[OpenGL::Dirty::ColorMasks] = true; - flags[OpenGL::Dirty::ColorMask0 + index] = true; + (*flags)[OpenGL::Dirty::ColorMasks] = true; + (*flags)[OpenGL::Dirty::ColorMask0 + index] = true; } void NotifyBlend0() { - flags[OpenGL::Dirty::BlendStates] = true; - flags[OpenGL::Dirty::BlendState0] = true; + (*flags)[OpenGL::Dirty::BlendStates] = true; + (*flags)[OpenGL::Dirty::BlendState0] = true; } void NotifyFramebuffer() { - flags[VideoCommon::Dirty::RenderTargets] = true; + (*flags)[VideoCommon::Dirty::RenderTargets] = true; } void NotifyFrontFace() { - flags[OpenGL::Dirty::FrontFace] = true; + (*flags)[OpenGL::Dirty::FrontFace] = true; } void NotifyCullTest() { - flags[OpenGL::Dirty::CullTest] = true; + (*flags)[OpenGL::Dirty::CullTest] = true; } void NotifyDepthMask() { - flags[OpenGL::Dirty::DepthMask] = true; + (*flags)[OpenGL::Dirty::DepthMask] = true; } void NotifyDepthTest() { - flags[OpenGL::Dirty::DepthTest] = true; + (*flags)[OpenGL::Dirty::DepthTest] = true; } void NotifyStencilTest() { - flags[OpenGL::Dirty::StencilTest] = true; + (*flags)[OpenGL::Dirty::StencilTest] = true; } void NotifyPolygonOffset() { - flags[OpenGL::Dirty::PolygonOffset] = true; + (*flags)[OpenGL::Dirty::PolygonOffset] = true; } void NotifyRasterizeEnable() { - flags[OpenGL::Dirty::RasterizeEnable] = true; + (*flags)[OpenGL::Dirty::RasterizeEnable] = true; } void NotifyFramebufferSRGB() { - flags[OpenGL::Dirty::FramebufferSRGB] = true; + (*flags)[OpenGL::Dirty::FramebufferSRGB] = true; } void NotifyLogicOp() { - flags[OpenGL::Dirty::LogicOp] = true; + (*flags)[OpenGL::Dirty::LogicOp] = true; } void NotifyClipControl() { - flags[OpenGL::Dirty::ClipControl] = true; + (*flags)[OpenGL::Dirty::ClipControl] = true; } void NotifyAlphaTest() { - flags[OpenGL::Dirty::AlphaTest] = true; + (*flags)[OpenGL::Dirty::AlphaTest] = true; } + void NotifyRange(u8 start, u8 end) { + for (auto flag = start; flag <= end; flag++) { + (*flags)[flag] = true; + } + } + + void SetupTables(Tegra::Control::ChannelState& channel_state); + + void ChangeChannel(Tegra::Control::ChannelState& channel_state); + + void InvalidateState(); + private: - Tegra::Engines::Maxwell3D::DirtyState::Flags& flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags* flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags default_flags{}; GLuint framebuffer = 0; GLuint index_buffer = 0; diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index 9a72d0d6d..004421236 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -87,7 +87,7 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TAB {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM - {GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // R4G4_UNORM + {GL_R8, GL_RED, GL_UNSIGNED_BYTE}, // G4R4_UNORM {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB @@ -99,6 +99,8 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> FORMAT_TAB {GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB {GL_COMPRESSED_RGBA_ASTC_10x6_KHR}, // ASTC_2D_10X6_UNORM + {GL_COMPRESSED_RGBA_ASTC_10x5_KHR}, // ASTC_2D_10X5_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x5_KHR}, // ASTC_2D_10X5_SRGB {GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB {GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 34f3f7a67..8bd5eba7e 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -131,7 +131,7 @@ RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_, Core::Memory::Memory& cpu_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_}, state_tracker{gpu}, + emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, state_tracker{}, program_manager{device}, rasterizer(emu_window, gpu, cpu_memory, device, screen_info, program_manager, state_tracker) { if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) { diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index bdb71dc53..e7104d377 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -184,7 +184,7 @@ struct FormatTuple { {VK_FORMAT_BC3_SRGB_BLOCK}, // BC3_SRGB {VK_FORMAT_BC7_SRGB_BLOCK}, // BC7_SRGB {VK_FORMAT_R4G4B4A4_UNORM_PACK16, Attachable}, // A4B4G4R4_UNORM - {VK_FORMAT_R4G4_UNORM_PACK8}, // R4G4_UNORM + {VK_FORMAT_R4G4_UNORM_PACK8}, // G4R4_UNORM {VK_FORMAT_ASTC_4x4_SRGB_BLOCK}, // ASTC_2D_4X4_SRGB {VK_FORMAT_ASTC_8x8_SRGB_BLOCK}, // ASTC_2D_8X8_SRGB {VK_FORMAT_ASTC_8x5_SRGB_BLOCK}, // ASTC_2D_8X5_SRGB @@ -196,6 +196,8 @@ struct FormatTuple { {VK_FORMAT_ASTC_6x6_UNORM_BLOCK}, // ASTC_2D_6X6_UNORM {VK_FORMAT_ASTC_6x6_SRGB_BLOCK}, // ASTC_2D_6X6_SRGB {VK_FORMAT_ASTC_10x6_UNORM_BLOCK}, // ASTC_2D_10X6_UNORM + {VK_FORMAT_ASTC_10x5_UNORM_BLOCK}, // ASTC_2D_10X5_UNORM + {VK_FORMAT_ASTC_10x5_SRGB_BLOCK}, // ASTC_2D_10X5_SRGB {VK_FORMAT_ASTC_10x10_UNORM_BLOCK}, // ASTC_2D_10X10_UNORM {VK_FORMAT_ASTC_10x10_SRGB_BLOCK}, // ASTC_2D_10X10_SRGB {VK_FORMAT_ASTC_12x12_UNORM_BLOCK}, // ASTC_2D_12X12_UNORM diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 7c78d0299..d8131232a 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -102,13 +102,13 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, debug_callback(Settings::values.renderer_debug ? CreateDebugCallback(instance) : nullptr), surface(CreateSurface(instance, render_window)), device(CreateDevice(instance, dld, *surface)), memory_allocator(device, false), - state_tracker(gpu), scheduler(device, state_tracker), + state_tracker(), scheduler(device, state_tracker), swapchain(*surface, device, scheduler, render_window.GetFramebufferLayout().width, render_window.GetFramebufferLayout().height, false), blit_screen(cpu_memory, render_window, device, memory_allocator, swapchain, scheduler, screen_info), - rasterizer(render_window, gpu, gpu.MemoryManager(), cpu_memory, screen_info, device, - memory_allocator, state_tracker, scheduler) { + rasterizer(render_window, gpu, cpu_memory, screen_info, device, memory_allocator, + state_tracker, scheduler) { Report(); } catch (const vk::Exception& exception) { LOG_ERROR(Render_Vulkan, "Vulkan initialization failed with error: {}", exception.what()); @@ -142,7 +142,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { const auto recreate_swapchain = [&] { if (!has_been_recreated) { has_been_recreated = true; - scheduler.WaitWorker(); + scheduler.Finish(); } const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout(); swapchain.Create(layout.width, layout.height, is_srgb); diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index 444c29f68..cb7fa2078 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -145,6 +145,11 @@ VkSemaphore BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, // Finish any pending renderpass scheduler.RequestOutsideRenderPassOperationContext(); + if (const auto swapchain_images = swapchain.GetImageCount(); swapchain_images != image_count) { + image_count = swapchain_images; + Recreate(); + } + const std::size_t image_index = swapchain.GetImageIndex(); scheduler.Wait(resource_ticks[image_index]); @@ -448,15 +453,15 @@ vk::Framebuffer BlitScreen::CreateFramebuffer(const VkImageView& image_view, VkE void BlitScreen::CreateStaticResources() { CreateShaders(); + CreateSampler(); +} + +void BlitScreen::CreateDynamicResources() { CreateSemaphores(); CreateDescriptorPool(); CreateDescriptorSetLayout(); CreateDescriptorSets(); CreatePipelineLayout(); - CreateSampler(); -} - -void BlitScreen::CreateDynamicResources() { CreateRenderPass(); CreateFramebuffers(); CreateGraphicsPipeline(); diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h index b8c67bef0..29e2ea925 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.h +++ b/src/video_core/renderer_vulkan/vk_blit_screen.h @@ -109,7 +109,7 @@ private: MemoryAllocator& memory_allocator; Swapchain& swapchain; Scheduler& scheduler; - const std::size_t image_count; + std::size_t image_count; const ScreenInfo& screen_info; vk::ShaderModule vertex_shader; diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index f17a5ccd6..241d7573e 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -26,8 +26,6 @@ namespace Vulkan { -using Tegra::Texture::SWIZZLE_TABLE; - namespace { constexpr u32 ASTC_BINDING_INPUT_BUFFER = 0; diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index 6447210e2..7906e11a8 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -126,8 +126,8 @@ void ComputePipeline::Configure(Tegra::Engines::KeplerCompute& kepler_compute, const u32 secondary_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].Address() + secondary_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory.Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr) << desc.secondary_shift_left}; return TexturePair(lhs_raw | rhs_raw, via_header_index); } } diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.cpp b/src/video_core/renderer_vulkan/vk_fence_manager.cpp index c249b34d4..0214b103a 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_fence_manager.cpp @@ -11,11 +11,8 @@ namespace Vulkan { -InnerFence::InnerFence(Scheduler& scheduler_, u32 payload_, bool is_stubbed_) - : FenceBase{payload_, is_stubbed_}, scheduler{scheduler_} {} - -InnerFence::InnerFence(Scheduler& scheduler_, GPUVAddr address_, u32 payload_, bool is_stubbed_) - : FenceBase{address_, payload_, is_stubbed_}, scheduler{scheduler_} {} +InnerFence::InnerFence(Scheduler& scheduler_, bool is_stubbed_) + : FenceBase{is_stubbed_}, scheduler{scheduler_} {} InnerFence::~InnerFence() = default; @@ -48,12 +45,8 @@ FenceManager::FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::G : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_}, scheduler{scheduler_} {} -Fence FenceManager::CreateFence(u32 value, bool is_stubbed) { - return std::make_shared<InnerFence>(scheduler, value, is_stubbed); -} - -Fence FenceManager::CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) { - return std::make_shared<InnerFence>(scheduler, addr, value, is_stubbed); +Fence FenceManager::CreateFence(bool is_stubbed) { + return std::make_shared<InnerFence>(scheduler, is_stubbed); } void FenceManager::QueueFence(Fence& fence) { diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h index 7c0bbd80a..7fe2afcd9 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.h +++ b/src/video_core/renderer_vulkan/vk_fence_manager.h @@ -25,8 +25,7 @@ class Scheduler; class InnerFence : public VideoCommon::FenceBase { public: - explicit InnerFence(Scheduler& scheduler_, u32 payload_, bool is_stubbed_); - explicit InnerFence(Scheduler& scheduler_, GPUVAddr address_, u32 payload_, bool is_stubbed_); + explicit InnerFence(Scheduler& scheduler_, bool is_stubbed_); ~InnerFence(); void Queue(); @@ -50,8 +49,7 @@ public: QueryCache& query_cache, const Device& device, Scheduler& scheduler); protected: - Fence CreateFence(u32 value, bool is_stubbed) override; - Fence CreateFence(GPUVAddr addr, u32 value, bool is_stubbed) override; + Fence CreateFence(bool is_stubbed) override; void QueueFence(Fence& fence) override; bool IsFenceSignaled(Fence& fence) const override; void WaitFence(Fence& fence) override; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 5aca8f038..f47786f48 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -215,15 +215,14 @@ ConfigureFuncPtr ConfigureFunc(const std::array<vk::ShaderModule, NUM_STAGES>& m } // Anonymous namespace GraphicsPipeline::GraphicsPipeline( - Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_, Scheduler& scheduler_, - BufferCache& buffer_cache_, TextureCache& texture_cache_, + Scheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_, VideoCore::ShaderNotify* shader_notify, const Device& device_, DescriptorPool& descriptor_pool, UpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread, PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages, const std::array<const Shader::Info*, NUM_STAGES>& infos) - : key{key_}, maxwell3d{maxwell3d_}, gpu_memory{gpu_memory_}, device{device_}, - texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, scheduler{scheduler_}, + : key{key_}, device{device_}, texture_cache{texture_cache_}, + buffer_cache{buffer_cache_}, scheduler{scheduler_}, update_descriptor_queue{update_descriptor_queue_}, spv_modules{std::move(stages)} { if (shader_notify) { shader_notify->MarkShaderBuilding(); @@ -288,7 +287,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { buffer_cache.SetUniformBuffersState(enabled_uniform_buffer_masks, &uniform_buffer_sizes); - const auto& regs{maxwell3d.regs}; + const auto& regs{maxwell3d->regs}; const bool via_header_index{regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex}; const auto config_stage{[&](size_t stage) LAMBDA_FORCEINLINE { const Shader::Info& info{stage_infos[stage]}; @@ -302,7 +301,7 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { ++ssbo_index; } } - const auto& cbufs{maxwell3d.state.shader_stages[stage].const_buffers}; + const auto& cbufs{maxwell3d->state.shader_stages[stage].const_buffers}; const auto read_handle{[&](const auto& desc, u32 index) { ASSERT(cbufs[desc.cbuf_index].enabled); const u32 index_offset{index << desc.size_shift}; @@ -315,13 +314,14 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { const u32 second_offset{desc.secondary_cbuf_offset + index_offset}; const GPUVAddr separate_addr{cbufs[desc.secondary_cbuf_index].address + second_offset}; - const u32 lhs_raw{gpu_memory.Read<u32>(addr)}; - const u32 rhs_raw{gpu_memory.Read<u32>(separate_addr)}; + const u32 lhs_raw{gpu_memory->Read<u32>(addr) << desc.shift_left}; + const u32 rhs_raw{gpu_memory->Read<u32>(separate_addr) + << desc.secondary_shift_left}; const u32 raw{lhs_raw | rhs_raw}; return TexturePair(raw, via_header_index); } } - return TexturePair(gpu_memory.Read<u32>(addr), via_header_index); + return TexturePair(gpu_memory->Read<u32>(addr), via_header_index); }}; const auto add_image{[&](const auto& desc, bool blacklist) LAMBDA_FORCEINLINE { for (u32 index = 0; index < desc.count; ++index) { diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h index e8949a9ab..85602592b 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.h +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.h @@ -69,15 +69,16 @@ class GraphicsPipeline { static constexpr size_t NUM_STAGES = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage; public: - explicit GraphicsPipeline( - Tegra::Engines::Maxwell3D& maxwell3d, Tegra::MemoryManager& gpu_memory, - Scheduler& scheduler, BufferCache& buffer_cache, TextureCache& texture_cache, - VideoCore::ShaderNotify* shader_notify, const Device& device, - DescriptorPool& descriptor_pool, UpdateDescriptorQueue& update_descriptor_queue, - Common::ThreadWorker* worker_thread, PipelineStatistics* pipeline_statistics, - RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key, - std::array<vk::ShaderModule, NUM_STAGES> stages, - const std::array<const Shader::Info*, NUM_STAGES>& infos); + explicit GraphicsPipeline(Scheduler& scheduler, BufferCache& buffer_cache, + TextureCache& texture_cache, VideoCore::ShaderNotify* shader_notify, + const Device& device, DescriptorPool& descriptor_pool, + UpdateDescriptorQueue& update_descriptor_queue, + Common::ThreadWorker* worker_thread, + PipelineStatistics* pipeline_statistics, + RenderPassCache& render_pass_cache, + const GraphicsPipelineCacheKey& key, + std::array<vk::ShaderModule, NUM_STAGES> stages, + const std::array<const Shader::Info*, NUM_STAGES>& infos); GraphicsPipeline& operator=(GraphicsPipeline&&) noexcept = delete; GraphicsPipeline(GraphicsPipeline&&) noexcept = delete; @@ -109,6 +110,11 @@ public: return [](GraphicsPipeline* pl, bool is_indexed) { pl->ConfigureImpl<Spec>(is_indexed); }; } + void SetEngine(Tegra::Engines::Maxwell3D* maxwell3d_, Tegra::MemoryManager* gpu_memory_) { + maxwell3d = maxwell3d_; + gpu_memory = gpu_memory_; + } + private: template <typename Spec> void ConfigureImpl(bool is_indexed); @@ -120,8 +126,8 @@ private: void Validate(); const GraphicsPipelineCacheKey key; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::MemoryManager& gpu_memory; + Tegra::Engines::Maxwell3D* maxwell3d; + Tegra::MemoryManager* gpu_memory; const Device& device; TextureCache& texture_cache; BufferCache& buffer_cache; diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 9708dc45e..732e7b6f2 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -259,20 +259,18 @@ bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) c return std::memcmp(&rhs, this, Size()) == 0; } -PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_, const Device& device_, +PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_, UpdateDescriptorQueue& update_descriptor_queue_, RenderPassCache& render_pass_cache_, BufferCache& buffer_cache_, TextureCache& texture_cache_, VideoCore::ShaderNotify& shader_notify_) - : VideoCommon::ShaderCache{rasterizer_, gpu_memory_, maxwell3d_, kepler_compute_}, - device{device_}, scheduler{scheduler_}, descriptor_pool{descriptor_pool_}, - update_descriptor_queue{update_descriptor_queue_}, render_pass_cache{render_pass_cache_}, - buffer_cache{buffer_cache_}, texture_cache{texture_cache_}, shader_notify{shader_notify_}, + : VideoCommon::ShaderCache{rasterizer_}, device{device_}, scheduler{scheduler_}, + descriptor_pool{descriptor_pool_}, update_descriptor_queue{update_descriptor_queue_}, + render_pass_cache{render_pass_cache_}, buffer_cache{buffer_cache_}, + texture_cache{texture_cache_}, shader_notify{shader_notify_}, use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()}, - workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "yuzu:PipelineBuilder"), - serialization_thread(1, "yuzu:PipelineSerialization") { + workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "VkPipelineBuilder"), + serialization_thread(1, "VkPipelineSerialization") { const auto& float_control{device.FloatControlProperties()}; const VkDriverIdKHR driver_id{device.GetDriverID()}; profile = Shader::Profile{ @@ -337,7 +335,7 @@ GraphicsPipeline* PipelineCache::CurrentGraphicsPipeline() { current_pipeline = nullptr; return nullptr; } - graphics_key.state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported(), + graphics_key.state.Refresh(*maxwell3d, device.IsExtExtendedDynamicStateSupported(), device.IsExtVertexInputDynamicStateSupported()); if (current_pipeline) { @@ -357,7 +355,7 @@ ComputePipeline* PipelineCache::CurrentComputePipeline() { if (!shader) { return nullptr; } - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const ComputePipelineCacheKey key{ .unique_hash = shader->unique_hash, .shared_memory_size = qmd.shared_alloc, @@ -486,13 +484,13 @@ GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const } // If something is using depth, we can assume that games are not rendering anything which // will be used one time. - if (maxwell3d.regs.zeta_enable) { + if (maxwell3d->regs.zeta_enable) { return nullptr; } // If games are using a small index count, we can assume these are full screen quads. // Usually these shaders are only used once for building textures so we can assume they // can't be built async - if (maxwell3d.regs.index_array.count <= 6 || maxwell3d.regs.vertex_buffer.count <= 6) { + if (maxwell3d->regs.index_array.count <= 6 || maxwell3d->regs.vertex_buffer.count <= 6) { return pipeline; } return nullptr; @@ -557,10 +555,10 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline( previous_stage = &program; } Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr}; - return std::make_unique<GraphicsPipeline>( - maxwell3d, gpu_memory, scheduler, buffer_cache, texture_cache, &shader_notify, device, - descriptor_pool, update_descriptor_queue, thread_worker, statistics, render_pass_cache, key, - std::move(modules), infos); + return std::make_unique<GraphicsPipeline>(scheduler, buffer_cache, texture_cache, + &shader_notify, device, descriptor_pool, + update_descriptor_queue, thread_worker, statistics, + render_pass_cache, key, std::move(modules), infos); } catch (const Shader::Exception& exception) { LOG_ERROR(Render_Vulkan, "{}", exception.what()); @@ -592,9 +590,9 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() { std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline( const ComputePipelineCacheKey& key, const ShaderInfo* shader) { - const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; - const auto& qmd{kepler_compute.launch_description}; - ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + const GPUVAddr program_base{kepler_compute->regs.code_loc.Address()}; + const auto& qmd{kepler_compute->launch_description}; + ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start}; env.SetCachedSize(shader->size_bytes); main_pools.ReleaseContents(); diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.h b/src/video_core/renderer_vulkan/vk_pipeline_cache.h index 127957dbf..61f9e9366 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.h +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.h @@ -100,10 +100,8 @@ struct ShaderPools { class PipelineCache : public VideoCommon::ShaderCache { public: - explicit PipelineCache(RasterizerVulkan& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::Engines::KeplerCompute& kepler_compute, - Tegra::MemoryManager& gpu_memory, const Device& device, - Scheduler& scheduler, DescriptorPool& descriptor_pool, + explicit PipelineCache(RasterizerVulkan& rasterizer, const Device& device, Scheduler& scheduler, + DescriptorPool& descriptor_pool, UpdateDescriptorQueue& update_descriptor_queue, RenderPassCache& render_pass_cache, BufferCache& buffer_cache, TextureCache& texture_cache, VideoCore::ShaderNotify& shader_notify_); diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp index 2b859c6b8..7cb02631c 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp @@ -65,10 +65,9 @@ void QueryPool::Reserve(std::pair<VkQueryPool, u32> query) { usage[pool_index * GROW_STEP + static_cast<std::ptrdiff_t>(query.second)] = false; } -QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_, - const Device& device_, Scheduler& scheduler_) - : QueryCacheBase{rasterizer_, maxwell3d_, gpu_memory_}, device{device_}, scheduler{scheduler_}, +QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_, + Scheduler& scheduler_) + : QueryCacheBase{rasterizer_}, device{device_}, scheduler{scheduler_}, query_pools{ QueryPool{device_, scheduler_, QueryType::SamplesPassed}, } {} diff --git a/src/video_core/renderer_vulkan/vk_query_cache.h b/src/video_core/renderer_vulkan/vk_query_cache.h index b0d86c4f8..26762ee09 100644 --- a/src/video_core/renderer_vulkan/vk_query_cache.h +++ b/src/video_core/renderer_vulkan/vk_query_cache.h @@ -52,9 +52,8 @@ private: class QueryCache final : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> { public: - explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::MemoryManager& gpu_memory_, - const Device& device_, Scheduler& scheduler_); + explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_, + Scheduler& scheduler_); ~QueryCache(); std::pair<VkQueryPool, u32> AllocateQuery(VideoCore::QueryType type); diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 7e40c2df1..acfd5da7d 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -11,6 +11,7 @@ #include "common/microprofile.h" #include "common/scope_exit.h" #include "common/settings.h" +#include "video_core/control/channel_state.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_vulkan/blit_image.h" @@ -148,14 +149,11 @@ DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instan } // Anonymous namespace RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, ScreenInfo& screen_info_, const Device& device_, MemoryAllocator& memory_allocator_, StateTracker& state_tracker_, Scheduler& scheduler_) - : RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, - gpu_memory{gpu_memory_}, maxwell3d{gpu.Maxwell3D()}, kepler_compute{gpu.KeplerCompute()}, - screen_info{screen_info_}, device{device_}, memory_allocator{memory_allocator_}, - state_tracker{state_tracker_}, scheduler{scheduler_}, + : RasterizerAccelerated{cpu_memory_}, gpu{gpu_}, 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), update_descriptor_queue(device, scheduler), blit_image(device, scheduler, state_tracker, descriptor_pool), @@ -165,14 +163,13 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra memory_allocator, staging_pool, blit_image, astc_decoder_pass, render_pass_cache}, - texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), + texture_cache(texture_cache_runtime, *this), buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool, update_descriptor_queue, descriptor_pool), - buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime), - pipeline_cache(*this, maxwell3d, kepler_compute, gpu_memory, device, scheduler, - descriptor_pool, update_descriptor_queue, render_pass_cache, buffer_cache, - texture_cache, gpu.ShaderNotify()), - query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, accelerate_dma{buffer_cache}, + buffer_cache(*this, cpu_memory_, buffer_cache_runtime), + pipeline_cache(*this, device, scheduler, descriptor_pool, update_descriptor_queue, + render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()), + query_cache{*this, device, scheduler}, accelerate_dma{buffer_cache}, fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler), wfi_event(device.GetLogical().CreateEvent()) { scheduler.SetQueryCache(query_cache); @@ -193,14 +190,16 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { return; } std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + // update engine as channel may be different. + pipeline->SetEngine(maxwell3d, gpu_memory); pipeline->Configure(is_indexed); BeginTransformFeedback(); UpdateDynamicStates(); - const auto& regs{maxwell3d.regs}; - const u32 num_instances{maxwell3d.mme_draw.instance_count}; + const auto& regs{maxwell3d->regs}; + const u32 num_instances{maxwell3d->mme_draw.instance_count}; const DrawParams draw_params{MakeDrawParams(regs, num_instances, is_instanced, is_indexed)}; scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) { if (draw_params.is_indexed) { @@ -218,14 +217,14 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { void RasterizerVulkan::Clear() { MICROPROFILE_SCOPE(Vulkan_Clearing); - if (!maxwell3d.ShouldExecute()) { + if (!maxwell3d->ShouldExecute()) { return; } FlushWork(); query_cache.UpdateCounters(); - auto& regs = maxwell3d.regs; + auto& regs = maxwell3d->regs; const bool use_color = regs.clear_buffers.R || regs.clear_buffers.G || regs.clear_buffers.B || regs.clear_buffers.A; const bool use_depth = regs.clear_buffers.Z; @@ -248,8 +247,15 @@ void RasterizerVulkan::Clear() { } UpdateViewportsState(regs); + VkRect2D default_scissor; + default_scissor.offset.x = 0; + default_scissor.offset.y = 0; + default_scissor.extent.width = std::numeric_limits<s32>::max(); + default_scissor.extent.height = std::numeric_limits<s32>::max(); + VkClearRect clear_rect{ - .rect = GetScissorState(regs, 0, up_scale, down_shift), + .rect = regs.clear_flags.scissor ? GetScissorState(regs, 0, up_scale, down_shift) + : default_scissor, .baseArrayLayer = regs.clear_buffers.layer, .layerCount = 1, }; @@ -339,9 +345,9 @@ void RasterizerVulkan::DispatchCompute() { return; } std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex}; - pipeline->Configure(kepler_compute, gpu_memory, scheduler, buffer_cache, texture_cache); + pipeline->Configure(*kepler_compute, *gpu_memory, scheduler, buffer_cache, texture_cache); - const auto& qmd{kepler_compute.launch_description}; + const auto& qmd{kepler_compute->launch_description}; const std::array<u32, 3> dim{qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z}; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); }); @@ -422,7 +428,7 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { } } -void RasterizerVulkan::SyncGuestHost() { +void RasterizerVulkan::InvalidateGPUCache() { pipeline_cache.SyncGuestHost(); { std::scoped_lock lock{buffer_cache.mutex}; @@ -442,40 +448,30 @@ void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) { pipeline_cache.OnCPUWrite(addr, size); } -void RasterizerVulkan::ModifyGPUMemory(GPUVAddr addr, u64 size) { +void RasterizerVulkan::ModifyGPUMemory(size_t as_id, GPUVAddr addr, u64 size) { { std::scoped_lock lock{texture_cache.mutex}; - texture_cache.UnmapGPUMemory(addr, size); + texture_cache.UnmapGPUMemory(as_id, addr, size); } } -void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) { - if (!gpu.IsAsync()) { - gpu_memory.Write<u32>(addr, value); - return; - } - fence_manager.SignalSemaphore(addr, value); +void RasterizerVulkan::SignalFence(std::function<void()>&& func) { + fence_manager.SignalFence(std::move(func)); +} + +void RasterizerVulkan::SyncOperation(std::function<void()>&& func) { + fence_manager.SyncOperation(std::move(func)); } void RasterizerVulkan::SignalSyncPoint(u32 value) { - if (!gpu.IsAsync()) { - gpu.IncrementSyncPoint(value); - return; - } fence_manager.SignalSyncPoint(value); } void RasterizerVulkan::SignalReference() { - if (!gpu.IsAsync()) { - return; - } fence_manager.SignalOrdering(); } void RasterizerVulkan::ReleaseFences() { - if (!gpu.IsAsync()) { - return; - } fence_manager.WaitPendingFences(); } @@ -552,13 +548,13 @@ Tegra::Engines::AccelerateDMAInterface& RasterizerVulkan::AccessAccelerateDMA() } void RasterizerVulkan::AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) { - auto cpu_addr = gpu_memory.GpuToCpuAddress(address); + std::span<const u8> memory) { + auto cpu_addr = gpu_memory->GpuToCpuAddress(address); if (!cpu_addr) [[unlikely]] { - gpu_memory.WriteBlock(address, memory.data(), copy_size); + gpu_memory->WriteBlock(address, memory.data(), copy_size); return; } - gpu_memory.WriteBlockUnsafe(address, memory.data(), copy_size); + gpu_memory->WriteBlockUnsafe(address, memory.data(), copy_size); { std::unique_lock<std::mutex> lock{buffer_cache.mutex}; if (!buffer_cache.InlineMemory(*cpu_addr, copy_size, memory)) { @@ -627,7 +623,7 @@ bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 } void RasterizerVulkan::UpdateDynamicStates() { - auto& regs = maxwell3d.regs; + auto& regs = maxwell3d->regs; UpdateViewportsState(regs); UpdateScissorsState(regs); UpdateDepthBias(regs); @@ -651,7 +647,7 @@ void RasterizerVulkan::UpdateDynamicStates() { } void RasterizerVulkan::BeginTransformFeedback() { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.tfb_enabled == 0) { return; } @@ -667,7 +663,7 @@ void RasterizerVulkan::BeginTransformFeedback() { } void RasterizerVulkan::EndTransformFeedback() { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (regs.tfb_enabled == 0) { return; } @@ -917,7 +913,7 @@ void RasterizerVulkan::UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& } void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) { - auto& dirty{maxwell3d.dirty.flags}; + auto& dirty{maxwell3d->dirty.flags}; if (!dirty[Dirty::VertexInput]) { return; } @@ -974,4 +970,41 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs) }); } +void RasterizerVulkan::InitializeChannel(Tegra::Control::ChannelState& channel) { + CreateChannel(channel); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.CreateChannel(channel); + buffer_cache.CreateChannel(channel); + } + pipeline_cache.CreateChannel(channel); + query_cache.CreateChannel(channel); + state_tracker.SetupTables(channel); +} + +void RasterizerVulkan::BindChannel(Tegra::Control::ChannelState& channel) { + const s32 channel_id = channel.bind_id; + BindToChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.BindToChannel(channel_id); + buffer_cache.BindToChannel(channel_id); + } + pipeline_cache.BindToChannel(channel_id); + query_cache.BindToChannel(channel_id); + state_tracker.ChangeChannel(channel); + state_tracker.InvalidateState(); +} + +void RasterizerVulkan::ReleaseChannel(s32 channel_id) { + EraseChannel(channel_id); + { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + texture_cache.EraseChannel(channel_id); + buffer_cache.EraseChannel(channel_id); + } + pipeline_cache.EraseChannel(channel_id); + query_cache.EraseChannel(channel_id); +} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 0370ea39b..4cde3c983 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -8,6 +8,7 @@ #include <boost/container/static_vector.hpp> #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" @@ -54,13 +55,13 @@ private: BufferCache& buffer_cache; }; -class RasterizerVulkan final : public VideoCore::RasterizerAccelerated { +class RasterizerVulkan final : public VideoCore::RasterizerAccelerated, + protected VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { public: explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - ScreenInfo& screen_info_, const Device& device_, - MemoryAllocator& memory_allocator_, StateTracker& state_tracker_, - Scheduler& scheduler_); + Core::Memory::Memory& cpu_memory_, ScreenInfo& screen_info_, + const Device& device_, MemoryAllocator& memory_allocator_, + StateTracker& state_tracker_, Scheduler& scheduler_); ~RasterizerVulkan() override; void Draw(bool is_indexed, bool is_instanced) override; @@ -75,10 +76,11 @@ public: bool MustFlushRegion(VAddr addr, u64 size) override; void InvalidateRegion(VAddr addr, u64 size) override; void OnCPUWrite(VAddr addr, u64 size) override; - void SyncGuestHost() override; + void InvalidateGPUCache() override; void UnmapMemory(VAddr addr, u64 size) override; - void ModifyGPUMemory(GPUVAddr addr, u64 size) override; - void SignalSemaphore(GPUVAddr addr, u32 value) 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; void SignalSyncPoint(u32 value) override; void SignalReference() override; void ReleaseFences() override; @@ -93,12 +95,18 @@ public: const Tegra::Engines::Fermi2D::Config& copy_config) override; Tegra::Engines::AccelerateDMAInterface& AccessAccelerateDMA() override; void AccelerateInlineToMemory(GPUVAddr address, size_t copy_size, - std::span<u8> memory) override; + std::span<const u8> memory) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; void LoadDiskResources(u64 title_id, std::stop_token stop_loading, const VideoCore::DiskResourceLoadCallback& callback) override; + void InitializeChannel(Tegra::Control::ChannelState& channel) override; + + void BindChannel(Tegra::Control::ChannelState& channel) override; + + void ReleaseChannel(s32 channel_id) override; + private: static constexpr size_t MAX_TEXTURES = 192; static constexpr size_t MAX_IMAGES = 48; @@ -134,9 +142,6 @@ private: void UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs); Tegra::GPU& gpu; - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; ScreenInfo& screen_info; const Device& device; diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 7a4868e58..7934f2a51 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -136,7 +136,7 @@ bool Scheduler::UpdateRescaling(bool is_rescaling) { } void Scheduler::WorkerThread(std::stop_token stop_token) { - Common::SetCurrentThreadName("yuzu:VulkanWorker"); + Common::SetCurrentThreadName("VulkanWorker"); do { std::unique_ptr<CommandChunk> work; bool has_submit{false}; diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp index 06f68d09a..7fb256953 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -26,20 +26,39 @@ using namespace Common::Literals; constexpr VkDeviceSize MAX_ALIGNMENT = 256; // Maximum size to put elements in the stream buffer constexpr VkDeviceSize MAX_STREAM_BUFFER_REQUEST_SIZE = 8_MiB; -// Stream buffer size in bytes -constexpr VkDeviceSize STREAM_BUFFER_SIZE = 128_MiB; -constexpr VkDeviceSize REGION_SIZE = STREAM_BUFFER_SIZE / StagingBufferPool::NUM_SYNCS; constexpr VkMemoryPropertyFlags HOST_FLAGS = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT; constexpr VkMemoryPropertyFlags STREAM_FLAGS = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | HOST_FLAGS; -bool IsStreamHeap(VkMemoryHeap heap) noexcept { - return STREAM_BUFFER_SIZE < (heap.size * 2) / 3; +static bool IsStreamHeap(VkMemoryHeap heap, size_t staging_buffer_size) noexcept { + return staging_buffer_size < (heap.size * 2) / 3; +} + +static bool HasLargeDeviceLocalHostVisibleMemory(const VkPhysicalDeviceMemoryProperties& props) { + const auto flags{VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT}; + + for (u32 type_index = 0; type_index < props.memoryTypeCount; ++type_index) { + const auto& memory_type{props.memoryTypes[type_index]}; + + if ((memory_type.propertyFlags & flags) != flags) { + // Memory must be device local and host visible + continue; + } + + const auto& heap{props.memoryHeaps[memory_type.heapIndex]}; + if (heap.size >= 7168_MiB) { + // This is the right type of memory + return true; + } + } + + return false; } std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask, - VkMemoryPropertyFlags flags) noexcept { + VkMemoryPropertyFlags flags, + size_t staging_buffer_size) noexcept { for (u32 type_index = 0; type_index < props.memoryTypeCount; ++type_index) { if (((type_mask >> type_index) & 1) == 0) { // Memory type is incompatible @@ -50,7 +69,7 @@ std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& p // Memory type doesn't have the flags we want continue; } - if (!IsStreamHeap(props.memoryHeaps[memory_type.heapIndex])) { + if (!IsStreamHeap(props.memoryHeaps[memory_type.heapIndex], staging_buffer_size)) { // Memory heap is not suitable for streaming continue; } @@ -61,17 +80,17 @@ std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& p } u32 FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask, - bool try_device_local) { + bool try_device_local, size_t staging_buffer_size) { std::optional<u32> type; if (try_device_local) { // Try to find a DEVICE_LOCAL_BIT type, Nvidia and AMD have a dedicated heap for this - type = FindMemoryTypeIndex(props, type_mask, STREAM_FLAGS); + type = FindMemoryTypeIndex(props, type_mask, STREAM_FLAGS, staging_buffer_size); if (type) { return *type; } } // Otherwise try without the DEVICE_LOCAL_BIT - type = FindMemoryTypeIndex(props, type_mask, HOST_FLAGS); + type = FindMemoryTypeIndex(props, type_mask, HOST_FLAGS, staging_buffer_size); if (type) { return *type; } @@ -79,20 +98,32 @@ u32 FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_ throw vk::Exception(VK_ERROR_OUT_OF_DEVICE_MEMORY); } -size_t Region(size_t iterator) noexcept { - return iterator / REGION_SIZE; +size_t Region(size_t iterator, size_t region_size) noexcept { + return iterator / region_size; } } // Anonymous namespace StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& memory_allocator_, Scheduler& scheduler_) : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_} { + + const auto memory_properties{device.GetPhysical().GetMemoryProperties().memoryProperties}; + if (HasLargeDeviceLocalHostVisibleMemory(memory_properties)) { + // Possible on many integrated and newer discrete cards + staging_buffer_size = 1_GiB; + } else { + // Well-supported default size used by most Vulkan PC games + staging_buffer_size = 256_MiB; + } + + region_size = staging_buffer_size / StagingBufferPool::NUM_SYNCS; + const vk::Device& dev = device.GetLogical(); stream_buffer = dev.CreateBuffer(VkBufferCreateInfo{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, - .size = STREAM_BUFFER_SIZE, + .size = staging_buffer_size, .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, @@ -117,19 +148,18 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem .image = nullptr, .buffer = *stream_buffer, }; - const auto memory_properties = device.GetPhysical().GetMemoryProperties().memoryProperties; VkMemoryAllocateInfo stream_memory_info{ .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, .pNext = make_dedicated ? &dedicated_info : nullptr, .allocationSize = requirements.size, - .memoryTypeIndex = - FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, true), + .memoryTypeIndex = FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, true, + staging_buffer_size), }; stream_memory = dev.TryAllocateMemory(stream_memory_info); if (!stream_memory) { LOG_INFO(Render_Vulkan, "Dynamic memory allocation failed, trying with system memory"); - stream_memory_info.memoryTypeIndex = - FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, false); + stream_memory_info.memoryTypeIndex = FindMemoryTypeIndex( + memory_properties, requirements.memoryTypeBits, false, staging_buffer_size); stream_memory = dev.AllocateMemory(stream_memory_info); } @@ -137,7 +167,7 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem stream_memory.SetObjectNameEXT("Stream Buffer Memory"); } stream_buffer.BindMemory(*stream_memory, 0); - stream_pointer = stream_memory.Map(0, STREAM_BUFFER_SIZE); + stream_pointer = stream_memory.Map(0, staging_buffer_size); } StagingBufferPool::~StagingBufferPool() = default; @@ -158,25 +188,25 @@ void StagingBufferPool::TickFrame() { } StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) { - if (AreRegionsActive(Region(free_iterator) + 1, - std::min(Region(iterator + size) + 1, NUM_SYNCS))) { + if (AreRegionsActive(Region(free_iterator, region_size) + 1, + std::min(Region(iterator + size, region_size) + 1, NUM_SYNCS))) { // Avoid waiting for the previous usages to be free return GetStagingBuffer(size, MemoryUsage::Upload); } const u64 current_tick = scheduler.CurrentTick(); - std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + Region(iterator), - current_tick); + std::fill(sync_ticks.begin() + Region(used_iterator, region_size), + sync_ticks.begin() + Region(iterator, region_size), current_tick); used_iterator = iterator; free_iterator = std::max(free_iterator, iterator + size); - if (iterator + size >= STREAM_BUFFER_SIZE) { - std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + NUM_SYNCS, - current_tick); + if (iterator + size >= staging_buffer_size) { + std::fill(sync_ticks.begin() + Region(used_iterator, region_size), + sync_ticks.begin() + NUM_SYNCS, current_tick); used_iterator = 0; iterator = 0; free_iterator = size; - if (AreRegionsActive(0, Region(size) + 1)) { + if (AreRegionsActive(0, Region(size, region_size) + 1)) { // Avoid waiting for the previous usages to be free return GetStagingBuffer(size, MemoryUsage::Upload); } diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h index 91dc84da8..90c67177f 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h @@ -93,6 +93,9 @@ private: size_t free_iterator = 0; std::array<u64, NUM_SYNCS> sync_ticks{}; + size_t staging_buffer_size = 0; + size_t region_size = 0; + StagingBuffersCache device_local_cache; StagingBuffersCache upload_cache; StagingBuffersCache download_cache; diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp index 9ad096431..f234e1a31 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp +++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp @@ -7,9 +7,9 @@ #include "common/common_types.h" #include "core/core.h" +#include "video_core/control/channel_state.h" #include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" -#include "video_core/gpu.h" #include "video_core/renderer_vulkan/vk_state_tracker.h" #define OFF(field_name) MAXWELL3D_REG_INDEX(field_name) @@ -174,9 +174,8 @@ void SetupDirtyVertexBindings(Tables& tables) { } } // Anonymous namespace -StateTracker::StateTracker(Tegra::GPU& gpu) - : flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} { - auto& tables{gpu.Maxwell3D().dirty.tables}; +void StateTracker::SetupTables(Tegra::Control::ChannelState& channel_state) { + auto& tables{channel_state.maxwell_3d->dirty.tables}; SetupDirtyFlags(tables); SetupDirtyViewports(tables); SetupDirtyScissors(tables); @@ -199,4 +198,15 @@ StateTracker::StateTracker(Tegra::GPU& gpu) SetupDirtyVertexBindings(tables); } +void StateTracker::ChangeChannel(Tegra::Control::ChannelState& channel_state) { + flags = &channel_state.maxwell_3d->dirty.flags; +} + +void StateTracker::InvalidateState() { + flags->set(); +} + +StateTracker::StateTracker() + : flags{&default_flags}, default_flags{}, invalidation_flags{MakeInvalidationFlags()} {} + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.h b/src/video_core/renderer_vulkan/vk_state_tracker.h index a85bc1c10..2296dea60 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -10,6 +10,12 @@ #include "video_core/dirty_flags.h" #include "video_core/engines/maxwell_3d.h" +namespace Tegra { +namespace Control { +struct ChannelState; +} +} // namespace Tegra + namespace Vulkan { namespace Dirty { @@ -53,19 +59,19 @@ class StateTracker { using Maxwell = Tegra::Engines::Maxwell3D::Regs; public: - explicit StateTracker(Tegra::GPU& gpu); + explicit StateTracker(); void InvalidateCommandBufferState() { - flags |= invalidation_flags; + (*flags) |= invalidation_flags; current_topology = INVALID_TOPOLOGY; } void InvalidateViewports() { - flags[Dirty::Viewports] = true; + (*flags)[Dirty::Viewports] = true; } void InvalidateScissors() { - flags[Dirty::Scissors] = true; + (*flags)[Dirty::Scissors] = true; } bool TouchViewports() { @@ -139,16 +145,23 @@ public: return has_changed; } + void SetupTables(Tegra::Control::ChannelState& channel_state); + + void ChangeChannel(Tegra::Control::ChannelState& channel_state); + + void InvalidateState(); + private: static constexpr auto INVALID_TOPOLOGY = static_cast<Maxwell::PrimitiveTopology>(~0u); bool Exchange(std::size_t id, bool new_value) const noexcept { - const bool is_dirty = flags[id]; - flags[id] = new_value; + const bool is_dirty = (*flags)[id]; + (*flags)[id] = new_value; return is_dirty; } - Tegra::Engines::Maxwell3D::DirtyState::Flags& flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags* flags; + Tegra::Engines::Maxwell3D::DirtyState::Flags default_flags; Tegra::Engines::Maxwell3D::DirtyState::Flags invalidation_flags; Maxwell::PrimitiveTopology current_topology = INVALID_TOPOLOGY; }; diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index a69ae7725..706d9ba74 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -36,7 +36,8 @@ VkPresentModeKHR ChooseSwapPresentMode(vk::Span<VkPresentModeKHR> modes) { // Mailbox (triple buffering) doesn't lock the application like fifo (vsync), // prefer it if vsync option is not selected const auto found_mailbox = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_MAILBOX_KHR); - if (found_mailbox != modes.end() && !Settings::values.use_vsync.GetValue()) { + if (Settings::values.fullscreen_mode.GetValue() == Settings::FullscreenMode::Borderless && + found_mailbox != modes.end() && !Settings::values.use_vsync.GetValue()) { return VK_PRESENT_MODE_MAILBOX_KHR; } if (!Settings::values.use_speed_limit.GetValue()) { @@ -156,8 +157,16 @@ void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u3 present_mode = ChooseSwapPresentMode(present_modes); u32 requested_image_count{capabilities.minImageCount + 1}; - if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) { - requested_image_count = capabilities.maxImageCount; + // Ensure Tripple buffering if possible. + if (capabilities.maxImageCount > 0) { + if (requested_image_count > capabilities.maxImageCount) { + requested_image_count = capabilities.maxImageCount; + } else { + requested_image_count = + std::max(requested_image_count, std::min(3U, capabilities.maxImageCount)); + } + } else { + requested_image_count = std::max(requested_image_count, 3U); } VkSwapchainCreateInfoKHR swapchain_ci{ .sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index caca79d79..305ad8aee 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -592,7 +592,7 @@ void TryTransformSwizzleIfNeeded(PixelFormat format, std::array<SwizzleSource, 4 case PixelFormat::A5B5G5R1_UNORM: std::ranges::transform(swizzle, swizzle.begin(), SwapSpecial); break; - case PixelFormat::R4G4_UNORM: + case PixelFormat::G4R4_UNORM: std::ranges::transform(swizzle, swizzle.begin(), SwapGreenRed); break; default: @@ -1474,13 +1474,14 @@ bool Image::BlitScaleHelper(bool scale_up) { }; const VkExtent2D extent{ .width = std::max(scaled_width, info.size.width), - .height = std::max(scaled_height, info.size.width), + .height = std::max(scaled_height, info.size.height), }; auto* view_ptr = blit_view.get(); if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) { if (!blit_framebuffer) { - blit_framebuffer = std::make_unique<Framebuffer>(*runtime, view_ptr, nullptr, extent); + blit_framebuffer = + std::make_unique<Framebuffer>(*runtime, view_ptr, nullptr, extent, scale_up); } const auto color_view = blit_view->Handle(Shader::TextureType::Color2D); @@ -1488,7 +1489,8 @@ bool Image::BlitScaleHelper(bool scale_up) { src_region, operation, BLIT_OPERATION); } else if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) { if (!blit_framebuffer) { - blit_framebuffer = std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent); + blit_framebuffer = + std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent, scale_up); } runtime->blit_image_helper.BlitDepthStencil(blit_framebuffer.get(), blit_view->DepthView(), blit_view->StencilView(), dst_region, @@ -1756,34 +1758,42 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM .width = key.size.width, .height = key.size.height, }} { - CreateFramebuffer(runtime, color_buffers, depth_buffer); + CreateFramebuffer(runtime, color_buffers, depth_buffer, key.is_rescaled); if (runtime.device.HasDebuggingToolAttached()) { framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str()); } } Framebuffer::Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, - ImageView* depth_buffer, VkExtent2D extent) + ImageView* depth_buffer, VkExtent2D extent, bool is_rescaled) : render_area{extent} { std::array<ImageView*, NUM_RT> color_buffers{color_buffer}; - CreateFramebuffer(runtime, color_buffers, depth_buffer); + CreateFramebuffer(runtime, color_buffers, depth_buffer, is_rescaled); } Framebuffer::~Framebuffer() = default; void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, - ImageView* depth_buffer) { + ImageView* depth_buffer, bool is_rescaled) { std::vector<VkImageView> attachments; RenderPassKey renderpass_key{}; s32 num_layers = 1; + const auto& resolution = runtime.resolution; + + u32 width = 0; + u32 height = 0; for (size_t index = 0; index < NUM_RT; ++index) { const ImageView* const color_buffer = color_buffers[index]; if (!color_buffer) { renderpass_key.color_formats[index] = PixelFormat::Invalid; continue; } + width = std::max(width, is_rescaled ? resolution.ScaleUp(color_buffer->size.width) + : color_buffer->size.width); + height = std::max(height, is_rescaled ? resolution.ScaleUp(color_buffer->size.height) + : color_buffer->size.height); attachments.push_back(color_buffer->RenderTarget()); renderpass_key.color_formats[index] = color_buffer->format; num_layers = std::max(num_layers, color_buffer->range.extent.layers); @@ -1794,6 +1804,10 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, } const size_t num_colors = attachments.size(); if (depth_buffer) { + width = std::max(width, is_rescaled ? resolution.ScaleUp(depth_buffer->size.width) + : depth_buffer->size.width); + height = std::max(height, is_rescaled ? resolution.ScaleUp(depth_buffer->size.height) + : depth_buffer->size.height); attachments.push_back(depth_buffer->RenderTarget()); renderpass_key.depth_format = depth_buffer->format; num_layers = std::max(num_layers, depth_buffer->range.extent.layers); @@ -1810,6 +1824,8 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime, renderpass_key.samples = samples; renderpass = runtime.render_pass_cache.Get(renderpass_key); + render_area.width = std::min(render_area.width, width); + render_area.height = std::min(render_area.height, height); num_color_buffers = static_cast<u32>(num_colors); framebuffer = runtime.device.GetLogical().CreateFramebuffer({ diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 69f06ee7b..0b7ac0df1 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -268,7 +268,7 @@ public: ImageView* depth_buffer, const VideoCommon::RenderTargets& key); explicit Framebuffer(TextureCacheRuntime& runtime, ImageView* color_buffer, - ImageView* depth_buffer, VkExtent2D extent); + ImageView* depth_buffer, VkExtent2D extent, bool is_rescaled); ~Framebuffer(); @@ -279,7 +279,8 @@ public: Framebuffer& operator=(Framebuffer&&) = default; void CreateFramebuffer(TextureCacheRuntime& runtime, - std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer); + std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer, + bool is_rescaled = false); [[nodiscard]] VkFramebuffer Handle() const noexcept { return *framebuffer; diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp index 164e4ee0e..f53066579 100644 --- a/src/video_core/shader_cache.cpp +++ b/src/video_core/shader_cache.cpp @@ -8,6 +8,7 @@ #include "common/assert.h" #include "shader_recompiler/frontend/maxwell/control_flow.h" #include "shader_recompiler/object_pool.h" +#include "video_core/control/channel_state.h" #include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/maxwell_3d.h" @@ -33,29 +34,25 @@ void ShaderCache::SyncGuestHost() { RemovePendingShaders(); } -ShaderCache::ShaderCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_) - : gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, - rasterizer{rasterizer_} {} +ShaderCache::ShaderCache(VideoCore::RasterizerInterface& rasterizer_) : rasterizer{rasterizer_} {} bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { - auto& dirty{maxwell3d.dirty.flags}; + auto& dirty{maxwell3d->dirty.flags}; if (!dirty[VideoCommon::Dirty::Shaders]) { return last_shaders_valid; } dirty[VideoCommon::Dirty::Shaders] = false; - const GPUVAddr base_addr{maxwell3d.regs.code_address.CodeAddress()}; + const GPUVAddr base_addr{maxwell3d->regs.code_address.CodeAddress()}; for (size_t index = 0; index < Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram; ++index) { - if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { + if (!maxwell3d->regs.IsShaderConfigEnabled(index)) { unique_hashes[index] = 0; continue; } - const auto& shader_config{maxwell3d.regs.shader_config[index]}; + const auto& shader_config{maxwell3d->regs.shader_config[index]}; const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderProgram>(index)}; const GPUVAddr shader_addr{base_addr + shader_config.offset}; - const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; + const std::optional<VAddr> cpu_shader_addr{gpu_memory->GpuToCpuAddress(shader_addr)}; if (!cpu_shader_addr) { LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); last_shaders_valid = false; @@ -64,7 +61,7 @@ bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { const ShaderInfo* shader_info{TryGet(*cpu_shader_addr)}; if (!shader_info) { const u32 start_address{shader_config.offset}; - GraphicsEnvironment env{maxwell3d, gpu_memory, program, base_addr, start_address}; + GraphicsEnvironment env{*maxwell3d, *gpu_memory, program, base_addr, start_address}; shader_info = MakeShaderInfo(env, *cpu_shader_addr); } shader_infos[index] = shader_info; @@ -75,10 +72,10 @@ bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { } const ShaderInfo* ShaderCache::ComputeShader() { - const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; - const auto& qmd{kepler_compute.launch_description}; + const GPUVAddr program_base{kepler_compute->regs.code_loc.Address()}; + const auto& qmd{kepler_compute->launch_description}; const GPUVAddr shader_addr{program_base + qmd.program_start}; - const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; + const std::optional<VAddr> cpu_shader_addr{gpu_memory->GpuToCpuAddress(shader_addr)}; if (!cpu_shader_addr) { LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); return nullptr; @@ -86,22 +83,22 @@ const ShaderInfo* ShaderCache::ComputeShader() { if (const ShaderInfo* const shader = TryGet(*cpu_shader_addr)) { return shader; } - ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; + ComputeEnvironment env{*kepler_compute, *gpu_memory, program_base, qmd.program_start}; return MakeShaderInfo(env, *cpu_shader_addr); } void ShaderCache::GetGraphicsEnvironments(GraphicsEnvironments& result, const std::array<u64, NUM_PROGRAMS>& unique_hashes) { size_t env_index{}; - const GPUVAddr base_addr{maxwell3d.regs.code_address.CodeAddress()}; + const GPUVAddr base_addr{maxwell3d->regs.code_address.CodeAddress()}; for (size_t index = 0; index < NUM_PROGRAMS; ++index) { if (unique_hashes[index] == 0) { continue; } const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderProgram>(index)}; auto& env{result.envs[index]}; - const u32 start_address{maxwell3d.regs.shader_config[index].offset}; - env = GraphicsEnvironment{maxwell3d, gpu_memory, program, base_addr, start_address}; + const u32 start_address{maxwell3d->regs.shader_config[index].offset}; + env = GraphicsEnvironment{*maxwell3d, *gpu_memory, program, base_addr, start_address}; env.SetCachedSize(shader_infos[index]->size_bytes); result.env_ptrs[env_index++] = &env; } diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h index f67cea8c4..a4391202d 100644 --- a/src/video_core/shader_cache.h +++ b/src/video_core/shader_cache.h @@ -12,6 +12,7 @@ #include <vector> #include "common/common_types.h" +#include "video_core/control/channel_state_cache.h" #include "video_core/rasterizer_interface.h" #include "video_core/shader_environment.h" @@ -19,6 +20,10 @@ namespace Tegra { class MemoryManager; } +namespace Tegra::Control { +struct ChannelState; +} + namespace VideoCommon { class GenericEnvironment; @@ -28,7 +33,7 @@ struct ShaderInfo { size_t size_bytes{}; }; -class ShaderCache { +class ShaderCache : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> { static constexpr u64 YUZU_PAGEBITS = 14; static constexpr u64 YUZU_PAGESIZE = u64(1) << YUZU_PAGEBITS; @@ -71,9 +76,7 @@ protected: } }; - explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_); + explicit ShaderCache(VideoCore::RasterizerInterface& rasterizer_); /// @brief Update the hashes and information of shader stages /// @param unique_hashes Shader hashes to store into when a stage is enabled @@ -88,10 +91,6 @@ protected: void GetGraphicsEnvironments(GraphicsEnvironments& result, const std::array<u64, NUM_PROGRAMS>& unique_hashes); - Tegra::MemoryManager& gpu_memory; - Tegra::Engines::Maxwell3D& maxwell3d; - Tegra::Engines::KeplerCompute& kepler_compute; - std::array<const ShaderInfo*, NUM_PROGRAMS> shader_infos{}; bool last_shaders_valid = false; diff --git a/src/video_core/surface.cpp b/src/video_core/surface.cpp index 079d5f028..a2bf08294 100644 --- a/src/video_core/surface.cpp +++ b/src/video_core/surface.cpp @@ -247,6 +247,8 @@ bool IsPixelFormatASTC(PixelFormat format) { case PixelFormat::ASTC_2D_6X6_UNORM: case PixelFormat::ASTC_2D_6X6_SRGB: case PixelFormat::ASTC_2D_10X6_UNORM: + case PixelFormat::ASTC_2D_10X5_UNORM: + case PixelFormat::ASTC_2D_10X5_SRGB: case PixelFormat::ASTC_2D_10X10_UNORM: case PixelFormat::ASTC_2D_10X10_SRGB: case PixelFormat::ASTC_2D_12X12_UNORM: @@ -276,6 +278,7 @@ bool IsPixelFormatSRGB(PixelFormat format) { case PixelFormat::ASTC_2D_5X5_SRGB: case PixelFormat::ASTC_2D_10X8_SRGB: case PixelFormat::ASTC_2D_6X6_SRGB: + case PixelFormat::ASTC_2D_10X5_SRGB: case PixelFormat::ASTC_2D_10X10_SRGB: case PixelFormat::ASTC_2D_12X12_SRGB: case PixelFormat::ASTC_2D_8X6_SRGB: diff --git a/src/video_core/surface.h b/src/video_core/surface.h index 16273f185..57ca7f597 100644 --- a/src/video_core/surface.h +++ b/src/video_core/surface.h @@ -82,7 +82,7 @@ enum class PixelFormat { BC3_SRGB, BC7_SRGB, A4B4G4R4_UNORM, - R4G4_UNORM, + G4R4_UNORM, ASTC_2D_4X4_SRGB, ASTC_2D_8X8_SRGB, ASTC_2D_8X5_SRGB, @@ -94,6 +94,8 @@ enum class PixelFormat { ASTC_2D_6X6_UNORM, ASTC_2D_6X6_SRGB, ASTC_2D_10X6_UNORM, + ASTC_2D_10X5_UNORM, + ASTC_2D_10X5_SRGB, ASTC_2D_10X10_UNORM, ASTC_2D_10X10_SRGB, ASTC_2D_12X12_UNORM, @@ -216,7 +218,7 @@ constexpr std::array<u8, MaxPixelFormat> BLOCK_WIDTH_TABLE = {{ 4, // BC3_SRGB 4, // BC7_SRGB 1, // A4B4G4R4_UNORM - 1, // R4G4_UNORM + 1, // G4R4_UNORM 4, // ASTC_2D_4X4_SRGB 8, // ASTC_2D_8X8_SRGB 8, // ASTC_2D_8X5_SRGB @@ -228,6 +230,8 @@ constexpr std::array<u8, MaxPixelFormat> BLOCK_WIDTH_TABLE = {{ 6, // ASTC_2D_6X6_UNORM 6, // ASTC_2D_6X6_SRGB 10, // ASTC_2D_10X6_UNORM + 10, // ASTC_2D_10X5_UNORM + 10, // ASTC_2D_10X5_SRGB 10, // ASTC_2D_10X10_UNORM 10, // ASTC_2D_10X10_SRGB 12, // ASTC_2D_12X12_UNORM @@ -319,7 +323,7 @@ constexpr std::array<u8, MaxPixelFormat> BLOCK_HEIGHT_TABLE = {{ 4, // BC3_SRGB 4, // BC7_SRGB 1, // A4B4G4R4_UNORM - 1, // R4G4_UNORM + 1, // G4R4_UNORM 4, // ASTC_2D_4X4_SRGB 8, // ASTC_2D_8X8_SRGB 5, // ASTC_2D_8X5_SRGB @@ -331,6 +335,8 @@ constexpr std::array<u8, MaxPixelFormat> BLOCK_HEIGHT_TABLE = {{ 6, // ASTC_2D_6X6_UNORM 6, // ASTC_2D_6X6_SRGB 6, // ASTC_2D_10X6_UNORM + 5, // ASTC_2D_10X5_UNORM + 5, // ASTC_2D_10X5_SRGB 10, // ASTC_2D_10X10_UNORM 10, // ASTC_2D_10X10_SRGB 12, // ASTC_2D_12X12_UNORM @@ -422,7 +428,7 @@ constexpr std::array<u8, MaxPixelFormat> BITS_PER_BLOCK_TABLE = {{ 128, // BC3_SRGB 128, // BC7_UNORM 16, // A4B4G4R4_UNORM - 8, // R4G4_UNORM + 8, // G4R4_UNORM 128, // ASTC_2D_4X4_SRGB 128, // ASTC_2D_8X8_SRGB 128, // ASTC_2D_8X5_SRGB @@ -434,6 +440,8 @@ constexpr std::array<u8, MaxPixelFormat> BITS_PER_BLOCK_TABLE = {{ 128, // ASTC_2D_6X6_UNORM 128, // ASTC_2D_6X6_SRGB 128, // ASTC_2D_10X6_UNORM + 128, // ASTC_2D_10X5_UNORM + 128, // ASTC_2D_10X5_SRGB 128, // ASTC_2D_10X10_UNORM 128, // ASTC_2D_10X10_SRGB 128, // ASTC_2D_12X12_UNORM diff --git a/src/video_core/texture_cache/format_lookup_table.cpp b/src/video_core/texture_cache/format_lookup_table.cpp index 1412aa076..ad935d386 100644 --- a/src/video_core/texture_cache/format_lookup_table.cpp +++ b/src/video_core/texture_cache/format_lookup_table.cpp @@ -63,7 +63,7 @@ PixelFormat PixelFormatFromTextureInfo(TextureFormat format, ComponentType red, case Hash(TextureFormat::A4B4G4R4, UNORM): return PixelFormat::A4B4G4R4_UNORM; case Hash(TextureFormat::G4R4, UNORM): - return PixelFormat::R4G4_UNORM; + return PixelFormat::G4R4_UNORM; case Hash(TextureFormat::A5B5G5R1, UNORM): return PixelFormat::A5B5G5R1_UNORM; case Hash(TextureFormat::R8, UNORM): @@ -208,6 +208,10 @@ PixelFormat PixelFormatFromTextureInfo(TextureFormat format, ComponentType red, return PixelFormat::ASTC_2D_6X6_SRGB; case Hash(TextureFormat::ASTC_2D_10X6, UNORM, LINEAR): return PixelFormat::ASTC_2D_10X6_UNORM; + case Hash(TextureFormat::ASTC_2D_10X5, UNORM, LINEAR): + return PixelFormat::ASTC_2D_10X5_UNORM; + case Hash(TextureFormat::ASTC_2D_10X5, UNORM, SRGB): + return PixelFormat::ASTC_2D_10X5_SRGB; case Hash(TextureFormat::ASTC_2D_10X10, UNORM, LINEAR): return PixelFormat::ASTC_2D_10X10_UNORM; case Hash(TextureFormat::ASTC_2D_10X10, UNORM, SRGB): diff --git a/src/video_core/texture_cache/formatter.h b/src/video_core/texture_cache/formatter.h index 95a572604..acc854715 100644 --- a/src/video_core/texture_cache/formatter.h +++ b/src/video_core/texture_cache/formatter.h @@ -153,8 +153,8 @@ struct fmt::formatter<VideoCore::Surface::PixelFormat> : fmt::formatter<fmt::str return "BC7_SRGB"; case PixelFormat::A4B4G4R4_UNORM: return "A4B4G4R4_UNORM"; - case PixelFormat::R4G4_UNORM: - return "R4G4_UNORM"; + case PixelFormat::G4R4_UNORM: + return "G4R4_UNORM"; case PixelFormat::ASTC_2D_4X4_SRGB: return "ASTC_2D_4X4_SRGB"; case PixelFormat::ASTC_2D_8X8_SRGB: @@ -177,6 +177,10 @@ struct fmt::formatter<VideoCore::Surface::PixelFormat> : fmt::formatter<fmt::str return "ASTC_2D_6X6_SRGB"; case PixelFormat::ASTC_2D_10X6_UNORM: return "ASTC_2D_10X6_UNORM"; + case PixelFormat::ASTC_2D_10X5_UNORM: + return "ASTC_2D_10X5_UNORM"; + case PixelFormat::ASTC_2D_10X5_SRGB: + return "ASTC_2D_10X5_SRGB"; case PixelFormat::ASTC_2D_10X10_UNORM: return "ASTC_2D_10X10_UNORM"; case PixelFormat::ASTC_2D_10X10_SRGB: diff --git a/src/video_core/texture_cache/image_base.cpp b/src/video_core/texture_cache/image_base.cpp index f61e09ac7..91512022f 100644 --- a/src/video_core/texture_cache/image_base.cpp +++ b/src/video_core/texture_cache/image_base.cpp @@ -7,6 +7,7 @@ #include <vector> #include "common/common_types.h" +#include "common/div_ceil.h" #include "video_core/surface.h" #include "video_core/texture_cache/formatter.h" #include "video_core/texture_cache/image_base.h" @@ -182,10 +183,6 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i }; const bool is_lhs_compressed = lhs_block.width > 1 || lhs_block.height > 1; const bool is_rhs_compressed = rhs_block.width > 1 || rhs_block.height > 1; - if (is_lhs_compressed && is_rhs_compressed) { - LOG_ERROR(HW_GPU, "Compressed to compressed image aliasing is not implemented"); - return; - } const s32 lhs_mips = lhs.info.resources.levels; const s32 rhs_mips = rhs.info.resources.levels; const s32 num_mips = std::min(lhs_mips - base->level, rhs_mips); @@ -199,12 +196,12 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i Extent3D lhs_size = MipSize(lhs.info.size, base->level + mip_level); Extent3D rhs_size = MipSize(rhs.info.size, mip_level); if (is_lhs_compressed) { - lhs_size.width /= lhs_block.width; - lhs_size.height /= lhs_block.height; + lhs_size.width = Common::DivCeil(lhs_size.width, lhs_block.width); + lhs_size.height = Common::DivCeil(lhs_size.height, lhs_block.height); } if (is_rhs_compressed) { - rhs_size.width /= rhs_block.width; - rhs_size.height /= rhs_block.height; + rhs_size.width = Common::DivCeil(rhs_size.width, rhs_block.width); + rhs_size.height = Common::DivCeil(rhs_size.height, rhs_block.height); } const Extent3D copy_size{ .width = std::min(lhs_size.width, rhs_size.width), diff --git a/src/video_core/texture_cache/image_base.h b/src/video_core/texture_cache/image_base.h index 1f85ec9da..620565684 100644 --- a/src/video_core/texture_cache/image_base.h +++ b/src/video_core/texture_cache/image_base.h @@ -88,6 +88,9 @@ struct ImageBase { u32 scale_rating = 0; u64 scale_tick = 0; bool has_scaled = false; + + size_t channel = 0; + ImageFlagBits flags = ImageFlagBits::CpuModified; GPUVAddr gpu_addr = 0; diff --git a/src/video_core/texture_cache/render_targets.h b/src/video_core/texture_cache/render_targets.h index da8ffe9ec..1efbd6507 100644 --- a/src/video_core/texture_cache/render_targets.h +++ b/src/video_core/texture_cache/render_targets.h @@ -26,6 +26,7 @@ struct RenderTargets { ImageViewId depth_buffer_id{}; std::array<u8, NUM_RT> draw_buffers{}; Extent2D size{}; + bool is_rescaled{}; }; } // namespace VideoCommon diff --git a/src/video_core/texture_cache/texture_cache.cpp b/src/video_core/texture_cache/texture_cache.cpp new file mode 100644 index 000000000..8a9a32f44 --- /dev/null +++ b/src/video_core/texture_cache/texture_cache.cpp @@ -0,0 +1,15 @@ +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later + +#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 diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 1dbe01bc0..eaf4a1c95 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -1,5 +1,5 @@ -// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once @@ -7,6 +7,7 @@ #include "common/alignment.h" #include "common/settings.h" +#include "video_core/control/channel_state.h" #include "video_core/dirty_flags.h" #include "video_core/engines/kepler_compute.h" #include "video_core/texture_cache/image_view_base.h" @@ -29,12 +30,8 @@ using VideoCore::Surface::SurfaceType; using namespace Common::Literals; template <class P> -TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& rasterizer_, - Tegra::Engines::Maxwell3D& maxwell3d_, - Tegra::Engines::KeplerCompute& kepler_compute_, - Tegra::MemoryManager& gpu_memory_) - : runtime{runtime_}, rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, - kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_} { +TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& rasterizer_) + : runtime{runtime_}, rasterizer{rasterizer_} { // Configure null sampler TSCEntry sampler_descriptor{}; sampler_descriptor.min_filter.Assign(Tegra::Texture::TextureFilter::Linear); @@ -93,7 +90,7 @@ void TextureCache<P>::RunGarbageCollector() { const auto copies = FullDownloadCopies(image.info); image.DownloadMemory(map, copies); runtime.Finish(); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); } if (True(image.flags & ImageFlagBits::Tracked)) { UntrackImage(image, image_id); @@ -152,22 +149,24 @@ void TextureCache<P>::MarkModification(ImageId id) noexcept { template <class P> template <bool has_blacklists> void TextureCache<P>::FillGraphicsImageViews(std::span<ImageViewInOut> views) { - FillImageViews<has_blacklists>(graphics_image_table, graphics_image_view_ids, views); + FillImageViews<has_blacklists>(channel_state->graphics_image_table, + channel_state->graphics_image_view_ids, views); } template <class P> void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) { - FillImageViews<true>(compute_image_table, compute_image_view_ids, views); + FillImageViews<true>(channel_state->compute_image_table, channel_state->compute_image_view_ids, + views); } template <class P> typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) { - if (index > graphics_sampler_table.Limit()) { + if (index > channel_state->graphics_sampler_table.Limit()) { LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); return &slot_samplers[NULL_SAMPLER_ID]; } - const auto [descriptor, is_new] = graphics_sampler_table.Read(index); - SamplerId& id = graphics_sampler_ids[index]; + const auto [descriptor, is_new] = channel_state->graphics_sampler_table.Read(index); + SamplerId& id = channel_state->graphics_sampler_ids[index]; if (is_new) { id = FindSampler(descriptor); } @@ -176,12 +175,12 @@ typename P::Sampler* TextureCache<P>::GetGraphicsSampler(u32 index) { template <class P> typename P::Sampler* TextureCache<P>::GetComputeSampler(u32 index) { - if (index > compute_sampler_table.Limit()) { + if (index > channel_state->compute_sampler_table.Limit()) { LOG_DEBUG(HW_GPU, "Invalid sampler index={}", index); return &slot_samplers[NULL_SAMPLER_ID]; } - const auto [descriptor, is_new] = compute_sampler_table.Read(index); - SamplerId& id = compute_sampler_ids[index]; + const auto [descriptor, is_new] = channel_state->compute_sampler_table.Read(index); + SamplerId& id = channel_state->compute_sampler_ids[index]; if (is_new) { id = FindSampler(descriptor); } @@ -191,34 +190,36 @@ typename P::Sampler* TextureCache<P>::GetComputeSampler(u32 index) { template <class P> void TextureCache<P>::SynchronizeGraphicsDescriptors() { using SamplerIndex = Tegra::Engines::Maxwell3D::Regs::SamplerIndex; - const bool linked_tsc = maxwell3d.regs.sampler_index == SamplerIndex::ViaHeaderIndex; - const u32 tic_limit = maxwell3d.regs.tic.limit; - const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d.regs.tsc.limit; - if (graphics_sampler_table.Synchornize(maxwell3d.regs.tsc.Address(), tsc_limit)) { - graphics_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); + const bool linked_tsc = maxwell3d->regs.sampler_index == SamplerIndex::ViaHeaderIndex; + const u32 tic_limit = maxwell3d->regs.tic.limit; + const u32 tsc_limit = linked_tsc ? tic_limit : maxwell3d->regs.tsc.limit; + if (channel_state->graphics_sampler_table.Synchornize(maxwell3d->regs.tsc.Address(), + tsc_limit)) { + channel_state->graphics_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); } - if (graphics_image_table.Synchornize(maxwell3d.regs.tic.Address(), tic_limit)) { - graphics_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); + if (channel_state->graphics_image_table.Synchornize(maxwell3d->regs.tic.Address(), tic_limit)) { + channel_state->graphics_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); } } template <class P> void TextureCache<P>::SynchronizeComputeDescriptors() { - const bool linked_tsc = kepler_compute.launch_description.linked_tsc; - const u32 tic_limit = kepler_compute.regs.tic.limit; - const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute.regs.tsc.limit; - const GPUVAddr tsc_gpu_addr = kepler_compute.regs.tsc.Address(); - if (compute_sampler_table.Synchornize(tsc_gpu_addr, tsc_limit)) { - compute_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); + const bool linked_tsc = kepler_compute->launch_description.linked_tsc; + const u32 tic_limit = kepler_compute->regs.tic.limit; + const u32 tsc_limit = linked_tsc ? tic_limit : kepler_compute->regs.tsc.limit; + const GPUVAddr tsc_gpu_addr = kepler_compute->regs.tsc.Address(); + if (channel_state->compute_sampler_table.Synchornize(tsc_gpu_addr, tsc_limit)) { + channel_state->compute_sampler_ids.resize(tsc_limit + 1, CORRUPT_ID); } - if (compute_image_table.Synchornize(kepler_compute.regs.tic.Address(), tic_limit)) { - compute_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); + if (channel_state->compute_image_table.Synchornize(kepler_compute->regs.tic.Address(), + tic_limit)) { + channel_state->compute_image_view_ids.resize(tic_limit + 1, CORRUPT_ID); } } template <class P> bool TextureCache<P>::RescaleRenderTargets(bool is_clear) { - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; u32 scale_rating = 0; bool rescaled = false; std::array<ImageId, NUM_RT> tmp_color_images{}; @@ -315,7 +316,7 @@ bool TextureCache<P>::RescaleRenderTargets(bool is_clear) { template <class P> void TextureCache<P>::UpdateRenderTargets(bool is_clear) { using namespace VideoCommon::Dirty; - auto& flags = maxwell3d.dirty.flags; + auto& flags = maxwell3d->dirty.flags; if (!flags[Dirty::RenderTargets]) { for (size_t index = 0; index < NUM_RT; ++index) { ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index]; @@ -342,7 +343,7 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) { PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id)); for (size_t index = 0; index < NUM_RT; ++index) { - render_targets.draw_buffers[index] = static_cast<u8>(maxwell3d.regs.rt_control.Map(index)); + render_targets.draw_buffers[index] = static_cast<u8>(maxwell3d->regs.rt_control.Map(index)); } u32 up_scale = 1; u32 down_shift = 0; @@ -351,9 +352,10 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) { down_shift = Settings::values.resolution_info.down_shift; } render_targets.size = Extent2D{ - (maxwell3d.regs.render_area.width * up_scale) >> down_shift, - (maxwell3d.regs.render_area.height * up_scale) >> down_shift, + (maxwell3d->regs.render_area.width * up_scale) >> down_shift, + (maxwell3d->regs.render_area.height * up_scale) >> down_shift, }; + render_targets.is_rescaled = is_rescaling; flags[Dirty::DepthBiasGlobal] = true; } @@ -458,7 +460,7 @@ void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) { const auto copies = FullDownloadCopies(image.info); image.DownloadMemory(map, copies); runtime.Finish(); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); } } @@ -477,12 +479,20 @@ void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) { } template <class P> -void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) { +void TextureCache<P>::UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size) { std::vector<ImageId> deleted_images; - ForEachImageInRegionGPU(gpu_addr, size, + ForEachImageInRegionGPU(as_id, gpu_addr, size, [&](ImageId id, Image&) { deleted_images.push_back(id); }); for (const ImageId id : deleted_images) { Image& image = slot_images[id]; + if (True(image.flags & ImageFlagBits::CpuModified)) { + return; + } + image.flags |= ImageFlagBits::CpuModified; + if (True(image.flags & ImageFlagBits::Tracked)) { + UntrackImage(image, id); + } + /* if (True(image.flags & ImageFlagBits::Remapped)) { continue; } @@ -490,6 +500,7 @@ void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) { if (True(image.flags & ImageFlagBits::Tracked)) { UntrackImage(image, id); } + */ } } @@ -655,7 +666,7 @@ void TextureCache<P>::PopAsyncFlushes() { for (const ImageId image_id : download_ids) { const ImageBase& image = slot_images[image_id]; const auto copies = FullDownloadCopies(image.info); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, download_span); + SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, download_span); download_map.offset += image.unswizzled_size_bytes; download_span = download_span.subspan(image.unswizzled_size_bytes); } @@ -714,26 +725,26 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging) const GPUVAddr gpu_addr = image.gpu_addr; if (True(image.flags & ImageFlagBits::AcceleratedUpload)) { - gpu_memory.ReadBlockUnsafe(gpu_addr, mapped_span.data(), mapped_span.size_bytes()); + gpu_memory->ReadBlockUnsafe(gpu_addr, mapped_span.data(), mapped_span.size_bytes()); const auto uploads = FullUploadSwizzles(image.info); runtime.AccelerateImageUpload(image, staging, uploads); } else if (True(image.flags & ImageFlagBits::Converted)) { std::vector<u8> unswizzled_data(image.unswizzled_size_bytes); - auto copies = UnswizzleImage(gpu_memory, gpu_addr, image.info, unswizzled_data); + auto copies = UnswizzleImage(*gpu_memory, gpu_addr, image.info, unswizzled_data); ConvertImage(unswizzled_data, image.info, mapped_span, copies); image.UploadMemory(staging, copies); } else { - const auto copies = UnswizzleImage(gpu_memory, gpu_addr, image.info, mapped_span); + const auto copies = UnswizzleImage(*gpu_memory, gpu_addr, image.info, mapped_span); image.UploadMemory(staging, copies); } } template <class P> ImageViewId TextureCache<P>::FindImageView(const TICEntry& config) { - if (!IsValidEntry(gpu_memory, config)) { + if (!IsValidEntry(*gpu_memory, config)) { return NULL_IMAGE_VIEW_ID; } - const auto [pair, is_new] = image_views.try_emplace(config); + const auto [pair, is_new] = channel_state->image_views.try_emplace(config); ImageViewId& image_view_id = pair->second; if (is_new) { image_view_id = CreateImageView(config); @@ -777,9 +788,9 @@ 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<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (!cpu_addr) { - cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr, CalculateGuestSizeInBytes(info)); + cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, CalculateGuestSizeInBytes(info)); if (!cpu_addr) { return ImageId{}; } @@ -860,7 +871,7 @@ void TextureCache<P>::InvalidateScale(Image& image) { image.scale_tick = frame_tick + 1; } const std::span<const ImageViewId> image_view_ids = image.image_view_ids; - auto& dirty = maxwell3d.dirty.flags; + auto& dirty = maxwell3d->dirty.flags; dirty[Dirty::RenderTargets] = true; dirty[Dirty::ZetaBuffer] = true; for (size_t rt = 0; rt < NUM_RT; ++rt) { @@ -880,12 +891,15 @@ void TextureCache<P>::InvalidateScale(Image& image) { } image.image_view_ids.clear(); image.image_view_infos.clear(); - if constexpr (ENABLE_VALIDATION) { - std::ranges::fill(graphics_image_view_ids, CORRUPT_ID); - std::ranges::fill(compute_image_view_ids, CORRUPT_ID); + for (size_t c : active_channel_ids) { + auto& channel_info = channel_storage[c]; + if constexpr (ENABLE_VALIDATION) { + std::ranges::fill(channel_info.graphics_image_view_ids, CORRUPT_ID); + std::ranges::fill(channel_info.compute_image_view_ids, CORRUPT_ID); + } + channel_info.graphics_image_table.Invalidate(); + channel_info.compute_image_table.Invalidate(); } - graphics_image_table.Invalidate(); - compute_image_table.Invalidate(); has_deleted_images = true; } @@ -929,10 +943,10 @@ 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<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (!cpu_addr) { const auto size = CalculateGuestSizeInBytes(info); - cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr, size); + cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, size); if (!cpu_addr) { const VAddr fake_addr = ~(1ULL << 40ULL) + virtual_invalid_space; virtual_invalid_space += Common::AlignUp(size, 32); @@ -1050,7 +1064,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr); Image& new_image = slot_images[new_image_id]; - if (!gpu_memory.IsContinousRange(new_image.gpu_addr, new_image.guest_size_bytes)) { + if (!gpu_memory->IsContinousRange(new_image.gpu_addr, new_image.guest_size_bytes)) { new_image.flags |= ImageFlagBits::Sparse; } @@ -1192,7 +1206,7 @@ SamplerId TextureCache<P>::FindSampler(const TSCEntry& config) { if (std::ranges::all_of(config.raw, [](u64 value) { return value == 0; })) { return NULL_SAMPLER_ID; } - const auto [pair, is_new] = samplers.try_emplace(config); + const auto [pair, is_new] = channel_state->samplers.try_emplace(config); if (is_new) { pair->second = slot_samplers.insert(runtime, config); } @@ -1201,7 +1215,7 @@ SamplerId TextureCache<P>::FindSampler(const TSCEntry& config) { template <class P> ImageViewId TextureCache<P>::FindColorBuffer(size_t index, bool is_clear) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (index >= regs.rt_control.count) { return ImageViewId{}; } @@ -1219,7 +1233,7 @@ ImageViewId TextureCache<P>::FindColorBuffer(size_t index, bool is_clear) { template <class P> ImageViewId TextureCache<P>::FindDepthBuffer(bool is_clear) { - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; if (!regs.zeta_enable) { return ImageViewId{}; } @@ -1316,11 +1330,17 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f template <class P> template <typename Func> -void TextureCache<P>::ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func) { +void TextureCache<P>::ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_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, 8> images; - ForEachGPUPage(gpu_addr, size, [this, &images, gpu_addr, size, func](u64 page) { + auto storage_id = getStorageID(as_id); + if (!storage_id) { + return; + } + auto& gpu_page_table = gpu_page_table_storage[*storage_id]; + ForEachGPUPage(gpu_addr, size, [this, gpu_page_table, &images, gpu_addr, size, func](u64 page) { const auto it = gpu_page_table.find(page); if (it == gpu_page_table.end()) { if constexpr (BOOL_BREAK) { @@ -1403,9 +1423,9 @@ template <typename Func> void TextureCache<P>::ForEachSparseSegment(ImageBase& image, Func&& func) { using FuncReturn = typename std::invoke_result<Func, GPUVAddr, VAddr, 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); + 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<VAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); ASSERT(cpu_addr); if constexpr (RETURNS_BOOL) { if (func(gpu_addr, *cpu_addr, size)) { @@ -1448,8 +1468,9 @@ void TextureCache<P>::RegisterImage(ImageId image_id) { } image.lru_index = lru_cache.Insert(image_id, frame_tick); - ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, - [this, image_id](u64 page) { gpu_page_table[page].push_back(image_id); }); + ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) { + (*channel_state->gpu_page_table)[page].push_back(image_id); + }); if (False(image.flags & ImageFlagBits::Sparse)) { auto map_id = slot_map_views.insert(image.gpu_addr, image.cpu_addr, image.guest_size_bytes, image_id); @@ -1480,9 +1501,9 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) { image.flags &= ~ImageFlagBits::BadOverlap; lru_cache.Free(image.lru_index); const auto& clear_page_table = - [this, image_id]( - u64 page, - std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>>& selected_page_table) { + [this, image_id](u64 page, + std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>>& + selected_page_table) { const auto page_it = selected_page_table.find(page); if (page_it == selected_page_table.end()) { ASSERT_MSG(false, "Unregistering unregistered page=0x{:x}", page << YUZU_PAGEBITS); @@ -1497,8 +1518,9 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) { } image_ids.erase(vector_it); }; - ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, - [this, &clear_page_table](u64 page) { clear_page_table(page, gpu_page_table); }); + ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, &clear_page_table](u64 page) { + clear_page_table(page, (*channel_state->gpu_page_table)); + }); if (False(image.flags & ImageFlagBits::Sparse)) { const auto map_id = image.map_view_id; ForEachCPUPage(image.cpu_addr, image.guest_size_bytes, [this, map_id](u64 page) { @@ -1631,7 +1653,7 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) { ASSERT_MSG(False(image.flags & ImageFlagBits::Registered), "Image was not unregistered"); // Mark render targets as dirty - auto& dirty = maxwell3d.dirty.flags; + auto& dirty = maxwell3d->dirty.flags; dirty[Dirty::RenderTargets] = true; dirty[Dirty::ZetaBuffer] = true; for (size_t rt = 0; rt < NUM_RT; ++rt) { @@ -1681,24 +1703,30 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) { if (alloc_images.empty()) { image_allocs_table.erase(alloc_it); } - if constexpr (ENABLE_VALIDATION) { - std::ranges::fill(graphics_image_view_ids, CORRUPT_ID); - std::ranges::fill(compute_image_view_ids, CORRUPT_ID); + for (size_t c : active_channel_ids) { + auto& channel_info = channel_storage[c]; + if constexpr (ENABLE_VALIDATION) { + std::ranges::fill(channel_info.graphics_image_view_ids, CORRUPT_ID); + std::ranges::fill(channel_info.compute_image_view_ids, CORRUPT_ID); + } + channel_info.graphics_image_table.Invalidate(); + channel_info.compute_image_table.Invalidate(); } - graphics_image_table.Invalidate(); - compute_image_table.Invalidate(); has_deleted_images = true; } template <class P> void TextureCache<P>::RemoveImageViewReferences(std::span<const ImageViewId> removed_views) { - auto it = image_views.begin(); - while (it != image_views.end()) { - const auto found = std::ranges::find(removed_views, it->second); - if (found != removed_views.end()) { - it = image_views.erase(it); - } else { - ++it; + for (size_t c : active_channel_ids) { + auto& channel_info = channel_storage[c]; + auto it = channel_info.image_views.begin(); + while (it != channel_info.image_views.end()) { + const auto found = std::ranges::find(removed_views, it->second); + if (found != removed_views.end()) { + it = channel_info.image_views.erase(it); + } else { + ++it; + } } } } @@ -1729,6 +1757,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { boost::container::small_vector<const AliasedImage*, 1> aliased_images; Image& image = slot_images[image_id]; bool any_rescaled = True(image.flags & ImageFlagBits::Rescaled); + bool any_modified = True(image.flags & ImageFlagBits::GpuModified); u64 most_recent_tick = image.modification_tick; for (const AliasedImage& aliased : image.aliased_images) { ImageBase& aliased_image = slot_images[aliased.id]; @@ -1736,9 +1765,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { most_recent_tick = std::max(most_recent_tick, aliased_image.modification_tick); aliased_images.push_back(&aliased); any_rescaled |= True(aliased_image.flags & ImageFlagBits::Rescaled); - if (True(aliased_image.flags & ImageFlagBits::GpuModified)) { - image.flags |= ImageFlagBits::GpuModified; - } + any_modified |= True(aliased_image.flags & ImageFlagBits::GpuModified); } } if (aliased_images.empty()) { @@ -1753,6 +1780,9 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) { } } image.modification_tick = most_recent_tick; + if (any_modified) { + image.flags |= ImageFlagBits::GpuModified; + } std::ranges::sort(aliased_images, [this](const AliasedImage* lhs, const AliasedImage* rhs) { const ImageBase& lhs_image = slot_images[lhs->id]; const ImageBase& rhs_image = slot_images[rhs->id]; @@ -1931,6 +1961,7 @@ std::pair<FramebufferId, ImageViewId> TextureCache<P>::RenderTargetFromImage( .color_buffer_ids = {color_view_id}, .depth_buffer_id = depth_view_id, .size = {extent.width >> samples_x, extent.height >> samples_y}, + .is_rescaled = is_rescaled, }); return {framebuffer_id, view_id}; } @@ -1943,7 +1974,7 @@ bool TextureCache<P>::IsFullClear(ImageViewId id) { const ImageViewBase& image_view = slot_image_views[id]; const ImageBase& image = slot_images[image_view.image_id]; const Extent3D size = image_view.size; - const auto& regs = maxwell3d.regs; + const auto& regs = maxwell3d->regs; const auto& scissor = regs.scissor_test[0]; if (image.info.resources.levels > 1 || image.info.resources.layers > 1) { // Images with multiple resources can't be cleared in a single call @@ -1958,4 +1989,19 @@ bool TextureCache<P>::IsFullClear(ImageViewId id) { scissor.max_y >= size.height; } +template <class P> +void TextureCache<P>::CreateChannel(struct Tegra::Control::ChannelState& channel) { + VideoCommon::ChannelSetupCaches<TextureCacheChannelInfo>::CreateChannel(channel); + const auto it = channel_map.find(channel.bind_id); + auto* this_state = &channel_storage[it->second]; + const auto& this_as_ref = address_spaces[channel.memory_manager->GetID()]; + this_state->gpu_page_table = &gpu_page_table_storage[this_as_ref.storage_id]; +} + +/// Bind a channel for execution. +template <class P> +void TextureCache<P>::OnGPUASRegister([[maybe_unused]] size_t map_id) { + gpu_page_table_storage.emplace_back(); +} + } // namespace VideoCommon diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index 7e6c6cef2..2fa8445eb 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -1,8 +1,10 @@ -// SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project -// SPDX-License-Identifier: GPL-2.0-or-later +// SPDX-FileCopyrightText: 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-3.0-or-later #pragma once +#include <deque> +#include <limits> #include <mutex> #include <span> #include <type_traits> @@ -11,9 +13,11 @@ #include <queue> #include "common/common_types.h" +#include "common/hash.h" #include "common/literals.h" #include "common/lru_cache.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/engines/fermi_2d.h" #include "video_core/surface.h" @@ -26,6 +30,10 @@ #include "video_core/texture_cache/types.h" #include "video_core/textures/texture.h" +namespace Tegra::Control { +struct ChannelState; +} + namespace VideoCommon { using Tegra::Texture::SwizzleSource; @@ -44,8 +52,35 @@ struct ImageViewInOut { 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> -class TextureCache { +class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelInfo> { /// Address shift for caching images into a hash table static constexpr u64 YUZU_PAGEBITS = 20; @@ -58,6 +93,8 @@ class TextureCache { /// True when the API can provide info about the memory of the device. static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO; + static constexpr size_t UNSET_CHANNEL{std::numeric_limits<size_t>::max()}; + static constexpr s64 TARGET_THRESHOLD = 4_GiB; static constexpr s64 DEFAULT_EXPECTED_MEMORY = 1_GiB + 125_MiB; static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB + 625_MiB; @@ -77,16 +114,8 @@ class TextureCache { PixelFormat src_format; }; - template <typename T> - struct IdentityHash { - [[nodiscard]] size_t operator()(T value) const noexcept { - return static_cast<size_t>(value); - } - }; - public: - explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&, Tegra::Engines::Maxwell3D&, - Tegra::Engines::KeplerCompute&, Tegra::MemoryManager&); + explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&); /// Notify the cache that a new frame has been queued void TickFrame(); @@ -142,7 +171,7 @@ public: void UnmapMemory(VAddr cpu_addr, size_t size); /// Remove images in a region - void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size); + void UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size); /// Blit an image with the given parameters void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst, @@ -171,6 +200,9 @@ public: [[nodiscard]] bool IsRescaling(const ImageViewBase& image_view) const noexcept; + /// Create channel state. + void CreateChannel(Tegra::Control::ChannelState& channel) final override; + std::mutex mutex; private: @@ -205,6 +237,8 @@ private: } } + void OnGPUASRegister(size_t map_id) final override; + /// Runs the Garbage Collector. void RunGarbageCollector(); @@ -273,7 +307,7 @@ private: void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func); template <typename Func> - void ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func); + void ForEachImageInRegionGPU(size_t as_id, GPUVAddr gpu_addr, size_t size, Func&& func); template <typename Func> void ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func); @@ -338,31 +372,16 @@ private: u64 GetScaledImageSizeBytes(ImageBase& image); Runtime& runtime; - VideoCore::RasterizerInterface& rasterizer; - 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; + VideoCore::RasterizerInterface& rasterizer; + std::deque<TextureCacheGPUMap> gpu_page_table_storage; RenderTargets render_targets; - std::unordered_map<TICEntry, ImageViewId> image_views; - std::unordered_map<TSCEntry, SamplerId> samplers; std::unordered_map<RenderTargets, FramebufferId> framebuffers; - std::unordered_map<u64, std::vector<ImageMapId>, IdentityHash<u64>> page_table; - std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> gpu_page_table; - std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> sparse_page_table; - + std::unordered_map<u64, std::vector<ImageMapId>, Common::IdentityHash<u64>> page_table; + std::unordered_map<u64, std::vector<ImageId>, Common::IdentityHash<u64>> sparse_page_table; std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views; VAddr virtual_invalid_space{}; diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp index 1820823b2..1223df5a0 100644 --- a/src/video_core/texture_cache/util.cpp +++ b/src/video_core/texture_cache/util.cpp @@ -517,7 +517,6 @@ void SwizzleBlockLinearImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr const u32 host_bytes_per_layer = num_blocks_per_layer * bytes_per_block; UNIMPLEMENTED_IF(info.tile_width_spacing > 0); - UNIMPLEMENTED_IF(copy.image_offset.x != 0); UNIMPLEMENTED_IF(copy.image_offset.y != 0); UNIMPLEMENTED_IF(copy.image_offset.z != 0); @@ -755,7 +754,7 @@ bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config if (address == 0) { return false; } - if (address > (1ULL << 48)) { + if (address >= (1ULL << 40)) { return false; } if (gpu_memory.GpuToCpuAddress(address).has_value()) { diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index e3742ddf5..15b9d4182 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -1656,7 +1656,7 @@ void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, const u32 cols = Common::DivideUp(width, block_width); Common::ThreadWorker workers{std::max(std::thread::hardware_concurrency(), 2U) / 2, - "yuzu:ASTCDecompress"}; + "ASTCDecompress"}; for (u32 z = 0; z < depth; ++z) { const u32 depth_offset = z * height * width * 4; diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 913f8ebcb..52d067a2d 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -35,7 +35,7 @@ void incrpdep(u32& value) { template <bool TO_LINEAR, u32 BYTES_PER_PIXEL> void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, u32 depth, - u32 block_height, u32 block_depth, u32 stride_alignment) { + u32 block_height, u32 block_depth, u32 stride) { // The origin of the transformation can be configured here, leave it as zero as the current API // doesn't expose it. static constexpr u32 origin_x = 0; @@ -45,7 +45,6 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 // We can configure here a custom pitch // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch. const u32 pitch = width * BYTES_PER_PIXEL; - const u32 stride = Common::AlignUpLog2(width, stride_alignment) * BYTES_PER_PIXEL; const u32 gobs_in_x = Common::DivCeilLog2(stride, GOB_SIZE_X_SHIFT); const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth); @@ -89,6 +88,69 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 } } +template <bool TO_LINEAR, u32 BYTES_PER_PIXEL> +void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, + u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 num_lines, + u32 block_height, u32 block_depth, u32 pitch_linear) { + // The origin of the transformation can be configured here, leave it as zero as the current API + // doesn't expose it. + static constexpr u32 origin_z = 0; + + // We can configure here a custom pitch + // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch. + const u32 pitch = pitch_linear; + const u32 stride = Common::AlignUpLog2(width * BYTES_PER_PIXEL, GOB_SIZE_X_SHIFT); + + const u32 gobs_in_x = Common::DivCeilLog2(stride, GOB_SIZE_X_SHIFT); + const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth); + const u32 slice_size = + Common::DivCeilLog2(height, block_height + GOB_SIZE_Y_SHIFT) * block_size; + + const u32 block_height_mask = (1U << block_height) - 1; + const u32 block_depth_mask = (1U << block_depth) - 1; + const u32 x_shift = GOB_SIZE_SHIFT + block_height + block_depth; + + u32 unprocessed_lines = num_lines; + u32 extent_y = std::min(num_lines, height - origin_y); + + for (u32 slice = 0; slice < depth; ++slice) { + const u32 z = slice + origin_z; + const u32 offset_z = (z >> block_depth) * slice_size + + ((z & block_depth_mask) << (GOB_SIZE_SHIFT + block_height)); + const u32 lines_in_y = std::min(unprocessed_lines, extent_y); + for (u32 line = 0; line < lines_in_y; ++line) { + const u32 y = line + origin_y; + const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(y); + + const u32 block_y = y >> GOB_SIZE_Y_SHIFT; + const u32 offset_y = (block_y >> block_height) * block_size + + ((block_y & block_height_mask) << GOB_SIZE_SHIFT); + + u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL); + for (u32 column = 0; column < extent_x; + ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) { + const u32 x = (column + origin_x) * BYTES_PER_PIXEL; + const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift; + + const u32 base_swizzled_offset = offset_z + offset_y + offset_x; + const u32 swizzled_offset = base_swizzled_offset + (swizzled_x | swizzled_y); + + const u32 unswizzled_offset = + slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL; + + u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset]; + const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset]; + + std::memcpy(dst, src, BYTES_PER_PIXEL); + } + } + unprocessed_lines -= lines_in_y; + if (unprocessed_lines == 0) { + return; + } + } +} + template <bool TO_LINEAR> void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) { @@ -111,122 +173,39 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe } } -template <u32 BYTES_PER_PIXEL> -void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u8* swizzled_data, const u8* unswizzled_data, u32 block_height_bit, - u32 offset_x, u32 offset_y) { - const u32 block_height = 1U << block_height_bit; - const u32 image_width_in_gobs = - (swizzled_width * BYTES_PER_PIXEL + (GOB_SIZE_X - 1)) / GOB_SIZE_X; - for (u32 line = 0; line < subrect_height; ++line) { - const u32 dst_y = line + offset_y; - const u32 gob_address_y = - (dst_y / (GOB_SIZE_Y * block_height)) * GOB_SIZE * block_height * image_width_in_gobs + - ((dst_y % (GOB_SIZE_Y * block_height)) / GOB_SIZE_Y) * GOB_SIZE; - - const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(dst_y); - u32 swizzled_x = pdep<SWIZZLE_X_BITS>(offset_x * BYTES_PER_PIXEL); - for (u32 x = 0; x < subrect_width; - ++x, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) { - const u32 dst_x = x + offset_x; - const u32 gob_address = - gob_address_y + (dst_x * BYTES_PER_PIXEL / GOB_SIZE_X) * GOB_SIZE * block_height; - const u32 swizzled_offset = gob_address + (swizzled_x | swizzled_y); - const u32 unswizzled_offset = line * source_pitch + x * BYTES_PER_PIXEL; - - const u8* const source_line = unswizzled_data + unswizzled_offset; - u8* const dest_addr = swizzled_data + swizzled_offset; - std::memcpy(dest_addr, source_line, BYTES_PER_PIXEL); - } - } -} - -template <u32 BYTES_PER_PIXEL> -void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 block_height, - u32 origin_x, u32 origin_y, u8* output, const u8* input) { - const u32 stride = width * BYTES_PER_PIXEL; - const u32 gobs_in_x = (stride + GOB_SIZE_X - 1) / GOB_SIZE_X; - const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height); - - const u32 block_height_mask = (1U << block_height) - 1; - const u32 x_shift = GOB_SIZE_SHIFT + block_height; - - for (u32 line = 0; line < line_count; ++line) { - const u32 src_y = line + origin_y; - const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(src_y); - - const u32 block_y = src_y >> GOB_SIZE_Y_SHIFT; - const u32 src_offset_y = (block_y >> block_height) * block_size + - ((block_y & block_height_mask) << GOB_SIZE_SHIFT); - - u32 swizzled_x = pdep<SWIZZLE_X_BITS>(origin_x * BYTES_PER_PIXEL); - for (u32 column = 0; column < line_length_in; - ++column, incrpdep<SWIZZLE_X_BITS, BYTES_PER_PIXEL>(swizzled_x)) { - const u32 src_x = (column + origin_x) * BYTES_PER_PIXEL; - const u32 src_offset_x = (src_x >> GOB_SIZE_X_SHIFT) << x_shift; - - const u32 swizzled_offset = src_offset_y + src_offset_x + (swizzled_x | swizzled_y); - const u32 unswizzled_offset = line * pitch + column * BYTES_PER_PIXEL; - - std::memcpy(output + unswizzled_offset, input + swizzled_offset, BYTES_PER_PIXEL); - } - } -} - -template <u32 BYTES_PER_PIXEL> -void SwizzleSliceToVoxel(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 height, - u32 block_height, u32 block_depth, u32 origin_x, u32 origin_y, u8* output, - const u8* input) { - UNIMPLEMENTED_IF(origin_x > 0); - UNIMPLEMENTED_IF(origin_y > 0); - - const u32 stride = width * BYTES_PER_PIXEL; - const u32 gobs_in_x = (stride + GOB_SIZE_X - 1) / GOB_SIZE_X; - const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth); - - const u32 block_height_mask = (1U << block_height) - 1; - const u32 x_shift = static_cast<u32>(GOB_SIZE_SHIFT) + block_height + block_depth; - - for (u32 line = 0; line < line_count; ++line) { - const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(line); - const u32 block_y = line / GOB_SIZE_Y; - const u32 dst_offset_y = - (block_y >> block_height) * block_size + (block_y & block_height_mask) * GOB_SIZE; - - u32 swizzled_x = 0; - for (u32 x = 0; x < line_length_in; ++x, incrpdep<SWIZZLE_X_BITS, 1>(swizzled_x)) { - const u32 dst_offset = - ((x / GOB_SIZE_X) << x_shift) + dst_offset_y + (swizzled_x | swizzled_y); - const u32 src_offset = x * BYTES_PER_PIXEL + line * pitch; - std::memcpy(output + dst_offset, input + src_offset, BYTES_PER_PIXEL); - } - } -} } // Anonymous namespace void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) { + const u32 stride = Common::AlignUpLog2(width, stride_alignment) * bytes_per_pixel; + const u32 new_bpp = std::min(4U, static_cast<u32>(std::countr_zero(width * bytes_per_pixel))); + width = (width * bytes_per_pixel) >> new_bpp; + bytes_per_pixel = 1U << new_bpp; Swizzle<false>(output, input, bytes_per_pixel, width, height, depth, block_height, block_depth, - stride_alignment); + stride); } void SwizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) { + const u32 stride = Common::AlignUpLog2(width, stride_alignment) * bytes_per_pixel; + const u32 new_bpp = std::min(4U, static_cast<u32>(std::countr_zero(width * bytes_per_pixel))); + width = (width * bytes_per_pixel) >> new_bpp; + bytes_per_pixel = 1U << new_bpp; Swizzle<true>(output, input, bytes_per_pixel, width, height, depth, block_height, block_depth, - stride_alignment); + stride); } -void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u32 bytes_per_pixel, u8* swizzled_data, const u8* unswizzled_data, - u32 block_height_bit, u32 offset_x, u32 offset_y) { +void SwizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, + u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 extent_y, + u32 block_height, u32 block_depth, u32 pitch_linear) { switch (bytes_per_pixel) { #define BPP_CASE(x) \ case x: \ - return SwizzleSubrect<x>(subrect_width, subrect_height, source_pitch, swizzled_width, \ - swizzled_data, unswizzled_data, block_height_bit, offset_x, \ - offset_y); + return SwizzleSubrectImpl<true, x>(output, input, width, height, depth, origin_x, \ + origin_y, extent_x, extent_y, block_height, \ + block_depth, pitch_linear); BPP_CASE(1) BPP_CASE(2) BPP_CASE(3) @@ -241,13 +220,15 @@ void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 } } -void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 bytes_per_pixel, - u32 block_height, u32 origin_x, u32 origin_y, u8* output, const u8* input) { +void UnswizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, + u32 width, u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, + u32 extent_y, u32 block_height, u32 block_depth, u32 pitch_linear) { switch (bytes_per_pixel) { #define BPP_CASE(x) \ case x: \ - return UnswizzleSubrect<x>(line_length_in, line_count, pitch, width, block_height, \ - origin_x, origin_y, output, input); + return SwizzleSubrectImpl<false, x>(output, input, width, height, depth, origin_x, \ + origin_y, extent_x, extent_y, block_height, \ + block_depth, pitch_linear); BPP_CASE(1) BPP_CASE(2) BPP_CASE(3) @@ -262,55 +243,6 @@ void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, } } -void SwizzleSliceToVoxel(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 height, - u32 bytes_per_pixel, u32 block_height, u32 block_depth, u32 origin_x, - u32 origin_y, u8* output, const u8* input) { - switch (bytes_per_pixel) { -#define BPP_CASE(x) \ - case x: \ - return SwizzleSliceToVoxel<x>(line_length_in, line_count, pitch, width, height, \ - block_height, block_depth, origin_x, origin_y, output, \ - input); - BPP_CASE(1) - BPP_CASE(2) - BPP_CASE(3) - BPP_CASE(4) - BPP_CASE(6) - BPP_CASE(8) - BPP_CASE(12) - BPP_CASE(16) -#undef BPP_CASE - default: - ASSERT_MSG(false, "Invalid bytes_per_pixel={}", bytes_per_pixel); - } -} - -void SwizzleKepler(const u32 width, const u32 height, const u32 dst_x, const u32 dst_y, - const u32 block_height_bit, const std::size_t copy_size, const u8* source_data, - u8* swizzle_data) { - const u32 block_height = 1U << block_height_bit; - const u32 image_width_in_gobs{(width + GOB_SIZE_X - 1) / GOB_SIZE_X}; - std::size_t count = 0; - for (std::size_t y = dst_y; y < height && count < copy_size; ++y) { - const std::size_t gob_address_y = - (y / (GOB_SIZE_Y * block_height)) * GOB_SIZE * block_height * image_width_in_gobs + - ((y % (GOB_SIZE_Y * block_height)) / GOB_SIZE_Y) * GOB_SIZE; - const u32 swizzled_y = pdep<SWIZZLE_Y_BITS>(static_cast<u32>(y)); - u32 swizzled_x = pdep<SWIZZLE_X_BITS>(dst_x); - for (std::size_t x = dst_x; x < width && count < copy_size; - ++x, incrpdep<SWIZZLE_X_BITS, 1>(swizzled_x)) { - const std::size_t gob_address = - gob_address_y + (x / GOB_SIZE_X) * GOB_SIZE * block_height; - const std::size_t swizzled_offset = gob_address + (swizzled_x | swizzled_y); - const u8* source_line = source_data + count; - u8* dest_addr = swizzle_data + swizzled_offset; - count++; - - *dest_addr = *source_line; - } - } -} - std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth) { if (tiled) { diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index 31a11708f..e70407692 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -40,7 +40,6 @@ constexpr SwizzleTable MakeSwizzleTable() { } return table; } -constexpr SwizzleTable SWIZZLE_TABLE = MakeSwizzleTable(); /// Unswizzles a block linear texture into linear memory. void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, @@ -57,34 +56,14 @@ std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height u32 block_height, u32 block_depth); /// Copies an untiled subrectangle into a tiled surface. -void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u32 bytes_per_pixel, u8* swizzled_data, const u8* unswizzled_data, - u32 block_height_bit, u32 offset_x, u32 offset_y); +void SwizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width, + u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 extent_y, + u32 block_height, u32 block_depth, u32 pitch_linear); /// Copies a tiled subrectangle into a linear surface. -void UnswizzleSubrect(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 bytes_per_pixel, - u32 block_height, u32 origin_x, u32 origin_y, u8* output, const u8* input); - -/// @brief Swizzles a 2D array of pixels into a 3D texture -/// @param line_length_in Number of pixels per line -/// @param line_count Number of lines -/// @param pitch Number of bytes per line -/// @param width Width of the swizzled texture -/// @param height Height of the swizzled texture -/// @param bytes_per_pixel Number of bytes used per pixel -/// @param block_height Block height shift -/// @param block_depth Block depth shift -/// @param origin_x Column offset in pixels of the swizzled texture -/// @param origin_y Row offset in pixels of the swizzled texture -/// @param output Pointer to the pixels of the swizzled texture -/// @param input Pointer to the 2D array of pixels used as input -/// @pre input and output points to an array large enough to hold the number of bytes used -void SwizzleSliceToVoxel(u32 line_length_in, u32 line_count, u32 pitch, u32 width, u32 height, - u32 bytes_per_pixel, u32 block_height, u32 block_depth, u32 origin_x, - u32 origin_y, u8* output, const u8* input); - -void SwizzleKepler(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height, - std::size_t copy_size, const u8* source_data, u8* swizzle_data); +void UnswizzleSubrect(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, + u32 width, u32 height, u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, + u32 extent_y, u32 block_height, u32 block_depth, u32 pitch_linear); /// Obtains the offset of the gob for positions 'dst_x' & 'dst_y' u64 GetGOBOffset(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height, diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 795f16bfb..1b3f493bd 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -519,9 +519,7 @@ public: dld{rhs.dld} {} /// Assign an allocation transfering ownership from another allocation. - /// Releases any previously held allocation. PoolAllocations& operator=(PoolAllocations&& rhs) noexcept { - Release(); allocations = std::move(rhs.allocations); num = rhs.num; device = rhs.device; @@ -530,11 +528,6 @@ public: return *this; } - /// Destroys any held allocation. - ~PoolAllocations() { - Release(); - } - /// Returns the number of allocations. std::size_t size() const noexcept { return num; @@ -557,19 +550,6 @@ public: } private: - /// Destroys the held allocations if they exist. - void Release() noexcept { - if (!allocations) { - return; - } - const Span<AllocationType> span(allocations.get(), num); - const VkResult result = Free(device, pool, span, *dld); - // There's no way to report errors from a destructor. - if (result != VK_SUCCESS) { - std::terminate(); - } - } - std::unique_ptr<AllocationType[]> allocations; std::size_t num = 0; VkDevice device = nullptr; diff --git a/src/web_service/web_backend.cpp b/src/web_service/web_backend.cpp index 378804c08..12a7e4922 100644 --- a/src/web_service/web_backend.cpp +++ b/src/web_service/web_backend.cpp @@ -111,7 +111,8 @@ struct Client::Impl { httplib::Error error; if (!cli->send(request, response, error)) { - LOG_ERROR(WebService, "{} to {} returned null", method, host + path); + LOG_ERROR(WebService, "{} to {} returned null (httplib Error: {})", method, host + path, + httplib::to_string(error)); return WebResult{WebResult::Code::LibError, "Null response", ""}; } diff --git a/src/yuzu/bootmanager.cpp b/src/yuzu/bootmanager.cpp index d3fbdb09d..24251247d 100644 --- a/src/yuzu/bootmanager.cpp +++ b/src/yuzu/bootmanager.cpp @@ -47,7 +47,7 @@ EmuThread::EmuThread(Core::System& system_) : system{system_} {} EmuThread::~EmuThread() = default; void EmuThread::run() { - std::string name = "yuzu:EmuControlThread"; + std::string name = "EmuControlThread"; MicroProfileOnThreadCreate(name.c_str()); Common::SetCurrentThreadName(name.c_str()); diff --git a/src/yuzu/configuration/configure_graphics.cpp b/src/yuzu/configuration/configure_graphics.cpp index 87e5d0f48..bd69d04a6 100644 --- a/src/yuzu/configuration/configure_graphics.cpp +++ b/src/yuzu/configuration/configure_graphics.cpp @@ -57,9 +57,10 @@ ConfigureGraphics::ConfigureGraphics(const Core::System& system_, QWidget* paren UpdateBackgroundColorButton(new_bg_color); }); - ui->api->setEnabled(!UISettings::values.has_broken_vulkan); - ui->api_widget->setEnabled(!UISettings::values.has_broken_vulkan || - Settings::IsConfiguringGlobal()); + ui->api->setEnabled(!UISettings::values.has_broken_vulkan && ui->api->isEnabled()); + ui->api_widget->setEnabled( + (!UISettings::values.has_broken_vulkan || Settings::IsConfiguringGlobal()) && + ui->api_widget->isEnabled()); ui->bg_label->setVisible(Settings::IsConfiguringGlobal()); ui->bg_combobox->setVisible(!Settings::IsConfiguringGlobal()); } diff --git a/src/yuzu/main.cpp b/src/yuzu/main.cpp index c63ce3a30..f45a25410 100644 --- a/src/yuzu/main.cpp +++ b/src/yuzu/main.cpp @@ -3,6 +3,7 @@ #include <cinttypes> #include <clocale> +#include <cmath> #include <memory> #include <thread> #ifdef __APPLE__ @@ -261,6 +262,18 @@ static QString PrettyProductName() { return QSysInfo::prettyProductName(); } +#ifdef _WIN32 +static void OverrideWindowsFont() { + // Qt5 chooses these fonts on Windows and they have fairly ugly alphanumeric/cyrllic characters + // Asking to use "MS Shell Dlg 2" gives better other chars while leaving the Chinese Characters. + const QString startup_font = QApplication::font().family(); + const QStringList ugly_fonts = {QStringLiteral("SimSun"), QStringLiteral("PMingLiU")}; + if (ugly_fonts.contains(startup_font)) { + QApplication::setFont(QFont(QStringLiteral("MS Shell Dlg 2"), 9, QFont::Normal)); + } +} +#endif + bool GMainWindow::CheckDarkMode() { #ifdef __linux__ const QPalette test_palette(qApp->palette()); @@ -3451,9 +3464,10 @@ void GMainWindow::UpdateStatusBar() { } if (!Settings::values.use_speed_limit) { game_fps_label->setText( - tr("Game: %1 FPS (Unlocked)").arg(results.average_game_fps, 0, 'f', 0)); + tr("Game: %1 FPS (Unlocked)").arg(std::round(results.average_game_fps), 0, 'f', 0)); } else { - game_fps_label->setText(tr("Game: %1 FPS").arg(results.average_game_fps, 0, 'f', 0)); + game_fps_label->setText( + tr("Game: %1 FPS").arg(std::round(results.average_game_fps), 0, 'f', 0)); } emu_frametime_label->setText(tr("Frame: %1 ms").arg(results.frametime * 1000.0, 0, 'f', 2)); @@ -4135,6 +4149,10 @@ int main(int argc, char* argv[]) { QCoreApplication::setAttribute(Qt::AA_DontCheckOpenGLContextThreadAffinity); QApplication app(argc, argv); +#ifdef _WIN32 + OverrideWindowsFont(); +#endif + // Workaround for QTBUG-85409, for Suzhou numerals the number 1 is actually \u3021 // so we can see if we get \u3008 instead // TL;DR all other number formats are consecutive in unicode code points |