diff options
Diffstat (limited to 'src/video_core')
99 files changed, 3462 insertions, 3216 deletions
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index dd4c29ed3..9b931976a 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -2,10 +2,8 @@ add_subdirectory(host_shaders) add_library(video_core STATIC buffer_cache/buffer_base.h - buffer_cache/buffer_block.h + buffer_cache/buffer_cache.cpp buffer_cache/buffer_cache.h - buffer_cache/map_interval.cpp - buffer_cache/map_interval.h cdma_pusher.cpp cdma_pusher.h command_classes/codecs/codec.cpp @@ -152,8 +150,6 @@ add_library(video_core STATIC renderer_vulkan/vk_staging_buffer_pool.h renderer_vulkan/vk_state_tracker.cpp renderer_vulkan/vk_state_tracker.h - renderer_vulkan/vk_stream_buffer.cpp - renderer_vulkan/vk_stream_buffer.h renderer_vulkan/vk_swapchain.cpp renderer_vulkan/vk_swapchain.h renderer_vulkan/vk_texture_cache.cpp diff --git a/src/video_core/buffer_cache/buffer_base.h b/src/video_core/buffer_cache/buffer_base.h index ee8602ce9..0c00ae280 100644 --- a/src/video_core/buffer_cache/buffer_base.h +++ b/src/video_core/buffer_cache/buffer_base.h @@ -19,6 +19,7 @@ namespace VideoCommon { enum class BufferFlagBits { Picked = 1 << 0, + CachedWrites = 1 << 1, }; DECLARE_ENUM_FLAG_OPERATORS(BufferFlagBits) @@ -40,7 +41,7 @@ class BufferBase { static constexpr u64 BYTES_PER_WORD = PAGES_PER_WORD * BYTES_PER_PAGE; /// Vector tracking modified pages tightly packed with small vector optimization - union WrittenWords { + union WordsArray { /// Returns the pointer to the words state [[nodiscard]] const u64* Pointer(bool is_short) const noexcept { return is_short ? &stack : heap; @@ -55,49 +56,59 @@ class BufferBase { u64* heap; ///< Not-small buffers pointer to the storage }; - struct GpuCpuWords { - explicit GpuCpuWords() = default; - explicit GpuCpuWords(u64 size_bytes_) : size_bytes{size_bytes_} { + struct Words { + explicit Words() = default; + explicit Words(u64 size_bytes_) : size_bytes{size_bytes_} { if (IsShort()) { cpu.stack = ~u64{0}; gpu.stack = 0; + cached_cpu.stack = 0; + untracked.stack = ~u64{0}; } else { // Share allocation between CPU and GPU pages and set their default values const size_t num_words = NumWords(); - u64* const alloc = new u64[num_words * 2]; + u64* const alloc = new u64[num_words * 4]; cpu.heap = alloc; gpu.heap = alloc + num_words; + cached_cpu.heap = alloc + num_words * 2; + untracked.heap = alloc + num_words * 3; std::fill_n(cpu.heap, num_words, ~u64{0}); std::fill_n(gpu.heap, num_words, 0); + std::fill_n(cached_cpu.heap, num_words, 0); + std::fill_n(untracked.heap, num_words, ~u64{0}); } // Clean up tailing bits - const u64 last_local_page = - Common::DivCeil(size_bytes % BYTES_PER_WORD, BYTES_PER_PAGE); + const u64 last_word_size = size_bytes % BYTES_PER_WORD; + const u64 last_local_page = Common::DivCeil(last_word_size, BYTES_PER_PAGE); const u64 shift = (PAGES_PER_WORD - last_local_page) % PAGES_PER_WORD; - u64& last_word = cpu.Pointer(IsShort())[NumWords() - 1]; - last_word = (last_word << shift) >> shift; + const u64 last_word = (~u64{0} << shift) >> shift; + cpu.Pointer(IsShort())[NumWords() - 1] = last_word; + untracked.Pointer(IsShort())[NumWords() - 1] = last_word; } - ~GpuCpuWords() { + ~Words() { Release(); } - GpuCpuWords& operator=(GpuCpuWords&& rhs) noexcept { + Words& operator=(Words&& rhs) noexcept { Release(); size_bytes = rhs.size_bytes; cpu = rhs.cpu; gpu = rhs.gpu; + cached_cpu = rhs.cached_cpu; + untracked = rhs.untracked; rhs.cpu.heap = nullptr; return *this; } - GpuCpuWords(GpuCpuWords&& rhs) noexcept - : size_bytes{rhs.size_bytes}, cpu{rhs.cpu}, gpu{rhs.gpu} { + Words(Words&& rhs) noexcept + : size_bytes{rhs.size_bytes}, cpu{rhs.cpu}, gpu{rhs.gpu}, + cached_cpu{rhs.cached_cpu}, untracked{rhs.untracked} { rhs.cpu.heap = nullptr; } - GpuCpuWords& operator=(const GpuCpuWords&) = delete; - GpuCpuWords(const GpuCpuWords&) = delete; + Words& operator=(const Words&) = delete; + Words(const Words&) = delete; /// Returns true when the buffer fits in the small vector optimization [[nodiscard]] bool IsShort() const noexcept { @@ -118,8 +129,17 @@ class BufferBase { } u64 size_bytes = 0; - WrittenWords cpu; - WrittenWords gpu; + WordsArray cpu; + WordsArray gpu; + WordsArray cached_cpu; + WordsArray untracked; + }; + + enum class Type { + CPU, + GPU, + CachedCPU, + Untracked, }; public: @@ -132,68 +152,93 @@ public: BufferBase& operator=(const BufferBase&) = delete; BufferBase(const BufferBase&) = delete; + BufferBase& operator=(BufferBase&&) = default; + BufferBase(BufferBase&&) = default; + /// Returns the inclusive CPU modified range in a begin end pair [[nodiscard]] std::pair<u64, u64> ModifiedCpuRegion(VAddr query_cpu_addr, u64 query_size) const noexcept { const u64 offset = query_cpu_addr - cpu_addr; - return ModifiedRegion<false>(offset, query_size); + return ModifiedRegion<Type::CPU>(offset, query_size); } /// Returns the inclusive GPU modified range in a begin end pair [[nodiscard]] std::pair<u64, u64> ModifiedGpuRegion(VAddr query_cpu_addr, u64 query_size) const noexcept { const u64 offset = query_cpu_addr - cpu_addr; - return ModifiedRegion<true>(offset, query_size); + return ModifiedRegion<Type::GPU>(offset, query_size); } /// Returns true if a region has been modified from the CPU [[nodiscard]] bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) const noexcept { const u64 offset = query_cpu_addr - cpu_addr; - return IsRegionModified<false>(offset, query_size); + return IsRegionModified<Type::CPU>(offset, query_size); } /// Returns true if a region has been modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) const noexcept { const u64 offset = query_cpu_addr - cpu_addr; - return IsRegionModified<true>(offset, query_size); + return IsRegionModified<Type::GPU>(offset, query_size); } /// Mark region as CPU modified, notifying the rasterizer about this change void MarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 size) { - ChangeRegionState<true, true>(words.cpu, dirty_cpu_addr, size); + ChangeRegionState<Type::CPU, true>(dirty_cpu_addr, size); } /// Unmark region as CPU modified, notifying the rasterizer about this change void UnmarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 size) { - ChangeRegionState<false, true>(words.cpu, dirty_cpu_addr, size); + ChangeRegionState<Type::CPU, false>(dirty_cpu_addr, size); } /// Mark region as modified from the host GPU void MarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 size) noexcept { - ChangeRegionState<true, false>(words.gpu, dirty_cpu_addr, size); + ChangeRegionState<Type::GPU, true>(dirty_cpu_addr, size); } /// Unmark region as modified from the host GPU void UnmarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 size) noexcept { - ChangeRegionState<false, false>(words.gpu, dirty_cpu_addr, size); + ChangeRegionState<Type::GPU, false>(dirty_cpu_addr, size); + } + + /// Mark region as modified from the CPU + /// but don't mark it as modified until FlusHCachedWrites is called. + void CachedCpuWrite(VAddr dirty_cpu_addr, u64 size) { + flags |= BufferFlagBits::CachedWrites; + ChangeRegionState<Type::CachedCPU, true>(dirty_cpu_addr, size); + } + + /// Flushes cached CPU writes, and notify the rasterizer about the deltas + void FlushCachedWrites() noexcept { + flags &= ~BufferFlagBits::CachedWrites; + const u64 num_words = NumWords(); + const u64* const cached_words = Array<Type::CachedCPU>(); + u64* const untracked_words = Array<Type::Untracked>(); + u64* const cpu_words = Array<Type::CPU>(); + for (u64 word_index = 0; word_index < num_words; ++word_index) { + const u64 cached_bits = cached_words[word_index]; + NotifyRasterizer<false>(word_index, untracked_words[word_index], cached_bits); + untracked_words[word_index] |= cached_bits; + cpu_words[word_index] |= cached_bits; + } } /// Call 'func' for each CPU modified range and unmark those pages as CPU modified template <typename Func> void ForEachUploadRange(VAddr query_cpu_range, u64 size, Func&& func) { - ForEachModifiedRange<false, true>(query_cpu_range, size, func); + ForEachModifiedRange<Type::CPU>(query_cpu_range, size, func); } /// Call 'func' for each GPU modified range and unmark those pages as GPU modified template <typename Func> void ForEachDownloadRange(VAddr query_cpu_range, u64 size, Func&& func) { - ForEachModifiedRange<true, false>(query_cpu_range, size, func); + ForEachModifiedRange<Type::GPU>(query_cpu_range, size, func); } /// Call 'func' for each GPU modified range and unmark those pages as GPU modified template <typename Func> void ForEachDownloadRange(Func&& func) { - ForEachModifiedRange<true, false>(cpu_addr, SizeBytes(), func); + ForEachModifiedRange<Type::GPU>(cpu_addr, SizeBytes(), func); } /// Mark buffer as picked @@ -206,6 +251,16 @@ public: flags &= ~BufferFlagBits::Picked; } + /// Increases the likeliness of this being a stream buffer + void IncreaseStreamScore(int score) noexcept { + stream_score += score; + } + + /// Returns the likeliness of this being a stream buffer + [[nodiscard]] int StreamScore() const noexcept { + return stream_score; + } + /// Returns true when vaddr -> vaddr+size is fully contained in the buffer [[nodiscard]] bool IsInBounds(VAddr addr, u64 size) const noexcept { return addr >= cpu_addr && addr + size <= cpu_addr + SizeBytes(); @@ -216,6 +271,11 @@ public: return True(flags & BufferFlagBits::Picked); } + /// Returns true when the buffer has pending cached writes + [[nodiscard]] bool HasCachedWrites() const noexcept { + return True(flags & BufferFlagBits::CachedWrites); + } + /// Returns the base CPU address of the buffer [[nodiscard]] VAddr CpuAddr() const noexcept { return cpu_addr; @@ -233,26 +293,48 @@ public: } private: + template <Type type> + u64* Array() noexcept { + if constexpr (type == Type::CPU) { + return words.cpu.Pointer(IsShort()); + } else if constexpr (type == Type::GPU) { + return words.gpu.Pointer(IsShort()); + } else if constexpr (type == Type::CachedCPU) { + return words.cached_cpu.Pointer(IsShort()); + } else if constexpr (type == Type::Untracked) { + return words.untracked.Pointer(IsShort()); + } + } + + template <Type type> + const u64* Array() const noexcept { + if constexpr (type == Type::CPU) { + return words.cpu.Pointer(IsShort()); + } else if constexpr (type == Type::GPU) { + return words.gpu.Pointer(IsShort()); + } else if constexpr (type == Type::CachedCPU) { + return words.cached_cpu.Pointer(IsShort()); + } else if constexpr (type == Type::Untracked) { + return words.untracked.Pointer(IsShort()); + } + } + /** * Change the state of a range of pages * - * @param written_words Pages to be marked or unmarked as modified * @param dirty_addr Base address to mark or unmark as modified * @param size Size in bytes to mark or unmark as modified - * - * @tparam enable True when the bits will be set to one, false for zero - * @tparam notify_rasterizer True when the rasterizer has to be notified about the changes */ - template <bool enable, bool notify_rasterizer> - void ChangeRegionState(WrittenWords& written_words, u64 dirty_addr, - s64 size) noexcept(!notify_rasterizer) { + template <Type type, bool enable> + void ChangeRegionState(u64 dirty_addr, s64 size) noexcept(type == Type::GPU) { const s64 difference = dirty_addr - cpu_addr; const u64 offset = std::max<s64>(difference, 0); size += std::min<s64>(difference, 0); if (offset >= SizeBytes() || size < 0) { return; } - u64* const state_words = written_words.Pointer(IsShort()); + u64* const untracked_words = Array<Type::Untracked>(); + u64* const state_words = Array<type>(); const u64 offset_end = std::min(offset + size, SizeBytes()); const u64 begin_page_index = offset / BYTES_PER_PAGE; const u64 begin_word_index = begin_page_index / PAGES_PER_WORD; @@ -268,13 +350,19 @@ private: u64 bits = ~u64{0}; bits = (bits >> right_offset) << right_offset; bits = (bits << left_offset) >> left_offset; - if constexpr (notify_rasterizer) { - NotifyRasterizer<!enable>(word_index, state_words[word_index], bits); + if constexpr (type == Type::CPU || type == Type::CachedCPU) { + NotifyRasterizer<!enable>(word_index, untracked_words[word_index], bits); } if constexpr (enable) { state_words[word_index] |= bits; + if constexpr (type == Type::CPU || type == Type::CachedCPU) { + untracked_words[word_index] |= bits; + } } else { state_words[word_index] &= ~bits; + if constexpr (type == Type::CPU || type == Type::CachedCPU) { + untracked_words[word_index] &= ~bits; + } } page_index = 0; ++word_index; @@ -291,7 +379,7 @@ private: * @tparam add_to_rasterizer True when the rasterizer should start tracking the new pages */ template <bool add_to_rasterizer> - void NotifyRasterizer(u64 word_index, u64 current_bits, u64 new_bits) { + void NotifyRasterizer(u64 word_index, u64 current_bits, u64 new_bits) const { u64 changed_bits = (add_to_rasterizer ? current_bits : ~current_bits) & new_bits; VAddr addr = cpu_addr + word_index * BYTES_PER_WORD; while (changed_bits != 0) { @@ -315,21 +403,20 @@ private: * @param query_cpu_range Base CPU address to loop over * @param size Size in bytes of the CPU range to loop over * @param func Function to call for each turned off region - * - * @tparam gpu True for host GPU pages, false for CPU pages - * @tparam notify_rasterizer True when the rasterizer should be notified about state changes */ - template <bool gpu, bool notify_rasterizer, typename Func> + template <Type type, typename Func> void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) { + static_assert(type != Type::Untracked); + const s64 difference = query_cpu_range - cpu_addr; const u64 query_begin = std::max<s64>(difference, 0); size += std::min<s64>(difference, 0); if (query_begin >= SizeBytes() || size < 0) { return; } - const u64* const cpu_words = words.cpu.Pointer(IsShort()); + u64* const untracked_words = Array<Type::Untracked>(); + u64* const state_words = Array<type>(); const u64 query_end = query_begin + std::min(static_cast<u64>(size), SizeBytes()); - u64* const state_words = (gpu ? words.gpu : words.cpu).Pointer(IsShort()); u64* const words_begin = state_words + query_begin / BYTES_PER_WORD; u64* const words_end = state_words + Common::DivCeil(query_end, BYTES_PER_WORD); @@ -345,7 +432,8 @@ private: const u64 word_index_end = std::distance(state_words, last_modified_word); const unsigned local_page_begin = std::countr_zero(*first_modified_word); - const unsigned local_page_end = PAGES_PER_WORD - std::countl_zero(last_modified_word[-1]); + const unsigned local_page_end = + static_cast<unsigned>(PAGES_PER_WORD) - std::countl_zero(last_modified_word[-1]); const u64 word_page_begin = word_index_begin * PAGES_PER_WORD; const u64 word_page_end = (word_index_end - 1) * PAGES_PER_WORD; const u64 query_page_begin = query_begin / BYTES_PER_PAGE; @@ -371,11 +459,13 @@ private: const u64 current_word = state_words[word_index] & bits; state_words[word_index] &= ~bits; - // Exclude CPU modified pages when visiting GPU pages - const u64 word = current_word & ~(gpu ? cpu_words[word_index] : 0); - if constexpr (notify_rasterizer) { - NotifyRasterizer<true>(word_index, word, ~u64{0}); + if constexpr (type == Type::CPU) { + const u64 current_bits = untracked_words[word_index] & bits; + untracked_words[word_index] &= ~bits; + NotifyRasterizer<true>(word_index, current_bits, ~u64{0}); } + // Exclude CPU modified pages when visiting GPU pages + const u64 word = current_word & ~(type == Type::GPU ? untracked_words[word_index] : 0); u64 page = page_begin; page_begin = 0; @@ -416,17 +506,20 @@ private: * @param offset Offset in bytes from the start of the buffer * @param size Size in bytes of the region to query for modifications */ - template <bool gpu> + template <Type type> [[nodiscard]] bool IsRegionModified(u64 offset, u64 size) const noexcept { - const u64* const cpu_words = words.cpu.Pointer(IsShort()); - const u64* const state_words = (gpu ? words.gpu : words.cpu).Pointer(IsShort()); + static_assert(type != Type::Untracked); + + const u64* const untracked_words = Array<Type::Untracked>(); + const u64* const state_words = Array<type>(); const u64 num_query_words = size / BYTES_PER_WORD + 1; const u64 word_begin = offset / BYTES_PER_WORD; const u64 word_end = std::min(word_begin + num_query_words, NumWords()); const u64 page_limit = Common::DivCeil(offset + size, BYTES_PER_PAGE); u64 page_index = (offset / BYTES_PER_PAGE) % PAGES_PER_WORD; for (u64 word_index = word_begin; word_index < word_end; ++word_index, page_index = 0) { - const u64 word = state_words[word_index] & ~(gpu ? cpu_words[word_index] : 0); + const u64 off_word = type == Type::GPU ? untracked_words[word_index] : 0; + const u64 word = state_words[word_index] & ~off_word; if (word == 0) { continue; } @@ -445,13 +538,13 @@ private: * * @param offset Offset in bytes from the start of the buffer * @param size Size in bytes of the region to query for modifications - * - * @tparam gpu True to query GPU modified pages, false for CPU pages */ - template <bool gpu> + template <Type type> [[nodiscard]] std::pair<u64, u64> ModifiedRegion(u64 offset, u64 size) const noexcept { - const u64* const cpu_words = words.cpu.Pointer(IsShort()); - const u64* const state_words = (gpu ? words.gpu : words.cpu).Pointer(IsShort()); + static_assert(type != Type::Untracked); + + const u64* const untracked_words = Array<Type::Untracked>(); + const u64* const state_words = Array<type>(); const u64 num_query_words = size / BYTES_PER_WORD + 1; const u64 word_begin = offset / BYTES_PER_WORD; const u64 word_end = std::min(word_begin + num_query_words, NumWords()); @@ -460,7 +553,8 @@ private: u64 begin = std::numeric_limits<u64>::max(); u64 end = 0; for (u64 word_index = word_begin; word_index < word_end; ++word_index) { - const u64 word = state_words[word_index] & ~(gpu ? cpu_words[word_index] : 0); + const u64 off_word = type == Type::GPU ? untracked_words[word_index] : 0; + const u64 word = state_words[word_index] & ~off_word; if (word == 0) { continue; } @@ -488,8 +582,9 @@ private: RasterizerInterface* rasterizer = nullptr; VAddr cpu_addr = 0; - GpuCpuWords words; + Words words; BufferFlagBits flags{}; + int stream_score = 0; }; } // namespace VideoCommon diff --git a/src/video_core/buffer_cache/buffer_block.h b/src/video_core/buffer_cache/buffer_block.h deleted file mode 100644 index e9306194a..000000000 --- a/src/video_core/buffer_cache/buffer_block.h +++ /dev/null @@ -1,62 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include "common/common_types.h" - -namespace VideoCommon { - -class BufferBlock { -public: - [[nodiscard]] bool Overlaps(VAddr start, VAddr end) const { - return (cpu_addr < end) && (cpu_addr_end > start); - } - - [[nodiscard]] bool IsInside(VAddr other_start, VAddr other_end) const { - return cpu_addr <= other_start && other_end <= cpu_addr_end; - } - - [[nodiscard]] std::size_t Offset(VAddr in_addr) const { - return static_cast<std::size_t>(in_addr - cpu_addr); - } - - [[nodiscard]] VAddr CpuAddr() const { - return cpu_addr; - } - - [[nodiscard]] VAddr CpuAddrEnd() const { - return cpu_addr_end; - } - - void SetCpuAddr(VAddr new_addr) { - cpu_addr = new_addr; - cpu_addr_end = new_addr + size; - } - - [[nodiscard]] std::size_t Size() const { - return size; - } - - [[nodiscard]] u64 Epoch() const { - return epoch; - } - - void SetEpoch(u64 new_epoch) { - epoch = new_epoch; - } - -protected: - explicit BufferBlock(VAddr cpu_addr_, std::size_t size_) : size{size_} { - SetCpuAddr(cpu_addr_); - } - -private: - VAddr cpu_addr{}; - VAddr cpu_addr_end{}; - std::size_t size{}; - u64 epoch{}; -}; - -} // namespace VideoCommon diff --git a/src/video_core/buffer_cache/buffer_cache.cpp b/src/video_core/buffer_cache/buffer_cache.cpp new file mode 100644 index 000000000..ab32294c8 --- /dev/null +++ b/src/video_core/buffer_cache/buffer_cache.cpp @@ -0,0 +1,13 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/microprofile.h" + +namespace VideoCommon { + +MICROPROFILE_DEFINE(GPU_PrepareBuffers, "GPU", "Prepare buffers", MP_RGB(224, 128, 128)); +MICROPROFILE_DEFINE(GPU_BindUploadBuffers, "GPU", "Bind and upload buffers", MP_RGB(224, 128, 128)); +MICROPROFILE_DEFINE(GPU_DownloadMemory, "GPU", "Download buffers", MP_RGB(224, 128, 128)); + +} // namespace VideoCommon diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 83b9ee871..2a6844ab1 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -4,591 +4,1289 @@ #pragma once -#include <list> +#include <algorithm> +#include <array> +#include <deque> #include <memory> #include <mutex> +#include <span> #include <unordered_map> -#include <unordered_set> -#include <utility> #include <vector> #include <boost/container/small_vector.hpp> -#include <boost/icl/interval_set.hpp> -#include <boost/intrusive/set.hpp> -#include "common/alignment.h" -#include "common/assert.h" #include "common/common_types.h" -#include "common/logging/log.h" -#include "core/core.h" +#include "common/div_ceil.h" +#include "common/microprofile.h" +#include "common/scope_exit.h" #include "core/memory.h" #include "core/settings.h" -#include "video_core/buffer_cache/buffer_block.h" -#include "video_core/buffer_cache/map_interval.h" +#include "video_core/buffer_cache/buffer_base.h" +#include "video_core/delayed_destruction_ring.h" +#include "video_core/dirty_flags.h" +#include "video_core/engines/kepler_compute.h" +#include "video_core/engines/maxwell_3d.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" +#include "video_core/texture_cache/slot_vector.h" +#include "video_core/texture_cache/types.h" namespace VideoCommon { -template <typename Buffer, typename BufferType, typename StreamBuffer> +MICROPROFILE_DECLARE(GPU_PrepareBuffers); +MICROPROFILE_DECLARE(GPU_BindUploadBuffers); +MICROPROFILE_DECLARE(GPU_DownloadMemory); + +using BufferId = SlotId; + +constexpr u32 NUM_VERTEX_BUFFERS = 32; +constexpr u32 NUM_TRANSFORM_FEEDBACK_BUFFERS = 4; +constexpr u32 NUM_GRAPHICS_UNIFORM_BUFFERS = 18; +constexpr u32 NUM_COMPUTE_UNIFORM_BUFFERS = 8; +constexpr u32 NUM_STORAGE_BUFFERS = 16; +constexpr u32 NUM_STAGES = 5; + +template <typename P> class BufferCache { - using IntervalSet = boost::icl::interval_set<VAddr>; - using IntervalType = typename IntervalSet::interval_type; - using VectorMapInterval = boost::container::small_vector<MapInterval*, 1>; + // Page size for caching purposes. + // This is unrelated to the CPU page size and it can be changed as it seems optimal. + static constexpr u32 PAGE_BITS = 16; + static constexpr u64 PAGE_SIZE = u64{1} << PAGE_BITS; + + static constexpr bool IS_OPENGL = P::IS_OPENGL; + static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = + P::HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS; + static constexpr bool HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT = + P::HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT; + static constexpr bool NEEDS_BIND_UNIFORM_INDEX = P::NEEDS_BIND_UNIFORM_INDEX; + static constexpr bool NEEDS_BIND_STORAGE_INDEX = P::NEEDS_BIND_STORAGE_INDEX; + static constexpr bool USE_MEMORY_MAPS = P::USE_MEMORY_MAPS; + + static constexpr BufferId NULL_BUFFER_ID{0}; + + using Maxwell = Tegra::Engines::Maxwell3D::Regs; + + using Runtime = typename P::Runtime; + using Buffer = typename P::Buffer; + + struct Empty {}; + + struct OverlapResult { + std::vector<BufferId> ids; + VAddr begin; + VAddr end; + bool has_stream_leap = false; + }; - static constexpr u64 WRITE_PAGE_BIT = 11; - static constexpr u64 BLOCK_PAGE_BITS = 21; - static constexpr u64 BLOCK_PAGE_SIZE = 1ULL << BLOCK_PAGE_BITS; + struct Binding { + VAddr cpu_addr{}; + u32 size{}; + BufferId buffer_id; + }; -public: - struct BufferInfo { - BufferType handle; - u64 offset; - u64 address; + static constexpr Binding NULL_BINDING{ + .cpu_addr = 0, + .size = 0, + .buffer_id = NULL_BUFFER_ID, }; - BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4, - bool is_written = false, bool use_fast_cbuf = false) { - std::lock_guard lock{mutex}; +public: + static constexpr u32 SKIP_CACHE_SIZE = 4096; - const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr); - if (!cpu_addr) { - return GetEmptyBuffer(size); - } + 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_); - // Cache management is a big overhead, so only cache entries with a given size. - // TODO: Figure out which size is the best for given games. - constexpr std::size_t max_stream_size = 0x800; - if (use_fast_cbuf || size < max_stream_size) { - if (!is_written && !IsRegionWritten(*cpu_addr, *cpu_addr + size - 1)) { - const bool is_granular = gpu_memory.IsGranularRange(gpu_addr, size); - if (use_fast_cbuf) { - u8* dest; - if (is_granular) { - dest = gpu_memory.GetPointer(gpu_addr); - } else { - staging_buffer.resize(size); - dest = staging_buffer.data(); - gpu_memory.ReadBlockUnsafe(gpu_addr, dest, size); - } - return ConstBufferUpload(dest, size); - } - if (is_granular) { - u8* const host_ptr = gpu_memory.GetPointer(gpu_addr); - return StreamBufferUpload(size, alignment, [host_ptr, size](u8* dest) { - std::memcpy(dest, host_ptr, size); - }); - } else { - return StreamBufferUpload(size, alignment, [this, gpu_addr, size](u8* dest) { - gpu_memory.ReadBlockUnsafe(gpu_addr, dest, size); - }); - } - } - } + void TickFrame(); - Buffer* const block = GetBlock(*cpu_addr, size); - MapInterval* const map = MapAddress(block, gpu_addr, *cpu_addr, size); - if (!map) { - return GetEmptyBuffer(size); - } - if (is_written) { - map->MarkAsModified(true, GetModifiedTicks()); - if (Settings::IsGPULevelHigh() && - Settings::values.use_asynchronous_gpu_emulation.GetValue()) { - MarkForAsyncFlush(map); - } - if (!map->is_written) { - map->is_written = true; - MarkRegionAsWritten(map->start, map->end - 1); - } - } + void WriteMemory(VAddr cpu_addr, u64 size); - return BufferInfo{block->Handle(), block->Offset(*cpu_addr), block->Address()}; - } + void CachedWriteMemory(VAddr cpu_addr, u64 size); - /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset. - BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size, - std::size_t alignment = 4) { - std::lock_guard lock{mutex}; - return StreamBufferUpload(size, alignment, [raw_pointer, size](u8* dest) { - std::memcpy(dest, raw_pointer, size); - }); - } + void DownloadMemory(VAddr cpu_addr, u64 size); - /// Prepares the buffer cache for data uploading - /// @param max_size Maximum number of bytes that will be uploaded - /// @return True when a stream buffer invalidation was required, false otherwise - void Map(std::size_t max_size) { - std::lock_guard lock{mutex}; + void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size); - std::tie(buffer_ptr, buffer_offset_base) = stream_buffer.Map(max_size, 4); - buffer_offset = buffer_offset_base; - } + void UpdateGraphicsBuffers(bool is_indexed); - /// Finishes the upload stream - void Unmap() { - std::lock_guard lock{mutex}; - stream_buffer.Unmap(buffer_offset - buffer_offset_base); - } + void UpdateComputeBuffers(); - /// Function called at the end of each frame, inteded for deferred operations - void TickFrame() { - ++epoch; + void BindHostGeometryBuffers(bool is_indexed); - while (!pending_destruction.empty()) { - // Delay at least 4 frames before destruction. - // This is due to triple buffering happening on some drivers. - static constexpr u64 epochs_to_destroy = 5; - if (pending_destruction.front()->Epoch() + epochs_to_destroy > epoch) { - break; - } - pending_destruction.pop(); - } - } + void BindHostStageBuffers(size_t stage); - /// Write any cached resources overlapping the specified region back to memory - void FlushRegion(VAddr addr, std::size_t size) { - std::lock_guard lock{mutex}; + void BindHostComputeBuffers(); - VectorMapInterval objects = GetMapsInRange(addr, size); - std::sort(objects.begin(), objects.end(), - [](MapInterval* lhs, MapInterval* rhs) { return lhs->ticks < rhs->ticks; }); - for (MapInterval* object : objects) { - if (object->is_modified && object->is_registered) { - mutex.unlock(); - FlushMap(object); - mutex.lock(); - } - } - } + void SetEnabledUniformBuffers(size_t stage, u32 enabled); - bool MustFlushRegion(VAddr addr, std::size_t size) { - std::lock_guard lock{mutex}; + void SetEnabledComputeUniformBuffers(u32 enabled); - const VectorMapInterval objects = GetMapsInRange(addr, size); - return std::any_of(objects.cbegin(), objects.cend(), [](const MapInterval* map) { - return map->is_modified && map->is_registered; - }); - } + void UnbindGraphicsStorageBuffers(size_t stage); - /// Mark the specified region as being invalidated - void InvalidateRegion(VAddr addr, u64 size) { - std::lock_guard lock{mutex}; + void BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset, + bool is_written); - for (auto& object : GetMapsInRange(addr, size)) { - if (object->is_registered) { - Unregister(object); - } - } - } + void UnbindComputeStorageBuffers(); - void OnCPUWrite(VAddr addr, std::size_t size) { - std::lock_guard lock{mutex}; + void BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset, + bool is_written); - for (MapInterval* object : GetMapsInRange(addr, size)) { - if (object->is_memory_marked && object->is_registered) { - UnmarkMemory(object); - object->is_sync_pending = true; - marked_for_unregister.emplace_back(object); - } - } - } + void FlushCachedWrites(); - void SyncGuestHost() { - std::lock_guard lock{mutex}; + /// Return true when there are uncommitted buffers to be downloaded + [[nodiscard]] bool HasUncommittedFlushes() const noexcept; - for (auto& object : marked_for_unregister) { - if (object->is_registered) { - object->is_sync_pending = false; - Unregister(object); - } + /// Return true when the caller should wait for async downloads + [[nodiscard]] bool ShouldWaitAsyncFlushes() const noexcept; + + /// Commit asynchronous downloads + void CommitAsyncFlushes(); + + /// Pop asynchronous downloads + void PopAsyncFlushes(); + + /// Return true when a CPU region is modified from the GPU + [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); + + std::mutex mutex; + +private: + template <typename Func> + static void ForEachEnabledBit(u32 enabled_mask, Func&& func) { + for (u32 index = 0; enabled_mask != 0; ++index, enabled_mask >>= 1) { + const int disabled_bits = std::countr_zero(enabled_mask); + index += disabled_bits; + enabled_mask >>= disabled_bits; + func(index); } - marked_for_unregister.clear(); } - void CommitAsyncFlushes() { - if (uncommitted_flushes) { - auto commit_list = std::make_shared<std::list<MapInterval*>>(); - for (MapInterval* map : *uncommitted_flushes) { - if (map->is_registered && map->is_modified) { - // TODO(Blinkhawk): Implement backend asynchronous flushing - // AsyncFlushMap(map) - commit_list->push_back(map); - } - } - if (!commit_list->empty()) { - committed_flushes.push_back(commit_list); - } else { - committed_flushes.emplace_back(); + template <typename Func> + void ForEachBufferInRange(VAddr cpu_addr, u64 size, Func&& func) { + const u64 page_end = Common::DivCeil(cpu_addr + size, PAGE_SIZE); + for (u64 page = cpu_addr >> PAGE_BITS; page < page_end;) { + const BufferId buffer_id = page_table[page]; + if (!buffer_id) { + ++page; + continue; } - } else { - committed_flushes.emplace_back(); + Buffer& buffer = slot_buffers[buffer_id]; + func(buffer_id, buffer); + + const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes(); + page = Common::DivCeil(end_addr, PAGE_SIZE); } - uncommitted_flushes.reset(); } - bool ShouldWaitAsyncFlushes() const { - return !committed_flushes.empty() && committed_flushes.front() != nullptr; + static bool IsRangeGranular(VAddr cpu_addr, size_t size) { + return (cpu_addr & ~Core::Memory::PAGE_MASK) == + ((cpu_addr + size) & ~Core::Memory::PAGE_MASK); } - bool HasUncommittedFlushes() const { - return uncommitted_flushes != nullptr; - } + void BindHostIndexBuffer(); - void PopAsyncFlushes() { - if (committed_flushes.empty()) { - return; - } - auto& flush_list = committed_flushes.front(); - if (!flush_list) { - committed_flushes.pop_front(); - return; - } - for (MapInterval* map : *flush_list) { - if (map->is_registered) { - // TODO(Blinkhawk): Replace this for reading the asynchronous flush - FlushMap(map); - } - } - committed_flushes.pop_front(); - } + void BindHostVertexBuffers(); - virtual BufferInfo GetEmptyBuffer(std::size_t size) = 0; + void BindHostGraphicsUniformBuffers(size_t stage); -protected: - explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - StreamBuffer& stream_buffer_) - : rasterizer{rasterizer_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_}, - stream_buffer{stream_buffer_} {} + void BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 binding_index, bool needs_bind); - ~BufferCache() = default; + void BindHostGraphicsStorageBuffers(size_t stage); - virtual std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) = 0; + void BindHostTransformFeedbackBuffers(); - virtual BufferInfo ConstBufferUpload(const void* raw_pointer, std::size_t size) { - return {}; - } + void BindHostComputeUniformBuffers(); - /// Register an object into the cache - MapInterval* Register(MapInterval new_map, bool inherit_written = false) { - const VAddr cpu_addr = new_map.start; - if (!cpu_addr) { - LOG_CRITICAL(HW_GPU, "Failed to register buffer with unmapped gpu_address 0x{:016x}", - new_map.gpu_addr); - return nullptr; - } - const std::size_t size = new_map.end - new_map.start; - new_map.is_registered = true; - rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1); - new_map.is_memory_marked = true; - if (inherit_written) { - MarkRegionAsWritten(new_map.start, new_map.end - 1); - new_map.is_written = true; - } - MapInterval* const storage = mapped_addresses_allocator.Allocate(); - *storage = new_map; - mapped_addresses.insert(*storage); - return storage; - } + void BindHostComputeStorageBuffers(); - void UnmarkMemory(MapInterval* map) { - if (!map->is_memory_marked) { - return; - } - const std::size_t size = map->end - map->start; - rasterizer.UpdatePagesCachedCount(map->start, size, -1); - map->is_memory_marked = false; - } - - /// Unregisters an object from the cache - void Unregister(MapInterval* map) { - UnmarkMemory(map); - map->is_registered = false; - if (map->is_sync_pending) { - map->is_sync_pending = false; - marked_for_unregister.remove(map); + void DoUpdateGraphicsBuffers(bool is_indexed); + + void DoUpdateComputeBuffers(); + + void UpdateIndexBuffer(); + + void UpdateVertexBuffers(); + + void UpdateVertexBuffer(u32 index); + + void UpdateUniformBuffers(size_t stage); + + void UpdateStorageBuffers(size_t stage); + + void UpdateTransformFeedbackBuffers(); + + void UpdateTransformFeedbackBuffer(u32 index); + + void UpdateComputeUniformBuffers(); + + void UpdateComputeStorageBuffers(); + + void MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size); + + [[nodiscard]] BufferId FindBuffer(VAddr cpu_addr, u32 size); + + [[nodiscard]] OverlapResult ResolveOverlaps(VAddr cpu_addr, u32 wanted_size); + + void JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, bool accumulate_stream_score); + + [[nodiscard]] BufferId CreateBuffer(VAddr cpu_addr, u32 wanted_size); + + void Register(BufferId buffer_id); + + void Unregister(BufferId buffer_id); + + template <bool insert> + void ChangeRegister(BufferId buffer_id); + + void SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size); + + void SynchronizeBufferImpl(Buffer& buffer, VAddr cpu_addr, u32 size); + + void UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy, + std::span<BufferCopy> copies); + + void ImmediateUploadMemory(Buffer& buffer, u64 largest_copy, + std::span<const BufferCopy> copies); + + void MappedUploadMemory(Buffer& buffer, u64 total_size_bytes, std::span<BufferCopy> copies); + + void DeleteBuffer(BufferId buffer_id); + + void ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id); + + void NotifyBufferDeletion(); + + [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr) const; + + [[nodiscard]] std::span<const u8> ImmediateBufferWithData(VAddr cpu_addr, size_t size); + + [[nodiscard]] std::span<u8> ImmediateBuffer(size_t wanted_capacity); + + [[nodiscard]] bool HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept; + + VideoCore::RasterizerInterface& rasterizer; + Tegra::Engines::Maxwell3D& maxwell3d; + Tegra::Engines::KeplerCompute& kepler_compute; + Tegra::MemoryManager& gpu_memory; + Core::Memory::Memory& cpu_memory; + Runtime& runtime; + + SlotVector<Buffer> slot_buffers; + DelayedDestructionRing<Buffer, 8> delayed_destruction_ring; + + u32 last_index_count = 0; + + Binding index_buffer; + std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers; + std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers; + std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers; + std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers; + + std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers; + std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers; + + std::array<u32, NUM_STAGES> enabled_uniform_buffers{}; + u32 enabled_compute_uniform_buffers = 0; + + std::array<u32, NUM_STAGES> enabled_storage_buffers{}; + std::array<u32, NUM_STAGES> written_storage_buffers{}; + u32 enabled_compute_storage_buffers = 0; + u32 written_compute_storage_buffers = 0; + + std::array<u32, NUM_STAGES> fast_bound_uniform_buffers{}; + + bool has_deleted_buffers = false; + + std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS, std::array<u32, NUM_STAGES>, Empty> + dirty_uniform_buffers{}; + + std::vector<BufferId> cached_write_buffer_ids; + + // TODO: This data structure is not optimal and it should be reworked + std::vector<BufferId> uncommitted_downloads; + std::deque<std::vector<BufferId>> committed_downloads; + + size_t immediate_buffer_capacity = 0; + std::unique_ptr<u8[]> immediate_buffer_alloc; + + std::array<BufferId, ((1ULL << 39) >> PAGE_BITS)> page_table; +}; + +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_) + : rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, + gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_}, runtime{runtime_} { + // Ensure the first slot is used for the null buffer + void(slot_buffers.insert(runtime, NullBufferParams{})); +} + +template <class P> +void BufferCache<P>::TickFrame() { + delayed_destruction_ring.Tick(); +} + +template <class P> +void BufferCache<P>::WriteMemory(VAddr cpu_addr, u64 size) { + ForEachBufferInRange(cpu_addr, size, [&](BufferId, Buffer& buffer) { + buffer.MarkRegionAsCpuModified(cpu_addr, size); + }); +} + +template <class P> +void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) { + ForEachBufferInRange(cpu_addr, size, [&](BufferId buffer_id, Buffer& buffer) { + if (!buffer.HasCachedWrites()) { + cached_write_buffer_ids.push_back(buffer_id); } - if (map->is_written) { - UnmarkRegionAsWritten(map->start, map->end - 1); + buffer.CachedCpuWrite(cpu_addr, size); + }); +} + +template <class P> +void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) { + ForEachBufferInRange(cpu_addr, size, [&](BufferId, Buffer& buffer) { + boost::container::small_vector<BufferCopy, 1> copies; + u64 total_size_bytes = 0; + u64 largest_copy = 0; + buffer.ForEachDownloadRange(cpu_addr, size, [&](u64 range_offset, u64 range_size) { + copies.push_back(BufferCopy{ + .src_offset = range_offset, + .dst_offset = total_size_bytes, + .size = range_size, + }); + total_size_bytes += range_size; + largest_copy = std::max(largest_copy, range_size); + }); + if (total_size_bytes == 0) { + return; } - const auto it = mapped_addresses.find(*map); - ASSERT(it != mapped_addresses.end()); - mapped_addresses.erase(it); - mapped_addresses_allocator.Release(map); - } - -private: - MapInterval* MapAddress(Buffer* block, GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size) { - const VectorMapInterval overlaps = GetMapsInRange(cpu_addr, size); - if (overlaps.empty()) { - const VAddr cpu_addr_end = cpu_addr + size; - if (gpu_memory.IsGranularRange(gpu_addr, size)) { - u8* const host_ptr = gpu_memory.GetPointer(gpu_addr); - block->Upload(block->Offset(cpu_addr), size, host_ptr); - } else { - staging_buffer.resize(size); - gpu_memory.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size); - block->Upload(block->Offset(cpu_addr), size, staging_buffer.data()); + MICROPROFILE_SCOPE(GPU_DownloadMemory); + + if constexpr (USE_MEMORY_MAPS) { + auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes); + const u8* const mapped_memory = download_staging.mapped_span.data(); + const std::span<BufferCopy> copies_span(copies.data(), copies.data() + copies.size()); + for (BufferCopy& copy : copies) { + // Modify copies to have the staging offset in mind + copy.dst_offset += download_staging.offset; } - return Register(MapInterval(cpu_addr, cpu_addr_end, gpu_addr)); - } - - const VAddr cpu_addr_end = cpu_addr + size; - if (overlaps.size() == 1) { - MapInterval* const current_map = overlaps[0]; - if (current_map->IsInside(cpu_addr, cpu_addr_end)) { - return current_map; + runtime.CopyBuffer(download_staging.buffer, buffer, copies_span); + runtime.Finish(); + for (const BufferCopy& copy : copies) { + const VAddr copy_cpu_addr = buffer.CpuAddr() + copy.src_offset; + // Undo the modified offset + const u64 dst_offset = copy.dst_offset - download_staging.offset; + const u8* copy_mapped_memory = mapped_memory + dst_offset; + cpu_memory.WriteBlockUnsafe(copy_cpu_addr, copy_mapped_memory, copy.size); + } + } else { + const std::span<u8> immediate_buffer = ImmediateBuffer(largest_copy); + for (const BufferCopy& copy : copies) { + buffer.ImmediateDownload(copy.src_offset, immediate_buffer.subspan(0, copy.size)); + const VAddr copy_cpu_addr = buffer.CpuAddr() + copy.src_offset; + cpu_memory.WriteBlockUnsafe(copy_cpu_addr, immediate_buffer.data(), copy.size); } } - VAddr new_start = cpu_addr; - VAddr new_end = cpu_addr_end; - bool write_inheritance = false; - bool modified_inheritance = false; - // Calculate new buffer parameters - for (MapInterval* overlap : overlaps) { - new_start = std::min(overlap->start, new_start); - new_end = std::max(overlap->end, new_end); - write_inheritance |= overlap->is_written; - modified_inheritance |= overlap->is_modified; + }); +} + +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); + if (!cpu_addr) { + uniform_buffers[stage][index] = NULL_BINDING; + return; + } + const Binding binding{ + .cpu_addr = *cpu_addr, + .size = size, + .buffer_id = BufferId{}, + }; + uniform_buffers[stage][index] = binding; +} + +template <class P> +void BufferCache<P>::UpdateGraphicsBuffers(bool is_indexed) { + MICROPROFILE_SCOPE(GPU_PrepareBuffers); + do { + has_deleted_buffers = false; + DoUpdateGraphicsBuffers(is_indexed); + } while (has_deleted_buffers); +} + +template <class P> +void BufferCache<P>::UpdateComputeBuffers() { + MICROPROFILE_SCOPE(GPU_PrepareBuffers); + do { + has_deleted_buffers = false; + DoUpdateComputeBuffers(); + } while (has_deleted_buffers); +} + +template <class P> +void BufferCache<P>::BindHostGeometryBuffers(bool is_indexed) { + MICROPROFILE_SCOPE(GPU_BindUploadBuffers); + if (is_indexed) { + BindHostIndexBuffer(); + } else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { + const auto& regs = maxwell3d.regs; + if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) { + runtime.BindQuadArrayIndexBuffer(regs.vertex_buffer.first, regs.vertex_buffer.count); } - GPUVAddr new_gpu_addr = gpu_addr + new_start - cpu_addr; - for (auto& overlap : overlaps) { - Unregister(overlap); + } + BindHostVertexBuffers(); + BindHostTransformFeedbackBuffers(); +} + +template <class P> +void BufferCache<P>::BindHostStageBuffers(size_t stage) { + MICROPROFILE_SCOPE(GPU_BindUploadBuffers); + BindHostGraphicsUniformBuffers(stage); + BindHostGraphicsStorageBuffers(stage); +} + +template <class P> +void BufferCache<P>::BindHostComputeBuffers() { + MICROPROFILE_SCOPE(GPU_BindUploadBuffers); + BindHostComputeUniformBuffers(); + BindHostComputeStorageBuffers(); +} + +template <class P> +void BufferCache<P>::SetEnabledUniformBuffers(size_t stage, u32 enabled) { + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { + if (enabled_uniform_buffers[stage] != enabled) { + dirty_uniform_buffers[stage] = ~u32{0}; } - UpdateBlock(block, new_start, new_end, overlaps); - - const MapInterval new_map{new_start, new_end, new_gpu_addr}; - MapInterval* const map = Register(new_map, write_inheritance); - if (!map) { - return nullptr; + } + enabled_uniform_buffers[stage] = enabled; +} + +template <class P> +void BufferCache<P>::SetEnabledComputeUniformBuffers(u32 enabled) { + enabled_compute_uniform_buffers = enabled; +} + +template <class P> +void BufferCache<P>::UnbindGraphicsStorageBuffers(size_t stage) { + enabled_storage_buffers[stage] = 0; + written_storage_buffers[stage] = 0; +} + +template <class P> +void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, + u32 cbuf_offset, bool is_written) { + 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 GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset; + storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr); +} + +template <class P> +void BufferCache<P>::UnbindComputeStorageBuffers() { + enabled_compute_storage_buffers = 0; + written_compute_storage_buffers = 0; +} + +template <class P> +void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset, + bool is_written) { + 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; + 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); +} + +template <class P> +void BufferCache<P>::FlushCachedWrites() { + for (const BufferId buffer_id : cached_write_buffer_ids) { + slot_buffers[buffer_id].FlushCachedWrites(); + } + cached_write_buffer_ids.clear(); +} + +template <class P> +bool BufferCache<P>::HasUncommittedFlushes() const noexcept { + return !uncommitted_downloads.empty(); +} + +template <class P> +bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept { + return !committed_downloads.empty() && !committed_downloads.front().empty(); +} + +template <class P> +void BufferCache<P>::CommitAsyncFlushes() { + // This is intentionally passing the value by copy + committed_downloads.push_front(uncommitted_downloads); + uncommitted_downloads.clear(); +} + +template <class P> +void BufferCache<P>::PopAsyncFlushes() { + if (committed_downloads.empty()) { + return; + } + auto scope_exit_pop_download = detail::ScopeExit([this] { committed_downloads.pop_back(); }); + const std::span<const BufferId> download_ids = committed_downloads.back(); + if (download_ids.empty()) { + return; + } + MICROPROFILE_SCOPE(GPU_DownloadMemory); + + boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads; + u64 total_size_bytes = 0; + u64 largest_copy = 0; + for (const BufferId buffer_id : download_ids) { + slot_buffers[buffer_id].ForEachDownloadRange([&](u64 range_offset, u64 range_size) { + downloads.push_back({ + BufferCopy{ + .src_offset = range_offset, + .dst_offset = total_size_bytes, + .size = range_size, + }, + buffer_id, + }); + total_size_bytes += range_size; + largest_copy = std::max(largest_copy, range_size); + }); + } + if (downloads.empty()) { + return; + } + if constexpr (USE_MEMORY_MAPS) { + auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes); + for (auto& [copy, buffer_id] : downloads) { + // Have in mind the staging buffer offset for the copy + copy.dst_offset += download_staging.offset; + const std::array copies{copy}; + runtime.CopyBuffer(download_staging.buffer, slot_buffers[buffer_id], copies); } - if (modified_inheritance) { - map->MarkAsModified(true, GetModifiedTicks()); - if (Settings::IsGPULevelHigh() && - Settings::values.use_asynchronous_gpu_emulation.GetValue()) { - MarkForAsyncFlush(map); - } + runtime.Finish(); + for (const auto [copy, buffer_id] : downloads) { + const Buffer& buffer = slot_buffers[buffer_id]; + const VAddr cpu_addr = buffer.CpuAddr() + copy.src_offset; + // Undo the modified offset + const u64 dst_offset = copy.dst_offset - download_staging.offset; + const u8* read_mapped_memory = download_staging.mapped_span.data() + dst_offset; + cpu_memory.WriteBlockUnsafe(cpu_addr, read_mapped_memory, copy.size); + } + } else { + const std::span<u8> immediate_buffer = ImmediateBuffer(largest_copy); + for (const auto [copy, buffer_id] : downloads) { + Buffer& buffer = slot_buffers[buffer_id]; + buffer.ImmediateDownload(copy.src_offset, immediate_buffer.subspan(0, copy.size)); + const VAddr cpu_addr = buffer.CpuAddr() + copy.src_offset; + cpu_memory.WriteBlockUnsafe(cpu_addr, immediate_buffer.data(), copy.size); } - return map; } - - void UpdateBlock(Buffer* block, VAddr start, VAddr end, const VectorMapInterval& overlaps) { - const IntervalType base_interval{start, end}; - IntervalSet interval_set{}; - interval_set.add(base_interval); - for (auto& overlap : overlaps) { - const IntervalType subtract{overlap->start, overlap->end}; - interval_set.subtract(subtract); +} + +template <class P> +bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) { + const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); + for (u64 page = addr >> PAGE_BITS; page < page_end;) { + const BufferId image_id = page_table[page]; + if (!image_id) { + ++page; + continue; } - for (auto& interval : interval_set) { - const std::size_t size = interval.upper() - interval.lower(); - if (size == 0) { - continue; - } - staging_buffer.resize(size); - cpu_memory.ReadBlockUnsafe(interval.lower(), staging_buffer.data(), size); - block->Upload(block->Offset(interval.lower()), size, staging_buffer.data()); + Buffer& buffer = slot_buffers[image_id]; + if (buffer.IsRegionGpuModified(addr, size)) { + return true; } + const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes(); + page = Common::DivCeil(end_addr, PAGE_SIZE); } - - VectorMapInterval GetMapsInRange(VAddr addr, std::size_t size) { - VectorMapInterval result; - if (size == 0) { - return result; + return false; +} + +template <class P> +void BufferCache<P>::BindHostIndexBuffer() { + Buffer& buffer = slot_buffers[index_buffer.buffer_id]; + const u32 offset = buffer.Offset(index_buffer.cpu_addr); + const u32 size = index_buffer.size; + SynchronizeBuffer(buffer, index_buffer.cpu_addr, size); + if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { + runtime.BindIndexBuffer(buffer, 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); + } +} + +template <class P> +void BufferCache<P>::BindHostVertexBuffers() { + 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]; + SynchronizeBuffer(buffer, binding.cpu_addr, binding.size); + if (!flags[Dirty::VertexBuffer0 + index]) { + continue; } + flags[Dirty::VertexBuffer0 + index] = false; - const VAddr addr_end = addr + size; - auto it = mapped_addresses.lower_bound(addr); - if (it != mapped_addresses.begin()) { - --it; + 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); + } +} + +template <class P> +void BufferCache<P>::BindHostGraphicsUniformBuffers(size_t stage) { + u32 dirty = ~0U; + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { + dirty = std::exchange(dirty_uniform_buffers[stage], 0); + } + u32 binding_index = 0; + ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) { + const bool needs_bind = ((dirty >> index) & 1) != 0; + BindHostGraphicsUniformBuffer(stage, index, binding_index, needs_bind); + if constexpr (NEEDS_BIND_UNIFORM_INDEX) { + ++binding_index; } - while (it != mapped_addresses.end() && it->start < addr_end) { - if (it->Overlaps(addr, addr_end)) { - result.push_back(&*it); + }); +} + +template <class P> +void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 binding_index, + bool needs_bind) { + const Binding& binding = uniform_buffers[stage][index]; + const VAddr cpu_addr = binding.cpu_addr; + const u32 size = binding.size; + Buffer& buffer = slot_buffers[binding.buffer_id]; + if (size <= SKIP_CACHE_SIZE && !buffer.IsRegionGpuModified(cpu_addr, size)) { + if constexpr (IS_OPENGL) { + if (runtime.HasFastBufferSubData()) { + // Fast path for Nvidia + if (!HasFastUniformBufferBound(stage, binding_index)) { + // We only have to bind when the currently bound buffer is not the fast version + runtime.BindFastUniformBuffer(stage, binding_index, size); + } + const auto span = ImmediateBufferWithData(cpu_addr, size); + runtime.PushFastUniformBuffer(stage, binding_index, span); + return; } - ++it; } - return result; - } + fast_bound_uniform_buffers[stage] |= 1U << binding_index; - /// Returns a ticks counter used for tracking when cached objects were last modified - u64 GetModifiedTicks() { - return ++modified_ticks; + // Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan + const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size); + cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size); + return; } - - void FlushMap(MapInterval* map) { - const auto it = blocks.find(map->start >> BLOCK_PAGE_BITS); - ASSERT_OR_EXECUTE(it != blocks.end(), return;); - - std::shared_ptr<Buffer> block = it->second; - - const std::size_t size = map->end - map->start; - staging_buffer.resize(size); - block->Download(block->Offset(map->start), size, staging_buffer.data()); - cpu_memory.WriteBlockUnsafe(map->start, staging_buffer.data(), size); - map->MarkAsModified(false, 0); + // Classic cached path + SynchronizeBuffer(buffer, cpu_addr, size); + if (!needs_bind && !HasFastUniformBufferBound(stage, binding_index)) { + // Skip binding if it's not needed and if the bound buffer is not the fast version + // This exists to avoid instances where the fast buffer is bound and a GPU write happens + return; } + fast_bound_uniform_buffers[stage] &= ~(1U << binding_index); - template <typename Callable> - BufferInfo StreamBufferUpload(std::size_t size, std::size_t alignment, Callable&& callable) { - AlignBuffer(alignment); - const std::size_t uploaded_offset = buffer_offset; - callable(buffer_ptr); - - buffer_ptr += size; - buffer_offset += size; - return BufferInfo{stream_buffer.Handle(), uploaded_offset, stream_buffer.Address()}; + const u32 offset = buffer.Offset(cpu_addr); + if constexpr (NEEDS_BIND_UNIFORM_INDEX) { + runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size); + } else { + runtime.BindUniformBuffer(buffer, offset, size); } +} + +template <class P> +void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) { + u32 binding_index = 0; + ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) { + const Binding& binding = storage_buffers[stage][index]; + Buffer& buffer = slot_buffers[binding.buffer_id]; + const u32 size = binding.size; + SynchronizeBuffer(buffer, binding.cpu_addr, size); + + const u32 offset = buffer.Offset(binding.cpu_addr); + const bool is_written = ((written_storage_buffers[stage] >> index) & 1) != 0; + if constexpr (NEEDS_BIND_STORAGE_INDEX) { + runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written); + ++binding_index; + } else { + runtime.BindStorageBuffer(buffer, offset, size, is_written); + } + }); +} - void AlignBuffer(std::size_t alignment) { - // Align the offset, not the mapped pointer - const std::size_t offset_aligned = Common::AlignUp(buffer_offset, alignment); - buffer_ptr += offset_aligned - buffer_offset; - buffer_offset = offset_aligned; +template <class P> +void BufferCache<P>::BindHostTransformFeedbackBuffers() { + if (maxwell3d.regs.tfb_enabled == 0) { + return; } + for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) { + const Binding& binding = transform_feedback_buffers[index]; + Buffer& buffer = slot_buffers[binding.buffer_id]; + const u32 size = binding.size; + SynchronizeBuffer(buffer, binding.cpu_addr, size); + + const u32 offset = buffer.Offset(binding.cpu_addr); + runtime.BindTransformFeedbackBuffer(index, buffer, offset, size); + } +} - std::shared_ptr<Buffer> EnlargeBlock(std::shared_ptr<Buffer> buffer) { - const std::size_t old_size = buffer->Size(); - const std::size_t new_size = old_size + BLOCK_PAGE_SIZE; - const VAddr cpu_addr = buffer->CpuAddr(); - std::shared_ptr<Buffer> new_buffer = CreateBlock(cpu_addr, new_size); - new_buffer->CopyFrom(*buffer, 0, 0, old_size); - QueueDestruction(std::move(buffer)); - - const VAddr cpu_addr_end = cpu_addr + new_size - 1; - const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS; - for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) { - blocks.insert_or_assign(page_start, new_buffer); +template <class P> +void BufferCache<P>::BindHostComputeUniformBuffers() { + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { + // Mark all uniform buffers as dirty + dirty_uniform_buffers.fill(~u32{0}); + } + u32 binding_index = 0; + ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) { + const Binding& binding = compute_uniform_buffers[index]; + Buffer& buffer = slot_buffers[binding.buffer_id]; + const u32 size = binding.size; + SynchronizeBuffer(buffer, binding.cpu_addr, size); + + const u32 offset = buffer.Offset(binding.cpu_addr); + if constexpr (NEEDS_BIND_UNIFORM_INDEX) { + runtime.BindComputeUniformBuffer(binding_index, buffer, offset, size); + ++binding_index; + } else { + runtime.BindUniformBuffer(buffer, offset, size); } + }); +} + +template <class P> +void BufferCache<P>::BindHostComputeStorageBuffers() { + u32 binding_index = 0; + ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) { + const Binding& binding = compute_storage_buffers[index]; + Buffer& buffer = slot_buffers[binding.buffer_id]; + const u32 size = binding.size; + SynchronizeBuffer(buffer, binding.cpu_addr, size); + + const u32 offset = buffer.Offset(binding.cpu_addr); + const bool is_written = ((written_compute_storage_buffers >> index) & 1) != 0; + if constexpr (NEEDS_BIND_STORAGE_INDEX) { + runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written); + ++binding_index; + } else { + runtime.BindStorageBuffer(buffer, offset, size, is_written); + } + }); +} - return new_buffer; +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); + } +} + +template <class P> +void BufferCache<P>::DoUpdateComputeBuffers() { + UpdateComputeUniformBuffers(); + UpdateComputeStorageBuffers(); +} + +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; + if (!flags[Dirty::IndexBuffer] && last_index_count == index_array.count) { + return; + } + flags[Dirty::IndexBuffer] = false; + last_index_count = index_array.count; + + 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 u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin); + const u32 draw_size = index_array.count * index_array.FormatSizeInBytes(); + const u32 size = std::min(address_size, draw_size); + if (size == 0 || !cpu_addr) { + index_buffer = NULL_BINDING; + return; + } + index_buffer = Binding{ + .cpu_addr = *cpu_addr, + .size = size, + .buffer_id = FindBuffer(*cpu_addr, size), + }; +} - std::shared_ptr<Buffer> MergeBlocks(std::shared_ptr<Buffer> first, - std::shared_ptr<Buffer> second) { - const std::size_t size_1 = first->Size(); - const std::size_t size_2 = second->Size(); - const VAddr first_addr = first->CpuAddr(); - const VAddr second_addr = second->CpuAddr(); - const VAddr new_addr = std::min(first_addr, second_addr); - const std::size_t new_size = size_1 + size_2; - - std::shared_ptr<Buffer> new_buffer = CreateBlock(new_addr, new_size); - new_buffer->CopyFrom(*first, 0, new_buffer->Offset(first_addr), size_1); - new_buffer->CopyFrom(*second, 0, new_buffer->Offset(second_addr), size_2); - QueueDestruction(std::move(first)); - QueueDestruction(std::move(second)); +template <class P> +void BufferCache<P>::UpdateVertexBuffers() { + auto& flags = maxwell3d.dirty.flags; + if (!maxwell3d.dirty.flags[Dirty::VertexBuffers]) { + return; + } + flags[Dirty::VertexBuffers] = false; - const VAddr cpu_addr_end = new_addr + new_size - 1; - const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS; - for (u64 page_start = new_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) { - blocks.insert_or_assign(page_start, new_buffer); - } - return new_buffer; + for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { + UpdateVertexBuffer(index); } +} - Buffer* GetBlock(VAddr cpu_addr, std::size_t size) { - std::shared_ptr<Buffer> found; +template <class P> +void BufferCache<P>::UpdateVertexBuffer(u32 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 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); + const u32 address_size = static_cast<u32>(gpu_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) { + vertex_buffers[index] = NULL_BINDING; + return; + } + vertex_buffers[index] = Binding{ + .cpu_addr = *cpu_addr, + .size = size, + .buffer_id = FindBuffer(*cpu_addr, size), + }; +} + +template <class P> +void BufferCache<P>::UpdateUniformBuffers(size_t stage) { + ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) { + Binding& binding = uniform_buffers[stage][index]; + if (binding.buffer_id) { + // Already updated + return; + } + // Mark as dirty + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { + dirty_uniform_buffers[stage] |= 1U << index; + } + // Resolve buffer + binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); + }); +} + +template <class P> +void BufferCache<P>::UpdateStorageBuffers(size_t stage) { + const u32 written_mask = written_storage_buffers[stage]; + ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) { + // Resolve buffer + Binding& binding = storage_buffers[stage][index]; + const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size); + binding.buffer_id = buffer_id; + // Mark buffer as written if needed + if (((written_mask >> index) & 1) != 0) { + MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size); + } + }); +} - const VAddr cpu_addr_end = cpu_addr + size - 1; - const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS; - for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) { - auto it = blocks.find(page_start); - if (it == blocks.end()) { - if (found) { - found = EnlargeBlock(found); - continue; - } - const VAddr start_addr = page_start << BLOCK_PAGE_BITS; - found = CreateBlock(start_addr, BLOCK_PAGE_SIZE); - blocks.insert_or_assign(page_start, found); - continue; - } - if (!found) { - found = it->second; - continue; - } - if (found != it->second) { - found = MergeBlocks(std::move(found), it->second); +template <class P> +void BufferCache<P>::UpdateTransformFeedbackBuffers() { + if (maxwell3d.regs.tfb_enabled == 0) { + return; + } + for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) { + UpdateTransformFeedbackBuffer(index); + } +} + +template <class P> +void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 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); + if (binding.buffer_enable == 0 || size == 0 || !cpu_addr) { + transform_feedback_buffers[index] = NULL_BINDING; + return; + } + const BufferId buffer_id = FindBuffer(*cpu_addr, size); + transform_feedback_buffers[index] = Binding{ + .cpu_addr = *cpu_addr, + .size = size, + .buffer_id = buffer_id, + }; + MarkWrittenBuffer(buffer_id, *cpu_addr, size); +} + +template <class P> +void BufferCache<P>::UpdateComputeUniformBuffers() { + ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) { + Binding& binding = compute_uniform_buffers[index]; + binding = NULL_BINDING; + 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()); + if (cpu_addr) { + binding.cpu_addr = *cpu_addr; + binding.size = cbuf.size; } } - return found.get(); + binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); + }); +} + +template <class P> +void BufferCache<P>::UpdateComputeStorageBuffers() { + ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) { + // Resolve buffer + Binding& binding = compute_storage_buffers[index]; + const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size); + binding.buffer_id = buffer_id; + // Mark as written if needed + if (((written_compute_storage_buffers >> index) & 1) != 0) { + MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size); + } + }); +} + +template <class P> +void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size) { + Buffer& buffer = slot_buffers[buffer_id]; + buffer.MarkRegionAsGpuModified(cpu_addr, size); + + const bool is_accuracy_high = Settings::IsGPULevelHigh(); + const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue(); + if (!is_accuracy_high || !is_async) { + return; + } + if (std::ranges::find(uncommitted_downloads, buffer_id) != uncommitted_downloads.end()) { + // Already inserted + return; } + uncommitted_downloads.push_back(buffer_id); +} - void MarkRegionAsWritten(VAddr start, VAddr end) { - const u64 page_end = end >> WRITE_PAGE_BIT; - for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) { - if (const auto [it, inserted] = written_pages.emplace(page_start, 1); !inserted) { - ++it->second; - } +template <class P> +BufferId BufferCache<P>::FindBuffer(VAddr cpu_addr, u32 size) { + if (cpu_addr == 0) { + return NULL_BUFFER_ID; + } + const u64 page = cpu_addr >> PAGE_BITS; + const BufferId buffer_id = page_table[page]; + if (!buffer_id) { + return CreateBuffer(cpu_addr, size); + } + const Buffer& buffer = slot_buffers[buffer_id]; + if (buffer.IsInBounds(cpu_addr, size)) { + return buffer_id; + } + return CreateBuffer(cpu_addr, size); +} + +template <class P> +typename BufferCache<P>::OverlapResult BufferCache<P>::ResolveOverlaps(VAddr cpu_addr, + u32 wanted_size) { + static constexpr int STREAM_LEAP_THRESHOLD = 16; + std::vector<BufferId> overlap_ids; + VAddr begin = cpu_addr; + VAddr end = cpu_addr + wanted_size; + int stream_score = 0; + bool has_stream_leap = false; + for (; cpu_addr >> PAGE_BITS < Common::DivCeil(end, PAGE_SIZE); cpu_addr += PAGE_SIZE) { + const BufferId overlap_id = page_table[cpu_addr >> PAGE_BITS]; + if (!overlap_id) { + continue; + } + Buffer& overlap = slot_buffers[overlap_id]; + if (overlap.IsPicked()) { + continue; + } + overlap_ids.push_back(overlap_id); + overlap.Pick(); + const VAddr overlap_cpu_addr = overlap.CpuAddr(); + if (overlap_cpu_addr < begin) { + cpu_addr = begin = overlap_cpu_addr; + } + end = std::max(end, overlap_cpu_addr + overlap.SizeBytes()); + + stream_score += overlap.StreamScore(); + if (stream_score > STREAM_LEAP_THRESHOLD && !has_stream_leap) { + // When this memory region has been joined a bunch of times, we assume it's being used + // as a stream buffer. Increase the size to skip constantly recreating buffers. + has_stream_leap = true; + end += PAGE_SIZE * 256; } } - - void UnmarkRegionAsWritten(VAddr start, VAddr end) { - const u64 page_end = end >> WRITE_PAGE_BIT; - for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) { - auto it = written_pages.find(page_start); - if (it != written_pages.end()) { - if (it->second > 1) { - --it->second; - } else { - written_pages.erase(it); - } - } + return OverlapResult{ + .ids = std::move(overlap_ids), + .begin = begin, + .end = end, + .has_stream_leap = has_stream_leap, + }; +} + +template <class P> +void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, + bool accumulate_stream_score) { + Buffer& new_buffer = slot_buffers[new_buffer_id]; + Buffer& overlap = slot_buffers[overlap_id]; + if (accumulate_stream_score) { + new_buffer.IncreaseStreamScore(overlap.StreamScore() + 1); + } + std::vector<BufferCopy> copies; + const size_t dst_base_offset = overlap.CpuAddr() - new_buffer.CpuAddr(); + overlap.ForEachDownloadRange([&](u64 begin, u64 range_size) { + copies.push_back(BufferCopy{ + .src_offset = begin, + .dst_offset = dst_base_offset + begin, + .size = range_size, + }); + new_buffer.UnmarkRegionAsCpuModified(begin, range_size); + new_buffer.MarkRegionAsGpuModified(begin, range_size); + }); + if (!copies.empty()) { + runtime.CopyBuffer(slot_buffers[new_buffer_id], overlap, copies); + } + ReplaceBufferDownloads(overlap_id, new_buffer_id); + DeleteBuffer(overlap_id); +} + +template <class P> +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); + for (const BufferId overlap_id : overlap.ids) { + JoinOverlap(new_buffer_id, overlap_id, !overlap.has_stream_leap); + } + Register(new_buffer_id); + return new_buffer_id; +} + +template <class P> +void BufferCache<P>::Register(BufferId buffer_id) { + ChangeRegister<true>(buffer_id); +} + +template <class P> +void BufferCache<P>::Unregister(BufferId buffer_id) { + ChangeRegister<false>(buffer_id); +} + +template <class P> +template <bool insert> +void BufferCache<P>::ChangeRegister(BufferId buffer_id) { + const Buffer& buffer = slot_buffers[buffer_id]; + const VAddr cpu_addr_begin = buffer.CpuAddr(); + const VAddr cpu_addr_end = cpu_addr_begin + buffer.SizeBytes(); + const u64 page_begin = cpu_addr_begin / PAGE_SIZE; + const u64 page_end = Common::DivCeil(cpu_addr_end, PAGE_SIZE); + for (u64 page = page_begin; page != page_end; ++page) { + if constexpr (insert) { + page_table[page] = buffer_id; + } else { + page_table[page] = BufferId{}; } } +} - bool IsRegionWritten(VAddr start, VAddr end) const { - const u64 page_end = end >> WRITE_PAGE_BIT; - for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) { - if (written_pages.contains(page_start)) { - return true; +template <class P> +void BufferCache<P>::SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size) { + if (buffer.CpuAddr() == 0) { + return; + } + SynchronizeBufferImpl(buffer, cpu_addr, size); +} + +template <class P> +void BufferCache<P>::SynchronizeBufferImpl(Buffer& buffer, VAddr cpu_addr, u32 size) { + boost::container::small_vector<BufferCopy, 4> copies; + u64 total_size_bytes = 0; + u64 largest_copy = 0; + buffer.ForEachUploadRange(cpu_addr, size, [&](u64 range_offset, u64 range_size) { + copies.push_back(BufferCopy{ + .src_offset = total_size_bytes, + .dst_offset = range_offset, + .size = range_size, + }); + total_size_bytes += range_size; + largest_copy = std::max(largest_copy, range_size); + }); + if (total_size_bytes == 0) { + return; + } + const std::span<BufferCopy> copies_span(copies.data(), copies.size()); + UploadMemory(buffer, total_size_bytes, largest_copy, copies_span); +} + +template <class P> +void BufferCache<P>::UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy, + std::span<BufferCopy> copies) { + if constexpr (USE_MEMORY_MAPS) { + MappedUploadMemory(buffer, total_size_bytes, copies); + } else { + ImmediateUploadMemory(buffer, largest_copy, copies); + } +} + +template <class P> +void BufferCache<P>::ImmediateUploadMemory(Buffer& buffer, u64 largest_copy, + std::span<const BufferCopy> copies) { + std::span<u8> immediate_buffer; + for (const BufferCopy& copy : copies) { + std::span<const u8> upload_span; + const VAddr cpu_addr = buffer.CpuAddr() + copy.dst_offset; + if (IsRangeGranular(cpu_addr, copy.size)) { + upload_span = std::span(cpu_memory.GetPointer(cpu_addr), copy.size); + } else { + if (immediate_buffer.empty()) { + immediate_buffer = ImmediateBuffer(largest_copy); } + cpu_memory.ReadBlockUnsafe(cpu_addr, immediate_buffer.data(), copy.size); + upload_span = immediate_buffer.subspan(0, copy.size); } - return false; + buffer.ImmediateUpload(copy.dst_offset, upload_span); } - - void QueueDestruction(std::shared_ptr<Buffer> buffer) { - buffer->SetEpoch(epoch); - pending_destruction.push(std::move(buffer)); +} + +template <class P> +void BufferCache<P>::MappedUploadMemory(Buffer& buffer, u64 total_size_bytes, + std::span<BufferCopy> copies) { + auto upload_staging = runtime.UploadStagingBuffer(total_size_bytes); + const std::span<u8> staging_pointer = upload_staging.mapped_span; + for (BufferCopy& copy : copies) { + u8* const src_pointer = staging_pointer.data() + copy.src_offset; + const VAddr cpu_addr = buffer.CpuAddr() + copy.dst_offset; + cpu_memory.ReadBlockUnsafe(cpu_addr, src_pointer, copy.size); + + // Apply the staging offset + copy.src_offset += upload_staging.offset; } - - void MarkForAsyncFlush(MapInterval* map) { - if (!uncommitted_flushes) { - uncommitted_flushes = std::make_shared<std::unordered_set<MapInterval*>>(); + runtime.CopyBuffer(buffer, upload_staging.buffer, copies); +} + +template <class P> +void BufferCache<P>::DeleteBuffer(BufferId buffer_id) { + const auto scalar_replace = [buffer_id](Binding& binding) { + if (binding.buffer_id == buffer_id) { + binding.buffer_id = BufferId{}; + } + }; + const auto replace = [scalar_replace](std::span<Binding> bindings) { + std::ranges::for_each(bindings, scalar_replace); + }; + scalar_replace(index_buffer); + replace(vertex_buffers); + std::ranges::for_each(uniform_buffers, replace); + std::ranges::for_each(storage_buffers, replace); + replace(transform_feedback_buffers); + replace(compute_uniform_buffers); + replace(compute_storage_buffers); + std::erase(cached_write_buffer_ids, buffer_id); + + // Mark the whole buffer as CPU written to stop tracking CPU writes + Buffer& buffer = slot_buffers[buffer_id]; + buffer.MarkRegionAsCpuModified(buffer.CpuAddr(), buffer.SizeBytes()); + + Unregister(buffer_id); + delayed_destruction_ring.Push(std::move(slot_buffers[buffer_id])); + + NotifyBufferDeletion(); +} + +template <class P> +void BufferCache<P>::ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id) { + const auto replace = [old_buffer_id, new_buffer_id](std::vector<BufferId>& buffers) { + std::ranges::replace(buffers, old_buffer_id, new_buffer_id); + if (auto it = std::ranges::find(buffers, new_buffer_id); it != buffers.end()) { + buffers.erase(std::remove(it + 1, buffers.end(), new_buffer_id), buffers.end()); } - uncommitted_flushes->insert(map); + }; + replace(uncommitted_downloads); + std::ranges::for_each(committed_downloads, replace); +} + +template <class P> +void BufferCache<P>::NotifyBufferDeletion() { + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { + dirty_uniform_buffers.fill(~u32{0}); } + auto& flags = maxwell3d.dirty.flags; + flags[Dirty::IndexBuffer] = true; + flags[Dirty::VertexBuffers] = true; + for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { + flags[Dirty::VertexBuffer0 + index] = true; + } + has_deleted_buffers = true; +} + +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); + if (!cpu_addr || size == 0) { + return NULL_BINDING; + } + // HACK(Rodrigo): This is the number of bytes bound in host beyond the guest API's range. + // It exists due to some games like Astral Chain operate out of bounds. + // Binding the whole map range would be technically correct, but games have large maps that make + // this approach unaffordable for now. + static constexpr u32 arbitrary_extra_bytes = 0xc000; + const u32 bytes_to_map_end = static_cast<u32>(gpu_memory.BytesToMapEnd(gpu_addr)); + const Binding binding{ + .cpu_addr = *cpu_addr, + .size = std::min(size + arbitrary_extra_bytes, bytes_to_map_end), + .buffer_id = BufferId{}, + }; + return binding; +} + +template <class P> +std::span<const u8> BufferCache<P>::ImmediateBufferWithData(VAddr cpu_addr, size_t size) { + u8* const base_pointer = cpu_memory.GetPointer(cpu_addr); + if (IsRangeGranular(cpu_addr, size) || + base_pointer + size == cpu_memory.GetPointer(cpu_addr + size)) { + return std::span(base_pointer, size); + } else { + const std::span<u8> span = ImmediateBuffer(size); + cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size); + return span; + } +} - VideoCore::RasterizerInterface& rasterizer; - Tegra::MemoryManager& gpu_memory; - Core::Memory::Memory& cpu_memory; - StreamBuffer& stream_buffer; - - u8* buffer_ptr = nullptr; - u64 buffer_offset = 0; - u64 buffer_offset_base = 0; - - MapIntervalAllocator mapped_addresses_allocator; - boost::intrusive::set<MapInterval, boost::intrusive::compare<MapIntervalCompare>> - mapped_addresses; - - std::unordered_map<u64, u32> written_pages; - std::unordered_map<u64, std::shared_ptr<Buffer>> blocks; - - std::queue<std::shared_ptr<Buffer>> pending_destruction; - u64 epoch = 0; - u64 modified_ticks = 0; - - std::vector<u8> staging_buffer; - - std::list<MapInterval*> marked_for_unregister; - - std::shared_ptr<std::unordered_set<MapInterval*>> uncommitted_flushes; - std::list<std::shared_ptr<std::list<MapInterval*>>> committed_flushes; - - std::recursive_mutex mutex; -}; +template <class P> +std::span<u8> BufferCache<P>::ImmediateBuffer(size_t wanted_capacity) { + if (wanted_capacity > immediate_buffer_capacity) { + immediate_buffer_capacity = wanted_capacity; + immediate_buffer_alloc = std::make_unique<u8[]>(wanted_capacity); + } + return std::span<u8>(immediate_buffer_alloc.get(), wanted_capacity); +} + +template <class P> +bool BufferCache<P>::HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept { + if constexpr (IS_OPENGL) { + return ((fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0; + } else { + // Only OpenGL has fast uniform buffers + return false; + } +} } // namespace VideoCommon diff --git a/src/video_core/buffer_cache/map_interval.cpp b/src/video_core/buffer_cache/map_interval.cpp deleted file mode 100644 index 62587e18a..000000000 --- a/src/video_core/buffer_cache/map_interval.cpp +++ /dev/null @@ -1,33 +0,0 @@ -// Copyright 2020 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <array> -#include <cstddef> -#include <memory> - -#include "video_core/buffer_cache/map_interval.h" - -namespace VideoCommon { - -MapIntervalAllocator::MapIntervalAllocator() { - FillFreeList(first_chunk); -} - -MapIntervalAllocator::~MapIntervalAllocator() = default; - -void MapIntervalAllocator::AllocateNewChunk() { - *new_chunk = std::make_unique<Chunk>(); - FillFreeList(**new_chunk); - new_chunk = &(*new_chunk)->next; -} - -void MapIntervalAllocator::FillFreeList(Chunk& chunk) { - const std::size_t old_size = free_list.size(); - free_list.resize(old_size + chunk.data.size()); - std::transform(chunk.data.rbegin(), chunk.data.rend(), free_list.begin() + old_size, - [](MapInterval& interval) { return &interval; }); -} - -} // namespace VideoCommon diff --git a/src/video_core/buffer_cache/map_interval.h b/src/video_core/buffer_cache/map_interval.h deleted file mode 100644 index ef974b08a..000000000 --- a/src/video_core/buffer_cache/map_interval.h +++ /dev/null @@ -1,93 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <cstddef> -#include <memory> -#include <vector> - -#include <boost/intrusive/set_hook.hpp> - -#include "common/common_types.h" -#include "video_core/gpu.h" - -namespace VideoCommon { - -struct MapInterval : public boost::intrusive::set_base_hook<boost::intrusive::optimize_size<true>> { - MapInterval() = default; - - /*implicit*/ MapInterval(VAddr start_) noexcept : start{start_} {} - - explicit MapInterval(VAddr start_, VAddr end_, GPUVAddr gpu_addr_) noexcept - : start{start_}, end{end_}, gpu_addr{gpu_addr_} {} - - bool IsInside(VAddr other_start, VAddr other_end) const noexcept { - return start <= other_start && other_end <= end; - } - - bool Overlaps(VAddr other_start, VAddr other_end) const noexcept { - return start < other_end && other_start < end; - } - - void MarkAsModified(bool is_modified_, u64 ticks_) noexcept { - is_modified = is_modified_; - ticks = ticks_; - } - - boost::intrusive::set_member_hook<> member_hook_; - VAddr start = 0; - VAddr end = 0; - GPUVAddr gpu_addr = 0; - u64 ticks = 0; - bool is_written = false; - bool is_modified = false; - bool is_registered = false; - bool is_memory_marked = false; - bool is_sync_pending = false; -}; - -struct MapIntervalCompare { - constexpr bool operator()(const MapInterval& lhs, const MapInterval& rhs) const noexcept { - return lhs.start < rhs.start; - } -}; - -class MapIntervalAllocator { -public: - MapIntervalAllocator(); - ~MapIntervalAllocator(); - - MapInterval* Allocate() { - if (free_list.empty()) { - AllocateNewChunk(); - } - MapInterval* const interval = free_list.back(); - free_list.pop_back(); - return interval; - } - - void Release(MapInterval* interval) { - free_list.push_back(interval); - } - -private: - struct Chunk { - std::unique_ptr<Chunk> next; - std::array<MapInterval, 0x8000> data; - }; - - void AllocateNewChunk(); - - void FillFreeList(Chunk& chunk); - - std::vector<MapInterval*> free_list; - - Chunk first_chunk; - - std::unique_ptr<Chunk>* new_chunk = &first_chunk.next; -}; - -} // namespace VideoCommon diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/command_classes/vic.cpp index 55e632346..2b7569335 100644 --- a/src/video_core/command_classes/vic.cpp +++ b/src/video_core/command_classes/vic.cpp @@ -110,12 +110,10 @@ void Vic::Execute() { converted_frame_buffer.get(), block_height, 0, 0); gpu.MemoryManager().WriteBlock(output_surface_luma_address, swizzled_data.data(), size); - gpu.Maxwell3D().OnMemoryWrite(); } else { // send pitch linear frame gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr, linear_size); - gpu.Maxwell3D().OnMemoryWrite(); } break; } @@ -163,7 +161,6 @@ void Vic::Execute() { } gpu.MemoryManager().WriteBlock(output_surface_chroma_u_address, chroma_buffer.data(), chroma_buffer.size()); - gpu.Maxwell3D().OnMemoryWrite(); break; } default: diff --git a/src/video_core/dirty_flags.cpp b/src/video_core/dirty_flags.cpp index b1eaac00c..7149af290 100644 --- a/src/video_core/dirty_flags.cpp +++ b/src/video_core/dirty_flags.cpp @@ -12,13 +12,30 @@ #define NUM(field_name) (sizeof(::Tegra::Engines::Maxwell3D::Regs::field_name) / (sizeof(u32))) namespace VideoCommon::Dirty { - +namespace { using Tegra::Engines::Maxwell3D; -void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables) { +void SetupDirtyVertexBuffers(Maxwell3D::DirtyState::Tables& tables) { + static constexpr std::size_t num_array = 3; + for (std::size_t i = 0; i < Maxwell3D::Regs::NumVertexArrays; ++i) { + const std::size_t array_offset = OFF(vertex_array) + i * NUM(vertex_array[0]); + const std::size_t limit_offset = OFF(vertex_array_limit) + i * NUM(vertex_array_limit[0]); + + FillBlock(tables, array_offset, num_array, VertexBuffer0 + i, VertexBuffers); + FillBlock(tables, limit_offset, NUM(vertex_array_limit), VertexBuffer0 + i, VertexBuffers); + } +} + +void SetupIndexBuffer(Maxwell3D::DirtyState::Tables& tables) { + FillBlock(tables[0], OFF(index_array), NUM(index_array), IndexBuffer); +} + +void SetupDirtyDescriptors(Maxwell3D::DirtyState::Tables& tables) { FillBlock(tables[0], OFF(tic), NUM(tic), Descriptors); FillBlock(tables[0], OFF(tsc), NUM(tsc), Descriptors); +} +void SetupDirtyRenderTargets(Maxwell3D::DirtyState::Tables& tables) { static constexpr std::size_t num_per_rt = NUM(rt[0]); static constexpr std::size_t begin = OFF(rt); static constexpr std::size_t num = num_per_rt * Maxwell3D::Regs::NumRenderTargets; @@ -41,5 +58,13 @@ void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tabl FillBlock(table, OFF(zeta), NUM(zeta), flag); } } +} // Anonymous namespace + +void SetupDirtyFlags(Maxwell3D::DirtyState::Tables& tables) { + SetupDirtyVertexBuffers(tables); + SetupIndexBuffer(tables); + SetupDirtyDescriptors(tables); + SetupDirtyRenderTargets(tables); +} } // namespace VideoCommon::Dirty diff --git a/src/video_core/dirty_flags.h b/src/video_core/dirty_flags.h index 875527ddd..702688ace 100644 --- a/src/video_core/dirty_flags.h +++ b/src/video_core/dirty_flags.h @@ -30,6 +30,12 @@ enum : u8 { ColorBuffer7, ZetaBuffer, + VertexBuffers, + VertexBuffer0, + VertexBuffer31 = VertexBuffer0 + 31, + + IndexBuffer, + LastCommonEntry, }; @@ -47,6 +53,6 @@ void FillBlock(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables, std::size_ FillBlock(tables[1], begin, num, index_b); } -void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables); +void SetupDirtyFlags(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables); } // namespace VideoCommon::Dirty diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 2c8b20024..8b33c04ab 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -23,8 +23,6 @@ void DmaPusher::DispatchCalls() { MICROPROFILE_SCOPE(DispatchCalls); gpu.SyncGuestHost(); - // On entering GPU code, assume all memory may be touched by the ARM core. - gpu.Maxwell3D().OnMemoryWrite(); dma_pushbuffer_subindex = 0; diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp index a01d334ad..0f640fdae 100644 --- a/src/video_core/engines/fermi_2d.cpp +++ b/src/video_core/engines/fermi_2d.cpp @@ -18,8 +18,8 @@ Fermi2D::Fermi2D() { Fermi2D::~Fermi2D() = default; -void Fermi2D::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) { - rasterizer = &rasterizer_; +void Fermi2D::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; } void Fermi2D::CallMethod(u32 method, u32 method_argument, bool is_last_call) { diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h index 0de3280a2..c808a577d 100644 --- a/src/video_core/engines/fermi_2d.h +++ b/src/video_core/engines/fermi_2d.h @@ -38,7 +38,7 @@ public: ~Fermi2D(); /// Binds a rasterizer to this engine. - void BindRasterizer(VideoCore::RasterizerInterface& rasterizer); + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); /// Write the value to the register identified by method. void CallMethod(u32 method, u32 method_argument, bool is_last_call) override; diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp index ba387506e..a9b75091e 100644 --- a/src/video_core/engines/kepler_compute.cpp +++ b/src/video_core/engines/kepler_compute.cpp @@ -21,8 +21,8 @@ KeplerCompute::KeplerCompute(Core::System& system_, MemoryManager& memory_manage KeplerCompute::~KeplerCompute() = default; -void KeplerCompute::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) { - rasterizer = &rasterizer_; +void KeplerCompute::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; } void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_call) { @@ -39,7 +39,6 @@ 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) { - system.GPU().Maxwell3D().OnMemoryWrite(); } break; } diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h index 9f0a7b76d..7c40cba38 100644 --- a/src/video_core/engines/kepler_compute.h +++ b/src/video_core/engines/kepler_compute.h @@ -46,7 +46,7 @@ public: ~KeplerCompute(); /// Binds a rasterizer to this engine. - void BindRasterizer(VideoCore::RasterizerInterface& rasterizer); + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); static constexpr std::size_t NumConstBuffers = 8; diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp index 9911140e9..560551157 100644 --- a/src/video_core/engines/kepler_memory.cpp +++ b/src/video_core/engines/kepler_memory.cpp @@ -33,7 +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) { - system.GPU().Maxwell3D().OnMemoryWrite(); } break; } diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 116ad1722..75517a4f7 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -30,8 +30,8 @@ Maxwell3D::Maxwell3D(Core::System& system_, MemoryManager& memory_manager_) Maxwell3D::~Maxwell3D() = default; -void Maxwell3D::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) { - rasterizer = &rasterizer_; +void Maxwell3D::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; } void Maxwell3D::InitializeRegisterDefaults() { @@ -223,7 +223,6 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume case MAXWELL3D_REG_INDEX(data_upload): upload_state.ProcessData(argument, is_last_call); if (is_last_call) { - OnMemoryWrite(); } return; case MAXWELL3D_REG_INDEX(fragment_barrier): @@ -570,17 +569,18 @@ std::optional<u64> Maxwell3D::GetQueryResult() { } } -void Maxwell3D::ProcessCBBind(std::size_t stage_index) { +void Maxwell3D::ProcessCBBind(size_t stage_index) { // Bind the buffer currently in CB_ADDRESS to the specified index in the desired shader stage. - auto& shader = state.shader_stages[stage_index]; - auto& bind_data = regs.cb_bind[stage_index]; - - ASSERT(bind_data.index < Regs::MaxConstBuffers); - auto& buffer = shader.const_buffers[bind_data.index]; - + const auto& bind_data = regs.cb_bind[stage_index]; + auto& buffer = state.shader_stages[stage_index].const_buffers[bind_data.index]; buffer.enabled = bind_data.valid.Value() != 0; buffer.address = regs.const_buffer.BufferAddress(); buffer.size = regs.const_buffer.cb_size; + + const bool is_enabled = bind_data.valid.Value() != 0; + const GPUVAddr gpu_addr = is_enabled ? regs.const_buffer.BufferAddress() : 0; + const u32 size = is_enabled ? regs.const_buffer.cb_size : 0; + rasterizer->BindGraphicsUniformBuffer(stage_index, bind_data.index, gpu_addr, size); } void Maxwell3D::ProcessCBData(u32 value) { @@ -635,7 +635,6 @@ void Maxwell3D::FinishCBData() { const u32 id = cb_data_state.id; memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size); - OnMemoryWrite(); cb_data_state.id = null_cb_data; cb_data_state.current = null_cb_data; diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 002d1b3f9..ffed42a29 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -55,7 +55,7 @@ public: ~Maxwell3D(); /// Binds a rasterizer to this engine. - void BindRasterizer(VideoCore::RasterizerInterface& rasterizer); + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); /// Register structure of the Maxwell3D engine. /// TODO(Subv): This structure will need to be made bigger as more registers are discovered. @@ -1314,8 +1314,7 @@ public: GPUVAddr LimitAddress() const { return static_cast<GPUVAddr>((static_cast<GPUVAddr>(limit_high) << 32) | - limit_low) + - 1; + limit_low); } } vertex_array_limit[NumVertexArrays]; @@ -1403,6 +1402,7 @@ public: }; std::array<ShaderStageInfo, Regs::MaxShaderStage> shader_stages; + u32 current_instance = 0; ///< Current instance to be used to simulate instanced rendering. }; @@ -1452,11 +1452,6 @@ public: return *rasterizer; } - /// Notify a memory write has happened. - void OnMemoryWrite() { - dirty.flags |= dirty.on_write_stores; - } - enum class MMEDrawMode : u32 { Undefined, Array, @@ -1478,7 +1473,6 @@ public: using Tables = std::array<Table, 2>; Flags flags; - Flags on_write_stores; Tables tables{}; } dirty; @@ -1541,7 +1535,7 @@ private: void FinishCBData(); /// Handles a write to the CB_BIND register. - void ProcessCBBind(std::size_t stage_index); + void ProcessCBBind(size_t stage_index); /// Handles a write to the VERTEX_END_GL register, triggering a draw. void DrawArrays(); diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index ba750748c..a2f19559f 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -60,9 +60,6 @@ void MaxwellDMA::Launch() { return; } - // All copies here update the main memory, so mark all rasterizer states as invalid. - system.GPU().Maxwell3D().OnMemoryWrite(); - if (is_src_pitch && is_dst_pitch) { CopyPitchToPitch(); } else { diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h index 3512283ff..f055b61e9 100644 --- a/src/video_core/fence_manager.h +++ b/src/video_core/fence_manager.h @@ -143,22 +143,26 @@ private: } bool ShouldWait() const { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; return texture_cache.ShouldWaitAsyncFlushes() || buffer_cache.ShouldWaitAsyncFlushes() || query_cache.ShouldWaitAsyncFlushes(); } bool ShouldFlush() const { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; return texture_cache.HasUncommittedFlushes() || buffer_cache.HasUncommittedFlushes() || query_cache.HasUncommittedFlushes(); } void 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(); query_cache.CommitAsyncFlushes(); diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 6ab06775f..2a9bd4121 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -44,8 +44,8 @@ GPU::~GPU() = default; void GPU::BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer_) { renderer = std::move(renderer_); + rasterizer = renderer->ReadRasterizer(); - VideoCore::RasterizerInterface& rasterizer = renderer->Rasterizer(); memory_manager->BindRasterizer(rasterizer); maxwell_3d->BindRasterizer(rasterizer); fermi_2d->BindRasterizer(rasterizer); @@ -171,7 +171,7 @@ void GPU::TickWork() { const std::size_t size = request.size; flush_requests.pop_front(); flush_request_mutex.unlock(); - renderer->Rasterizer().FlushRegion(addr, size); + rasterizer->FlushRegion(addr, size); current_flush_fence.store(fence); flush_request_mutex.lock(); } @@ -193,11 +193,11 @@ u64 GPU::GetTicks() const { } void GPU::FlushCommands() { - renderer->Rasterizer().FlushCommands(); + rasterizer->FlushCommands(); } void GPU::SyncGuestHost() { - renderer->Rasterizer().SyncGuestHost(); + rasterizer->SyncGuestHost(); } enum class GpuSemaphoreOperation { diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index b4ce6b154..b2ee45496 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -366,6 +366,7 @@ protected: std::unique_ptr<Tegra::DmaPusher> dma_pusher; std::unique_ptr<Tegra::CDmaPusher> cdma_pusher; std::unique_ptr<VideoCore::RendererBase> renderer; + VideoCore::RasterizerInterface* rasterizer = nullptr; const bool use_nvdec; private: diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp index 7e490bcc3..50319f1d5 100644 --- a/src/video_core/gpu_thread.cpp +++ b/src/video_core/gpu_thread.cpp @@ -38,6 +38,7 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, } auto current_context = context.Acquire(); + VideoCore::RasterizerInterface* const rasterizer = renderer.ReadRasterizer(); CommandDataContainer next; while (state.is_running) { @@ -52,13 +53,13 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer, } 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)) { - renderer.Rasterizer().ReleaseFences(); + rasterizer->ReleaseFences(); } else if (std::holds_alternative<GPUTickCommand>(next.data)) { system.GPU().TickWork(); } else if (const auto* flush = std::get_if<FlushRegionCommand>(&next.data)) { - renderer.Rasterizer().FlushRegion(flush->addr, flush->size); + rasterizer->FlushRegion(flush->addr, flush->size); } else if (const auto* invalidate = std::get_if<InvalidateRegionCommand>(&next.data)) { - renderer.Rasterizer().OnCPUWrite(invalidate->addr, invalidate->size); + rasterizer->OnCPUWrite(invalidate->addr, invalidate->size); } else if (std::holds_alternative<EndProcessingCommand>(next.data)) { return; } else { @@ -84,6 +85,7 @@ ThreadManager::~ThreadManager() { void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Frontend::GraphicsContext& context, Tegra::DmaPusher& dma_pusher, Tegra::CDmaPusher& cdma_pusher) { + rasterizer = renderer.ReadRasterizer(); thread = std::thread(RunThread, std::ref(system), std::ref(renderer), std::ref(context), std::ref(dma_pusher), std::ref(state), std::ref(cdma_pusher)); } @@ -129,12 +131,12 @@ void ThreadManager::FlushRegion(VAddr addr, u64 size) { } void ThreadManager::InvalidateRegion(VAddr addr, u64 size) { - system.Renderer().Rasterizer().OnCPUWrite(addr, size); + rasterizer->OnCPUWrite(addr, size); } void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) { // Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important - system.Renderer().Rasterizer().OnCPUWrite(addr, size); + rasterizer->OnCPUWrite(addr, size); } void ThreadManager::WaitIdle() const { diff --git a/src/video_core/gpu_thread.h b/src/video_core/gpu_thread.h index 2775629e7..4cd951169 100644 --- a/src/video_core/gpu_thread.h +++ b/src/video_core/gpu_thread.h @@ -27,6 +27,7 @@ class System; } // namespace Core namespace VideoCore { +class RasterizerInterface; class RendererBase; } // namespace VideoCore @@ -151,11 +152,12 @@ private: /// Pushes a command to be executed by the GPU thread u64 PushCommand(CommandData&& command_data); - SynchState state; Core::System& system; - std::thread thread; - std::thread::id thread_id; const bool is_async; + VideoCore::RasterizerInterface* rasterizer = nullptr; + + SynchState state; + std::thread thread; }; } // namespace VideoCommon::GPUThread diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index 28f2b8614..970120acc 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -12,7 +12,6 @@ set(SHADER_FILES vulkan_blit_depth_stencil.frag vulkan_present.frag vulkan_present.vert - vulkan_quad_array.comp vulkan_quad_indexed.comp vulkan_uint8.comp ) diff --git a/src/video_core/host_shaders/vulkan_quad_array.comp b/src/video_core/host_shaders/vulkan_quad_array.comp deleted file mode 100644 index 212f4e998..000000000 --- a/src/video_core/host_shaders/vulkan_quad_array.comp +++ /dev/null @@ -1,28 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#version 460 core - -layout (local_size_x = 1024) in; - -layout (std430, set = 0, binding = 0) buffer OutputBuffer { - uint output_indexes[]; -}; - -layout (push_constant) uniform PushConstants { - uint first; -}; - -void main() { - uint primitive = gl_GlobalInvocationID.x; - if (primitive * 6 >= output_indexes.length()) { - return; - } - - const uint quad_map[6] = uint[](0, 1, 2, 0, 2, 3); - for (uint vertex = 0; vertex < 6; ++vertex) { - uint index = first + primitive * 4 + quad_map[vertex]; - output_indexes[primitive * 6 + vertex] = index; - } -} diff --git a/src/video_core/host_shaders/vulkan_uint8.comp b/src/video_core/host_shaders/vulkan_uint8.comp index ad74d7af9..872291670 100644 --- a/src/video_core/host_shaders/vulkan_uint8.comp +++ b/src/video_core/host_shaders/vulkan_uint8.comp @@ -16,9 +16,16 @@ layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer { uint16_t output_indexes[]; }; +uint AssembleIndex(uint id) { + // Most primitive restart indices are 0xFF + // Hardcode this to 0xFF for now + uint index = uint(input_indexes[id]); + return index == 0xFF ? 0xFFFF : index; +} + void main() { uint id = gl_GlobalInvocationID.x; if (id < input_indexes.length()) { - output_indexes[id] = uint16_t(input_indexes[id]); + output_indexes[id] = uint16_t(AssembleIndex(id)); } } diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index c841f3cd7..44240a9c4 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -21,8 +21,8 @@ MemoryManager::MemoryManager(Core::System& system_) MemoryManager::~MemoryManager() = default; -void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface& rasterizer_) { - rasterizer = &rasterizer_; +void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) { + rasterizer = rasterizer_; } GPUVAddr MemoryManager::UpdateRange(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size) { diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index b468a67de..b3538d503 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -72,7 +72,7 @@ public: ~MemoryManager(); /// Binds a renderer to the memory manager. - void BindRasterizer(VideoCore::RasterizerInterface& rasterizer); + void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); [[nodiscard]] std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr) const; @@ -157,6 +157,8 @@ private: using MapRange = std::pair<GPUVAddr, size_t>; std::vector<MapRange> map_ranges; + + std::vector<std::pair<VAddr, std::size_t>> cache_invalidate_queue; }; } // namespace Tegra diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h index 0cb0f387d..50491b758 100644 --- a/src/video_core/rasterizer_interface.h +++ b/src/video_core/rasterizer_interface.h @@ -7,6 +7,7 @@ #include <atomic> #include <functional> #include <optional> +#include <span> #include "common/common_types.h" #include "video_core/engines/fermi_2d.h" #include "video_core/gpu.h" @@ -49,6 +50,10 @@ public: /// Records a GPU query and caches it virtual void Query(GPUVAddr gpu_addr, QueryType type, std::optional<u64> timestamp) = 0; + /// Signal an uniform buffer binding + virtual void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, + u32 size) = 0; + /// Signal a GPU based semaphore as a fence virtual void SignalSemaphore(GPUVAddr addr, u32 value) = 0; diff --git a/src/video_core/renderer_base.h b/src/video_core/renderer_base.h index 51dde8eb5..320ee8d30 100644 --- a/src/video_core/renderer_base.h +++ b/src/video_core/renderer_base.h @@ -37,15 +37,11 @@ public: std::unique_ptr<Core::Frontend::GraphicsContext> context); virtual ~RendererBase(); - /// Initialize the renderer - [[nodiscard]] virtual bool Init() = 0; - - /// Shutdown the renderer - virtual void ShutDown() = 0; - /// Finalize rendering the guest frame and draw into the presentation texture virtual void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) = 0; + [[nodiscard]] virtual RasterizerInterface* ReadRasterizer() = 0; + // Getter/setter functions: // ------------------------ @@ -57,14 +53,6 @@ public: return m_current_frame; } - [[nodiscard]] RasterizerInterface& Rasterizer() { - return *rasterizer; - } - - [[nodiscard]] const RasterizerInterface& Rasterizer() const { - return *rasterizer; - } - [[nodiscard]] Core::Frontend::GraphicsContext& Context() { return *context; } @@ -98,7 +86,6 @@ public: protected: Core::Frontend::EmuWindow& render_window; ///< Reference to the render window handle. - std::unique_ptr<RasterizerInterface> rasterizer; std::unique_ptr<Core::Frontend::GraphicsContext> context; f32 m_current_fps = 0.0f; ///< Current framerate, should be set by the renderer int m_current_frame = 0; ///< Current frame, should be set by the renderer diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index 5772cad87..6da3906a4 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -2,98 +2,208 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include <memory> +#include <span> -#include <glad/glad.h> - -#include "common/assert.h" -#include "common/microprofile.h" #include "video_core/buffer_cache/buffer_cache.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/rasterizer_interface.h" #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_device.h" -#include "video_core/renderer_opengl/gl_rasterizer.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" namespace OpenGL { +namespace { +struct BindlessSSBO { + GLuint64EXT address; + GLsizei length; + GLsizei padding; +}; +static_assert(sizeof(BindlessSSBO) == sizeof(GLuint) * 4); + +constexpr std::array PROGRAM_LUT{ + GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV, + GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV, +}; +} // Anonymous namespace + +Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params) + : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {} + +Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_, + VAddr cpu_addr_, u64 size_bytes_) + : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) { + buffer.Create(); + const std::string name = fmt::format("Buffer 0x{:x}", CpuAddr()); + glObjectLabel(GL_BUFFER, buffer.handle, static_cast<GLsizei>(name.size()), name.data()); + glNamedBufferData(buffer.handle, SizeBytes(), nullptr, GL_DYNAMIC_DRAW); + + if (runtime.has_unified_vertex_buffers) { + glGetNamedBufferParameterui64vNV(buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &address); + } +} -using Maxwell = Tegra::Engines::Maxwell3D::Regs; +void Buffer::ImmediateUpload(size_t offset, std::span<const u8> data) noexcept { + glNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset), + static_cast<GLsizeiptr>(data.size_bytes()), data.data()); +} -MICROPROFILE_DEFINE(OpenGL_Buffer_Download, "OpenGL", "Buffer Download", MP_RGB(192, 192, 128)); +void Buffer::ImmediateDownload(size_t offset, std::span<u8> data) noexcept { + glGetNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset), + static_cast<GLsizeiptr>(data.size_bytes()), data.data()); +} -Buffer::Buffer(const Device& device_, VAddr cpu_addr_, std::size_t size_) - : BufferBlock{cpu_addr_, size_} { - gl_buffer.Create(); - glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size_), nullptr, GL_DYNAMIC_DRAW); - if (device_.UseAssemblyShaders() || device_.HasVertexBufferUnifiedMemory()) { - glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_WRITE); - glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address); +void Buffer::MakeResident(GLenum access) noexcept { + // Abuse GLenum's order to exit early + // GL_NONE (default) < GL_READ_ONLY < GL_READ_WRITE + if (access <= current_residency_access || buffer.handle == 0) { + return; + } + if (std::exchange(current_residency_access, access) != GL_NONE) { + // If the buffer is already resident, remove its residency before promoting it + glMakeNamedBufferNonResidentNV(buffer.handle); } + glMakeNamedBufferResidentNV(buffer.handle, access); } -Buffer::~Buffer() = default; - -void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) { - glNamedBufferSubData(Handle(), static_cast<GLintptr>(offset), - static_cast<GLsizeiptr>(data_size), data); +BufferCacheRuntime::BufferCacheRuntime(const Device& device_) + : device{device_}, has_fast_buffer_sub_data{device.HasFastBufferSubData()}, + use_assembly_shaders{device.UseAssemblyShaders()}, + has_unified_vertex_buffers{device.HasVertexBufferUnifiedMemory()}, + stream_buffer{has_fast_buffer_sub_data ? std::nullopt : std::make_optional<StreamBuffer>()} { + GLint gl_max_attributes; + glGetIntegerv(GL_MAX_VERTEX_ATTRIBS, &gl_max_attributes); + max_attributes = static_cast<u32>(gl_max_attributes); + for (auto& stage_uniforms : fast_uniforms) { + for (OGLBuffer& buffer : stage_uniforms) { + buffer.Create(); + glNamedBufferData(buffer.handle, BufferCache::SKIP_CACHE_SIZE, nullptr, GL_STREAM_DRAW); + } + } + for (auto& stage_uniforms : copy_uniforms) { + for (OGLBuffer& buffer : stage_uniforms) { + buffer.Create(); + glNamedBufferData(buffer.handle, 0x10'000, nullptr, GL_STREAM_COPY); + } + } + for (OGLBuffer& buffer : copy_compute_uniforms) { + buffer.Create(); + glNamedBufferData(buffer.handle, 0x10'000, nullptr, GL_STREAM_COPY); + } } -void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) { - MICROPROFILE_SCOPE(OpenGL_Buffer_Download); - const GLsizeiptr gl_size = static_cast<GLsizeiptr>(data_size); - const GLintptr gl_offset = static_cast<GLintptr>(offset); - if (read_buffer.handle == 0) { - read_buffer.Create(); - glNamedBufferData(read_buffer.handle, static_cast<GLsizeiptr>(Size()), nullptr, - GL_STREAM_READ); +void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, + std::span<const VideoCommon::BufferCopy> copies) { + for (const VideoCommon::BufferCopy& copy : copies) { + glCopyNamedBufferSubData( + src_buffer.Handle(), dst_buffer.Handle(), static_cast<GLintptr>(copy.src_offset), + static_cast<GLintptr>(copy.dst_offset), static_cast<GLsizeiptr>(copy.size)); } - glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT); - glCopyNamedBufferSubData(gl_buffer.handle, read_buffer.handle, gl_offset, gl_offset, gl_size); - glGetNamedBufferSubData(read_buffer.handle, gl_offset, gl_size, data); } -void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset, - std::size_t copy_size) { - glCopyNamedBufferSubData(src.Handle(), Handle(), static_cast<GLintptr>(src_offset), - static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(copy_size)); +void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) { + if (has_unified_vertex_buffers) { + buffer.MakeResident(GL_READ_ONLY); + glBufferAddressRangeNV(GL_ELEMENT_ARRAY_ADDRESS_NV, 0, buffer.HostGpuAddr() + offset, + static_cast<GLsizeiptr>(size)); + } else { + glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer.Handle()); + index_buffer_offset = offset; + } } -OGLBufferCache::OGLBufferCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - const Device& device_, OGLStreamBuffer& stream_buffer_, - StateTracker& state_tracker) - : GenericBufferCache{rasterizer_, gpu_memory_, cpu_memory_, stream_buffer_}, device{device_} { - if (!device.HasFastBufferSubData()) { +void BufferCacheRuntime::BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, + u32 stride) { + if (index >= max_attributes) { return; } - - static constexpr GLsizeiptr size = static_cast<GLsizeiptr>(Maxwell::MaxConstBufferSize); - glCreateBuffers(static_cast<GLsizei>(std::size(cbufs)), std::data(cbufs)); - for (const GLuint cbuf : cbufs) { - glNamedBufferData(cbuf, size, nullptr, GL_STREAM_DRAW); + if (has_unified_vertex_buffers) { + buffer.MakeResident(GL_READ_ONLY); + glBindVertexBuffer(index, 0, 0, static_cast<GLsizei>(stride)); + glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, index, + buffer.HostGpuAddr() + offset, static_cast<GLsizeiptr>(size)); + } else { + glBindVertexBuffer(index, buffer.Handle(), static_cast<GLintptr>(offset), + static_cast<GLsizei>(stride)); } } -OGLBufferCache::~OGLBufferCache() { - glDeleteBuffers(static_cast<GLsizei>(std::size(cbufs)), std::data(cbufs)); +void BufferCacheRuntime::BindUniformBuffer(size_t stage, u32 binding_index, Buffer& buffer, + u32 offset, u32 size) { + if (use_assembly_shaders) { + GLuint handle; + if (offset != 0) { + handle = copy_uniforms[stage][binding_index].handle; + glCopyNamedBufferSubData(buffer.Handle(), handle, offset, 0, size); + } else { + handle = buffer.Handle(); + } + glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, + static_cast<GLsizeiptr>(size)); + } else { + const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer; + const GLuint binding = base_binding + binding_index; + glBindBufferRange(GL_UNIFORM_BUFFER, binding, buffer.Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); + } } -std::shared_ptr<Buffer> OGLBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) { - return std::make_shared<Buffer>(device, cpu_addr, size); +void BufferCacheRuntime::BindComputeUniformBuffer(u32 binding_index, Buffer& buffer, u32 offset, + u32 size) { + if (use_assembly_shaders) { + GLuint handle; + if (offset != 0) { + handle = copy_compute_uniforms[binding_index].handle; + glCopyNamedBufferSubData(buffer.Handle(), handle, offset, 0, size); + } else { + handle = buffer.Handle(); + } + glBindBufferRangeNV(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding_index, handle, 0, + static_cast<GLsizeiptr>(size)); + } else { + glBindBufferRange(GL_UNIFORM_BUFFER, binding_index, buffer.Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); + } } -OGLBufferCache::BufferInfo OGLBufferCache::GetEmptyBuffer(std::size_t) { - return {0, 0, 0}; +void BufferCacheRuntime::BindStorageBuffer(size_t stage, u32 binding_index, Buffer& buffer, + u32 offset, u32 size, bool is_written) { + if (use_assembly_shaders) { + const BindlessSSBO ssbo{ + .address = buffer.HostGpuAddr() + offset, + .length = static_cast<GLsizei>(size), + .padding = 0, + }; + buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY); + glProgramLocalParametersI4uivNV(PROGRAM_LUT[stage], binding_index, 1, + reinterpret_cast<const GLuint*>(&ssbo)); + } else { + const GLuint base_binding = device.GetBaseBindings(stage).shader_storage_buffer; + const GLuint binding = base_binding + binding_index; + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, buffer.Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); + } } -OGLBufferCache::BufferInfo OGLBufferCache::ConstBufferUpload(const void* raw_pointer, - std::size_t size) { - DEBUG_ASSERT(cbuf_cursor < std::size(cbufs)); - const GLuint cbuf = cbufs[cbuf_cursor++]; +void BufferCacheRuntime::BindComputeStorageBuffer(u32 binding_index, Buffer& buffer, u32 offset, + u32 size, bool is_written) { + if (use_assembly_shaders) { + const BindlessSSBO ssbo{ + .address = buffer.HostGpuAddr() + offset, + .length = static_cast<GLsizei>(size), + .padding = 0, + }; + buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY); + glProgramLocalParametersI4uivNV(GL_COMPUTE_PROGRAM_NV, binding_index, 1, + reinterpret_cast<const GLuint*>(&ssbo)); + } else if (size == 0) { + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, 0, 0, 0); + } else { + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, buffer.Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); + } +} - glNamedBufferSubData(cbuf, 0, static_cast<GLsizeiptr>(size), raw_pointer); - return {cbuf, 0, 0}; +void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, Buffer& buffer, u32 offset, + u32 size) { + glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, index, buffer.Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index 17ee90316..d8b20a9af 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -5,79 +5,157 @@ #pragma once #include <array> -#include <memory> +#include <span> +#include "common/alignment.h" #include "common/common_types.h" +#include "common/dynamic_library.h" #include "video_core/buffer_cache/buffer_cache.h" -#include "video_core/engines/maxwell_3d.h" +#include "video_core/rasterizer_interface.h" +#include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_stream_buffer.h" -namespace Core { -class System; -} - namespace OpenGL { -class Device; -class OGLStreamBuffer; -class RasterizerOpenGL; -class StateTracker; +class BufferCacheRuntime; -class Buffer : public VideoCommon::BufferBlock { +class Buffer : public VideoCommon::BufferBase<VideoCore::RasterizerInterface> { public: - explicit Buffer(const Device& device_, VAddr cpu_addr_, std::size_t size_); - ~Buffer(); + explicit Buffer(BufferCacheRuntime&, VideoCore::RasterizerInterface& rasterizer, VAddr cpu_addr, + u64 size_bytes); + explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams); - void Upload(std::size_t offset, std::size_t data_size, const u8* data); + void ImmediateUpload(size_t offset, std::span<const u8> data) noexcept; - void Download(std::size_t offset, std::size_t data_size, u8* data); + void ImmediateDownload(size_t offset, std::span<u8> data) noexcept; - void CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset, - std::size_t copy_size); + void MakeResident(GLenum access) noexcept; - GLuint Handle() const noexcept { - return gl_buffer.handle; + [[nodiscard]] GLuint64EXT HostGpuAddr() const noexcept { + return address; } - u64 Address() const noexcept { - return gpu_address; + [[nodiscard]] GLuint Handle() const noexcept { + return buffer.handle; } private: - OGLBuffer gl_buffer; - OGLBuffer read_buffer; - u64 gpu_address = 0; + GLuint64EXT address = 0; + OGLBuffer buffer; + GLenum current_residency_access = GL_NONE; }; -using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>; -class OGLBufferCache final : public GenericBufferCache { +class BufferCacheRuntime { + friend Buffer; + public: - explicit OGLBufferCache(VideoCore::RasterizerInterface& rasterizer, - Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, - const Device& device, OGLStreamBuffer& stream_buffer, - StateTracker& state_tracker); - ~OGLBufferCache(); + static constexpr u8 INVALID_BINDING = std::numeric_limits<u8>::max(); + + explicit BufferCacheRuntime(const Device& device_); + + void CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer, + std::span<const VideoCommon::BufferCopy> copies); + + void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size); + + void BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride); + + void BindUniformBuffer(size_t stage, u32 binding_index, Buffer& buffer, u32 offset, u32 size); + + void BindComputeUniformBuffer(u32 binding_index, Buffer& buffer, u32 offset, u32 size); + + void BindStorageBuffer(size_t stage, u32 binding_index, Buffer& buffer, u32 offset, u32 size, + bool is_written); + + void BindComputeStorageBuffer(u32 binding_index, Buffer& buffer, u32 offset, u32 size, + bool is_written); + + void BindTransformFeedbackBuffer(u32 index, Buffer& buffer, u32 offset, u32 size); + + void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) { + if (use_assembly_shaders) { + const GLuint handle = fast_uniforms[stage][binding_index].handle; + const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size); + glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, gl_size); + } else { + const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer; + const GLuint binding = base_binding + binding_index; + glBindBufferRange(GL_UNIFORM_BUFFER, binding, + fast_uniforms[stage][binding_index].handle, 0, + static_cast<GLsizeiptr>(size)); + } + } - BufferInfo GetEmptyBuffer(std::size_t) override; + void PushFastUniformBuffer(size_t stage, u32 binding_index, std::span<const u8> data) { + if (use_assembly_shaders) { + glProgramBufferParametersIuivNV( + PABO_LUT[stage], binding_index, 0, + static_cast<GLsizei>(data.size_bytes() / sizeof(GLuint)), + reinterpret_cast<const GLuint*>(data.data())); + } else { + glNamedBufferSubData(fast_uniforms[stage][binding_index].handle, 0, + static_cast<GLsizeiptr>(data.size_bytes()), data.data()); + } + } - void Acquire() noexcept { - cbuf_cursor = 0; + std::span<u8> BindMappedUniformBuffer(size_t stage, u32 binding_index, u32 size) noexcept { + const auto [mapped_span, offset] = stream_buffer->Request(static_cast<size_t>(size)); + const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer; + const GLuint binding = base_binding + binding_index; + glBindBufferRange(GL_UNIFORM_BUFFER, binding, stream_buffer->Handle(), + static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size)); + return mapped_span; } -protected: - std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override; + [[nodiscard]] const GLvoid* IndexOffset() const noexcept { + return reinterpret_cast<const GLvoid*>(static_cast<uintptr_t>(index_buffer_offset)); + } - BufferInfo ConstBufferUpload(const void* raw_pointer, std::size_t size) override; + [[nodiscard]] bool HasFastBufferSubData() const noexcept { + return has_fast_buffer_sub_data; + } private: - static constexpr std::size_t NUM_CBUFS = Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers * - Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram; + static constexpr std::array PABO_LUT{ + GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, + GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV, + GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV, + }; const Device& device; - std::size_t cbuf_cursor = 0; - std::array<GLuint, NUM_CBUFS> cbufs{}; + bool has_fast_buffer_sub_data = false; + bool use_assembly_shaders = false; + bool has_unified_vertex_buffers = false; + + u32 max_attributes = 0; + + std::optional<StreamBuffer> stream_buffer; + + std::array<std::array<OGLBuffer, VideoCommon::NUM_GRAPHICS_UNIFORM_BUFFERS>, + VideoCommon::NUM_STAGES> + fast_uniforms; + std::array<std::array<OGLBuffer, VideoCommon::NUM_GRAPHICS_UNIFORM_BUFFERS>, + VideoCommon::NUM_STAGES> + copy_uniforms; + std::array<OGLBuffer, VideoCommon::NUM_COMPUTE_UNIFORM_BUFFERS> copy_compute_uniforms; + + u32 index_buffer_offset = 0; +}; + +struct BufferCacheParams { + using Runtime = OpenGL::BufferCacheRuntime; + using Buffer = OpenGL::Buffer; + + static constexpr bool IS_OPENGL = true; + static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = true; + static constexpr bool HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT = true; + static constexpr bool NEEDS_BIND_UNIFORM_INDEX = true; + static constexpr bool NEEDS_BIND_STORAGE_INDEX = true; + static constexpr bool USE_MEMORY_MAPS = false; }; +using BufferCache = VideoCommon::BufferCache<BufferCacheParams>; + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index 04c267ee4..48d5c4a5e 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -21,9 +21,7 @@ #include "video_core/renderer_opengl/gl_resource_manager.h" namespace OpenGL { - namespace { - // One uniform block is reserved for emulation purposes constexpr u32 ReservedUniformBlocks = 1; @@ -197,11 +195,13 @@ bool IsASTCSupported() { const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED"); return nsight || HasExtension(extensions, "GL_EXT_debug_tool"); } - } // Anonymous namespace -Device::Device() - : max_uniform_buffers{BuildMaxUniformBuffers()}, base_bindings{BuildBaseBindings()} { +Device::Device() { + if (!GLAD_GL_VERSION_4_6) { + LOG_ERROR(Render_OpenGL, "OpenGL 4.6 is not available"); + throw std::runtime_error{"Insufficient version"}; + } const std::string_view vendor = reinterpret_cast<const char*>(glGetString(GL_VENDOR)); const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION)); const std::vector extensions = GetExtensions(); @@ -217,6 +217,9 @@ Device::Device() "Beta driver 443.24 is known to have issues. There might be performance issues."); disable_fast_buffer_sub_data = true; } + + max_uniform_buffers = BuildMaxUniformBuffers(); + base_bindings = BuildBaseBindings(); uniform_buffer_alignment = GetInteger<size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT); shader_storage_alignment = GetInteger<size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT); max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS); diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index 9141de635..ee053776d 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -10,11 +10,9 @@ namespace OpenGL { -static constexpr u32 EmulationUniformBlockBinding = 0; - -class Device final { +class Device { public: - struct BaseBindings final { + struct BaseBindings { u32 uniform_buffer{}; u32 shader_storage_buffer{}; u32 sampler{}; diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp index 3e9c922f5..151290101 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.cpp +++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp @@ -47,7 +47,7 @@ void GLInnerFence::Wait() { FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, TextureCache& texture_cache_, - OGLBufferCache& buffer_cache_, QueryCache& query_cache_) + BufferCache& buffer_cache_, QueryCache& query_cache_) : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_} {} Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) { diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h index 30dbee613..e714aa115 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.h +++ b/src/video_core/renderer_opengl/gl_fence_manager.h @@ -32,14 +32,13 @@ private: }; using Fence = std::shared_ptr<GLInnerFence>; -using GenericFenceManager = - VideoCommon::FenceManager<Fence, TextureCache, OGLBufferCache, QueryCache>; +using GenericFenceManager = VideoCommon::FenceManager<Fence, TextureCache, BufferCache, QueryCache>; class FenceManagerOpenGL final : public GenericFenceManager { public: - explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, - TextureCache& texture_cache_, OGLBufferCache& buffer_cache_, - QueryCache& query_cache_); + explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, + TextureCache& texture_cache, BufferCache& buffer_cache, + QueryCache& query_cache); protected: Fence CreateFence(u32 value, bool is_stubbed) override; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index ea4ca9a82..418644108 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -38,34 +38,21 @@ namespace OpenGL { using Maxwell = Tegra::Engines::Maxwell3D::Regs; +using GLvec4 = std::array<GLfloat, 4>; using Tegra::Engines::ShaderType; using VideoCore::Surface::PixelFormat; using VideoCore::Surface::SurfaceTarget; using VideoCore::Surface::SurfaceType; -MICROPROFILE_DEFINE(OpenGL_VAO, "OpenGL", "Vertex Format Setup", MP_RGB(128, 128, 192)); -MICROPROFILE_DEFINE(OpenGL_VB, "OpenGL", "Vertex Buffer Setup", MP_RGB(128, 128, 192)); -MICROPROFILE_DEFINE(OpenGL_Shader, "OpenGL", "Shader Setup", MP_RGB(128, 128, 192)); -MICROPROFILE_DEFINE(OpenGL_UBO, "OpenGL", "Const Buffer Setup", MP_RGB(128, 128, 192)); -MICROPROFILE_DEFINE(OpenGL_Index, "OpenGL", "Index Buffer Setup", MP_RGB(128, 128, 192)); -MICROPROFILE_DEFINE(OpenGL_Texture, "OpenGL", "Texture Setup", MP_RGB(128, 128, 192)); -MICROPROFILE_DEFINE(OpenGL_Framebuffer, "OpenGL", "Framebuffer Setup", MP_RGB(128, 128, 192)); MICROPROFILE_DEFINE(OpenGL_Drawing, "OpenGL", "Drawing", MP_RGB(128, 128, 192)); +MICROPROFILE_DEFINE(OpenGL_Clears, "OpenGL", "Clears", MP_RGB(128, 128, 192)); MICROPROFILE_DEFINE(OpenGL_Blits, "OpenGL", "Blits", MP_RGB(128, 128, 192)); -MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Mgmt", MP_RGB(100, 255, 100)); -MICROPROFILE_DEFINE(OpenGL_PrimitiveAssembly, "OpenGL", "Prim Asmbl", MP_RGB(255, 100, 100)); +MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Management", MP_RGB(100, 255, 100)); namespace { -constexpr size_t NUM_CONST_BUFFERS_PER_STAGE = 18; -constexpr size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE = - NUM_CONST_BUFFERS_PER_STAGE * Maxwell::MaxConstBufferSize; -constexpr size_t TOTAL_CONST_BUFFER_BYTES = - NUM_CONST_BUFFERS_BYTES_PER_STAGE * Maxwell::MaxShaderStage; - constexpr size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16; -constexpr size_t NUM_SUPPORTED_VERTEX_BINDINGS = 16; struct TextureHandle { constexpr TextureHandle(u32 data, bool via_header_index) { @@ -101,20 +88,6 @@ TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index); } -std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer, - const ConstBufferEntry& entry) { - if (!entry.IsIndirect()) { - return entry.GetSize(); - } - if (buffer.size > Maxwell::MaxConstBufferSize) { - LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", buffer.size, - Maxwell::MaxConstBufferSize); - return Maxwell::MaxConstBufferSize; - } - - return buffer.size; -} - /// Translates hardware transform feedback indices /// @param location Hardware location /// @return Pair of ARB_transform_feedback3 token stream first and third arguments @@ -147,14 +120,6 @@ void oglEnable(GLenum cap, bool state) { (state ? glEnable : glDisable)(cap); } -void UpdateBindlessSSBOs(GLenum target, const BindlessSSBO* ssbos, size_t num_ssbos) { - if (num_ssbos == 0) { - return; - } - glProgramLocalParametersI4uivNV(target, 0, static_cast<GLsizei>(num_ssbos), - reinterpret_cast<const GLuint*>(ssbos)); -} - ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) { if (entry.is_buffer) { return ImageViewType::Buffer; @@ -201,44 +166,28 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra : 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_), - stream_buffer(device, 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_, gpu, maxwell3d, kepler_compute, gpu_memory, device), query_cache(*this, maxwell3d, gpu_memory), - buffer_cache(*this, gpu_memory, cpu_memory_, device, stream_buffer, state_tracker), fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache), async_shaders(emu_window_) { - unified_uniform_buffer.Create(); - glNamedBufferStorage(unified_uniform_buffer.handle, TOTAL_CONST_BUFFER_BYTES, nullptr, 0); - - if (device.UseAssemblyShaders()) { - glCreateBuffers(static_cast<GLsizei>(staging_cbufs.size()), staging_cbufs.data()); - for (const GLuint cbuf : staging_cbufs) { - glNamedBufferStorage(cbuf, static_cast<GLsizeiptr>(Maxwell::MaxConstBufferSize), - nullptr, 0); - } - } if (device.UseAsynchronousShaders()) { async_shaders.AllocateWorkers(); } } -RasterizerOpenGL::~RasterizerOpenGL() { - if (device.UseAssemblyShaders()) { - glDeleteBuffers(static_cast<GLsizei>(staging_cbufs.size()), staging_cbufs.data()); - } -} +RasterizerOpenGL::~RasterizerOpenGL() = default; -void RasterizerOpenGL::SetupVertexFormat() { +void RasterizerOpenGL::SyncVertexFormats() { auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::VertexFormats]) { return; } flags[Dirty::VertexFormats] = false; - MICROPROFILE_SCOPE(OpenGL_VAO); - // Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL. Enables // the first 16 vertex attributes always, as we don't know which ones are actually used until // shader time. Note, Tegra technically supports 32, but we're capping this to 16 for now to @@ -274,55 +223,7 @@ void RasterizerOpenGL::SetupVertexFormat() { } } -void RasterizerOpenGL::SetupVertexBuffer() { - auto& flags = maxwell3d.dirty.flags; - if (!flags[Dirty::VertexBuffers]) { - return; - } - flags[Dirty::VertexBuffers] = false; - - MICROPROFILE_SCOPE(OpenGL_VB); - - const bool use_unified_memory = device.HasVertexBufferUnifiedMemory(); - - // Upload all guest vertex arrays sequentially to our buffer - const auto& regs = maxwell3d.regs; - for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_BINDINGS; ++index) { - if (!flags[Dirty::VertexBuffer0 + index]) { - continue; - } - flags[Dirty::VertexBuffer0 + index] = false; - - const auto& vertex_array = regs.vertex_array[index]; - if (!vertex_array.IsEnabled()) { - continue; - } - - const GPUVAddr start = vertex_array.StartAddress(); - const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress(); - ASSERT(end >= start); - - const GLuint gl_index = static_cast<GLuint>(index); - const u64 size = end - start; - if (size == 0) { - glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride); - if (use_unified_memory) { - glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index, 0, 0); - } - continue; - } - const auto info = buffer_cache.UploadMemory(start, size); - if (use_unified_memory) { - glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride); - glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index, - info.address + info.offset, size); - } else { - glBindVertexBuffer(gl_index, info.handle, info.offset, vertex_array.stride); - } - } -} - -void RasterizerOpenGL::SetupVertexInstances() { +void RasterizerOpenGL::SyncVertexInstances() { auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::VertexInstances]) { return; @@ -343,17 +244,7 @@ void RasterizerOpenGL::SetupVertexInstances() { } } -GLintptr RasterizerOpenGL::SetupIndexBuffer() { - MICROPROFILE_SCOPE(OpenGL_Index); - const auto& regs = maxwell3d.regs; - const std::size_t size = CalculateIndexBufferSize(); - const auto info = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size); - glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, info.handle); - return info.offset; -} - -void RasterizerOpenGL::SetupShaders() { - MICROPROFILE_SCOPE(OpenGL_Shader); +void RasterizerOpenGL::SetupShaders(bool is_indexed) { u32 clip_distances = 0; std::array<Shader*, Maxwell::MaxShaderStage> shaders{}; @@ -410,11 +301,19 @@ void RasterizerOpenGL::SetupShaders() { const size_t stage = index == 0 ? 0 : index - 1; shaders[stage] = shader; - SetupDrawConstBuffers(stage, shader); - SetupDrawGlobalMemory(stage, shader); SetupDrawTextures(shader, stage); SetupDrawImages(shader, stage); + buffer_cache.SetEnabledUniformBuffers(stage, shader->GetEntries().enabled_uniform_buffers); + + buffer_cache.UnbindGraphicsStorageBuffers(stage); + u32 ssbo_index = 0; + for (const auto& buffer : shader->GetEntries().global_memory_entries) { + buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index, + buffer.cbuf_offset, buffer.is_written); + ++ssbo_index; + } + // Workaround for Intel drivers. // When a clip distance is enabled but not set in the shader it crops parts of the screen // (sometimes it's half the screen, sometimes three quarters). To avoid this, enable the @@ -430,43 +329,26 @@ void RasterizerOpenGL::SetupShaders() { SyncClipEnabled(clip_distances); maxwell3d.dirty.flags[Dirty::Shaders] = false; + buffer_cache.UpdateGraphicsBuffers(is_indexed); + const std::span indices_span(image_view_indices.data(), image_view_indices.size()); texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + buffer_cache.BindHostGeometryBuffers(is_indexed); + size_t image_view_index = 0; size_t texture_index = 0; size_t image_index = 0; for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { const Shader* const shader = shaders[stage]; - if (shader) { - const auto base = device.GetBaseBindings(stage); - BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index, - texture_index, image_index); - } - } -} - -std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const { - const auto& regs = maxwell3d.regs; - - std::size_t size = 0; - for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) { - if (!regs.vertex_array[index].IsEnabled()) + if (!shader) { continue; - - const GPUVAddr start = regs.vertex_array[index].StartAddress(); - const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress(); - - size += end - start; - ASSERT(end >= start); + } + buffer_cache.BindHostStageBuffers(stage); + const auto& base = device.GetBaseBindings(stage); + BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index, + texture_index, image_index); } - - return size; -} - -std::size_t RasterizerOpenGL::CalculateIndexBufferSize() const { - return static_cast<std::size_t>(maxwell3d.regs.index_array.count) * - static_cast<std::size_t>(maxwell3d.regs.index_array.FormatSizeInBytes()); } void RasterizerOpenGL::LoadDiskResources(u64 title_id, const std::atomic_bool& stop_loading, @@ -475,6 +357,7 @@ void RasterizerOpenGL::LoadDiskResources(u64 title_id, const std::atomic_bool& s } void RasterizerOpenGL::Clear() { + MICROPROFILE_SCOPE(OpenGL_Clears); if (!maxwell3d.ShouldExecute()) { return; } @@ -525,11 +408,9 @@ void RasterizerOpenGL::Clear() { } UNIMPLEMENTED_IF(regs.clear_flags.viewport); - { - auto lock = texture_cache.AcquireLock(); - texture_cache.UpdateRenderTargets(true); - state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); - } + std::scoped_lock lock{texture_cache.mutex}; + texture_cache.UpdateRenderTargets(true); + state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); if (use_color) { glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color); @@ -541,7 +422,6 @@ void RasterizerOpenGL::Clear() { } else if (use_stencil) { glClearBufferiv(GL_STENCIL, 0, ®s.clear_stencil); } - ++num_queued_commands; } @@ -550,75 +430,12 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { query_cache.UpdateCounters(); - SyncViewport(); - SyncRasterizeEnable(); - SyncPolygonModes(); - SyncColorMask(); - SyncFragmentColorClampState(); - SyncMultiSampleState(); - SyncDepthTestState(); - SyncDepthClamp(); - SyncStencilTestState(); - SyncBlendState(); - SyncLogicOpState(); - SyncCullMode(); - SyncPrimitiveRestart(); - SyncScissorTest(); - SyncPointState(); - SyncLineState(); - SyncPolygonOffset(); - SyncAlphaTest(); - SyncFramebufferSRGB(); - - buffer_cache.Acquire(); - current_cbuf = 0; - - std::size_t buffer_size = CalculateVertexArraysSize(); - - // Add space for index buffer - if (is_indexed) { - buffer_size = Common::AlignUp(buffer_size, 4) + CalculateIndexBufferSize(); - } - - // Uniform space for the 5 shader stages - buffer_size = - Common::AlignUp<std::size_t>(buffer_size, 4) + - (sizeof(MaxwellUniformData) + device.GetUniformBufferAlignment()) * Maxwell::MaxShaderStage; - - // Add space for at least 18 constant buffers - buffer_size += Maxwell::MaxConstBuffers * - (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment()); - - // Prepare the vertex array. - buffer_cache.Map(buffer_size); - - // Prepare vertex array format. - SetupVertexFormat(); - - // Upload vertex and index data. - SetupVertexBuffer(); - SetupVertexInstances(); - GLintptr index_buffer_offset = 0; - if (is_indexed) { - index_buffer_offset = SetupIndexBuffer(); - } - - // Setup emulation uniform buffer. - if (!device.UseAssemblyShaders()) { - MaxwellUniformData ubo; - ubo.SetFromRegs(maxwell3d); - const auto info = - buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment()); - glBindBufferRange(GL_UNIFORM_BUFFER, EmulationUniformBlockBinding, info.handle, info.offset, - static_cast<GLsizeiptr>(sizeof(ubo))); - } + SyncState(); // Setup shaders and their used resources. - auto lock = texture_cache.AcquireLock(); - SetupShaders(); + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; + SetupShaders(is_indexed); - // Signal the buffer cache that we are not going to upload more things. - buffer_cache.Unmap(); texture_cache.UpdateRenderTargets(false); state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); program_manager.BindGraphicsPipeline(); @@ -632,7 +449,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { 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 GLvoid* offset = reinterpret_cast<const GLvoid*>(index_buffer_offset); + const GLvoid* const offset = buffer_cache_runtime.IndexOffset(); 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); @@ -672,22 +489,22 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { } void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { - buffer_cache.Acquire(); - current_cbuf = 0; - Shader* const kernel = shader_cache.GetComputeKernel(code_addr); - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; BindComputeTextures(kernel); - const size_t buffer_size = Tegra::Engines::KeplerCompute::NumConstBuffers * - (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment()); - buffer_cache.Map(buffer_size); - - SetupComputeConstBuffers(kernel); - SetupComputeGlobalMemory(kernel); - - buffer_cache.Unmap(); + const auto& entries = kernel->GetEntries(); + buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers); + buffer_cache.UnbindComputeStorageBuffers(); + u32 ssbo_index = 0; + for (const auto& buffer : entries.global_memory_entries) { + buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset, + buffer.is_written); + ++ssbo_index; + } + buffer_cache.UpdateComputeBuffers(); + buffer_cache.BindHostComputeBuffers(); const auto& launch_desc = kepler_compute.launch_description; glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z); @@ -703,6 +520,12 @@ void RasterizerOpenGL::Query(GPUVAddr gpu_addr, VideoCore::QueryType type, query_cache.Query(gpu_addr, type, timestamp); } +void RasterizerOpenGL::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, + u32 size) { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.BindGraphicsUniformBuffer(stage, index, gpu_addr, size); +} + void RasterizerOpenGL::FlushAll() {} void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) { @@ -711,19 +534,23 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) { return; } { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.DownloadMemory(addr, size); } - buffer_cache.FlushRegion(addr, size); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.DownloadMemory(addr, size); + } query_cache.FlushRegion(addr, size); } bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) { + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; if (!Settings::IsGPULevelHigh()) { - return buffer_cache.MustFlushRegion(addr, size); + return buffer_cache.IsRegionGpuModified(addr, size); } return texture_cache.IsRegionGpuModified(addr, size) || - buffer_cache.MustFlushRegion(addr, size); + buffer_cache.IsRegionGpuModified(addr, size); } void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) { @@ -732,11 +559,14 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) { return; } { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); } + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.WriteMemory(addr, size); + } shader_cache.InvalidateRegion(addr, size); - buffer_cache.InvalidateRegion(addr, size); query_cache.InvalidateRegion(addr, size); } @@ -745,26 +575,35 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { if (addr == 0 || size == 0) { return; } + shader_cache.OnCPUWrite(addr, size); { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); } - shader_cache.OnCPUWrite(addr, size); - buffer_cache.OnCPUWrite(addr, size); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.CachedWriteMemory(addr, size); + } } void RasterizerOpenGL::SyncGuestHost() { MICROPROFILE_SCOPE(OpenGL_CacheManagement); - buffer_cache.SyncGuestHost(); shader_cache.SyncGuestHost(); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.FlushCachedWrites(); + } } void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.UnmapMemory(addr, size); } - buffer_cache.OnCPUWrite(addr, size); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.WriteMemory(addr, size); + } shader_cache.OnCPUWrite(addr, size); } @@ -799,14 +638,7 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) { } void RasterizerOpenGL::WaitForIdle() { - // Place a barrier on everything that is not framebuffer related. - // This is related to another flag that is not currently implemented. - glMemoryBarrier(GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT | GL_ELEMENT_ARRAY_BARRIER_BIT | - GL_UNIFORM_BARRIER_BIT | GL_TEXTURE_FETCH_BARRIER_BIT | - GL_SHADER_IMAGE_ACCESS_BARRIER_BIT | GL_COMMAND_BARRIER_BIT | - GL_PIXEL_BUFFER_BARRIER_BIT | GL_TEXTURE_UPDATE_BARRIER_BIT | - GL_BUFFER_UPDATE_BARRIER_BIT | GL_TRANSFORM_FEEDBACK_BARRIER_BIT | - GL_SHADER_STORAGE_BARRIER_BIT | GL_QUERY_BUFFER_BARRIER_BIT); + glMemoryBarrier(GL_ALL_BARRIER_BITS); } void RasterizerOpenGL::FragmentBarrier() { @@ -831,18 +663,21 @@ void RasterizerOpenGL::TickFrame() { num_queued_commands = 0; fence_manager.TickFrame(); - buffer_cache.TickFrame(); { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.TickFrame(); } + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.TickFrame(); + } } bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src, const Tegra::Engines::Fermi2D::Surface& dst, const Tegra::Engines::Fermi2D::Config& copy_config) { MICROPROFILE_SCOPE(OpenGL_Blits); - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.BlitImage(dst, src, copy_config); return true; } @@ -854,7 +689,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, } MICROPROFILE_SCOPE(OpenGL_CacheManagement); - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; ImageView* const image_view{texture_cache.TryFindFramebufferImageView(framebuffer_addr)}; if (!image_view) { return false; @@ -921,166 +756,6 @@ void RasterizerOpenGL::BindTextures(const ShaderEntries& entries, GLuint base_te } } -void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* shader) { - static constexpr std::array PARAMETER_LUT{ - GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, - GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV, - GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV, - }; - MICROPROFILE_SCOPE(OpenGL_UBO); - const auto& stages = maxwell3d.state.shader_stages; - const auto& shader_stage = stages[stage_index]; - const auto& entries = shader->GetEntries(); - const bool use_unified = entries.use_unified_uniforms; - const std::size_t base_unified_offset = stage_index * NUM_CONST_BUFFERS_BYTES_PER_STAGE; - - const auto base_bindings = device.GetBaseBindings(stage_index); - u32 binding = device.UseAssemblyShaders() ? 0 : base_bindings.uniform_buffer; - for (const auto& entry : entries.const_buffers) { - const u32 index = entry.GetIndex(); - const auto& buffer = shader_stage.const_buffers[index]; - SetupConstBuffer(PARAMETER_LUT[stage_index], binding, buffer, entry, use_unified, - base_unified_offset + index * Maxwell::MaxConstBufferSize); - ++binding; - } - if (use_unified) { - const u32 index = static_cast<u32>(base_bindings.shader_storage_buffer + - entries.global_memory_entries.size()); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, - base_unified_offset, NUM_CONST_BUFFERS_BYTES_PER_STAGE); - } -} - -void RasterizerOpenGL::SetupComputeConstBuffers(Shader* kernel) { - MICROPROFILE_SCOPE(OpenGL_UBO); - const auto& launch_desc = kepler_compute.launch_description; - const auto& entries = kernel->GetEntries(); - const bool use_unified = entries.use_unified_uniforms; - - u32 binding = 0; - for (const auto& entry : entries.const_buffers) { - const auto& config = launch_desc.const_buffer_config[entry.GetIndex()]; - const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value(); - Tegra::Engines::ConstBufferInfo buffer; - buffer.address = config.Address(); - buffer.size = config.size; - buffer.enabled = mask[entry.GetIndex()]; - SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding, buffer, entry, - use_unified, entry.GetIndex() * Maxwell::MaxConstBufferSize); - ++binding; - } - if (use_unified) { - const GLuint index = static_cast<GLuint>(entries.global_memory_entries.size()); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, 0, - NUM_CONST_BUFFERS_BYTES_PER_STAGE); - } -} - -void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding, - const Tegra::Engines::ConstBufferInfo& buffer, - const ConstBufferEntry& entry, bool use_unified, - std::size_t unified_offset) { - if (!buffer.enabled) { - // Set values to zero to unbind buffers - if (device.UseAssemblyShaders()) { - glBindBufferRangeNV(stage, entry.GetIndex(), 0, 0, 0); - } else { - glBindBufferRange(GL_UNIFORM_BUFFER, binding, 0, 0, sizeof(float)); - } - return; - } - - // Align the actual size so it ends up being a multiple of vec4 to meet the OpenGL std140 - // UBO alignment requirements. - const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4)); - - const bool fast_upload = !use_unified && device.HasFastBufferSubData(); - - const std::size_t alignment = use_unified ? 4 : device.GetUniformBufferAlignment(); - const GPUVAddr gpu_addr = buffer.address; - auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, false, fast_upload); - - if (device.UseAssemblyShaders()) { - UNIMPLEMENTED_IF(use_unified); - if (info.offset != 0) { - const GLuint staging_cbuf = staging_cbufs[current_cbuf++]; - glCopyNamedBufferSubData(info.handle, staging_cbuf, info.offset, 0, size); - info.handle = staging_cbuf; - info.offset = 0; - } - glBindBufferRangeNV(stage, binding, info.handle, info.offset, size); - return; - } - - if (use_unified) { - glCopyNamedBufferSubData(info.handle, unified_uniform_buffer.handle, info.offset, - unified_offset, size); - } else { - glBindBufferRange(GL_UNIFORM_BUFFER, binding, info.handle, info.offset, size); - } -} - -void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader) { - static constexpr std::array TARGET_LUT = { - GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV, - GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV, - }; - const auto& cbufs{maxwell3d.state.shader_stages[stage_index]}; - const auto& entries{shader->GetEntries().global_memory_entries}; - - std::array<BindlessSSBO, 32> ssbos; - ASSERT(entries.size() < ssbos.size()); - - const bool assembly_shaders = device.UseAssemblyShaders(); - u32 binding = assembly_shaders ? 0 : device.GetBaseBindings(stage_index).shader_storage_buffer; - for (const auto& entry : entries) { - const GPUVAddr addr{cbufs.const_buffers[entry.cbuf_index].address + entry.cbuf_offset}; - const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)}; - const u32 size{gpu_memory.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding, entry, gpu_addr, size, &ssbos[binding]); - ++binding; - } - if (assembly_shaders) { - UpdateBindlessSSBOs(TARGET_LUT[stage_index], ssbos.data(), entries.size()); - } -} - -void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) { - const auto& cbufs{kepler_compute.launch_description.const_buffer_config}; - const auto& entries{kernel->GetEntries().global_memory_entries}; - - std::array<BindlessSSBO, 32> ssbos; - ASSERT(entries.size() < ssbos.size()); - - u32 binding = 0; - for (const auto& entry : entries) { - const GPUVAddr addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset}; - const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)}; - const u32 size{gpu_memory.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding, entry, gpu_addr, size, &ssbos[binding]); - ++binding; - } - if (device.UseAssemblyShaders()) { - UpdateBindlessSSBOs(GL_COMPUTE_PROGRAM_NV, ssbos.data(), ssbos.size()); - } -} - -void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, - GPUVAddr gpu_addr, size_t size, BindlessSSBO* ssbo) { - const size_t alignment{device.GetShaderStorageBufferAlignment()}; - const auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written); - if (device.UseAssemblyShaders()) { - *ssbo = BindlessSSBO{ - .address = static_cast<GLuint64EXT>(info.address + info.offset), - .length = static_cast<GLsizei>(size), - .padding = 0, - }; - } else { - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset, - static_cast<GLsizeiptr>(size)); - } -} - void RasterizerOpenGL::SetupDrawTextures(const Shader* shader, size_t stage_index) { const bool via_header_index = maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; @@ -1128,6 +803,30 @@ void RasterizerOpenGL::SetupComputeImages(const Shader* shader) { } } +void RasterizerOpenGL::SyncState() { + SyncViewport(); + SyncRasterizeEnable(); + SyncPolygonModes(); + SyncColorMask(); + SyncFragmentColorClampState(); + SyncMultiSampleState(); + SyncDepthTestState(); + SyncDepthClamp(); + SyncStencilTestState(); + SyncBlendState(); + SyncLogicOpState(); + SyncCullMode(); + SyncPrimitiveRestart(); + SyncScissorTest(); + SyncPointState(); + SyncLineState(); + SyncPolygonOffset(); + SyncAlphaTest(); + SyncFramebufferSRGB(); + SyncVertexFormats(); + SyncVertexInstances(); +} + void RasterizerOpenGL::SyncViewport() { auto& flags = maxwell3d.dirty.flags; const auto& regs = maxwell3d.regs; @@ -1163,9 +862,11 @@ void RasterizerOpenGL::SyncViewport() { if (regs.screen_y_control.y_negate != 0) { flip_y = !flip_y; } - glClipControl(flip_y ? GL_UPPER_LEFT : GL_LOWER_LEFT, - regs.depth_mode == Maxwell::DepthMode::ZeroToOne ? GL_ZERO_TO_ONE - : GL_NEGATIVE_ONE_TO_ONE); + const bool is_zero_to_one = regs.depth_mode == Maxwell::DepthMode::ZeroToOne; + const GLenum origin = flip_y ? GL_UPPER_LEFT : GL_LOWER_LEFT; + const GLenum depth = is_zero_to_one ? GL_ZERO_TO_ONE : GL_NEGATIVE_ONE_TO_ONE; + state_tracker.ClipControl(origin, depth); + state_tracker.SetYNegate(regs.screen_y_control.y_negate != 0); } if (dirty_viewport) { @@ -1649,36 +1350,13 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) { if (regs.tfb_enabled == 0) { return; } - if (device.UseAssemblyShaders()) { SyncTransformFeedback(); } - UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry)); - - for (std::size_t index = 0; index < Maxwell::NumTransformFeedbackBuffers; ++index) { - const auto& binding = regs.tfb_bindings[index]; - if (!binding.buffer_enable) { - if (enabled_transform_feedback_buffers[index]) { - glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, static_cast<GLuint>(index), 0, 0, - 0); - } - enabled_transform_feedback_buffers[index] = false; - continue; - } - enabled_transform_feedback_buffers[index] = true; - - auto& tfb_buffer = transform_feedback_buffers[index]; - tfb_buffer.Create(); - - const GLuint handle = tfb_buffer.handle; - const std::size_t size = binding.buffer_size; - glNamedBufferData(handle, static_cast<GLsizeiptr>(size), nullptr, GL_STREAM_COPY); - glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, static_cast<GLuint>(index), handle, 0, - static_cast<GLsizeiptr>(size)); - } + UNIMPLEMENTED_IF(primitive_mode != GL_POINTS); // We may have to call BeginTransformFeedbackNV here since they seem to call different // implementations on Nvidia's driver (the pointer is different) but we are using @@ -1692,23 +1370,7 @@ void RasterizerOpenGL::EndTransformFeedback() { if (regs.tfb_enabled == 0) { return; } - glEndTransformFeedback(); - - for (std::size_t index = 0; index < Maxwell::NumTransformFeedbackBuffers; ++index) { - const auto& binding = regs.tfb_bindings[index]; - if (!binding.buffer_enable) { - continue; - } - UNIMPLEMENTED_IF(binding.buffer_offset != 0); - - const GLuint handle = transform_feedback_buffers[index].handle; - const GPUVAddr gpu_addr = binding.Address(); - const std::size_t size = binding.buffer_size; - const auto info = buffer_cache.UploadMemory(gpu_addr, size, 4, true); - glCopyNamedBufferSubData(handle, info.handle, 0, info.offset, - static_cast<GLsizeiptr>(size)); - } } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 82e03e677..3745cf637 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -30,7 +30,6 @@ #include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" -#include "video_core/renderer_opengl/gl_stream_buffer.h" #include "video_core/renderer_opengl/gl_texture_cache.h" #include "video_core/shader/async_shaders.h" #include "video_core/textures/texture.h" @@ -72,6 +71,7 @@ public: void DispatchCompute(GPUVAddr code_addr) override; void ResetCounter(VideoCore::QueryType type) override; void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override; + void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override; void FlushAll() override; void FlushRegion(VAddr addr, u64 size) override; bool MustFlushRegion(VAddr addr, u64 size) override; @@ -119,27 +119,6 @@ private: void BindTextures(const ShaderEntries& entries, GLuint base_texture, GLuint base_image, size_t& image_view_index, size_t& texture_index, size_t& image_index); - /// Configures the current constbuffers to use for the draw command. - void SetupDrawConstBuffers(std::size_t stage_index, Shader* shader); - - /// Configures the current constbuffers to use for the kernel invocation. - void SetupComputeConstBuffers(Shader* kernel); - - /// Configures a constant buffer. - void SetupConstBuffer(GLenum stage, u32 binding, const Tegra::Engines::ConstBufferInfo& buffer, - const ConstBufferEntry& entry, bool use_unified, - std::size_t unified_offset); - - /// Configures the current global memory entries to use for the draw command. - void SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader); - - /// Configures the current global memory entries to use for the kernel invocation. - void SetupComputeGlobalMemory(Shader* kernel); - - /// Configures a global memory buffer. - void SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr, - size_t size, BindlessSSBO* ssbo); - /// Configures the current textures to use for the draw command. void SetupDrawTextures(const Shader* shader, size_t stage_index); @@ -152,6 +131,9 @@ private: /// Configures images in a compute shader. void SetupComputeImages(const Shader* shader); + /// Syncs state to match guest's + void SyncState(); + /// Syncs the viewport and depth range to match the guest state void SyncViewport(); @@ -215,6 +197,12 @@ private: /// Syncs the framebuffer sRGB state to match the guest state void SyncFramebufferSRGB(); + /// Syncs vertex formats to match the guest state + void SyncVertexFormats(); + + /// Syncs vertex instances to match the guest state + void SyncVertexInstances(); + /// Syncs transform feedback state to match guest state /// @note Only valid on assembly shaders void SyncTransformFeedback(); @@ -225,19 +213,7 @@ private: /// End a transform feedback void EndTransformFeedback(); - std::size_t CalculateVertexArraysSize() const; - - std::size_t CalculateIndexBufferSize() const; - - /// Updates the current vertex format - void SetupVertexFormat(); - - void SetupVertexBuffer(); - void SetupVertexInstances(); - - GLintptr SetupIndexBuffer(); - - void SetupShaders(); + void SetupShaders(bool is_indexed); Tegra::GPU& gpu; Tegra::Engines::Maxwell3D& maxwell3d; @@ -249,12 +225,12 @@ private: ProgramManager& program_manager; StateTracker& state_tracker; - OGLStreamBuffer stream_buffer; TextureCacheRuntime texture_cache_runtime; TextureCache texture_cache; + BufferCacheRuntime buffer_cache_runtime; + BufferCache buffer_cache; ShaderCacheOpenGL shader_cache; QueryCache query_cache; - OGLBufferCache buffer_cache; FenceManagerOpenGL fence_manager; VideoCommon::Shader::AsyncShaders async_shaders; @@ -262,20 +238,8 @@ private: boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices; std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids; boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles; - std::array<GLuint, MAX_TEXTURES> texture_handles; - std::array<GLuint, MAX_IMAGES> image_handles; - - std::array<OGLBuffer, Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers> - transform_feedback_buffers; - std::bitset<Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers> - enabled_transform_feedback_buffers; - - static constexpr std::size_t NUM_CONSTANT_BUFFERS = - Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers * - Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram; - std::array<GLuint, NUM_CONSTANT_BUFFERS> staging_cbufs{}; - std::size_t current_cbuf = 0; - OGLBuffer unified_uniform_buffer; + std::array<GLuint, MAX_TEXTURES> texture_handles{}; + std::array<GLuint, MAX_IMAGES> image_handles{}; /// Number of commands queued to the OpenGL driver. Resetted on flush. std::size_t num_queued_commands = 0; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp index 0e34a0f20..3428e5e21 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.cpp +++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp @@ -171,12 +171,6 @@ void OGLBuffer::Release() { handle = 0; } -void OGLBuffer::MakeStreamCopy(std::size_t buffer_size) { - ASSERT_OR_EXECUTE((handle != 0 && buffer_size != 0), { return; }); - - glNamedBufferData(handle, buffer_size, nullptr, GL_STREAM_COPY); -} - void OGLSync::Create() { if (handle != 0) return; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.h b/src/video_core/renderer_opengl/gl_resource_manager.h index f48398669..552d79db4 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.h +++ b/src/video_core/renderer_opengl/gl_resource_manager.h @@ -234,9 +234,6 @@ public: /// Deletes the internal OpenGL resource void Release(); - // Converts the buffer into a stream copy buffer with a fixed size - void MakeStreamCopy(std::size_t buffer_size); - GLuint handle = 0; }; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index c35b71b6b..ac78d344c 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -64,7 +64,7 @@ using TextureIR = std::variant<TextureOffset, TextureDerivates, TextureArgument> constexpr u32 MAX_CONSTBUFFER_SCALARS = static_cast<u32>(Maxwell::MaxConstBufferSize) / sizeof(u32); constexpr u32 MAX_CONSTBUFFER_ELEMENTS = MAX_CONSTBUFFER_SCALARS / sizeof(u32); -constexpr std::string_view CommonDeclarations = R"(#define ftoi floatBitsToInt +constexpr std::string_view COMMON_DECLARATIONS = R"(#define ftoi floatBitsToInt #define ftou floatBitsToUint #define itof intBitsToFloat #define utof uintBitsToFloat @@ -77,10 +77,6 @@ bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {{ const float fswzadd_modifiers_a[] = float[4](-1.0f, 1.0f, -1.0f, 0.0f ); const float fswzadd_modifiers_b[] = float[4](-1.0f, -1.0f, 1.0f, -1.0f ); - -layout (std140, binding = {}) uniform vs_config {{ - float y_direction; -}}; )"; class ShaderWriter final { @@ -402,13 +398,6 @@ std::string FlowStackTopName(MetaStackClass stack) { return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack)); } -bool UseUnifiedUniforms(const Device& device, const ShaderIR& ir, ShaderType stage) { - const u32 num_ubos = static_cast<u32>(ir.GetConstantBuffers().size()); - // We waste one UBO for emulation - const u32 num_available_ubos = device.GetMaxUniformBuffers(stage) - 1; - return num_ubos > num_available_ubos; -} - struct GenericVaryingDescription { std::string name; u8 first_element = 0; @@ -420,9 +409,8 @@ public: explicit GLSLDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_, ShaderType stage_, std::string_view identifier_, std::string_view suffix_) - : device{device_}, ir{ir_}, registry{registry_}, stage{stage_}, identifier{identifier_}, - suffix{suffix_}, header{ir.GetHeader()}, use_unified_uniforms{ - UseUnifiedUniforms(device_, ir_, stage_)} { + : device{device_}, ir{ir_}, registry{registry_}, stage{stage_}, + identifier{identifier_}, suffix{suffix_}, header{ir.GetHeader()} { if (stage != ShaderType::Compute) { transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo()); } @@ -516,7 +504,8 @@ private: if (!identifier.empty()) { code.AddLine("// {}", identifier); } - code.AddLine("#version 440 {}", ir.UsesLegacyVaryings() ? "compatibility" : "core"); + const bool use_compatibility = ir.UsesLegacyVaryings() || ir.UsesYNegate(); + code.AddLine("#version 440 {}", use_compatibility ? "compatibility" : "core"); code.AddLine("#extension GL_ARB_separate_shader_objects : enable"); if (device.HasShaderBallot()) { code.AddLine("#extension GL_ARB_shader_ballot : require"); @@ -542,7 +531,7 @@ private: code.AddNewLine(); - code.AddLine(CommonDeclarations, EmulationUniformBlockBinding); + code.AddLine(COMMON_DECLARATIONS); } void DeclareVertex() { @@ -865,17 +854,6 @@ private: } void DeclareConstantBuffers() { - if (use_unified_uniforms) { - const u32 binding = device.GetBaseBindings(stage).shader_storage_buffer + - static_cast<u32>(ir.GetGlobalMemory().size()); - code.AddLine("layout (std430, binding = {}) readonly buffer UnifiedUniforms {{", - binding); - code.AddLine(" uint cbufs[];"); - code.AddLine("}};"); - code.AddNewLine(); - return; - } - u32 binding = device.GetBaseBindings(stage).uniform_buffer; for (const auto& [index, info] : ir.GetConstantBuffers()) { const u32 num_elements = Common::DivCeil(info.GetSize(), 4 * sizeof(u32)); @@ -1081,29 +1059,17 @@ private: if (const auto cbuf = std::get_if<CbufNode>(&*node)) { const Node offset = cbuf->GetOffset(); - const u32 base_unified_offset = cbuf->GetIndex() * MAX_CONSTBUFFER_SCALARS; if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) { // Direct access const u32 offset_imm = immediate->GetValue(); ASSERT_MSG(offset_imm % 4 == 0, "Unaligned cbuf direct access"); - if (use_unified_uniforms) { - return {fmt::format("cbufs[{}]", base_unified_offset + offset_imm / 4), - Type::Uint}; - } else { - return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()), - offset_imm / (4 * 4), (offset_imm / 4) % 4), - Type::Uint}; - } - } - - // Indirect access - if (use_unified_uniforms) { - return {fmt::format("cbufs[{} + ({} >> 2)]", base_unified_offset, - Visit(offset).AsUint()), + return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()), + offset_imm / (4 * 4), (offset_imm / 4) % 4), Type::Uint}; } + // Indirect access const std::string final_offset = code.GenerateTemporary(); code.AddLine("uint {} = {} >> 2;", final_offset, Visit(offset).AsUint()); @@ -2293,7 +2259,6 @@ private: } } } - if (header.ps.omap.depth) { // The depth output is always 2 registers after the last color output, and current_reg // already contains one past the last color register. @@ -2337,7 +2302,8 @@ private: } Expression YNegate(Operation operation) { - return {"y_direction", Type::Float}; + // Y_NEGATE is mapped to this uniform value + return {"gl_FrontMaterial.ambient.a", Type::Float}; } template <u32 element> @@ -2787,7 +2753,6 @@ private: const std::string_view identifier; const std::string_view suffix; const Header header; - const bool use_unified_uniforms; std::unordered_map<u8, VaryingTFB> transform_feedback; ShaderWriter code; @@ -3003,8 +2968,10 @@ ShaderEntries MakeEntries(const Device& device, const ShaderIR& ir, ShaderType s for (std::size_t i = 0; i < std::size(clip_distances); ++i) { entries.clip_distances = (clip_distances[i] ? 1U : 0U) << i; } + for (const auto& buffer : entries.const_buffers) { + entries.enabled_uniform_buffers |= 1U << buffer.GetIndex(); + } entries.shader_length = ir.GetLength(); - entries.use_unified_uniforms = UseUnifiedUniforms(device, ir, stage); return entries; } diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h index be68994bb..0397a000c 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.h +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h @@ -55,7 +55,7 @@ struct ShaderEntries { std::vector<ImageEntry> images; std::size_t shader_length{}; u32 clip_distances{}; - bool use_unified_uniforms{}; + u32 enabled_uniform_buffers{}; }; ShaderEntries MakeEntries(const Device& device, const VideoCommon::Shader::ShaderIR& ir, diff --git a/src/video_core/renderer_opengl/gl_state_tracker.cpp b/src/video_core/renderer_opengl/gl_state_tracker.cpp index 60e6fa39f..dbdf5230f 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.cpp +++ b/src/video_core/renderer_opengl/gl_state_tracker.cpp @@ -36,16 +36,10 @@ void SetupDirtyColorMasks(Tables& tables) { FillBlock(tables[1], OFF(color_mask), NUM(color_mask), ColorMasks); } -void SetupDirtyVertexArrays(Tables& tables) { - static constexpr std::size_t num_array = 3; +void SetupDirtyVertexInstances(Tables& tables) { static constexpr std::size_t instance_base_offset = 3; for (std::size_t i = 0; i < Regs::NumVertexArrays; ++i) { const std::size_t array_offset = OFF(vertex_array) + i * NUM(vertex_array[0]); - const std::size_t limit_offset = OFF(vertex_array_limit) + i * NUM(vertex_array_limit[0]); - - FillBlock(tables, array_offset, num_array, VertexBuffer0 + i, VertexBuffers); - FillBlock(tables, limit_offset, NUM(vertex_array_limit), VertexBuffer0 + i, VertexBuffers); - const std::size_t instance_array_offset = array_offset + instance_base_offset; tables[0][instance_array_offset] = static_cast<u8>(VertexInstance0 + i); tables[1][instance_array_offset] = VertexInstances; @@ -217,11 +211,11 @@ void SetupDirtyMisc(Tables& tables) { StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} { auto& dirty = gpu.Maxwell3D().dirty; auto& tables = dirty.tables; - SetupDirtyRenderTargets(tables); + SetupDirtyFlags(tables); SetupDirtyColorMasks(tables); SetupDirtyViewports(tables); SetupDirtyScissors(tables); - SetupDirtyVertexArrays(tables); + SetupDirtyVertexInstances(tables); SetupDirtyVertexFormat(tables); SetupDirtyShaders(tables); SetupDirtyPolygonModes(tables); @@ -241,19 +235,6 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} SetupDirtyClipControl(tables); SetupDirtyDepthClampEnabled(tables); SetupDirtyMisc(tables); - - auto& store = dirty.on_write_stores; - store[VertexBuffers] = true; - for (std::size_t i = 0; i < Regs::NumVertexArrays; ++i) { - store[VertexBuffer0 + i] = true; - } -} - -void StateTracker::InvalidateStreamBuffer() { - flags[Dirty::VertexBuffers] = true; - for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) { - flags[index] = true; - } } } // 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 574615d3c..94c905116 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.h +++ b/src/video_core/renderer_opengl/gl_state_tracker.h @@ -28,10 +28,6 @@ enum : u8 { VertexFormat0, VertexFormat31 = VertexFormat0 + 31, - VertexBuffers, - VertexBuffer0, - VertexBuffer31 = VertexBuffer0 + 31, - VertexInstances, VertexInstance0, VertexInstance31 = VertexInstance0 + 31, @@ -92,8 +88,6 @@ class StateTracker { public: explicit StateTracker(Tegra::GPU& gpu); - void InvalidateStreamBuffer(); - void BindIndexBuffer(GLuint new_index_buffer) { if (index_buffer == new_index_buffer) { return; @@ -110,13 +104,32 @@ public: glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer); } + void ClipControl(GLenum new_origin, GLenum new_depth) { + if (new_origin == origin && new_depth == depth) { + return; + } + origin = new_origin; + depth = new_depth; + glClipControl(origin, depth); + } + + void SetYNegate(bool new_y_negate) { + if (new_y_negate == y_negate) { + return; + } + // Y_NEGATE is mapped to gl_FrontMaterial.ambient.a + y_negate = new_y_negate; + const std::array ambient{0.0f, 0.0f, 0.0f, y_negate ? -1.0f : 1.0f}; + glMaterialfv(GL_FRONT, GL_AMBIENT, ambient.data()); + } + void NotifyScreenDrawVertexArray() { flags[OpenGL::Dirty::VertexFormats] = true; flags[OpenGL::Dirty::VertexFormat0 + 0] = true; flags[OpenGL::Dirty::VertexFormat0 + 1] = true; - flags[OpenGL::Dirty::VertexBuffers] = true; - flags[OpenGL::Dirty::VertexBuffer0] = true; + flags[VideoCommon::Dirty::VertexBuffers] = true; + flags[VideoCommon::Dirty::VertexBuffer0] = true; flags[OpenGL::Dirty::VertexInstances] = true; flags[OpenGL::Dirty::VertexInstance0 + 0] = true; @@ -202,6 +215,9 @@ private: GLuint framebuffer = 0; GLuint index_buffer = 0; + GLenum origin = GL_LOWER_LEFT; + GLenum depth = GL_NEGATIVE_ONE_TO_ONE; + bool y_negate = false; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp index e0819cdf2..77b3ee0fe 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.cpp +++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp @@ -1,70 +1,64 @@ -// Copyright 2018 Citra Emulator Project +// Copyright 2021 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include <tuple> -#include <vector> +#include <array> +#include <memory> +#include <span> + +#include <glad/glad.h> #include "common/alignment.h" #include "common/assert.h" -#include "common/microprofile.h" -#include "video_core/renderer_opengl/gl_device.h" -#include "video_core/renderer_opengl/gl_state_tracker.h" #include "video_core/renderer_opengl/gl_stream_buffer.h" -MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning", - MP_RGB(128, 128, 192)); - namespace OpenGL { -OGLStreamBuffer::OGLStreamBuffer(const Device& device, StateTracker& state_tracker_) - : state_tracker{state_tracker_} { - gl_buffer.Create(); - - static constexpr GLbitfield flags = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT; - glNamedBufferStorage(gl_buffer.handle, BUFFER_SIZE, nullptr, flags); - mapped_ptr = static_cast<u8*>( - glMapNamedBufferRange(gl_buffer.handle, 0, BUFFER_SIZE, flags | GL_MAP_FLUSH_EXPLICIT_BIT)); - - if (device.UseAssemblyShaders() || device.HasVertexBufferUnifiedMemory()) { - glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_ONLY); - glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address); +StreamBuffer::StreamBuffer() { + static constexpr GLenum flags = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_COHERENT_BIT; + buffer.Create(); + glObjectLabel(GL_BUFFER, buffer.handle, -1, "Stream Buffer"); + glNamedBufferStorage(buffer.handle, STREAM_BUFFER_SIZE, nullptr, flags); + mapped_pointer = + static_cast<u8*>(glMapNamedBufferRange(buffer.handle, 0, STREAM_BUFFER_SIZE, flags)); + for (OGLSync& sync : fences) { + sync.Create(); } } -OGLStreamBuffer::~OGLStreamBuffer() { - glUnmapNamedBuffer(gl_buffer.handle); - gl_buffer.Release(); -} - -std::pair<u8*, GLintptr> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) { - ASSERT(size <= BUFFER_SIZE); - ASSERT(alignment <= BUFFER_SIZE); - mapped_size = size; - - if (alignment > 0) { - buffer_pos = Common::AlignUp<std::size_t>(buffer_pos, alignment); +std::pair<std::span<u8>, size_t> StreamBuffer::Request(size_t size) noexcept { + ASSERT(size < REGION_SIZE); + for (size_t region = Region(used_iterator), region_end = Region(iterator); region < region_end; + ++region) { + fences[region].Create(); } + used_iterator = iterator; - if (buffer_pos + size > BUFFER_SIZE) { - MICROPROFILE_SCOPE(OpenGL_StreamBuffer); - glInvalidateBufferData(gl_buffer.handle); - state_tracker.InvalidateStreamBuffer(); - - buffer_pos = 0; + for (size_t region = Region(free_iterator) + 1, + region_end = std::min(Region(iterator + size) + 1, NUM_SYNCS); + region < region_end; ++region) { + glClientWaitSync(fences[region].handle, 0, GL_TIMEOUT_IGNORED); + fences[region].Release(); } - - return std::make_pair(mapped_ptr + buffer_pos, buffer_pos); -} - -void OGLStreamBuffer::Unmap(GLsizeiptr size) { - ASSERT(size <= mapped_size); - - if (size > 0) { - glFlushMappedNamedBufferRange(gl_buffer.handle, buffer_pos, size); + if (iterator + size >= free_iterator) { + free_iterator = iterator + size; } - - buffer_pos += size; + if (iterator + size > STREAM_BUFFER_SIZE) { + for (size_t region = Region(used_iterator); region < NUM_SYNCS; ++region) { + fences[region].Create(); + } + used_iterator = 0; + iterator = 0; + free_iterator = size; + + for (size_t region = 0, region_end = Region(size); region <= region_end; ++region) { + glClientWaitSync(fences[region].handle, 0, GL_TIMEOUT_IGNORED); + fences[region].Release(); + } + } + const size_t offset = iterator; + iterator = Common::AlignUp(iterator + size, MAX_ALIGNMENT); + return {std::span(mapped_pointer + offset, size), offset}; } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.h b/src/video_core/renderer_opengl/gl_stream_buffer.h index dd9cf67eb..6dbb6bfba 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.h +++ b/src/video_core/renderer_opengl/gl_stream_buffer.h @@ -1,9 +1,12 @@ -// Copyright 2018 Citra Emulator Project +// Copyright 2021 yuzu Emulator Project // Licensed under GPLv2 or any later version // Refer to the license.txt file included. #pragma once +#include <array> +#include <memory> +#include <span> #include <utility> #include <glad/glad.h> @@ -13,48 +16,35 @@ namespace OpenGL { -class Device; -class StateTracker; +class StreamBuffer { + static constexpr size_t STREAM_BUFFER_SIZE = 64 * 1024 * 1024; + static constexpr size_t NUM_SYNCS = 16; + static constexpr size_t REGION_SIZE = STREAM_BUFFER_SIZE / NUM_SYNCS; + static constexpr size_t MAX_ALIGNMENT = 256; + static_assert(STREAM_BUFFER_SIZE % MAX_ALIGNMENT == 0); + static_assert(STREAM_BUFFER_SIZE % NUM_SYNCS == 0); + static_assert(REGION_SIZE % MAX_ALIGNMENT == 0); -class OGLStreamBuffer : private NonCopyable { public: - explicit OGLStreamBuffer(const Device& device, StateTracker& state_tracker_); - ~OGLStreamBuffer(); - - /* - * Allocates a linear chunk of memory in the GPU buffer with at least "size" bytes - * and the optional alignment requirement. - * If the buffer is full, the whole buffer is reallocated which invalidates old chunks. - * The return values are the pointer to the new chunk, and the offset within the buffer. - * The actual used size must be specified on unmapping the chunk. - */ - std::pair<u8*, GLintptr> Map(GLsizeiptr size, GLintptr alignment = 0); - - void Unmap(GLsizeiptr size); - - GLuint Handle() const { - return gl_buffer.handle; - } + explicit StreamBuffer(); - u64 Address() const { - return gpu_address; - } + [[nodiscard]] std::pair<std::span<u8>, size_t> Request(size_t size) noexcept; - GLsizeiptr Size() const noexcept { - return BUFFER_SIZE; + [[nodiscard]] GLuint Handle() const noexcept { + return buffer.handle; } private: - static constexpr GLsizeiptr BUFFER_SIZE = 256 * 1024 * 1024; - - StateTracker& state_tracker; - - OGLBuffer gl_buffer; + [[nodiscard]] static size_t Region(size_t offset) noexcept { + return offset / REGION_SIZE; + } - GLuint64EXT gpu_address = 0; - GLintptr buffer_pos = 0; - GLsizeiptr mapped_size = 0; - u8* mapped_ptr = nullptr; + size_t iterator = 0; + size_t used_iterator = 0; + size_t free_iterator = 0; + u8* mapped_pointer = nullptr; + OGLBuffer buffer; + std::array<OGLSync, NUM_SYNCS> fences; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 546cb6d00..12434db67 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -398,9 +398,6 @@ void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) { } // Anonymous namespace -ImageBufferMap::ImageBufferMap(GLuint handle_, u8* map, size_t size, OGLSync* sync_) - : span(map, size), sync{sync_}, handle{handle_} {} - ImageBufferMap::~ImageBufferMap() { if (sync) { sync->Create(); @@ -487,11 +484,11 @@ void TextureCacheRuntime::Finish() { glFinish(); } -ImageBufferMap TextureCacheRuntime::MapUploadBuffer(size_t size) { +ImageBufferMap TextureCacheRuntime::UploadStagingBuffer(size_t size) { return upload_buffers.RequestMap(size, true); } -ImageBufferMap TextureCacheRuntime::MapDownloadBuffer(size_t size) { +ImageBufferMap TextureCacheRuntime::DownloadStagingBuffer(size_t size) { return download_buffers.RequestMap(size, false); } @@ -553,15 +550,14 @@ void TextureCacheRuntime::BlitFramebuffer(Framebuffer* dst, Framebuffer* src, } void TextureCacheRuntime::AccelerateImageUpload(Image& image, const ImageBufferMap& map, - size_t buffer_offset, std::span<const SwizzleParameters> swizzles) { switch (image.info.type) { case ImageType::e2D: - return util_shaders.BlockLinearUpload2D(image, map, buffer_offset, swizzles); + return util_shaders.BlockLinearUpload2D(image, map, swizzles); case ImageType::e3D: - return util_shaders.BlockLinearUpload3D(image, map, buffer_offset, swizzles); + return util_shaders.BlockLinearUpload3D(image, map, swizzles); case ImageType::Linear: - return util_shaders.PitchUpload(image, map, buffer_offset, swizzles); + return util_shaders.PitchUpload(image, map, swizzles); default: UNREACHABLE(); break; @@ -596,7 +592,11 @@ ImageBufferMap TextureCacheRuntime::StagingBuffers::RequestMap(size_t requested_ bool insert_fence) { const size_t index = RequestBuffer(requested_size); OGLSync* const sync = insert_fence ? &syncs[index] : nullptr; - return ImageBufferMap(buffers[index].handle, maps[index], requested_size, sync); + return ImageBufferMap{ + .mapped_span = std::span(maps[index], requested_size), + .sync = sync, + .buffer = buffers[index].handle, + }; } size_t TextureCacheRuntime::StagingBuffers::RequestBuffer(size_t requested_size) { @@ -709,10 +709,10 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_, } } -void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, +void Image::UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies) { - glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.Handle()); - glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, buffer_offset, unswizzled_size_bytes); + glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.buffer); + glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, map.offset, unswizzled_size_bytes); glPixelStorei(GL_UNPACK_ALIGNMENT, 1); @@ -728,23 +728,23 @@ void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, current_image_height = copy.buffer_image_height; glPixelStorei(GL_UNPACK_IMAGE_HEIGHT, current_image_height); } - CopyBufferToImage(copy, buffer_offset); + CopyBufferToImage(copy, map.offset); } } -void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, +void Image::UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferCopy> copies) { for (const VideoCommon::BufferCopy& copy : copies) { - glCopyNamedBufferSubData(map.Handle(), buffer.handle, copy.src_offset + buffer_offset, + glCopyNamedBufferSubData(map.buffer, buffer.handle, copy.src_offset + map.offset, copy.dst_offset, copy.size); } } -void Image::DownloadMemory(ImageBufferMap& map, size_t buffer_offset, +void Image::DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies) { glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API - glBindBuffer(GL_PIXEL_PACK_BUFFER, map.Handle()); + glBindBuffer(GL_PIXEL_PACK_BUFFER, map.buffer); glPixelStorei(GL_PACK_ALIGNMENT, 1); u32 current_row_length = std::numeric_limits<u32>::max(); @@ -759,7 +759,38 @@ void Image::DownloadMemory(ImageBufferMap& map, size_t buffer_offset, current_image_height = copy.buffer_image_height; glPixelStorei(GL_PACK_IMAGE_HEIGHT, current_image_height); } - CopyImageToBuffer(copy, buffer_offset); + CopyImageToBuffer(copy, map.offset); + } +} + +GLuint Image::StorageHandle() noexcept { + switch (info.format) { + case PixelFormat::A8B8G8R8_SRGB: + case PixelFormat::B8G8R8A8_SRGB: + case PixelFormat::BC1_RGBA_SRGB: + case PixelFormat::BC2_SRGB: + case PixelFormat::BC3_SRGB: + case PixelFormat::BC7_SRGB: + case PixelFormat::ASTC_2D_4X4_SRGB: + case PixelFormat::ASTC_2D_8X8_SRGB: + case PixelFormat::ASTC_2D_8X5_SRGB: + case PixelFormat::ASTC_2D_5X4_SRGB: + case PixelFormat::ASTC_2D_5X5_SRGB: + case PixelFormat::ASTC_2D_10X8_SRGB: + case PixelFormat::ASTC_2D_6X6_SRGB: + case PixelFormat::ASTC_2D_10X10_SRGB: + case PixelFormat::ASTC_2D_12X12_SRGB: + case PixelFormat::ASTC_2D_8X6_SRGB: + case PixelFormat::ASTC_2D_6X5_SRGB: + if (store_view.handle != 0) { + return store_view.handle; + } + store_view.Create(); + glTextureView(store_view.handle, ImageTarget(info), texture.handle, GL_RGBA8, 0, + info.resources.levels, 0, info.resources.layers); + return store_view.handle; + default: + return texture.handle; } } diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 15b7c3676..a6172f009 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -31,23 +31,13 @@ using VideoCommon::NUM_RT; using VideoCommon::Offset2D; using VideoCommon::RenderTargets; -class ImageBufferMap { -public: - explicit ImageBufferMap(GLuint handle, u8* map, size_t size, OGLSync* sync); +struct ImageBufferMap { ~ImageBufferMap(); - GLuint Handle() const noexcept { - return handle; - } - - std::span<u8> Span() const noexcept { - return span; - } - -private: - std::span<u8> span; + std::span<u8> mapped_span; + size_t offset = 0; OGLSync* sync; - GLuint handle; + GLuint buffer; }; struct FormatProperties { @@ -69,9 +59,9 @@ public: void Finish(); - ImageBufferMap MapUploadBuffer(size_t size); + ImageBufferMap UploadStagingBuffer(size_t size); - ImageBufferMap MapDownloadBuffer(size_t size); + ImageBufferMap DownloadStagingBuffer(size_t size); void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies); @@ -89,7 +79,7 @@ public: Tegra::Engines::Fermi2D::Filter filter, Tegra::Engines::Fermi2D::Operation operation); - void AccelerateImageUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset, + void AccelerateImageUpload(Image& image, const ImageBufferMap& map, std::span<const VideoCommon::SwizzleParameters> swizzles); void InsertUploadMemoryBarrier(); @@ -148,14 +138,14 @@ public: explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr); - void UploadMemory(const ImageBufferMap& map, size_t buffer_offset, + void UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies); - void UploadMemory(const ImageBufferMap& map, size_t buffer_offset, - std::span<const VideoCommon::BufferCopy> copies); + void UploadMemory(const ImageBufferMap& map, std::span<const VideoCommon::BufferCopy> copies); + + void DownloadMemory(ImageBufferMap& map, std::span<const VideoCommon::BufferImageCopy> copies); - void DownloadMemory(ImageBufferMap& map, size_t buffer_offset, - std::span<const VideoCommon::BufferImageCopy> copies); + GLuint StorageHandle() noexcept; GLuint Handle() const noexcept { return texture.handle; @@ -167,8 +157,8 @@ private: void CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); OGLTexture texture; - OGLTextureView store_view; OGLBuffer buffer; + OGLTextureView store_view; GLenum gl_internal_format = GL_NONE; GLenum gl_format = GL_NONE; GLenum gl_type = GL_NONE; diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index cbccfdeb4..f7ad8f370 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -4,23 +4,10 @@ #pragma once -#include <array> #include <glad/glad.h> -#include "common/common_types.h" -#include "common/logging/log.h" #include "video_core/engines/maxwell_3d.h" -namespace OpenGL { - -using GLvec2 = std::array<GLfloat, 2>; -using GLvec3 = std::array<GLfloat, 3>; -using GLvec4 = std::array<GLfloat, 4>; - -using GLuvec2 = std::array<GLuint, 2>; -using GLuvec3 = std::array<GLuint, 3>; -using GLuvec4 = std::array<GLuint, 4>; - -namespace MaxwellToGL { +namespace OpenGL::MaxwellToGL { using Maxwell = Tegra::Engines::Maxwell3D::Regs; @@ -317,26 +304,6 @@ inline GLenum BlendFunc(Maxwell::Blend::Factor factor) { return GL_ZERO; } -inline GLenum SwizzleSource(Tegra::Texture::SwizzleSource source) { - switch (source) { - case Tegra::Texture::SwizzleSource::Zero: - return GL_ZERO; - case Tegra::Texture::SwizzleSource::R: - return GL_RED; - case Tegra::Texture::SwizzleSource::G: - return GL_GREEN; - case Tegra::Texture::SwizzleSource::B: - return GL_BLUE; - case Tegra::Texture::SwizzleSource::A: - return GL_ALPHA; - case Tegra::Texture::SwizzleSource::OneInt: - case Tegra::Texture::SwizzleSource::OneFloat: - return GL_ONE; - } - UNIMPLEMENTED_MSG("Unimplemented swizzle source={}", source); - return GL_ZERO; -} - inline GLenum ComparisonOp(Maxwell::ComparisonOp comparison) { switch (comparison) { case Maxwell::ComparisonOp::Never: @@ -493,5 +460,4 @@ inline GLenum ViewportSwizzle(Maxwell::ViewportSwizzle swizzle) { return GL_VIEWPORT_SWIZZLE_POSITIVE_X_NV + static_cast<GLenum>(swizzle); } -} // namespace MaxwellToGL -} // namespace OpenGL +} // namespace OpenGL::MaxwellToGL diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 21159e498..9d2acd4d9 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -29,9 +29,7 @@ #include "video_core/textures/decoders.h" namespace OpenGL { - namespace { - constexpr GLint PositionLocation = 0; constexpr GLint TexCoordLocation = 1; constexpr GLint ModelViewMatrixLocation = 0; @@ -124,7 +122,6 @@ void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum severit break; } } - } // Anonymous namespace RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_, @@ -132,7 +129,17 @@ 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_}, program_manager{device} {} + emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, state_tracker{gpu}, + 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) { + glEnable(GL_DEBUG_OUTPUT); + glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS); + glDebugMessageCallback(DebugHandler, nullptr); + } + AddTelemetryFields(); + InitOpenGLObjects(); +} RendererOpenGL::~RendererOpenGL() = default; @@ -148,7 +155,7 @@ void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { ++m_current_frame; - rasterizer->TickFrame(); + rasterizer.TickFrame(); context->SwapBuffers(); render_window.OnFrameDisplayed(); @@ -179,7 +186,7 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf framebuffer_crop_rect = framebuffer.crop_rect; const VAddr framebuffer_addr{framebuffer.address + framebuffer.offset}; - if (rasterizer->AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride)) { + if (rasterizer.AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride)) { return; } @@ -267,6 +274,7 @@ void RendererOpenGL::InitOpenGLObjects() { // Enable unified vertex attributes and query vertex buffer address when the driver supports it if (device.HasVertexBufferUnifiedMemory()) { glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV); + glEnableClientState(GL_ELEMENT_ARRAY_UNIFIED_NV); glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY); glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, @@ -289,14 +297,6 @@ void RendererOpenGL::AddTelemetryFields() { telemetry_session.AddField(user_system, "GPU_OpenGL_Version", std::string(gl_version)); } -void RendererOpenGL::CreateRasterizer() { - if (rasterizer) { - return; - } - rasterizer = std::make_unique<RasterizerOpenGL>(emu_window, gpu, cpu_memory, device, - screen_info, program_manager, state_tracker); -} - void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, const Tegra::FramebufferConfig& framebuffer) { texture.width = framebuffer.width; @@ -407,6 +407,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { program_manager.BindHostPipeline(pipeline.handle); + state_tracker.ClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE); glEnable(GL_CULL_FACE); if (screen_info.display_srgb) { glEnable(GL_FRAMEBUFFER_SRGB); @@ -425,7 +426,6 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { glCullFace(GL_BACK); glFrontFace(GL_CW); glColorMaski(0, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); - glClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE); glViewportIndexedf(0, 0.0f, 0.0f, static_cast<GLfloat>(layout.width), static_cast<GLfloat>(layout.height)); glDepthRangeIndexed(0, 0.0, 0.0); @@ -497,25 +497,4 @@ void RendererOpenGL::RenderScreenshot() { renderer_settings.screenshot_requested = false; } -bool RendererOpenGL::Init() { - if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) { - glEnable(GL_DEBUG_OUTPUT); - glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS); - glDebugMessageCallback(DebugHandler, nullptr); - } - - AddTelemetryFields(); - - if (!GLAD_GL_VERSION_4_6) { - return false; - } - - InitOpenGLObjects(); - CreateRasterizer(); - - return true; -} - -void RendererOpenGL::ShutDown() {} - } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index 44e109794..cc19a110f 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -10,6 +10,7 @@ #include "common/math_util.h" #include "video_core/renderer_base.h" #include "video_core/renderer_opengl/gl_device.h" +#include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" @@ -63,18 +64,18 @@ public: std::unique_ptr<Core::Frontend::GraphicsContext> context_); ~RendererOpenGL() override; - bool Init() override; - void ShutDown() override; void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override; + VideoCore::RasterizerInterface* ReadRasterizer() override { + return &rasterizer; + } + private: /// Initializes the OpenGL state and creates persistent objects. void InitOpenGLObjects(); void AddTelemetryFields(); - void CreateRasterizer(); - void ConfigureFramebufferTexture(TextureInfo& texture, const Tegra::FramebufferConfig& framebuffer); @@ -98,8 +99,10 @@ private: Core::Memory::Memory& cpu_memory; Tegra::GPU& gpu; - const Device device; - StateTracker state_tracker{gpu}; + Device device; + StateTracker state_tracker; + ProgramManager program_manager; + RasterizerOpenGL rasterizer; // OpenGL object IDs OGLSampler present_sampler; @@ -115,9 +118,6 @@ private: /// Display information for Switch screen ScreenInfo screen_info; - /// Global dummy shader pipeline - ProgramManager program_manager; - /// OpenGL framebuffer data std::vector<u8> gl_framebuffer_data; diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp index eb849cbf2..31ec68505 100644 --- a/src/video_core/renderer_opengl/util_shaders.cpp +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -63,7 +63,7 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_) UtilShaders::~UtilShaders() = default; -void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, size_t buffer_offset, +void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, std::span<const SwizzleParameters> swizzles) { static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1}; static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0; @@ -71,13 +71,13 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, s static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; program_manager.BindHostCompute(block_linear_unswizzle_2d_program.handle); - glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes); + glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes); glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle); const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format)); for (const SwizzleParameters& swizzle : swizzles) { const Extent3D num_tiles = swizzle.num_tiles; - const size_t input_offset = swizzle.buffer_offset + buffer_offset; + const size_t input_offset = swizzle.buffer_offset + map.offset; const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width); const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height); @@ -91,16 +91,16 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, s glUniform1ui(5, params.x_shift); glUniform1ui(6, params.block_height); glUniform1ui(7, params.block_height_mask); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(), - input_offset, image.guest_size_bytes - swizzle.buffer_offset); - glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0, + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0, GL_WRITE_ONLY, store_format); glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers); } program_manager.RestoreGuestCompute(); } -void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, size_t buffer_offset, +void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, std::span<const SwizzleParameters> swizzles) { static constexpr Extent3D WORKGROUP_SIZE{16, 8, 8}; @@ -108,14 +108,14 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, s static constexpr GLuint BINDING_INPUT_BUFFER = 1; static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; - glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes); + glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes); program_manager.BindHostCompute(block_linear_unswizzle_3d_program.handle); glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle); const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format)); for (const SwizzleParameters& swizzle : swizzles) { const Extent3D num_tiles = swizzle.num_tiles; - const size_t input_offset = swizzle.buffer_offset + buffer_offset; + const size_t input_offset = swizzle.buffer_offset + map.offset; const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width); const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height); @@ -132,16 +132,16 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, s glUniform1ui(7, params.block_height_mask); glUniform1ui(8, params.block_depth); glUniform1ui(9, params.block_depth_mask); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(), - input_offset, image.guest_size_bytes - swizzle.buffer_offset); - glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0, + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); + glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0, GL_WRITE_ONLY, store_format); glDispatchCompute(num_dispatches_x, num_dispatches_y, num_dispatches_z); } program_manager.RestoreGuestCompute(); } -void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset, +void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, std::span<const SwizzleParameters> swizzles) { static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1}; static constexpr GLuint BINDING_INPUT_BUFFER = 0; @@ -159,21 +159,22 @@ void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, size_t bu "Non-power of two images are not implemented"); program_manager.BindHostCompute(pitch_unswizzle_program.handle); - glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes); + glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes); glUniform2ui(LOC_ORIGIN, 0, 0); glUniform2i(LOC_DESTINATION, 0, 0); glUniform1ui(LOC_BYTES_PER_BLOCK, bytes_per_block); glUniform1ui(LOC_PITCH, pitch); - glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), 0, GL_FALSE, 0, GL_WRITE_ONLY, format); + glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), 0, GL_FALSE, 0, GL_WRITE_ONLY, + format); for (const SwizzleParameters& swizzle : swizzles) { const Extent3D num_tiles = swizzle.num_tiles; - const size_t input_offset = swizzle.buffer_offset + buffer_offset; + const size_t input_offset = swizzle.buffer_offset + map.offset; const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width); const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height); - glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(), - input_offset, image.guest_size_bytes - swizzle.buffer_offset); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset, + image.guest_size_bytes - swizzle.buffer_offset); glDispatchCompute(num_dispatches_x, num_dispatches_y, 1); } program_manager.RestoreGuestCompute(); @@ -195,9 +196,9 @@ void UtilShaders::CopyBC4(Image& dst_image, Image& src_image, std::span<const Im glUniform3ui(LOC_SRC_OFFSET, copy.src_offset.x, copy.src_offset.y, copy.src_offset.z); glUniform3ui(LOC_DST_OFFSET, copy.dst_offset.x, copy.dst_offset.y, copy.dst_offset.z); - glBindImageTexture(BINDING_INPUT_IMAGE, src_image.Handle(), copy.src_subresource.base_level, - GL_FALSE, 0, GL_READ_ONLY, GL_RG32UI); - glBindImageTexture(BINDING_OUTPUT_IMAGE, dst_image.Handle(), + glBindImageTexture(BINDING_INPUT_IMAGE, src_image.StorageHandle(), + copy.src_subresource.base_level, GL_FALSE, 0, GL_READ_ONLY, GL_RG32UI); + glBindImageTexture(BINDING_OUTPUT_IMAGE, dst_image.StorageHandle(), copy.dst_subresource.base_level, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGBA8UI); glDispatchCompute(copy.extent.width, copy.extent.height, copy.extent.depth); } diff --git a/src/video_core/renderer_opengl/util_shaders.h b/src/video_core/renderer_opengl/util_shaders.h index 359997255..7b1d16b09 100644 --- a/src/video_core/renderer_opengl/util_shaders.h +++ b/src/video_core/renderer_opengl/util_shaders.h @@ -15,21 +15,22 @@ namespace OpenGL { class Image; -class ImageBufferMap; class ProgramManager; +struct ImageBufferMap; + class UtilShaders { public: explicit UtilShaders(ProgramManager& program_manager); ~UtilShaders(); - void BlockLinearUpload2D(Image& image, const ImageBufferMap& map, size_t buffer_offset, + void BlockLinearUpload2D(Image& image, const ImageBufferMap& map, std::span<const VideoCommon::SwizzleParameters> swizzles); - void BlockLinearUpload3D(Image& image, const ImageBufferMap& map, size_t buffer_offset, + void BlockLinearUpload3D(Image& image, const ImageBufferMap& map, std::span<const VideoCommon::SwizzleParameters> swizzles); - void PitchUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset, + void PitchUpload(Image& image, const ImageBufferMap& map, std::span<const VideoCommon::SwizzleParameters> swizzles); void CopyBC4(Image& dst_image, Image& src_image, diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp index 5be6dabd9..362278f01 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.cpp @@ -12,14 +12,15 @@ #include "common/cityhash.h" #include "common/common_types.h" #include "video_core/renderer_vulkan/fixed_pipeline_state.h" +#include "video_core/renderer_vulkan/vk_state_tracker.h" namespace Vulkan { namespace { -constexpr std::size_t POINT = 0; -constexpr std::size_t LINE = 1; -constexpr std::size_t POLYGON = 2; +constexpr size_t POINT = 0; +constexpr size_t LINE = 1; +constexpr size_t POLYGON = 2; constexpr std::array POLYGON_OFFSET_ENABLE_LUT = { POINT, // Points LINE, // Lines @@ -40,10 +41,14 @@ constexpr std::array POLYGON_OFFSET_ENABLE_LUT = { } // Anonymous namespace -void FixedPipelineState::Fill(const Maxwell& regs, bool has_extended_dynamic_state) { - const std::array enabled_lut = {regs.polygon_offset_point_enable, - regs.polygon_offset_line_enable, - regs.polygon_offset_fill_enable}; +void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, + bool has_extended_dynamic_state) { + const Maxwell& regs = maxwell3d.regs; + const std::array enabled_lut{ + regs.polygon_offset_point_enable, + regs.polygon_offset_line_enable, + regs.polygon_offset_fill_enable, + }; const u32 topology_index = static_cast<u32>(regs.draw.topology.Value()); raw1 = 0; @@ -64,45 +69,53 @@ void FixedPipelineState::Fill(const Maxwell& regs, bool has_extended_dynamic_sta raw2 = 0; const auto test_func = - regs.alpha_test_enabled == 1 ? regs.alpha_test_func : Maxwell::ComparisonOp::Always; + regs.alpha_test_enabled != 0 ? regs.alpha_test_func : Maxwell::ComparisonOp::Always; alpha_test_func.Assign(PackComparisonOp(test_func)); early_z.Assign(regs.force_early_fragment_tests != 0 ? 1 : 0); alpha_test_ref = Common::BitCast<u32>(regs.alpha_test_ref); point_size = Common::BitCast<u32>(regs.point_size); - for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { - binding_divisors[index] = - regs.instanced_arrays.IsInstancingEnabled(index) ? regs.vertex_array[index].divisor : 0; + if (maxwell3d.dirty.flags[Dirty::InstanceDivisors]) { + maxwell3d.dirty.flags[Dirty::InstanceDivisors] = false; + for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { + const bool is_enabled = regs.instanced_arrays.IsInstancingEnabled(index); + binding_divisors[index] = is_enabled ? regs.vertex_array[index].divisor : 0; + } } - - for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { - const auto& input = regs.vertex_attrib_format[index]; - auto& attribute = attributes[index]; - attribute.raw = 0; - attribute.enabled.Assign(input.IsConstant() ? 0 : 1); - attribute.buffer.Assign(input.buffer); - attribute.offset.Assign(input.offset); - attribute.type.Assign(static_cast<u32>(input.type.Value())); - attribute.size.Assign(static_cast<u32>(input.size.Value())); - attribute.binding_index_enabled.Assign(regs.vertex_array[index].IsEnabled() ? 1 : 0); + if (maxwell3d.dirty.flags[Dirty::VertexAttributes]) { + maxwell3d.dirty.flags[Dirty::VertexAttributes] = false; + for (size_t index = 0; index < Maxwell::NumVertexAttributes; ++index) { + const auto& input = regs.vertex_attrib_format[index]; + auto& attribute = attributes[index]; + attribute.raw = 0; + attribute.enabled.Assign(input.IsConstant() ? 0 : 1); + attribute.buffer.Assign(input.buffer); + attribute.offset.Assign(input.offset); + attribute.type.Assign(static_cast<u32>(input.type.Value())); + attribute.size.Assign(static_cast<u32>(input.size.Value())); + } } - - for (std::size_t index = 0; index < std::size(attachments); ++index) { - attachments[index].Fill(regs, index); + if (maxwell3d.dirty.flags[Dirty::Blending]) { + maxwell3d.dirty.flags[Dirty::Blending] = false; + for (size_t index = 0; index < attachments.size(); ++index) { + attachments[index].Refresh(regs, index); + } + } + if (maxwell3d.dirty.flags[Dirty::ViewportSwizzles]) { + maxwell3d.dirty.flags[Dirty::ViewportSwizzles] = false; + const auto& transform = regs.viewport_transform; + std::ranges::transform(transform, viewport_swizzles.begin(), [](const auto& viewport) { + return static_cast<u16>(viewport.swizzle.raw); + }); } - - const auto& transform = regs.viewport_transform; - std::transform(transform.begin(), transform.end(), viewport_swizzles.begin(), - [](const auto& viewport) { return static_cast<u16>(viewport.swizzle.raw); }); - if (!has_extended_dynamic_state) { no_extended_dynamic_state.Assign(1); - dynamic_state.Fill(regs); + dynamic_state.Refresh(regs); } } -void FixedPipelineState::BlendingAttachment::Fill(const Maxwell& regs, std::size_t index) { +void FixedPipelineState::BlendingAttachment::Refresh(const Maxwell& regs, size_t index) { const auto& mask = regs.color_mask[regs.color_mask_common ? 0 : index]; raw = 0; @@ -141,7 +154,7 @@ void FixedPipelineState::BlendingAttachment::Fill(const Maxwell& regs, std::size enable.Assign(1); } -void FixedPipelineState::DynamicState::Fill(const Maxwell& regs) { +void FixedPipelineState::DynamicState::Refresh(const Maxwell& regs) { u32 packed_front_face = PackFrontFace(regs.front_face); if (regs.screen_y_control.triangle_rast_flip != 0) { // Flip front face @@ -178,9 +191,9 @@ void FixedPipelineState::DynamicState::Fill(const Maxwell& regs) { }); } -std::size_t FixedPipelineState::Hash() const noexcept { +size_t FixedPipelineState::Hash() const noexcept { const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), Size()); - return static_cast<std::size_t>(hash); + return static_cast<size_t>(hash); } bool FixedPipelineState::operator==(const FixedPipelineState& rhs) const noexcept { diff --git a/src/video_core/renderer_vulkan/fixed_pipeline_state.h b/src/video_core/renderer_vulkan/fixed_pipeline_state.h index 465a55fdb..a0eb83a68 100644 --- a/src/video_core/renderer_vulkan/fixed_pipeline_state.h +++ b/src/video_core/renderer_vulkan/fixed_pipeline_state.h @@ -58,7 +58,7 @@ struct FixedPipelineState { BitField<30, 1, u32> enable; }; - void Fill(const Maxwell& regs, std::size_t index); + void Refresh(const Maxwell& regs, size_t index); constexpr std::array<bool, 4> Mask() const noexcept { return {mask_r != 0, mask_g != 0, mask_b != 0, mask_a != 0}; @@ -96,8 +96,6 @@ struct FixedPipelineState { BitField<6, 14, u32> offset; BitField<20, 3, u32> type; BitField<23, 6, u32> size; - // Not really an element of a vertex attribute, but it can be packed here - BitField<29, 1, u32> binding_index_enabled; constexpr Maxwell::VertexAttribute::Type Type() const noexcept { return static_cast<Maxwell::VertexAttribute::Type>(type.Value()); @@ -108,7 +106,7 @@ struct FixedPipelineState { } }; - template <std::size_t Position> + template <size_t Position> union StencilFace { BitField<Position + 0, 3, u32> action_stencil_fail; BitField<Position + 3, 3, u32> action_depth_fail; @@ -152,7 +150,7 @@ struct FixedPipelineState { // Vertex stride is a 12 bits value, we have 4 bits to spare per element std::array<u16, Maxwell::NumVertexArrays> vertex_strides; - void Fill(const Maxwell& regs); + void Refresh(const Maxwell& regs); Maxwell::ComparisonOp DepthTestFunc() const noexcept { return UnpackComparisonOp(depth_test_func); @@ -199,9 +197,9 @@ struct FixedPipelineState { std::array<u16, Maxwell::NumViewports> viewport_swizzles; DynamicState dynamic_state; - void Fill(const Maxwell& regs, bool has_extended_dynamic_state); + void Refresh(Tegra::Engines::Maxwell3D& maxwell3d, bool has_extended_dynamic_state); - std::size_t Hash() const noexcept; + size_t Hash() const noexcept; bool operator==(const FixedPipelineState& rhs) const noexcept; @@ -209,8 +207,8 @@ struct FixedPipelineState { return !operator==(rhs); } - std::size_t Size() const noexcept { - const std::size_t total_size = sizeof *this; + size_t Size() const noexcept { + const size_t total_size = sizeof *this; return total_size - (no_extended_dynamic_state != 0 ? 0 : sizeof(DynamicState)); } }; @@ -224,7 +222,7 @@ namespace std { template <> struct hash<Vulkan::FixedPipelineState> { - std::size_t operator()(const Vulkan::FixedPipelineState& k) const noexcept { + size_t operator()(const Vulkan::FixedPipelineState& k) const noexcept { return k.Hash(); } }; diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index 85121d9fd..19aaf034f 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -531,13 +531,9 @@ VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison) { return {}; } -VkIndexType IndexFormat(const Device& device, Maxwell::IndexFormat index_format) { +VkIndexType IndexFormat(Maxwell::IndexFormat index_format) { switch (index_format) { case Maxwell::IndexFormat::UnsignedByte: - if (!device.IsExtIndexTypeUint8Supported()) { - UNIMPLEMENTED_MSG("Native uint8 indices are not supported on this device"); - return VK_INDEX_TYPE_UINT16; - } return VK_INDEX_TYPE_UINT8_EXT; case Maxwell::IndexFormat::UnsignedShort: return VK_INDEX_TYPE_UINT16; diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.h b/src/video_core/renderer_vulkan/maxwell_to_vk.h index 7c34b47dc..e3e06ba38 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.h +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.h @@ -53,7 +53,7 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison); -VkIndexType IndexFormat(const Device& device, Maxwell::IndexFormat index_format); +VkIndexType IndexFormat(Maxwell::IndexFormat index_format); VkStencilOp StencilOp(Maxwell::StencilOp stencil_op); diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 61796e33a..1cc720ddd 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -80,17 +80,50 @@ std::string BuildCommaSeparatedExtensions(std::vector<std::string> available_ext return separated_extensions; } +Device CreateDevice(const vk::Instance& instance, const vk::InstanceDispatch& dld, + VkSurfaceKHR surface) { + const std::vector<VkPhysicalDevice> devices = instance.EnumeratePhysicalDevices(); + const s32 device_index = Settings::values.vulkan_device.GetValue(); + if (device_index < 0 || device_index >= static_cast<s32>(devices.size())) { + LOG_ERROR(Render_Vulkan, "Invalid device index {}!", device_index); + throw vk::Exception(VK_ERROR_INITIALIZATION_FAILED); + } + const vk::PhysicalDevice physical_device(devices[device_index], dld); + return Device(*instance, physical_device, surface, dld); +} } // Anonymous namespace RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, Core::Frontend::EmuWindow& emu_window, 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_}, - cpu_memory{cpu_memory_}, gpu{gpu_} {} + std::unique_ptr<Core::Frontend::GraphicsContext> context_) try + : RendererBase(emu_window, std::move(context_)), + telemetry_session(telemetry_session_), + cpu_memory(cpu_memory_), + gpu(gpu_), + library(OpenLibrary()), + instance(CreateInstance(library, dld, VK_API_VERSION_1_1, render_window.GetWindowInfo().type, + true, Settings::values.renderer_debug)), + 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), + 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) { + Report(); +} catch (const vk::Exception& exception) { + LOG_ERROR(Render_Vulkan, "Vulkan initialization failed with error: {}", exception.what()); + throw std::runtime_error{fmt::format("Vulkan initialization error {}", exception.what())}; +} RendererVulkan::~RendererVulkan() { - ShutDown(); + void(device.GetLogical().WaitIdle()); } void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { @@ -101,101 +134,38 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { if (layout.width > 0 && layout.height > 0 && render_window.IsShown()) { const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset; const bool use_accelerated = - rasterizer->AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride); + rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride); const bool is_srgb = use_accelerated && screen_info.is_srgb; - if (swapchain->HasFramebufferChanged(layout) || swapchain->GetSrgbState() != is_srgb) { - swapchain->Create(layout.width, layout.height, is_srgb); - blit_screen->Recreate(); + if (swapchain.HasFramebufferChanged(layout) || swapchain.GetSrgbState() != is_srgb) { + swapchain.Create(layout.width, layout.height, is_srgb); + blit_screen.Recreate(); } - scheduler->WaitWorker(); + scheduler.WaitWorker(); - swapchain->AcquireNextImage(); - const VkSemaphore render_semaphore = blit_screen->Draw(*framebuffer, use_accelerated); + swapchain.AcquireNextImage(); + const VkSemaphore render_semaphore = blit_screen.Draw(*framebuffer, use_accelerated); - scheduler->Flush(render_semaphore); + scheduler.Flush(render_semaphore); - if (swapchain->Present(render_semaphore)) { - blit_screen->Recreate(); + if (swapchain.Present(render_semaphore)) { + blit_screen.Recreate(); } - - rasterizer->TickFrame(); + rasterizer.TickFrame(); } render_window.OnFrameDisplayed(); } -bool RendererVulkan::Init() try { - library = OpenLibrary(); - instance = CreateInstance(library, dld, VK_API_VERSION_1_1, render_window.GetWindowInfo().type, - true, Settings::values.renderer_debug); - if (Settings::values.renderer_debug) { - debug_callback = CreateDebugCallback(instance); - } - surface = CreateSurface(instance, render_window); - - InitializeDevice(); - Report(); - - memory_allocator = std::make_unique<MemoryAllocator>(*device); - - state_tracker = std::make_unique<StateTracker>(gpu); - - scheduler = std::make_unique<VKScheduler>(*device, *state_tracker); - - const auto& framebuffer = render_window.GetFramebufferLayout(); - swapchain = std::make_unique<VKSwapchain>(*surface, *device, *scheduler); - swapchain->Create(framebuffer.width, framebuffer.height, false); - - rasterizer = std::make_unique<RasterizerVulkan>(render_window, gpu, gpu.MemoryManager(), - cpu_memory, screen_info, *device, - *memory_allocator, *state_tracker, *scheduler); - - blit_screen = - std::make_unique<VKBlitScreen>(cpu_memory, render_window, *rasterizer, *device, - *memory_allocator, *swapchain, *scheduler, screen_info); - return true; - -} catch (const vk::Exception& exception) { - LOG_ERROR(Render_Vulkan, "Vulkan initialization failed with error: {}", exception.what()); - return false; -} - -void RendererVulkan::ShutDown() { - if (!device) { - return; - } - if (const auto& dev = device->GetLogical()) { - dev.WaitIdle(); - } - rasterizer.reset(); - blit_screen.reset(); - scheduler.reset(); - swapchain.reset(); - memory_allocator.reset(); - device.reset(); -} - -void RendererVulkan::InitializeDevice() { - const std::vector<VkPhysicalDevice> devices = instance.EnumeratePhysicalDevices(); - const s32 device_index = Settings::values.vulkan_device.GetValue(); - if (device_index < 0 || device_index >= static_cast<s32>(devices.size())) { - LOG_ERROR(Render_Vulkan, "Invalid device index {}!", device_index); - throw vk::Exception(VK_ERROR_INITIALIZATION_FAILED); - } - const vk::PhysicalDevice physical_device(devices[static_cast<size_t>(device_index)], dld); - device = std::make_unique<Device>(*instance, physical_device, *surface, dld); -} - void RendererVulkan::Report() const { - const std::string vendor_name{device->GetVendorName()}; - const std::string model_name{device->GetModelName()}; - const std::string driver_version = GetDriverVersion(*device); + const std::string vendor_name{device.GetVendorName()}; + const std::string model_name{device.GetModelName()}; + const std::string driver_version = GetDriverVersion(device); const std::string driver_name = fmt::format("{} {}", vendor_name, driver_version); - const std::string api_version = GetReadableVersion(device->ApiVersion()); + const std::string api_version = GetReadableVersion(device.ApiVersion()); - const std::string extensions = BuildCommaSeparatedExtensions(device->GetAvailableExtensions()); + const std::string extensions = BuildCommaSeparatedExtensions(device.GetAvailableExtensions()); LOG_INFO(Render_Vulkan, "Driver: {}", driver_name); LOG_INFO(Render_Vulkan, "Device: {}", model_name); @@ -209,21 +179,4 @@ void RendererVulkan::Report() const { telemetry_session.AddField(field, "GPU_Vulkan_Extensions", extensions); } -std::vector<std::string> RendererVulkan::EnumerateDevices() try { - vk::InstanceDispatch dld; - const Common::DynamicLibrary library = OpenLibrary(); - const vk::Instance instance = CreateInstance(library, dld, VK_API_VERSION_1_0); - const std::vector<VkPhysicalDevice> physical_devices = instance.EnumeratePhysicalDevices(); - std::vector<std::string> names; - names.reserve(physical_devices.size()); - for (const VkPhysicalDevice device : physical_devices) { - names.push_back(vk::PhysicalDevice(device, dld).GetProperties().deviceName); - } - return names; - -} catch (const vk::Exception& exception) { - LOG_ERROR(Render_Vulkan, "Failed to enumerate devices with error: {}", exception.what()); - return {}; -} - } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h index daf55b9b4..72071316c 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.h +++ b/src/video_core/renderer_vulkan/renderer_vulkan.h @@ -9,8 +9,14 @@ #include <vector> #include "common/dynamic_library.h" - #include "video_core/renderer_base.h" +#include "video_core/renderer_vulkan/vk_blit_screen.h" +#include "video_core/renderer_vulkan/vk_rasterizer.h" +#include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_state_tracker.h" +#include "video_core/renderer_vulkan/vk_swapchain.h" +#include "video_core/vulkan_common/vulkan_device.h" +#include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Core { @@ -27,20 +33,6 @@ class GPU; namespace Vulkan { -class Device; -class StateTracker; -class MemoryAllocator; -class VKBlitScreen; -class VKSwapchain; -class VKScheduler; - -struct VKScreenInfo { - VkImageView image_view{}; - u32 width{}; - u32 height{}; - bool is_srgb{}; -}; - class RendererVulkan final : public VideoCore::RendererBase { public: explicit RendererVulkan(Core::TelemetrySession& telemtry_session, @@ -49,15 +41,13 @@ public: std::unique_ptr<Core::Frontend::GraphicsContext> context_); ~RendererVulkan() override; - bool Init() override; - void ShutDown() override; void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override; - static std::vector<std::string> EnumerateDevices(); + VideoCore::RasterizerInterface* ReadRasterizer() override { + return &rasterizer; + } private: - void InitializeDevice(); - void Report() const; Core::TelemetrySession& telemetry_session; @@ -68,18 +58,18 @@ private: vk::InstanceDispatch dld; vk::Instance instance; - + vk::DebugUtilsMessenger debug_callback; vk::SurfaceKHR surface; VKScreenInfo screen_info; - vk::DebugUtilsMessenger debug_callback; - std::unique_ptr<Device> device; - std::unique_ptr<MemoryAllocator> memory_allocator; - std::unique_ptr<StateTracker> state_tracker; - std::unique_ptr<VKScheduler> scheduler; - std::unique_ptr<VKSwapchain> swapchain; - std::unique_ptr<VKBlitScreen> blit_screen; + Device device; + MemoryAllocator memory_allocator; + StateTracker state_tracker; + VKScheduler scheduler; + VKSwapchain swapchain; + VKBlitScreen blit_screen; + RasterizerVulkan rasterizer; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp index 3e3b895e0..a1a32aabe 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp +++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp @@ -18,7 +18,6 @@ #include "video_core/gpu.h" #include "video_core/host_shaders/vulkan_present_frag_spv.h" #include "video_core/host_shaders/vulkan_present_vert_spv.h" -#include "video_core/rasterizer_interface.h" #include "video_core/renderer_vulkan/renderer_vulkan.h" #include "video_core/renderer_vulkan/vk_blit_screen.h" #include "video_core/renderer_vulkan/vk_master_semaphore.h" @@ -113,13 +112,12 @@ struct VKBlitScreen::BufferData { }; VKBlitScreen::VKBlitScreen(Core::Memory::Memory& cpu_memory_, - Core::Frontend::EmuWindow& render_window_, - VideoCore::RasterizerInterface& rasterizer_, const Device& device_, + Core::Frontend::EmuWindow& render_window_, const Device& device_, MemoryAllocator& memory_allocator_, VKSwapchain& swapchain_, VKScheduler& scheduler_, const VKScreenInfo& screen_info_) - : cpu_memory{cpu_memory_}, render_window{render_window_}, rasterizer{rasterizer_}, - device{device_}, memory_allocator{memory_allocator_}, swapchain{swapchain_}, - scheduler{scheduler_}, image_count{swapchain.GetImageCount()}, screen_info{screen_info_} { + : cpu_memory{cpu_memory_}, render_window{render_window_}, device{device_}, + memory_allocator{memory_allocator_}, swapchain{swapchain_}, scheduler{scheduler_}, + image_count{swapchain.GetImageCount()}, screen_info{screen_info_} { resource_ticks.resize(image_count); CreateStaticResources(); @@ -150,8 +148,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool SetUniformData(data, framebuffer); SetVertexData(data, framebuffer); - const std::span<u8> map = buffer_commit.Map(); - std::memcpy(map.data(), &data, sizeof(data)); + const std::span<u8> mapped_span = buffer_commit.Map(); + std::memcpy(mapped_span.data(), &data, sizeof(data)); if (!use_accelerated) { const u64 image_offset = GetRawImageOffset(framebuffer, image_index); @@ -159,14 +157,13 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool const VAddr framebuffer_addr = framebuffer.address + framebuffer.offset; const u8* const host_ptr = cpu_memory.GetPointer(framebuffer_addr); const size_t size_bytes = GetSizeInBytes(framebuffer); - rasterizer.FlushRegion(ToCacheAddr(host_ptr), size_bytes); // TODO(Rodrigo): Read this from HLE constexpr u32 block_height_log2 = 4; const u32 bytes_per_pixel = GetBytesPerPixel(framebuffer); Tegra::Texture::UnswizzleTexture( - map.subspan(image_offset, size_bytes), std::span(host_ptr, size_bytes), bytes_per_pixel, - framebuffer.width, framebuffer.height, 1, block_height_log2, 0); + mapped_span.subspan(image_offset, size_bytes), std::span(host_ptr, size_bytes), + bytes_per_pixel, framebuffer.width, framebuffer.height, 1, block_height_log2, 0); const VkBufferImageCopy copy{ .bufferOffset = image_offset, @@ -266,7 +263,6 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool cmdbuf.Draw(4, 1, 0, 0); cmdbuf.EndRenderPass(); }); - return *semaphores[image_index]; } diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.h b/src/video_core/renderer_vulkan/vk_blit_screen.h index b52576957..5e3177685 100644 --- a/src/video_core/renderer_vulkan/vk_blit_screen.h +++ b/src/video_core/renderer_vulkan/vk_blit_screen.h @@ -38,12 +38,18 @@ class RasterizerVulkan; class VKScheduler; class VKSwapchain; -class VKBlitScreen final { +struct VKScreenInfo { + VkImageView image_view{}; + u32 width{}; + u32 height{}; + bool is_srgb{}; +}; + +class VKBlitScreen { public: explicit VKBlitScreen(Core::Memory::Memory& cpu_memory, - Core::Frontend::EmuWindow& render_window, - VideoCore::RasterizerInterface& rasterizer, const Device& device, - MemoryAllocator& memory_allocator, VKSwapchain& swapchain, + Core::Frontend::EmuWindow& render_window, const Device& device, + MemoryAllocator& memory_manager, VKSwapchain& swapchain, VKScheduler& scheduler, const VKScreenInfo& screen_info); ~VKBlitScreen(); @@ -84,7 +90,6 @@ private: Core::Memory::Memory& cpu_memory; Core::Frontend::EmuWindow& render_window; - VideoCore::RasterizerInterface& rasterizer; const Device& device; MemoryAllocator& memory_allocator; VKSwapchain& swapchain; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index d8ad40a0f..848eedd66 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -3,188 +3,308 @@ // Refer to the license.txt file included. #include <algorithm> +#include <array> #include <cstring> -#include <memory> +#include <span> +#include <vector> -#include "core/core.h" #include "video_core/buffer_cache/buffer_cache.h" +#include "video_core/renderer_vulkan/maxwell_to_vk.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" -#include "video_core/renderer_vulkan/vk_stream_buffer.h" +#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" +#include "video_core/renderer_vulkan/vk_update_descriptor.h" #include "video_core/vulkan_common/vulkan_device.h" +#include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { - namespace { +VkBufferCopy MakeBufferCopy(const VideoCommon::BufferCopy& copy) { + return VkBufferCopy{ + .srcOffset = copy.src_offset, + .dstOffset = copy.dst_offset, + .size = copy.size, + }; +} -constexpr VkBufferUsageFlags BUFFER_USAGE = - VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | - VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT; - -constexpr VkPipelineStageFlags UPLOAD_PIPELINE_STAGE = - VK_PIPELINE_STAGE_TRANSFER_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | - VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; - -constexpr VkAccessFlags UPLOAD_ACCESS_BARRIERS = - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT | - VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT; +VkIndexType IndexTypeFromNumElements(const Device& device, u32 num_elements) { + if (num_elements <= 0xff && device.IsExtIndexTypeUint8Supported()) { + return VK_INDEX_TYPE_UINT8_EXT; + } + if (num_elements <= 0xffff) { + return VK_INDEX_TYPE_UINT16; + } + return VK_INDEX_TYPE_UINT32; +} -constexpr VkAccessFlags TRANSFORM_FEEDBACK_WRITE_ACCESS = - VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT | VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT; +size_t BytesPerIndex(VkIndexType index_type) { + switch (index_type) { + case VK_INDEX_TYPE_UINT8_EXT: + return 1; + case VK_INDEX_TYPE_UINT16: + return 2; + case VK_INDEX_TYPE_UINT32: + return 4; + default: + UNREACHABLE_MSG("Invalid index type={}", index_type); + return 1; + } +} +template <typename T> +std::array<T, 6> MakeQuadIndices(u32 quad, u32 first) { + std::array<T, 6> indices{0, 1, 2, 0, 2, 3}; + std::ranges::transform(indices, indices.begin(), + [quad, first](u32 index) { return first + index + quad * 4; }); + return indices; +} } // Anonymous namespace -Buffer::Buffer(const Device& device_, MemoryAllocator& memory_allocator, VKScheduler& scheduler_, - StagingBufferPool& staging_pool_, VAddr cpu_addr_, std::size_t size_) - : BufferBlock{cpu_addr_, size_}, device{device_}, scheduler{scheduler_}, staging_pool{ - staging_pool_} { - buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{ +Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params) + : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {} + +Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_, + VAddr cpu_addr_, u64 size_bytes_) + : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) { + buffer = runtime.device.GetLogical().CreateBuffer(VkBufferCreateInfo{ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, .pNext = nullptr, .flags = 0, - .size = static_cast<VkDeviceSize>(size_), - .usage = BUFFER_USAGE | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT, + .size = SizeBytes(), + .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | + VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | + VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | + VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | + VK_BUFFER_USAGE_VERTEX_BUFFER_BIT, .sharingMode = VK_SHARING_MODE_EXCLUSIVE, .queueFamilyIndexCount = 0, .pQueueFamilyIndices = nullptr, }); - commit = memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal); + if (runtime.device.HasDebuggingToolAttached()) { + buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str()); + } + commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal); } -Buffer::~Buffer() = default; +BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_, + VKScheduler& scheduler_, StagingBufferPool& staging_pool_, + VKUpdateDescriptorQueue& update_descriptor_queue_, + VKDescriptorPool& descriptor_pool) + : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_}, + staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_}, + uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), + quad_index_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue) {} -void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) { - const auto& staging = staging_pool.Request(data_size, MemoryUsage::Upload); - std::memcpy(staging.mapped_span.data(), data, data_size); +StagingBufferRef BufferCacheRuntime::UploadStagingBuffer(size_t size) { + return staging_pool.Request(size, MemoryUsage::Upload); +} - scheduler.RequestOutsideRenderPassOperationContext(); +StagingBufferRef BufferCacheRuntime::DownloadStagingBuffer(size_t size) { + return staging_pool.Request(size, MemoryUsage::Download); +} - const VkBuffer handle = Handle(); - scheduler.Record([staging = staging.buffer, handle, offset, data_size, - &device = device](vk::CommandBuffer cmdbuf) { - const VkBufferMemoryBarrier read_barrier{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = - VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | - VK_ACCESS_HOST_WRITE_BIT | - (device.IsExtTransformFeedbackSupported() ? TRANSFORM_FEEDBACK_WRITE_ACCESS : 0), - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .buffer = handle, - .offset = offset, - .size = data_size, - }; - const VkBufferMemoryBarrier write_barrier{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = UPLOAD_ACCESS_BARRIERS, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .buffer = handle, - .offset = offset, - .size = data_size, - }; +void BufferCacheRuntime::Finish() { + scheduler.Finish(); +} + +void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer, + std::span<const VideoCommon::BufferCopy> copies) { + static constexpr VkMemoryBarrier READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + }; + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + }; + // Measuring a popular game, this number never exceeds the specified size once data is warmed up + boost::container::small_vector<VkBufferCopy, 3> vk_copies(copies.size()); + std::ranges::transform(copies, vk_copies.begin(), MakeBufferCopy); + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, - 0, read_barrier); - cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, data_size}); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, - write_barrier); + 0, READ_BARRIER); + cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, WRITE_BARRIER); }); } -void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) { - auto staging = staging_pool.Request(data_size, MemoryUsage::Download); - scheduler.RequestOutsideRenderPassOperationContext(); +void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, + u32 base_vertex, u32 num_indices, VkBuffer buffer, + u32 offset, [[maybe_unused]] u32 size) { + VkIndexType vk_index_type = MaxwellToVK::IndexFormat(index_format); + VkDeviceSize vk_offset = offset; + VkBuffer vk_buffer = buffer; + if (topology == PrimitiveTopology::Quads) { + vk_index_type = VK_INDEX_TYPE_UINT32; + std::tie(vk_buffer, vk_offset) = + quad_index_pass.Assemble(index_format, num_indices, base_vertex, buffer, offset); + } else if (vk_index_type == VK_INDEX_TYPE_UINT8_EXT && !device.IsExtIndexTypeUint8Supported()) { + vk_index_type = VK_INDEX_TYPE_UINT16; + std::tie(vk_buffer, vk_offset) = uint8_pass.Assemble(num_indices, buffer, offset); + } + if (vk_buffer == VK_NULL_HANDLE) { + // Vulkan doesn't support null index buffers. Replace it with our own null buffer. + ReserveNullIndexBuffer(); + vk_buffer = *null_index_buffer; + } + scheduler.Record([vk_buffer, vk_offset, vk_index_type](vk::CommandBuffer cmdbuf) { + cmdbuf.BindIndexBuffer(vk_buffer, vk_offset, vk_index_type); + }); +} - const VkBuffer handle = Handle(); - scheduler.Record( - [staging = staging.buffer, handle, offset, data_size](vk::CommandBuffer cmdbuf) { - const VkBufferMemoryBarrier barrier{ - .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, - .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, - .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, - .buffer = handle, - .offset = offset, - .size = data_size, - }; - - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | - VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | - VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, barrier, {}); - cmdbuf.CopyBuffer(handle, staging, VkBufferCopy{offset, 0, data_size}); - }); - scheduler.Finish(); +void BufferCacheRuntime::BindQuadArrayIndexBuffer(u32 first, u32 count) { + ReserveQuadArrayLUT(first + count, true); - std::memcpy(data, staging.mapped_span.data(), data_size); + // The LUT has the indices 0, 1, 2, and 3 copied as an array + // To apply these 'first' offsets we can apply an offset based on the modulus. + const VkIndexType index_type = quad_array_lut_index_type; + const size_t sub_first_offset = static_cast<size_t>(first % 4) * (current_num_indices / 4); + const size_t offset = (sub_first_offset + first / 4) * 6ULL * BytesPerIndex(index_type); + scheduler.Record([buffer = *quad_array_lut, index_type, offset](vk::CommandBuffer cmdbuf) { + cmdbuf.BindIndexBuffer(buffer, offset, index_type); + }); } -void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset, - std::size_t copy_size) { - scheduler.RequestOutsideRenderPassOperationContext(); +void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, + u32 stride) { + if (device.IsExtExtendedDynamicStateSupported()) { + scheduler.Record([index, buffer, offset, size, stride](vk::CommandBuffer cmdbuf) { + const VkDeviceSize vk_offset = offset; + const VkDeviceSize vk_size = buffer != VK_NULL_HANDLE ? size : VK_WHOLE_SIZE; + const VkDeviceSize vk_stride = stride; + cmdbuf.BindVertexBuffers2EXT(index, 1, &buffer, &vk_offset, &vk_size, &vk_stride); + }); + } else { + scheduler.Record([index, buffer, offset](vk::CommandBuffer cmdbuf) { + cmdbuf.BindVertexBuffer(index, buffer, offset); + }); + } +} - const VkBuffer dst_buffer = Handle(); - scheduler.Record([src_buffer = src.Handle(), dst_buffer, src_offset, dst_offset, - copy_size](vk::CommandBuffer cmdbuf) { - cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, copy_size}); - - std::array<VkBufferMemoryBarrier, 2> barriers; - barriers[0].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; - barriers[0].pNext = nullptr; - barriers[0].srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT; - barriers[0].dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT; - barriers[0].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barriers[0].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barriers[0].buffer = src_buffer; - barriers[0].offset = src_offset; - barriers[0].size = copy_size; - barriers[1].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; - barriers[1].pNext = nullptr; - barriers[1].srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - barriers[1].dstAccessMask = UPLOAD_ACCESS_BARRIERS; - barriers[1].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barriers[1].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barriers[1].buffer = dst_buffer; - barriers[1].offset = dst_offset; - barriers[1].size = copy_size; - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {}, - barriers, {}); +void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset, + u32 size) { + if (!device.IsExtTransformFeedbackSupported()) { + // Already logged in the rasterizer + return; + } + scheduler.Record([index, buffer, offset, size](vk::CommandBuffer cmdbuf) { + const VkDeviceSize vk_offset = offset; + const VkDeviceSize vk_size = size; + cmdbuf.BindTransformFeedbackBuffersEXT(index, 1, &buffer, &vk_offset, &vk_size); }); } -VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer_, - Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, - const Device& device_, MemoryAllocator& memory_allocator_, - VKScheduler& scheduler_, VKStreamBuffer& stream_buffer_, - StagingBufferPool& staging_pool_) - : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer_, gpu_memory_, - cpu_memory_, stream_buffer_}, - device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_}, - staging_pool{staging_pool_} {} +void BufferCacheRuntime::BindBuffer(VkBuffer buffer, u32 offset, u32 size) { + update_descriptor_queue.AddBuffer(buffer, offset, size); +} -VKBufferCache::~VKBufferCache() = default; +void BufferCacheRuntime::ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle) { + if (num_indices <= current_num_indices) { + return; + } + if (wait_for_idle) { + scheduler.Finish(); + } + current_num_indices = num_indices; + quad_array_lut_index_type = IndexTypeFromNumElements(device, num_indices); -std::shared_ptr<Buffer> VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) { - return std::make_shared<Buffer>(device, memory_allocator, scheduler, staging_pool, cpu_addr, - size); + const u32 num_quads = num_indices / 4; + const u32 num_triangle_indices = num_quads * 6; + const u32 num_first_offset_copies = 4; + const size_t bytes_per_index = BytesPerIndex(quad_array_lut_index_type); + const size_t size_bytes = num_triangle_indices * bytes_per_index * num_first_offset_copies; + quad_array_lut = device.GetLogical().CreateBuffer(VkBufferCreateInfo{ + .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .size = size_bytes, + .usage = VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT, + .sharingMode = VK_SHARING_MODE_EXCLUSIVE, + .queueFamilyIndexCount = 0, + .pQueueFamilyIndices = nullptr, + }); + if (device.HasDebuggingToolAttached()) { + quad_array_lut.SetObjectNameEXT("Quad LUT"); + } + quad_array_lut_commit = memory_allocator.Commit(quad_array_lut, MemoryUsage::DeviceLocal); + + const StagingBufferRef staging = staging_pool.Request(size_bytes, MemoryUsage::Upload); + u8* staging_data = staging.mapped_span.data(); + const size_t quad_size = bytes_per_index * 6; + for (u32 first = 0; first < num_first_offset_copies; ++first) { + for (u32 quad = 0; quad < num_quads; ++quad) { + switch (quad_array_lut_index_type) { + case VK_INDEX_TYPE_UINT8_EXT: + std::memcpy(staging_data, MakeQuadIndices<u8>(quad, first).data(), quad_size); + break; + case VK_INDEX_TYPE_UINT16: + std::memcpy(staging_data, MakeQuadIndices<u16>(quad, first).data(), quad_size); + break; + case VK_INDEX_TYPE_UINT32: + std::memcpy(staging_data, MakeQuadIndices<u32>(quad, first).data(), quad_size); + break; + default: + UNREACHABLE(); + break; + } + staging_data += quad_size; + } + } + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([src_buffer = staging.buffer, src_offset = staging.offset, + dst_buffer = *quad_array_lut, size_bytes](vk::CommandBuffer cmdbuf) { + const VkBufferCopy copy{ + .srcOffset = src_offset, + .dstOffset = 0, + .size = size_bytes, + }; + const VkBufferMemoryBarrier write_barrier{ + .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_INDEX_READ_BIT, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .buffer = dst_buffer, + .offset = 0, + .size = size_bytes, + }; + cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, + 0, write_barrier); + }); } -VKBufferCache::BufferInfo VKBufferCache::GetEmptyBuffer(std::size_t size) { - size = std::max(size, std::size_t(4)); - const auto& empty = staging_pool.Request(size, MemoryUsage::DeviceLocal); +void BufferCacheRuntime::ReserveNullIndexBuffer() { + if (null_index_buffer) { + return; + } + null_index_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{ + .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .size = 4, + .usage = VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT, + .sharingMode = VK_SHARING_MODE_EXCLUSIVE, + .queueFamilyIndexCount = 0, + .pQueueFamilyIndices = nullptr, + }); + if (device.HasDebuggingToolAttached()) { + null_index_buffer.SetObjectNameEXT("Null index buffer"); + } + null_index_buffer_commit = memory_allocator.Commit(null_index_buffer, MemoryUsage::DeviceLocal); + scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([size, buffer = empty.buffer](vk::CommandBuffer cmdbuf) { - cmdbuf.FillBuffer(buffer, 0, size, 0); + scheduler.Record([buffer = *null_index_buffer](vk::CommandBuffer cmdbuf) { + cmdbuf.FillBuffer(buffer, 0, VK_WHOLE_SIZE, 0); }); - return {empty.buffer, 0, 0}; } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h index 41d577510..041e6515c 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.h +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h @@ -4,69 +4,124 @@ #pragma once -#include <memory> - -#include "common/common_types.h" #include "video_core/buffer_cache/buffer_cache.h" +#include "video_core/engines/maxwell_3d.h" +#include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" -#include "video_core/renderer_vulkan/vk_stream_buffer.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { class Device; +class VKDescriptorPool; class VKScheduler; +class VKUpdateDescriptorQueue; -class Buffer final : public VideoCommon::BufferBlock { -public: - explicit Buffer(const Device& device, MemoryAllocator& memory_allocator, VKScheduler& scheduler, - StagingBufferPool& staging_pool, VAddr cpu_addr_, std::size_t size_); - ~Buffer(); - - void Upload(std::size_t offset, std::size_t data_size, const u8* data); +class BufferCacheRuntime; - void Download(std::size_t offset, std::size_t data_size, u8* data); - - void CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset, - std::size_t copy_size); +class Buffer : public VideoCommon::BufferBase<VideoCore::RasterizerInterface> { +public: + explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params); + explicit Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_, + VAddr cpu_addr_, u64 size_bytes_); - VkBuffer Handle() const { + [[nodiscard]] VkBuffer Handle() const noexcept { return *buffer; } - u64 Address() const { - return 0; + operator VkBuffer() const noexcept { + return *buffer; } private: - const Device& device; - VKScheduler& scheduler; - StagingBufferPool& staging_pool; - vk::Buffer buffer; MemoryCommit commit; }; -class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> { +class BufferCacheRuntime { + friend Buffer; + + using PrimitiveTopology = Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology; + using IndexFormat = Tegra::Engines::Maxwell3D::Regs::IndexFormat; + public: - explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer, - Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, - const Device& device, MemoryAllocator& memory_allocator, - VKScheduler& scheduler, VKStreamBuffer& stream_buffer, - StagingBufferPool& staging_pool); - ~VKBufferCache(); + explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_, + VKScheduler& scheduler_, StagingBufferPool& staging_pool_, + VKUpdateDescriptorQueue& update_descriptor_queue_, + VKDescriptorPool& descriptor_pool); + + void Finish(); + + [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size); + + [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size); + + void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer, + std::span<const VideoCommon::BufferCopy> copies); + + void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices, + u32 base_vertex, VkBuffer buffer, u32 offset, u32 size); - BufferInfo GetEmptyBuffer(std::size_t size) override; + void BindQuadArrayIndexBuffer(u32 first, u32 count); -protected: - std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override; + void BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride); + + void BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size); + + std::span<u8> BindMappedUniformBuffer([[maybe_unused]] size_t stage, + [[maybe_unused]] u32 binding_index, u32 size) { + const StagingBufferRef ref = staging_pool.Request(size, MemoryUsage::Upload); + BindBuffer(ref.buffer, static_cast<u32>(ref.offset), size); + return ref.mapped_span; + } + + void BindUniformBuffer(VkBuffer buffer, u32 offset, u32 size) { + BindBuffer(buffer, offset, size); + } + + void BindStorageBuffer(VkBuffer buffer, u32 offset, u32 size, + [[maybe_unused]] bool is_written) { + BindBuffer(buffer, offset, size); + } private: + void BindBuffer(VkBuffer buffer, u32 offset, u32 size); + + void ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle); + + void ReserveNullIndexBuffer(); + const Device& device; MemoryAllocator& memory_allocator; VKScheduler& scheduler; StagingBufferPool& staging_pool; + VKUpdateDescriptorQueue& update_descriptor_queue; + + vk::Buffer quad_array_lut; + MemoryCommit quad_array_lut_commit; + VkIndexType quad_array_lut_index_type{}; + u32 current_num_indices = 0; + + vk::Buffer null_index_buffer; + MemoryCommit null_index_buffer_commit; + + Uint8Pass uint8_pass; + QuadIndexedPass quad_index_pass; }; +struct BufferCacheParams { + using Runtime = Vulkan::BufferCacheRuntime; + using Buffer = Vulkan::Buffer; + + static constexpr bool IS_OPENGL = false; + static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = false; + static constexpr bool HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT = false; + static constexpr bool NEEDS_BIND_UNIFORM_INDEX = false; + static constexpr bool NEEDS_BIND_STORAGE_INDEX = false; + static constexpr bool USE_MEMORY_MAPS = true; +}; + +using BufferCache = VideoCommon::BufferCache<BufferCacheParams>; + } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp index 5eb6a54be..2f9a7b028 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp @@ -10,7 +10,7 @@ #include "common/alignment.h" #include "common/assert.h" #include "common/common_types.h" -#include "video_core/host_shaders/vulkan_quad_array_comp_spv.h" +#include "common/div_ceil.h" #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h" #include "video_core/host_shaders/vulkan_uint8_comp_spv.h" #include "video_core/renderer_vulkan/vk_compute_pass.h" @@ -22,30 +22,7 @@ #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { - namespace { - -VkDescriptorSetLayoutBinding BuildQuadArrayPassDescriptorSetLayoutBinding() { - return { - .binding = 0, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .descriptorCount = 1, - .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, - .pImmutableSamplers = nullptr, - }; -} - -VkDescriptorUpdateTemplateEntryKHR BuildQuadArrayPassDescriptorUpdateTemplateEntry() { - return { - .dstBinding = 0, - .dstArrayElement = 0, - .descriptorCount = 1, - .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, - .offset = 0, - .stride = sizeof(DescriptorUpdateEntry), - }; -} - VkPushConstantRange BuildComputePushConstantRange(std::size_t size) { return { .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT, @@ -162,55 +139,6 @@ VkDescriptorSet VKComputePass::CommitDescriptorSet( return set; } -QuadArrayPass::QuadArrayPass(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_) - : VKComputePass(device_, descriptor_pool_, BuildQuadArrayPassDescriptorSetLayoutBinding(), - BuildQuadArrayPassDescriptorUpdateTemplateEntry(), - BuildComputePushConstantRange(sizeof(u32)), VULKAN_QUAD_ARRAY_COMP_SPV), - scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, - update_descriptor_queue{update_descriptor_queue_} {} - -QuadArrayPass::~QuadArrayPass() = default; - -std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) { - const u32 num_triangle_vertices = (num_vertices / 4) * 6; - const std::size_t staging_size = num_triangle_vertices * sizeof(u32); - const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); - - update_descriptor_queue.Acquire(); - update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size); - const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); - - scheduler.RequestOutsideRenderPassOperationContext(); - - ASSERT(num_vertices % 4 == 0); - const u32 num_quads = num_vertices / 4; - scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer, - num_quads, first, set](vk::CommandBuffer cmdbuf) { - constexpr u32 dispatch_size = 1024; - cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline); - cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {}); - cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(first), &first); - cmdbuf.Dispatch(Common::AlignUp(num_quads, dispatch_size) / dispatch_size, 1, 1); - - VkBufferMemoryBarrier barrier; - barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; - barrier.pNext = nullptr; - barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; - barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT; - barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barrier.buffer = buffer; - barrier.offset = 0; - barrier.size = static_cast<VkDeviceSize>(num_quads) * 6 * sizeof(u32); - cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, {barrier}, {}); - }); - return {staging_ref.buffer, 0}; -} - Uint8Pass::Uint8Pass(const Device& device, VKScheduler& scheduler_, VKDescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_, VKUpdateDescriptorQueue& update_descriptor_queue_) @@ -221,38 +149,33 @@ Uint8Pass::Uint8Pass(const Device& device, VKScheduler& scheduler_, Uint8Pass::~Uint8Pass() = default; -std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer, - u64 src_offset) { +std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer, + u32 src_offset) { const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16)); - const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); + const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); update_descriptor_queue.Acquire(); update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices); - update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size); + update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer, set, + scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set, num_vertices](vk::CommandBuffer cmdbuf) { - constexpr u32 dispatch_size = 1024; + static constexpr u32 DISPATCH_SIZE = 1024; + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, + }; cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline); cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {}); - cmdbuf.Dispatch(Common::AlignUp(num_vertices, dispatch_size) / dispatch_size, 1, 1); - - VkBufferMemoryBarrier barrier; - barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; - barrier.pNext = nullptr; - barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; - barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT; - barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barrier.buffer = buffer; - barrier.offset = 0; - barrier.size = static_cast<VkDeviceSize>(num_vertices * sizeof(u16)); + cmdbuf.Dispatch(Common::DivCeil(num_vertices, DISPATCH_SIZE), 1, 1); cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {}); + VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, WRITE_BARRIER); }); - return {staging_ref.buffer, 0}; + return {staging.buffer, staging.offset}; } QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_, @@ -267,9 +190,9 @@ QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_, QuadIndexedPass::~QuadIndexedPass() = default; -std::pair<VkBuffer, u64> QuadIndexedPass::Assemble( +std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, u32 base_vertex, - VkBuffer src_buffer, u64 src_offset) { + VkBuffer src_buffer, u32 src_offset) { const u32 index_shift = [index_format] { switch (index_format) { case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedByte: @@ -286,38 +209,33 @@ std::pair<VkBuffer, u64> QuadIndexedPass::Assemble( const u32 num_tri_vertices = (num_vertices / 4) * 6; const std::size_t staging_size = num_tri_vertices * sizeof(u32); - const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); + const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); update_descriptor_queue.Acquire(); update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size); - update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size); + update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size); const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue); scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer, set, + scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set, num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) { - static constexpr u32 dispatch_size = 1024; + static constexpr u32 DISPATCH_SIZE = 1024; + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT, + }; const std::array push_constants = {base_vertex, index_shift}; cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline); cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {}); cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants), &push_constants); - cmdbuf.Dispatch(Common::AlignUp(num_tri_vertices, dispatch_size) / dispatch_size, 1, 1); - - VkBufferMemoryBarrier barrier; - barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; - barrier.pNext = nullptr; - barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT; - barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT; - barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; - barrier.buffer = buffer; - barrier.offset = 0; - barrier.size = static_cast<VkDeviceSize>(num_tri_vertices * sizeof(u32)); + cmdbuf.Dispatch(Common::DivCeil(num_tri_vertices, DISPATCH_SIZE), 1, 1); cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, - VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {}); + VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, WRITE_BARRIER); }); - return {staging_ref.buffer, 0}; + return {staging.buffer, staging.offset}; } } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h index f5c6f5f17..17d781d99 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pass.h +++ b/src/video_core/renderer_vulkan/vk_compute_pass.h @@ -41,22 +41,6 @@ private: vk::ShaderModule module; }; -class QuadArrayPass final : public VKComputePass { -public: - explicit QuadArrayPass(const Device& device_, VKScheduler& scheduler_, - VKDescriptorPool& descriptor_pool_, - StagingBufferPool& staging_buffer_pool_, - VKUpdateDescriptorQueue& update_descriptor_queue_); - ~QuadArrayPass(); - - std::pair<VkBuffer, VkDeviceSize> Assemble(u32 num_vertices, u32 first); - -private: - VKScheduler& scheduler; - StagingBufferPool& staging_buffer_pool; - VKUpdateDescriptorQueue& update_descriptor_queue; -}; - class Uint8Pass final : public VKComputePass { public: explicit Uint8Pass(const Device& device_, VKScheduler& scheduler_, @@ -64,7 +48,10 @@ public: VKUpdateDescriptorQueue& update_descriptor_queue_); ~Uint8Pass(); - std::pair<VkBuffer, u64> Assemble(u32 num_vertices, VkBuffer src_buffer, u64 src_offset); + /// Assemble uint8 indices into an uint16 index buffer + /// Returns a pair with the staging buffer, and the offset where the assembled data is + std::pair<VkBuffer, VkDeviceSize> Assemble(u32 num_vertices, VkBuffer src_buffer, + u32 src_offset); private: VKScheduler& scheduler; @@ -80,9 +67,9 @@ public: VKUpdateDescriptorQueue& update_descriptor_queue_); ~QuadIndexedPass(); - std::pair<VkBuffer, u64> Assemble(Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, - u32 num_vertices, u32 base_vertex, VkBuffer src_buffer, - u64 src_offset); + std::pair<VkBuffer, VkDeviceSize> Assemble( + Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, + u32 base_vertex, VkBuffer src_buffer, u32 src_offset); private: VKScheduler& scheduler; diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.cpp b/src/video_core/renderer_vulkan/vk_fence_manager.cpp index 6cd00884d..3bec48d14 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.cpp +++ b/src/video_core/renderer_vulkan/vk_fence_manager.cpp @@ -45,8 +45,8 @@ void InnerFence::Wait() { } VKFenceManager::VKFenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, - Tegra::MemoryManager& memory_manager_, TextureCache& texture_cache_, - VKBufferCache& buffer_cache_, VKQueryCache& query_cache_, + TextureCache& texture_cache_, BufferCache& buffer_cache_, + VKQueryCache& query_cache_, const Device& device_, VKScheduler& scheduler_) : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_}, scheduler{scheduler_} {} diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h index 9c5e5aa8f..2f8322d29 100644 --- a/src/video_core/renderer_vulkan/vk_fence_manager.h +++ b/src/video_core/renderer_vulkan/vk_fence_manager.h @@ -22,7 +22,6 @@ class RasterizerInterface; namespace Vulkan { class Device; -class VKBufferCache; class VKQueryCache; class VKScheduler; @@ -45,14 +44,14 @@ private: using Fence = std::shared_ptr<InnerFence>; using GenericFenceManager = - VideoCommon::FenceManager<Fence, TextureCache, VKBufferCache, VKQueryCache>; + VideoCommon::FenceManager<Fence, TextureCache, BufferCache, VKQueryCache>; class VKFenceManager final : public GenericFenceManager { public: - explicit VKFenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, - Tegra::MemoryManager& memory_manager_, TextureCache& texture_cache_, - VKBufferCache& buffer_cache_, VKQueryCache& query_cache_, - VKScheduler& scheduler_); + explicit VKFenceManager(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, + TextureCache& texture_cache, BufferCache& buffer_cache, + VKQueryCache& query_cache, const Device& device, + VKScheduler& scheduler); protected: Fence CreateFence(u32 value, bool is_stubbed) override; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index d50dca604..fc6dd83eb 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -221,9 +221,6 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const SPIRVProgram& program, std::vector<VkVertexInputBindingDescription> vertex_bindings; std::vector<VkVertexInputBindingDivisorDescriptionEXT> vertex_binding_divisors; for (std::size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { - if (state.attributes[index].binding_index_enabled == 0) { - continue; - } const bool instanced = state.binding_divisors[index] != 0; const auto rate = instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX; vertex_bindings.push_back({ diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.h b/src/video_core/renderer_vulkan/vk_master_semaphore.h index f336f1862..2c7ed654d 100644 --- a/src/video_core/renderer_vulkan/vk_master_semaphore.h +++ b/src/video_core/renderer_vulkan/vk_master_semaphore.h @@ -21,7 +21,12 @@ public: /// Returns the current logical tick. [[nodiscard]] u64 CurrentTick() const noexcept { - return current_tick; + return current_tick.load(std::memory_order_relaxed); + } + + /// Returns the last known GPU tick. + [[nodiscard]] u64 KnownGpuTick() const noexcept { + return gpu_tick.load(std::memory_order_relaxed); } /// Returns the timeline semaphore handle. @@ -31,7 +36,7 @@ public: /// Returns true when a tick has been hit by the GPU. [[nodiscard]] bool IsFree(u64 tick) { - return gpu_tick >= tick; + return gpu_tick.load(std::memory_order_relaxed) >= tick; } /// Advance to the logical tick. @@ -41,7 +46,7 @@ public: /// Refresh the known GPU tick void Refresh() { - gpu_tick = semaphore.GetCounter(); + gpu_tick.store(semaphore.GetCounter(), std::memory_order_relaxed); } /// Waits for a tick to be hit on the GPU diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index f0a111829..dfd38f575 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -8,8 +8,6 @@ #include <mutex> #include <vector> -#include <boost/container/static_vector.hpp> - #include "common/alignment.h" #include "common/assert.h" #include "common/logging/log.h" @@ -24,7 +22,6 @@ #include "video_core/renderer_vulkan/maxwell_to_vk.h" #include "video_core/renderer_vulkan/renderer_vulkan.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" -#include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_compute_pipeline.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_graphics_pipeline.h" @@ -50,15 +47,16 @@ MICROPROFILE_DEFINE(Vulkan_WaitForWorker, "Vulkan", "Wait for worker", MP_RGB(25 MICROPROFILE_DEFINE(Vulkan_Drawing, "Vulkan", "Record drawing", MP_RGB(192, 128, 128)); MICROPROFILE_DEFINE(Vulkan_Compute, "Vulkan", "Record compute", MP_RGB(192, 128, 128)); MICROPROFILE_DEFINE(Vulkan_Clearing, "Vulkan", "Record clearing", MP_RGB(192, 128, 128)); -MICROPROFILE_DEFINE(Vulkan_Geometry, "Vulkan", "Setup geometry", MP_RGB(192, 128, 128)); -MICROPROFILE_DEFINE(Vulkan_ConstBuffers, "Vulkan", "Setup constant buffers", MP_RGB(192, 128, 128)); -MICROPROFILE_DEFINE(Vulkan_GlobalBuffers, "Vulkan", "Setup global buffers", MP_RGB(192, 128, 128)); -MICROPROFILE_DEFINE(Vulkan_RenderTargets, "Vulkan", "Setup render targets", MP_RGB(192, 128, 128)); -MICROPROFILE_DEFINE(Vulkan_Textures, "Vulkan", "Setup textures", MP_RGB(192, 128, 128)); -MICROPROFILE_DEFINE(Vulkan_Images, "Vulkan", "Setup images", MP_RGB(192, 128, 128)); MICROPROFILE_DEFINE(Vulkan_PipelineCache, "Vulkan", "Pipeline cache", MP_RGB(192, 128, 128)); namespace { +struct DrawParams { + u32 base_instance; + u32 num_instances; + u32 base_vertex; + u32 num_vertices; + bool is_indexed; +}; constexpr auto COMPUTE_SHADER_INDEX = static_cast<size_t>(Tegra::Engines::ShaderType::Compute); @@ -67,7 +65,6 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in const float width = src.scale_x * 2.0f; const float height = src.scale_y * 2.0f; const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1.0f : 0.0f; - VkViewport viewport{ .x = src.translate_x - src.scale_x, .y = src.translate_y - src.scale_y, @@ -76,12 +73,10 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in .minDepth = src.translate_z - src.scale_z * reduce_z, .maxDepth = src.translate_z + src.scale_z, }; - if (!device.IsExtDepthRangeUnrestrictedSupported()) { viewport.minDepth = std::clamp(viewport.minDepth, 0.0f, 1.0f); viewport.maxDepth = std::clamp(viewport.maxDepth, 0.0f, 1.0f); } - return viewport; } @@ -146,13 +141,6 @@ TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index); } -template <size_t N> -std::array<VkDeviceSize, N> ExpandStrides(const std::array<u16, N>& strides) { - std::array<VkDeviceSize, N> expanded; - std::copy(strides.begin(), strides.end(), expanded.begin()); - return expanded; -} - ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) { if (entry.is_buffer) { return ImageViewType::e2D; @@ -221,190 +209,25 @@ void PushImageDescriptors(const ShaderEntries& entries, TextureCache& texture_ca } } -} // Anonymous namespace - -class BufferBindings final { -public: - void AddVertexBinding(VkBuffer buffer, VkDeviceSize offset, VkDeviceSize size, u32 stride) { - vertex.buffers[vertex.num_buffers] = buffer; - vertex.offsets[vertex.num_buffers] = offset; - vertex.sizes[vertex.num_buffers] = size; - vertex.strides[vertex.num_buffers] = static_cast<u16>(stride); - ++vertex.num_buffers; - } - - void SetIndexBinding(VkBuffer buffer, VkDeviceSize offset, VkIndexType type) { - index.buffer = buffer; - index.offset = offset; - index.type = type; - } - - void Bind(const Device& device, VKScheduler& scheduler) const { - // Use this large switch case to avoid dispatching more memory in the record lambda than - // what we need. It looks horrible, but it's the best we can do on standard C++. - switch (vertex.num_buffers) { - case 0: - return BindStatic<0>(device, scheduler); - case 1: - return BindStatic<1>(device, scheduler); - case 2: - return BindStatic<2>(device, scheduler); - case 3: - return BindStatic<3>(device, scheduler); - case 4: - return BindStatic<4>(device, scheduler); - case 5: - return BindStatic<5>(device, scheduler); - case 6: - return BindStatic<6>(device, scheduler); - case 7: - return BindStatic<7>(device, scheduler); - case 8: - return BindStatic<8>(device, scheduler); - case 9: - return BindStatic<9>(device, scheduler); - case 10: - return BindStatic<10>(device, scheduler); - case 11: - return BindStatic<11>(device, scheduler); - case 12: - return BindStatic<12>(device, scheduler); - case 13: - return BindStatic<13>(device, scheduler); - case 14: - return BindStatic<14>(device, scheduler); - case 15: - return BindStatic<15>(device, scheduler); - case 16: - return BindStatic<16>(device, scheduler); - case 17: - return BindStatic<17>(device, scheduler); - case 18: - return BindStatic<18>(device, scheduler); - case 19: - return BindStatic<19>(device, scheduler); - case 20: - return BindStatic<20>(device, scheduler); - case 21: - return BindStatic<21>(device, scheduler); - case 22: - return BindStatic<22>(device, scheduler); - case 23: - return BindStatic<23>(device, scheduler); - case 24: - return BindStatic<24>(device, scheduler); - case 25: - return BindStatic<25>(device, scheduler); - case 26: - return BindStatic<26>(device, scheduler); - case 27: - return BindStatic<27>(device, scheduler); - case 28: - return BindStatic<28>(device, scheduler); - case 29: - return BindStatic<29>(device, scheduler); - case 30: - return BindStatic<30>(device, scheduler); - case 31: - return BindStatic<31>(device, scheduler); - case 32: - return BindStatic<32>(device, scheduler); - } - UNREACHABLE(); - } - -private: - // Some of these fields are intentionally left uninitialized to avoid initializing them twice. - struct { - size_t num_buffers = 0; - std::array<VkBuffer, Maxwell::NumVertexArrays> buffers; - std::array<VkDeviceSize, Maxwell::NumVertexArrays> offsets; - std::array<VkDeviceSize, Maxwell::NumVertexArrays> sizes; - std::array<u16, Maxwell::NumVertexArrays> strides; - } vertex; - - struct { - VkBuffer buffer = nullptr; - VkDeviceSize offset; - VkIndexType type; - } index; - - template <size_t N> - void BindStatic(const Device& device, VKScheduler& scheduler) const { - if (device.IsExtExtendedDynamicStateSupported()) { - if (index.buffer) { - BindStatic<N, true, true>(scheduler); - } else { - BindStatic<N, false, true>(scheduler); - } - } else { - if (index.buffer) { - BindStatic<N, true, false>(scheduler); - } else { - BindStatic<N, false, false>(scheduler); - } - } - } - - template <size_t N, bool is_indexed, bool has_extended_dynamic_state> - void BindStatic(VKScheduler& scheduler) const { - static_assert(N <= Maxwell::NumVertexArrays); - if constexpr (N == 0) { - return; - } - - std::array<VkBuffer, N> buffers; - std::array<VkDeviceSize, N> offsets; - std::copy(vertex.buffers.begin(), vertex.buffers.begin() + N, buffers.begin()); - std::copy(vertex.offsets.begin(), vertex.offsets.begin() + N, offsets.begin()); - - if constexpr (has_extended_dynamic_state) { - // With extended dynamic states we can specify the length and stride of a vertex buffer - std::array<VkDeviceSize, N> sizes; - std::array<u16, N> strides; - std::copy(vertex.sizes.begin(), vertex.sizes.begin() + N, sizes.begin()); - std::copy(vertex.strides.begin(), vertex.strides.begin() + N, strides.begin()); - - if constexpr (is_indexed) { - scheduler.Record( - [buffers, offsets, sizes, strides, index = index](vk::CommandBuffer cmdbuf) { - cmdbuf.BindIndexBuffer(index.buffer, index.offset, index.type); - cmdbuf.BindVertexBuffers2EXT(0, static_cast<u32>(N), buffers.data(), - offsets.data(), sizes.data(), - ExpandStrides(strides).data()); - }); - } else { - scheduler.Record([buffers, offsets, sizes, strides](vk::CommandBuffer cmdbuf) { - cmdbuf.BindVertexBuffers2EXT(0, static_cast<u32>(N), buffers.data(), - offsets.data(), sizes.data(), - ExpandStrides(strides).data()); - }); - } - return; - } - - if constexpr (is_indexed) { - // Indexed draw - scheduler.Record([buffers, offsets, index = index](vk::CommandBuffer cmdbuf) { - cmdbuf.BindIndexBuffer(index.buffer, index.offset, index.type); - cmdbuf.BindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data()); - }); - } else { - // Array draw - scheduler.Record([buffers, offsets](vk::CommandBuffer cmdbuf) { - cmdbuf.BindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data()); - }); - } - } -}; - -void RasterizerVulkan::DrawParameters::Draw(vk::CommandBuffer cmdbuf) const { - if (is_indexed) { - cmdbuf.DrawIndexed(num_vertices, num_instances, 0, base_vertex, base_instance); - } else { - cmdbuf.Draw(num_vertices, num_instances, base_vertex, base_instance); +DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instanced, + bool is_indexed) { + DrawParams params{ + .base_instance = regs.vb_base_instance, + .num_instances = is_instanced ? num_instances : 1, + .base_vertex = is_indexed ? regs.vb_element_base : regs.vertex_buffer.first, + .num_vertices = is_indexed ? regs.index_array.count : regs.vertex_buffer.count, + .is_indexed = is_indexed, + }; + if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) { + // 6 triangle vertices per quad, base vertex is part of the index + // See BindQuadArrayIndexBuffer for more details + params.num_vertices = (params.num_vertices / 4) * 6; + params.base_vertex = 0; + params.is_indexed = true; } + return params; } +} // Anonymous namespace RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, Tegra::MemoryManager& gpu_memory_, @@ -414,21 +237,19 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra : 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_}, stream_buffer(device, scheduler), + 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), - quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), - quad_indexed_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), - uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), texture_cache_runtime{device, scheduler, memory_allocator, staging_pool, blit_image}, texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), + 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, gpu, maxwell3d, kepler_compute, gpu_memory, device, scheduler, descriptor_pool, update_descriptor_queue), - buffer_cache(*this, gpu_memory, cpu_memory_, device, memory_allocator, scheduler, - stream_buffer, staging_pool), query_cache{*this, maxwell3d, gpu_memory, device, scheduler}, - fence_manager(*this, gpu, gpu_memory, texture_cache, buffer_cache, query_cache, scheduler), + fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler), wfi_event(device.GetLogical().CreateEvent()), async_shaders(emu_window_) { scheduler.SetQueryCache(query_cache); if (device.UseAsynchronousShaders()) { @@ -446,52 +267,51 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) { query_cache.UpdateCounters(); - GraphicsPipelineCacheKey key; - key.fixed_state.Fill(maxwell3d.regs, device.IsExtExtendedDynamicStateSupported()); - - buffer_cache.Map(CalculateGraphicsStreamBufferSize(is_indexed)); + graphics_key.fixed_state.Refresh(maxwell3d, device.IsExtExtendedDynamicStateSupported()); - BufferBindings buffer_bindings; - const DrawParameters draw_params = - SetupGeometry(key.fixed_state, buffer_bindings, is_indexed, is_instanced); + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; - auto lock = texture_cache.AcquireLock(); texture_cache.SynchronizeGraphicsDescriptors(); - texture_cache.UpdateRenderTargets(false); const auto shaders = pipeline_cache.GetShaders(); - key.shaders = GetShaderAddresses(shaders); - SetupShaderDescriptors(shaders); + graphics_key.shaders = GetShaderAddresses(shaders); - buffer_cache.Unmap(); + SetupShaderDescriptors(shaders, is_indexed); const Framebuffer* const framebuffer = texture_cache.GetFramebuffer(); - key.renderpass = framebuffer->RenderPass(); + graphics_key.renderpass = framebuffer->RenderPass(); - auto* const pipeline = - pipeline_cache.GetGraphicsPipeline(key, framebuffer->NumColorBuffers(), async_shaders); + VKGraphicsPipeline* const pipeline = pipeline_cache.GetGraphicsPipeline( + graphics_key, framebuffer->NumColorBuffers(), async_shaders); if (pipeline == nullptr || pipeline->GetHandle() == VK_NULL_HANDLE) { // Async graphics pipeline was not ready. return; } - buffer_bindings.Bind(device, scheduler); - BeginTransformFeedback(); scheduler.RequestRenderpass(framebuffer); scheduler.BindGraphicsPipeline(pipeline->GetHandle()); UpdateDynamicStates(); - const auto pipeline_layout = pipeline->GetLayout(); - const auto descriptor_set = pipeline->CommitDescriptorSet(); + 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); + const VkPipelineLayout pipeline_layout = pipeline->GetLayout(); + const VkDescriptorSet descriptor_set = pipeline->CommitDescriptorSet(); scheduler.Record([pipeline_layout, descriptor_set, draw_params](vk::CommandBuffer cmdbuf) { if (descriptor_set) { cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout, - DESCRIPTOR_SET, descriptor_set, {}); + DESCRIPTOR_SET, descriptor_set, nullptr); + } + if (draw_params.is_indexed) { + cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances, 0, + draw_params.base_vertex, draw_params.base_instance); + } else { + cmdbuf.Draw(draw_params.num_vertices, draw_params.num_instances, + draw_params.base_vertex, draw_params.base_instance); } - draw_params.Draw(cmdbuf); }); EndTransformFeedback(); @@ -515,7 +335,7 @@ void RasterizerVulkan::Clear() { return; } - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.UpdateRenderTargets(true); const Framebuffer* const framebuffer = texture_cache.GetFramebuffer(); const VkExtent2D render_area = framebuffer->RenderArea(); @@ -559,7 +379,6 @@ void RasterizerVulkan::Clear() { if (use_stencil) { aspect_flags |= VK_IMAGE_ASPECT_STENCIL_BIT; } - scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil, clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) { VkClearAttachment attachment; @@ -580,12 +399,11 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) { auto& pipeline = pipeline_cache.GetComputePipeline({ .shader = code_addr, .shared_memory_size = launch_desc.shared_alloc, - .workgroup_size = - { - launch_desc.block_dim_x, - launch_desc.block_dim_y, - launch_desc.block_dim_z, - }, + .workgroup_size{ + launch_desc.block_dim_x, + launch_desc.block_dim_y, + launch_desc.block_dim_z, + }, }); // Compute dispatches can't be executed inside a renderpass @@ -594,10 +412,21 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) { image_view_indices.clear(); sampler_handles.clear(); - auto lock = texture_cache.AcquireLock(); - texture_cache.SynchronizeComputeDescriptors(); + std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex}; const auto& entries = pipeline.GetEntries(); + buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers); + buffer_cache.UnbindComputeStorageBuffers(); + u32 ssbo_index = 0; + for (const auto& buffer : entries.global_buffers) { + buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset, + buffer.is_written); + ++ssbo_index; + } + buffer_cache.UpdateComputeBuffers(); + + texture_cache.SynchronizeComputeDescriptors(); + SetupComputeUniformTexels(entries); SetupComputeTextures(entries); SetupComputeStorageTexels(entries); @@ -606,20 +435,15 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) { const std::span indices_span(image_view_indices.data(), image_view_indices.size()); texture_cache.FillComputeImageViews(indices_span, image_view_ids); - buffer_cache.Map(CalculateComputeStreamBufferSize()); - update_descriptor_queue.Acquire(); - SetupComputeConstBuffers(entries); - SetupComputeGlobalBuffers(entries); + buffer_cache.BindHostComputeBuffers(); ImageViewId* image_view_id_ptr = image_view_ids.data(); VkSampler* sampler_ptr = sampler_handles.data(); PushImageDescriptors(entries, texture_cache, update_descriptor_queue, image_view_id_ptr, sampler_ptr); - buffer_cache.Unmap(); - const VkPipeline pipeline_handle = pipeline.GetHandle(); const VkPipelineLayout pipeline_layout = pipeline.GetLayout(); const VkDescriptorSet descriptor_set = pipeline.CommitDescriptorSet(); @@ -644,6 +468,11 @@ void RasterizerVulkan::Query(GPUVAddr gpu_addr, VideoCore::QueryType type, query_cache.Query(gpu_addr, type, timestamp); } +void RasterizerVulkan::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, + u32 size) { + buffer_cache.BindGraphicsUniformBuffer(stage, index, gpu_addr, size); +} + void RasterizerVulkan::FlushAll() {} void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) { @@ -651,19 +480,23 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) { return; } { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.DownloadMemory(addr, size); } - buffer_cache.FlushRegion(addr, size); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.DownloadMemory(addr, size); + } query_cache.FlushRegion(addr, size); } bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size) { + std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex}; if (!Settings::IsGPULevelHigh()) { - return buffer_cache.MustFlushRegion(addr, size); + return buffer_cache.IsRegionGpuModified(addr, size); } return texture_cache.IsRegionGpuModified(addr, size) || - buffer_cache.MustFlushRegion(addr, size); + buffer_cache.IsRegionGpuModified(addr, size); } void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) { @@ -671,11 +504,14 @@ void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) { return; } { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); } + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.WriteMemory(addr, size); + } pipeline_cache.InvalidateRegion(addr, size); - buffer_cache.InvalidateRegion(addr, size); query_cache.InvalidateRegion(addr, size); } @@ -683,25 +519,34 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) { if (addr == 0 || size == 0) { return; } + pipeline_cache.OnCPUWrite(addr, size); { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.WriteMemory(addr, size); } - pipeline_cache.OnCPUWrite(addr, size); - buffer_cache.OnCPUWrite(addr, size); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.CachedWriteMemory(addr, size); + } } void RasterizerVulkan::SyncGuestHost() { - buffer_cache.SyncGuestHost(); pipeline_cache.SyncGuestHost(); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.FlushCachedWrites(); + } } void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) { { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.UnmapMemory(addr, size); } - buffer_cache.OnCPUWrite(addr, size); + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.WriteMemory(addr, size); + } pipeline_cache.OnCPUWrite(addr, size); } @@ -774,18 +619,21 @@ void RasterizerVulkan::TickFrame() { draw_counter = 0; update_descriptor_queue.TickFrame(); fence_manager.TickFrame(); - buffer_cache.TickFrame(); staging_pool.TickFrame(); { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.TickFrame(); } + { + std::scoped_lock lock{buffer_cache.mutex}; + buffer_cache.TickFrame(); + } } bool RasterizerVulkan::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src, const Tegra::Engines::Fermi2D::Surface& dst, const Tegra::Engines::Fermi2D::Config& copy_config) { - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; texture_cache.BlitImage(dst, src, copy_config); return true; } @@ -795,13 +643,11 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config, if (!framebuffer_addr) { return false; } - - auto lock = texture_cache.AcquireLock(); + std::scoped_lock lock{texture_cache.mutex}; ImageView* const image_view = texture_cache.TryFindFramebufferImageView(framebuffer_addr); if (!image_view) { return false; } - screen_info.image_view = image_view->Handle(VideoCommon::ImageViewType::e2D); screen_info.width = image_view->size.width; screen_info.height = image_view->size.height; @@ -830,29 +676,8 @@ void RasterizerVulkan::FlushWork() { draw_counter = 0; } -RasterizerVulkan::DrawParameters RasterizerVulkan::SetupGeometry(FixedPipelineState& fixed_state, - BufferBindings& buffer_bindings, - bool is_indexed, - bool is_instanced) { - MICROPROFILE_SCOPE(Vulkan_Geometry); - - const auto& regs = maxwell3d.regs; - - SetupVertexArrays(buffer_bindings); - - const u32 base_instance = regs.vb_base_instance; - const u32 num_instances = is_instanced ? maxwell3d.mme_draw.instance_count : 1; - const u32 base_vertex = is_indexed ? regs.vb_element_base : regs.vertex_buffer.first; - const u32 num_vertices = is_indexed ? regs.index_array.count : regs.vertex_buffer.count; - - DrawParameters params{base_instance, num_instances, base_vertex, num_vertices, is_indexed}; - SetupIndexBuffer(buffer_bindings, params, is_indexed); - - return params; -} - void RasterizerVulkan::SetupShaderDescriptors( - const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) { + const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders, bool is_indexed) { image_view_indices.clear(); sampler_handles.clear(); for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { @@ -860,15 +685,27 @@ void RasterizerVulkan::SetupShaderDescriptors( if (!shader) { continue; } - const auto& entries = shader->GetEntries(); + const ShaderEntries& entries = shader->GetEntries(); SetupGraphicsUniformTexels(entries, stage); SetupGraphicsTextures(entries, stage); SetupGraphicsStorageTexels(entries, stage); SetupGraphicsImages(entries, stage); + + buffer_cache.SetEnabledUniformBuffers(stage, entries.enabled_uniform_buffers); + buffer_cache.UnbindGraphicsStorageBuffers(stage); + u32 ssbo_index = 0; + for (const auto& buffer : entries.global_buffers) { + buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index, + buffer.cbuf_offset, buffer.is_written); + ++ssbo_index; + } } const std::span indices_span(image_view_indices.data(), image_view_indices.size()); + buffer_cache.UpdateGraphicsBuffers(is_indexed); texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + buffer_cache.BindHostGeometryBuffers(is_indexed); + update_descriptor_queue.Acquire(); ImageViewId* image_view_id_ptr = image_view_ids.data(); @@ -879,11 +716,9 @@ void RasterizerVulkan::SetupShaderDescriptors( if (!shader) { continue; } - const auto& entries = shader->GetEntries(); - SetupGraphicsConstBuffers(entries, stage); - SetupGraphicsGlobalBuffers(entries, stage); - PushImageDescriptors(entries, texture_cache, update_descriptor_queue, image_view_id_ptr, - sampler_ptr); + buffer_cache.BindHostStageBuffers(stage); + PushImageDescriptors(shader->GetEntries(), texture_cache, update_descriptor_queue, + image_view_id_ptr, sampler_ptr); } } @@ -916,27 +751,11 @@ void RasterizerVulkan::BeginTransformFeedback() { LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported"); return; } - UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) || regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry)); - - UNIMPLEMENTED_IF(regs.tfb_bindings[1].buffer_enable); - UNIMPLEMENTED_IF(regs.tfb_bindings[2].buffer_enable); - UNIMPLEMENTED_IF(regs.tfb_bindings[3].buffer_enable); - - const auto& binding = regs.tfb_bindings[0]; - UNIMPLEMENTED_IF(binding.buffer_enable == 0); - UNIMPLEMENTED_IF(binding.buffer_offset != 0); - - const GPUVAddr gpu_addr = binding.Address(); - const VkDeviceSize size = static_cast<VkDeviceSize>(binding.buffer_size); - const auto info = buffer_cache.UploadMemory(gpu_addr, size, 4, true); - - scheduler.Record([buffer = info.handle, offset = info.offset, size](vk::CommandBuffer cmdbuf) { - cmdbuf.BindTransformFeedbackBuffersEXT(0, 1, &buffer, &offset, &size); - cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr); - }); + scheduler.Record( + [](vk::CommandBuffer cmdbuf) { cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr); }); } void RasterizerVulkan::EndTransformFeedback() { @@ -947,104 +766,11 @@ void RasterizerVulkan::EndTransformFeedback() { if (!device.IsExtTransformFeedbackSupported()) { return; } - scheduler.Record( [](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); }); } -void RasterizerVulkan::SetupVertexArrays(BufferBindings& buffer_bindings) { - const auto& regs = maxwell3d.regs; - - for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) { - const auto& vertex_array = regs.vertex_array[index]; - if (!vertex_array.IsEnabled()) { - continue; - } - const GPUVAddr start{vertex_array.StartAddress()}; - const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()}; - - ASSERT(end >= start); - const size_t size = end - start; - if (size == 0) { - buffer_bindings.AddVertexBinding(DefaultBuffer(), 0, DEFAULT_BUFFER_SIZE, 0); - continue; - } - const auto info = buffer_cache.UploadMemory(start, size); - buffer_bindings.AddVertexBinding(info.handle, info.offset, size, vertex_array.stride); - } -} - -void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawParameters& params, - bool is_indexed) { - if (params.num_vertices == 0) { - return; - } - const auto& regs = maxwell3d.regs; - switch (regs.draw.topology) { - case Maxwell::PrimitiveTopology::Quads: { - if (!params.is_indexed) { - const auto [buffer, offset] = - quad_array_pass.Assemble(params.num_vertices, params.base_vertex); - buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32); - params.base_vertex = 0; - params.num_vertices = params.num_vertices * 6 / 4; - params.is_indexed = true; - break; - } - const GPUVAddr gpu_addr = regs.index_array.IndexStart(); - const auto info = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize()); - VkBuffer buffer = info.handle; - u64 offset = info.offset; - std::tie(buffer, offset) = quad_indexed_pass.Assemble( - regs.index_array.format, params.num_vertices, params.base_vertex, buffer, offset); - - buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32); - params.num_vertices = (params.num_vertices / 4) * 6; - params.base_vertex = 0; - break; - } - default: { - if (!is_indexed) { - break; - } - const GPUVAddr gpu_addr = regs.index_array.IndexStart(); - const auto info = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize()); - VkBuffer buffer = info.handle; - u64 offset = info.offset; - - auto format = regs.index_array.format; - const bool is_uint8 = format == Maxwell::IndexFormat::UnsignedByte; - if (is_uint8 && !device.IsExtIndexTypeUint8Supported()) { - std::tie(buffer, offset) = uint8_pass.Assemble(params.num_vertices, buffer, offset); - format = Maxwell::IndexFormat::UnsignedShort; - } - - buffer_bindings.SetIndexBinding(buffer, offset, MaxwellToVK::IndexFormat(device, format)); - break; - } - } -} - -void RasterizerVulkan::SetupGraphicsConstBuffers(const ShaderEntries& entries, size_t stage) { - MICROPROFILE_SCOPE(Vulkan_ConstBuffers); - const auto& shader_stage = maxwell3d.state.shader_stages[stage]; - for (const auto& entry : entries.const_buffers) { - SetupConstBuffer(entry, shader_stage.const_buffers[entry.GetIndex()]); - } -} - -void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries, size_t stage) { - MICROPROFILE_SCOPE(Vulkan_GlobalBuffers); - const auto& cbufs{maxwell3d.state.shader_stages[stage]}; - - for (const auto& entry : entries.global_buffers) { - const auto addr = cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset(); - SetupGlobalBuffer(entry, addr); - } -} - void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, size_t stage) { - MICROPROFILE_SCOPE(Vulkan_Textures); const auto& regs = maxwell3d.regs; const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; for (const auto& entry : entries.uniform_texels) { @@ -1054,7 +780,6 @@ void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, } void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, size_t stage) { - MICROPROFILE_SCOPE(Vulkan_Textures); const auto& regs = maxwell3d.regs; const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; for (const auto& entry : entries.samplers) { @@ -1070,7 +795,6 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, size_ } void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, size_t stage) { - MICROPROFILE_SCOPE(Vulkan_Textures); const auto& regs = maxwell3d.regs; const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; for (const auto& entry : entries.storage_texels) { @@ -1080,7 +804,6 @@ void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, } void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, size_t stage) { - MICROPROFILE_SCOPE(Vulkan_Images); const auto& regs = maxwell3d.regs; const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; for (const auto& entry : entries.images) { @@ -1089,32 +812,7 @@ void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, size_t } } -void RasterizerVulkan::SetupComputeConstBuffers(const ShaderEntries& entries) { - MICROPROFILE_SCOPE(Vulkan_ConstBuffers); - const auto& launch_desc = kepler_compute.launch_description; - for (const auto& entry : entries.const_buffers) { - const auto& config = launch_desc.const_buffer_config[entry.GetIndex()]; - const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value(); - const Tegra::Engines::ConstBufferInfo info{ - .address = config.Address(), - .size = config.size, - .enabled = mask[entry.GetIndex()], - }; - SetupConstBuffer(entry, info); - } -} - -void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) { - MICROPROFILE_SCOPE(Vulkan_GlobalBuffers); - const auto& cbufs{kepler_compute.launch_description.const_buffer_config}; - for (const auto& entry : entries.global_buffers) { - const auto addr{cbufs[entry.GetCbufIndex()].Address() + entry.GetCbufOffset()}; - SetupGlobalBuffer(entry, addr); - } -} - void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) { - MICROPROFILE_SCOPE(Vulkan_Textures); const bool via_header_index = kepler_compute.launch_description.linked_tsc; for (const auto& entry : entries.uniform_texels) { const TextureHandle handle = @@ -1124,7 +822,6 @@ void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) { } void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) { - MICROPROFILE_SCOPE(Vulkan_Textures); const bool via_header_index = kepler_compute.launch_description.linked_tsc; for (const auto& entry : entries.samplers) { for (size_t index = 0; index < entry.size; ++index) { @@ -1139,7 +836,6 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) { } void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) { - MICROPROFILE_SCOPE(Vulkan_Textures); const bool via_header_index = kepler_compute.launch_description.linked_tsc; for (const auto& entry : entries.storage_texels) { const TextureHandle handle = @@ -1149,7 +845,6 @@ void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) { } void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) { - MICROPROFILE_SCOPE(Vulkan_Images); const bool via_header_index = kepler_compute.launch_description.linked_tsc; for (const auto& entry : entries.images) { const TextureHandle handle = @@ -1158,42 +853,6 @@ void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) { } } -void RasterizerVulkan::SetupConstBuffer(const ConstBufferEntry& entry, - const Tegra::Engines::ConstBufferInfo& buffer) { - if (!buffer.enabled) { - // Set values to zero to unbind buffers - update_descriptor_queue.AddBuffer(DefaultBuffer(), 0, DEFAULT_BUFFER_SIZE); - return; - } - // Align the size to avoid bad std140 interactions - const size_t size = Common::AlignUp(CalculateConstBufferSize(entry, buffer), 4 * sizeof(float)); - ASSERT(size <= MaxConstbufferSize); - - const u64 alignment = device.GetUniformBufferAlignment(); - const auto info = buffer_cache.UploadMemory(buffer.address, size, alignment); - update_descriptor_queue.AddBuffer(info.handle, info.offset, size); -} - -void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address) { - const u64 actual_addr = gpu_memory.Read<u64>(address); - const u32 size = gpu_memory.Read<u32>(address + 8); - - if (size == 0) { - // Sometimes global memory pointers don't have a proper size. Upload a dummy entry - // because Vulkan doesn't like empty buffers. - // Note: Do *not* use DefaultBuffer() here, storage buffers can be written breaking the - // default buffer. - static constexpr size_t dummy_size = 4; - const auto info = buffer_cache.GetEmptyBuffer(dummy_size); - update_descriptor_queue.AddBuffer(info.handle, info.offset, dummy_size); - return; - } - - const auto info = buffer_cache.UploadMemory( - actual_addr, size, device.GetStorageBufferAlignment(), entry.IsWritten()); - update_descriptor_queue.AddBuffer(info.handle, info.offset, size); -} - void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs) { if (!state_tracker.TouchViewports()) { return; @@ -1206,7 +865,8 @@ void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& reg GetViewportState(device, regs, 8), GetViewportState(device, regs, 9), GetViewportState(device, regs, 10), GetViewportState(device, regs, 11), GetViewportState(device, regs, 12), GetViewportState(device, regs, 13), - GetViewportState(device, regs, 14), GetViewportState(device, regs, 15)}; + GetViewportState(device, regs, 14), GetViewportState(device, regs, 15), + }; scheduler.Record([viewports](vk::CommandBuffer cmdbuf) { cmdbuf.SetViewport(0, viewports); }); } @@ -1214,13 +874,14 @@ void RasterizerVulkan::UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs if (!state_tracker.TouchScissors()) { return; } - const std::array scissors = { + const std::array scissors{ GetScissorState(regs, 0), GetScissorState(regs, 1), GetScissorState(regs, 2), GetScissorState(regs, 3), GetScissorState(regs, 4), GetScissorState(regs, 5), GetScissorState(regs, 6), GetScissorState(regs, 7), GetScissorState(regs, 8), GetScissorState(regs, 9), GetScissorState(regs, 10), GetScissorState(regs, 11), GetScissorState(regs, 12), GetScissorState(regs, 13), GetScissorState(regs, 14), - GetScissorState(regs, 15)}; + GetScissorState(regs, 15), + }; scheduler.Record([scissors](vk::CommandBuffer cmdbuf) { cmdbuf.SetScissor(0, scissors); }); } @@ -1385,73 +1046,4 @@ void RasterizerVulkan::UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& }); } -size_t RasterizerVulkan::CalculateGraphicsStreamBufferSize(bool is_indexed) const { - size_t size = CalculateVertexArraysSize(); - if (is_indexed) { - size = Common::AlignUp(size, 4) + CalculateIndexBufferSize(); - } - size += Maxwell::MaxConstBuffers * (MaxConstbufferSize + device.GetUniformBufferAlignment()); - return size; -} - -size_t RasterizerVulkan::CalculateComputeStreamBufferSize() const { - return Tegra::Engines::KeplerCompute::NumConstBuffers * - (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment()); -} - -size_t RasterizerVulkan::CalculateVertexArraysSize() const { - const auto& regs = maxwell3d.regs; - - size_t size = 0; - for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) { - // This implementation assumes that all attributes are used in the shader. - const GPUVAddr start{regs.vertex_array[index].StartAddress()}; - const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()}; - DEBUG_ASSERT(end >= start); - - size += (end - start) * regs.vertex_array[index].enable; - } - return size; -} - -size_t RasterizerVulkan::CalculateIndexBufferSize() const { - return static_cast<size_t>(maxwell3d.regs.index_array.count) * - static_cast<size_t>(maxwell3d.regs.index_array.FormatSizeInBytes()); -} - -size_t RasterizerVulkan::CalculateConstBufferSize( - const ConstBufferEntry& entry, const Tegra::Engines::ConstBufferInfo& buffer) const { - if (entry.IsIndirect()) { - // Buffer is accessed indirectly, so upload the entire thing - return buffer.size; - } else { - // Buffer is accessed directly, upload just what we use - return entry.GetSize(); - } -} - -VkBuffer RasterizerVulkan::DefaultBuffer() { - if (default_buffer) { - return *default_buffer; - } - default_buffer = device.GetLogical().CreateBuffer({ - .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, - .pNext = nullptr, - .flags = 0, - .size = DEFAULT_BUFFER_SIZE, - .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | - VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, - .sharingMode = VK_SHARING_MODE_EXCLUSIVE, - .queueFamilyIndexCount = 0, - .pQueueFamilyIndices = nullptr, - }); - default_buffer_commit = memory_allocator.Commit(default_buffer, MemoryUsage::DeviceLocal); - - scheduler.RequestOutsideRenderPassOperationContext(); - scheduler.Record([buffer = *default_buffer](vk::CommandBuffer cmdbuf) { - cmdbuf.FillBuffer(buffer, 0, DEFAULT_BUFFER_SIZE, 0); - }); - return *default_buffer; -} - } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 8e261b9bd..acea1ba2d 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -18,14 +18,13 @@ #include "video_core/renderer_vulkan/blit_image.h" #include "video_core/renderer_vulkan/fixed_pipeline_state.h" #include "video_core/renderer_vulkan/vk_buffer_cache.h" -#include "video_core/renderer_vulkan/vk_compute_pass.h" #include "video_core/renderer_vulkan/vk_descriptor_pool.h" #include "video_core/renderer_vulkan/vk_fence_manager.h" +#include "video_core/renderer_vulkan/vk_graphics_pipeline.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h" #include "video_core/renderer_vulkan/vk_query_cache.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" -#include "video_core/renderer_vulkan/vk_stream_buffer.h" #include "video_core/renderer_vulkan/vk_texture_cache.h" #include "video_core/renderer_vulkan/vk_update_descriptor.h" #include "video_core/shader/async_shaders.h" @@ -49,7 +48,6 @@ namespace Vulkan { struct VKScreenInfo; class StateTracker; -class BufferBindings; class RasterizerVulkan final : public VideoCore::RasterizerAccelerated { public: @@ -65,6 +63,7 @@ public: void DispatchCompute(GPUVAddr code_addr) override; void ResetCounter(VideoCore::QueryType type) override; void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override; + void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override; void FlushAll() override; void FlushRegion(VAddr addr, u64 size) override; bool MustFlushRegion(VAddr addr, u64 size) override; @@ -107,24 +106,11 @@ private: static constexpr VkDeviceSize DEFAULT_BUFFER_SIZE = 4 * sizeof(float); - struct DrawParameters { - void Draw(vk::CommandBuffer cmdbuf) const; - - u32 base_instance = 0; - u32 num_instances = 0; - u32 base_vertex = 0; - u32 num_vertices = 0; - bool is_indexed = 0; - }; - void FlushWork(); - /// Setups geometry buffers and state. - DrawParameters SetupGeometry(FixedPipelineState& fixed_state, BufferBindings& buffer_bindings, - bool is_indexed, bool is_instanced); - /// Setup descriptors in the graphics pipeline. - void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders); + void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders, + bool is_indexed); void UpdateDynamicStates(); @@ -132,16 +118,6 @@ private: void EndTransformFeedback(); - void SetupVertexArrays(BufferBindings& buffer_bindings); - - void SetupIndexBuffer(BufferBindings& buffer_bindings, DrawParameters& params, bool is_indexed); - - /// Setup constant buffers in the graphics pipeline. - void SetupGraphicsConstBuffers(const ShaderEntries& entries, std::size_t stage); - - /// Setup global buffers in the graphics pipeline. - void SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage); - /// Setup uniform texels in the graphics pipeline. void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage); @@ -154,12 +130,6 @@ private: /// Setup images in the graphics pipeline. void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage); - /// Setup constant buffers in the compute pipeline. - void SetupComputeConstBuffers(const ShaderEntries& entries); - - /// Setup global buffers in the compute pipeline. - void SetupComputeGlobalBuffers(const ShaderEntries& entries); - /// Setup texel buffers in the compute pipeline. void SetupComputeUniformTexels(const ShaderEntries& entries); @@ -172,11 +142,6 @@ private: /// Setup images in the compute pipeline. void SetupComputeImages(const ShaderEntries& entries); - void SetupConstBuffer(const ConstBufferEntry& entry, - const Tegra::Engines::ConstBufferInfo& buffer); - - void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address); - void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs); @@ -193,19 +158,6 @@ private: void UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs); void UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs); - size_t CalculateGraphicsStreamBufferSize(bool is_indexed) const; - - size_t CalculateComputeStreamBufferSize() const; - - size_t CalculateVertexArraysSize() const; - - size_t CalculateIndexBufferSize() const; - - size_t CalculateConstBufferSize(const ConstBufferEntry& entry, - const Tegra::Engines::ConstBufferInfo& buffer) const; - - VkBuffer DefaultBuffer(); - Tegra::GPU& gpu; Tegra::MemoryManager& gpu_memory; Tegra::Engines::Maxwell3D& maxwell3d; @@ -217,24 +169,21 @@ private: StateTracker& state_tracker; VKScheduler& scheduler; - VKStreamBuffer stream_buffer; StagingBufferPool staging_pool; VKDescriptorPool descriptor_pool; VKUpdateDescriptorQueue update_descriptor_queue; BlitImageHelper blit_image; - QuadArrayPass quad_array_pass; - QuadIndexedPass quad_indexed_pass; - Uint8Pass uint8_pass; + + GraphicsPipelineCacheKey graphics_key; TextureCacheRuntime texture_cache_runtime; TextureCache texture_cache; + BufferCacheRuntime buffer_cache_runtime; + BufferCache buffer_cache; VKPipelineCache pipeline_cache; - VKBufferCache buffer_cache; VKQueryCache query_cache; VKFenceManager fence_manager; - vk::Buffer default_buffer; - MemoryCommit default_buffer_commit; vk::Event wfi_event; VideoCommon::Shader::AsyncShaders async_shaders; diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.cpp b/src/video_core/renderer_vulkan/vk_resource_pool.cpp index ee274ac59..a8bf7bda8 100644 --- a/src/video_core/renderer_vulkan/vk_resource_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_resource_pool.cpp @@ -17,21 +17,21 @@ ResourcePool::~ResourcePool() = default; size_t ResourcePool::CommitResource() { // Refresh semaphore to query updated results master_semaphore.Refresh(); - - const auto search = [this](size_t begin, size_t end) -> std::optional<size_t> { + const u64 gpu_tick = master_semaphore.KnownGpuTick(); + const auto search = [this, gpu_tick](size_t begin, size_t end) -> std::optional<size_t> { for (size_t iterator = begin; iterator < end; ++iterator) { - if (master_semaphore.IsFree(ticks[iterator])) { + if (gpu_tick >= ticks[iterator]) { ticks[iterator] = master_semaphore.CurrentTick(); return iterator; } } - return {}; + return std::nullopt; }; // Try to find a free resource from the hinted position to the end. - auto found = search(free_iterator, ticks.size()); + std::optional<size_t> found = search(hint_iterator, ticks.size()); if (!found) { // Search from beginning to the hinted position. - found = search(0, free_iterator); + found = search(0, hint_iterator); if (!found) { // Both searches failed, the pool is full; handle it. const size_t free_resource = ManageOverflow(); @@ -41,7 +41,7 @@ size_t ResourcePool::CommitResource() { } } // Free iterator is hinted to the resource after the one that's been commited. - free_iterator = (*found + 1) % ticks.size(); + hint_iterator = (*found + 1) % ticks.size(); return *found; } diff --git a/src/video_core/renderer_vulkan/vk_resource_pool.h b/src/video_core/renderer_vulkan/vk_resource_pool.h index a018c7ec2..9d0bb3b4d 100644 --- a/src/video_core/renderer_vulkan/vk_resource_pool.h +++ b/src/video_core/renderer_vulkan/vk_resource_pool.h @@ -36,7 +36,7 @@ private: MasterSemaphore& master_semaphore; size_t grow_step = 0; ///< Number of new resources created after an overflow - size_t free_iterator = 0; ///< Hint to where the next free resources is likely to be found + size_t hint_iterator = 0; ///< Hint to where the next free resources is likely to be found std::vector<u64> ticks; ///< Ticks for each resource }; diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 66004f9c0..f35c120b0 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -52,18 +52,6 @@ VKScheduler::~VKScheduler() { worker_thread.join(); } -u64 VKScheduler::CurrentTick() const noexcept { - return master_semaphore->CurrentTick(); -} - -bool VKScheduler::IsFree(u64 tick) const noexcept { - return master_semaphore->IsFree(tick); -} - -void VKScheduler::Wait(u64 tick) { - master_semaphore->Wait(tick); -} - void VKScheduler::Flush(VkSemaphore semaphore) { SubmitExecution(semaphore); AllocateNewContext(); @@ -269,7 +257,7 @@ void VKScheduler::EndRenderPass() { cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, - VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, 0, nullptr, nullptr, + VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, nullptr, nullptr, vk::Span(barriers.data(), num_images)); }); state.renderpass = nullptr; diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 15f2987eb..3ce48e9d2 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -14,6 +14,7 @@ #include "common/alignment.h" #include "common/common_types.h" #include "common/threadsafe_queue.h" +#include "video_core/renderer_vulkan/vk_master_semaphore.h" #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { @@ -21,7 +22,6 @@ namespace Vulkan { class CommandPool; class Device; class Framebuffer; -class MasterSemaphore; class StateTracker; class VKQueryCache; @@ -32,15 +32,6 @@ public: explicit VKScheduler(const Device& device, StateTracker& state_tracker); ~VKScheduler(); - /// Returns the current command buffer tick. - [[nodiscard]] u64 CurrentTick() const noexcept; - - /// Returns true when a tick has been triggered by the GPU. - [[nodiscard]] bool IsFree(u64 tick) const noexcept; - - /// Waits for the given tick to trigger on the GPU. - void Wait(u64 tick); - /// Sends the current execution context to the GPU. void Flush(VkSemaphore semaphore = nullptr); @@ -82,6 +73,21 @@ public: (void)chunk->Record(command); } + /// Returns the current command buffer tick. + [[nodiscard]] u64 CurrentTick() const noexcept { + return master_semaphore->CurrentTick(); + } + + /// Returns true when a tick has been triggered by the GPU. + [[nodiscard]] bool IsFree(u64 tick) const noexcept { + return master_semaphore->IsFree(tick); + } + + /// Waits for the given tick to trigger on the GPU. + void Wait(u64 tick) { + master_semaphore->Wait(tick); + } + /// Returns the master timeline semaphore. [[nodiscard]] MasterSemaphore& GetMasterSemaphore() const noexcept { return *master_semaphore; diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp index 61d52b961..40e2e0d38 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp @@ -3106,7 +3106,11 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) { entries.const_buffers.emplace_back(cbuf.second, cbuf.first); } for (const auto& [base, usage] : ir.GetGlobalMemory()) { - entries.global_buffers.emplace_back(base.cbuf_index, base.cbuf_offset, usage.is_written); + entries.global_buffers.emplace_back(GlobalBufferEntry{ + .cbuf_index = base.cbuf_index, + .cbuf_offset = base.cbuf_offset, + .is_written = usage.is_written, + }); } for (const auto& sampler : ir.GetSamplers()) { if (sampler.is_buffer) { @@ -3127,6 +3131,9 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) { entries.attributes.insert(GetGenericAttributeLocation(attribute)); } } + for (const auto& buffer : entries.const_buffers) { + entries.enabled_uniform_buffers |= 1U << buffer.GetIndex(); + } entries.clip_distances = ir.GetClipDistances(); entries.shader_length = ir.GetLength(); entries.uses_warps = ir.UsesWarps(); diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h index 26381e444..5d94132a5 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h @@ -39,24 +39,7 @@ private: u32 index{}; }; -class GlobalBufferEntry { -public: - constexpr explicit GlobalBufferEntry(u32 cbuf_index_, u32 cbuf_offset_, bool is_written_) - : cbuf_index{cbuf_index_}, cbuf_offset{cbuf_offset_}, is_written{is_written_} {} - - constexpr u32 GetCbufIndex() const { - return cbuf_index; - } - - constexpr u32 GetCbufOffset() const { - return cbuf_offset; - } - - constexpr bool IsWritten() const { - return is_written; - } - -private: +struct GlobalBufferEntry { u32 cbuf_index{}; u32 cbuf_offset{}; bool is_written{}; @@ -78,6 +61,7 @@ struct ShaderEntries { std::set<u32> attributes; std::array<bool, Maxwell::NumClipDistances> clip_distances{}; std::size_t shader_length{}; + u32 enabled_uniform_buffers{}; bool uses_warps{}; }; 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 97fd41cc1..7a1232497 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp @@ -8,6 +8,7 @@ #include <fmt/format.h> +#include "common/alignment.h" #include "common/assert.h" #include "common/bit_util.h" #include "common/common_types.h" @@ -17,18 +18,119 @@ #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { +namespace { +// Maximum potential alignment of a Vulkan buffer +constexpr VkDeviceSize MAX_ALIGNMENT = 256; +// Maximum size to put elements in the stream buffer +constexpr VkDeviceSize MAX_STREAM_BUFFER_REQUEST_SIZE = 8 * 1024 * 1024; +// Stream buffer size in bytes +constexpr VkDeviceSize STREAM_BUFFER_SIZE = 128 * 1024 * 1024; +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; +} + +std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask, + VkMemoryPropertyFlags flags) noexcept { + for (u32 type_index = 0; type_index < props.memoryTypeCount; ++type_index) { + if (((type_mask >> type_index) & 1) == 0) { + // Memory type is incompatible + continue; + } + const VkMemoryType& memory_type = props.memoryTypes[type_index]; + if ((memory_type.propertyFlags & flags) != flags) { + // Memory type doesn't have the flags we want + continue; + } + if (!IsStreamHeap(props.memoryHeaps[memory_type.heapIndex])) { + // Memory heap is not suitable for streaming + continue; + } + // Success! + return type_index; + } + return std::nullopt; +} + +u32 FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask) { + // Try to find a DEVICE_LOCAL_BIT type, Nvidia and AMD have a dedicated heap for this + std::optional<u32> type = FindMemoryTypeIndex(props, type_mask, STREAM_FLAGS); + if (type) { + return *type; + } + // Otherwise try without the DEVICE_LOCAL_BIT + type = FindMemoryTypeIndex(props, type_mask, HOST_FLAGS); + if (type) { + return *type; + } + // This should never happen, and in case it does, signal it as an out of memory situation + throw vk::Exception(VK_ERROR_OUT_OF_DEVICE_MEMORY); +} + +size_t Region(size_t iterator) noexcept { + return iterator / REGION_SIZE; +} +} // Anonymous namespace StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& memory_allocator_, VKScheduler& scheduler_) - : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_} {} + : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_} { + 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, + .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | + VK_BUFFER_USAGE_INDEX_BUFFER_BIT, + .sharingMode = VK_SHARING_MODE_EXCLUSIVE, + .queueFamilyIndexCount = 0, + .pQueueFamilyIndices = nullptr, + }); + if (device.HasDebuggingToolAttached()) { + stream_buffer.SetObjectNameEXT("Stream Buffer"); + } + VkMemoryDedicatedRequirements dedicated_reqs{ + .sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS, + .pNext = nullptr, + .prefersDedicatedAllocation = VK_FALSE, + .requiresDedicatedAllocation = VK_FALSE, + }; + const auto requirements = dev.GetBufferMemoryRequirements(*stream_buffer, &dedicated_reqs); + const bool make_dedicated = dedicated_reqs.prefersDedicatedAllocation == VK_TRUE || + dedicated_reqs.requiresDedicatedAllocation == VK_TRUE; + const VkMemoryDedicatedAllocateInfo dedicated_info{ + .sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO, + .pNext = nullptr, + .image = nullptr, + .buffer = *stream_buffer, + }; + const auto memory_properties = device.GetPhysical().GetMemoryProperties(); + stream_memory = dev.AllocateMemory(VkMemoryAllocateInfo{ + .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, + .pNext = make_dedicated ? &dedicated_info : nullptr, + .allocationSize = requirements.size, + .memoryTypeIndex = FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits), + }); + if (device.HasDebuggingToolAttached()) { + stream_memory.SetObjectNameEXT("Stream Buffer Memory"); + } + stream_buffer.BindMemory(*stream_memory, 0); + stream_pointer = stream_memory.Map(0, STREAM_BUFFER_SIZE); +} StagingBufferPool::~StagingBufferPool() = default; StagingBufferRef StagingBufferPool::Request(size_t size, MemoryUsage usage) { - if (const std::optional<StagingBufferRef> ref = TryGetReservedBuffer(size, usage)) { - return *ref; + if (usage == MemoryUsage::Upload && size <= MAX_STREAM_BUFFER_REQUEST_SIZE) { + return GetStreamBuffer(size); } - return CreateStagingBuffer(size, usage); + return GetStagingBuffer(size, usage); } void StagingBufferPool::TickFrame() { @@ -39,6 +141,52 @@ void StagingBufferPool::TickFrame() { ReleaseCache(MemoryUsage::Download); } +StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) { + if (AreRegionsActive(Region(free_iterator) + 1, + std::min(Region(iterator + 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); + 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); + used_iterator = 0; + iterator = 0; + free_iterator = size; + + if (AreRegionsActive(0, Region(size) + 1)) { + // Avoid waiting for the previous usages to be free + return GetStagingBuffer(size, MemoryUsage::Upload); + } + } + const size_t offset = iterator; + iterator = Common::AlignUp(iterator + size, MAX_ALIGNMENT); + return StagingBufferRef{ + .buffer = *stream_buffer, + .offset = static_cast<VkDeviceSize>(offset), + .mapped_span = std::span<u8>(stream_pointer + offset, size), + }; +} + +bool StagingBufferPool::AreRegionsActive(size_t region_begin, size_t region_end) const { + const u64 gpu_tick = scheduler.GetMasterSemaphore().KnownGpuTick(); + return std::any_of(sync_ticks.begin() + region_begin, sync_ticks.begin() + region_end, + [gpu_tick](u64 sync_tick) { return gpu_tick < sync_tick; }); +}; + +StagingBufferRef StagingBufferPool::GetStagingBuffer(size_t size, MemoryUsage usage) { + if (const std::optional<StagingBufferRef> ref = TryGetReservedBuffer(size, usage)) { + return *ref; + } + return CreateStagingBuffer(size, usage); +} + std::optional<StagingBufferRef> StagingBufferPool::TryGetReservedBuffer(size_t size, MemoryUsage usage) { StagingBuffers& cache_level = GetCache(usage)[Common::Log2Ceil64(size)]; 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 d42918a47..69f7618de 100644 --- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h +++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h @@ -19,11 +19,14 @@ class VKScheduler; struct StagingBufferRef { VkBuffer buffer; + VkDeviceSize offset; std::span<u8> mapped_span; }; class StagingBufferPool { public: + static constexpr size_t NUM_SYNCS = 16; + explicit StagingBufferPool(const Device& device, MemoryAllocator& memory_allocator, VKScheduler& scheduler); ~StagingBufferPool(); @@ -33,6 +36,11 @@ public: void TickFrame(); private: + struct StreamBufferCommit { + size_t upper_bound; + u64 tick; + }; + struct StagingBuffer { vk::Buffer buffer; MemoryCommit commit; @@ -42,6 +50,7 @@ private: StagingBufferRef Ref() const noexcept { return { .buffer = *buffer, + .offset = 0, .mapped_span = mapped_span, }; } @@ -56,6 +65,12 @@ private: static constexpr size_t NUM_LEVELS = sizeof(size_t) * CHAR_BIT; using StagingBuffersCache = std::array<StagingBuffers, NUM_LEVELS>; + StagingBufferRef GetStreamBuffer(size_t size); + + bool AreRegionsActive(size_t region_begin, size_t region_end) const; + + StagingBufferRef GetStagingBuffer(size_t size, MemoryUsage usage); + std::optional<StagingBufferRef> TryGetReservedBuffer(size_t size, MemoryUsage usage); StagingBufferRef CreateStagingBuffer(size_t size, MemoryUsage usage); @@ -70,6 +85,15 @@ private: MemoryAllocator& memory_allocator; VKScheduler& scheduler; + vk::Buffer stream_buffer; + vk::DeviceMemory stream_memory; + u8* stream_pointer = nullptr; + + size_t iterator = 0; + size_t used_iterator = 0; + size_t free_iterator = 0; + std::array<u64, NUM_SYNCS> sync_ticks{}; + 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 1779a2e30..956f86845 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp +++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp @@ -18,9 +18,7 @@ #define NUM(field_name) (sizeof(Maxwell3D::Regs::field_name) / (sizeof(u32))) namespace Vulkan { - namespace { - using namespace Dirty; using namespace VideoCommon::Dirty; using Tegra::Engines::Maxwell3D; @@ -30,15 +28,18 @@ using Table = Maxwell3D::DirtyState::Table; using Flags = Maxwell3D::DirtyState::Flags; Flags MakeInvalidationFlags() { - static constexpr std::array INVALIDATION_FLAGS{ + static constexpr int INVALIDATION_FLAGS[]{ Viewports, Scissors, DepthBias, BlendConstants, DepthBounds, StencilProperties, CullMode, DepthBoundsEnable, DepthTestEnable, DepthWriteEnable, - DepthCompareOp, FrontFace, StencilOp, StencilTestEnable, + DepthCompareOp, FrontFace, StencilOp, StencilTestEnable, VertexBuffers, }; Flags flags{}; for (const int flag : INVALIDATION_FLAGS) { flags[flag] = true; } + for (int index = VertexBuffer0; index <= VertexBuffer31; ++index) { + flags[index] = true; + } return flags; } @@ -125,12 +126,40 @@ void SetupDirtyStencilTestEnable(Tables& tables) { tables[0][OFF(stencil_enable)] = StencilTestEnable; } +void SetupDirtyBlending(Tables& tables) { + tables[0][OFF(color_mask_common)] = Blending; + tables[0][OFF(independent_blend_enable)] = Blending; + FillBlock(tables[0], OFF(color_mask), NUM(color_mask), Blending); + FillBlock(tables[0], OFF(blend), NUM(blend), Blending); + FillBlock(tables[0], OFF(independent_blend), NUM(independent_blend), Blending); +} + +void SetupDirtyInstanceDivisors(Tables& tables) { + static constexpr size_t divisor_offset = 3; + for (size_t index = 0; index < Regs::NumVertexArrays; ++index) { + tables[0][OFF(instanced_arrays) + index] = InstanceDivisors; + tables[0][OFF(vertex_array) + index * NUM(vertex_array[0]) + divisor_offset] = + InstanceDivisors; + } +} + +void SetupDirtyVertexAttributes(Tables& tables) { + FillBlock(tables[0], OFF(vertex_attrib_format), NUM(vertex_attrib_format), VertexAttributes); +} + +void SetupDirtyViewportSwizzles(Tables& tables) { + static constexpr size_t swizzle_offset = 6; + for (size_t index = 0; index < Regs::NumViewports; ++index) { + tables[0][OFF(viewport_transform) + index * NUM(viewport_transform[0]) + swizzle_offset] = + ViewportSwizzles; + } +} } // Anonymous namespace StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} { auto& tables = gpu.Maxwell3D().dirty.tables; - SetupDirtyRenderTargets(tables); + SetupDirtyFlags(tables); SetupDirtyViewports(tables); SetupDirtyScissors(tables); SetupDirtyDepthBias(tables); @@ -145,6 +174,10 @@ StateTracker::StateTracker(Tegra::GPU& gpu) SetupDirtyFrontFace(tables); SetupDirtyStencilOp(tables); SetupDirtyStencilTestEnable(tables); + SetupDirtyBlending(tables); + SetupDirtyInstanceDivisors(tables); + SetupDirtyVertexAttributes(tables); + SetupDirtyViewportSwizzles(tables); } } // 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 c335d2bdf..84e918a71 100644 --- a/src/video_core/renderer_vulkan/vk_state_tracker.h +++ b/src/video_core/renderer_vulkan/vk_state_tracker.h @@ -35,6 +35,11 @@ enum : u8 { StencilOp, StencilTestEnable, + Blending, + InstanceDivisors, + VertexAttributes, + ViewportSwizzles, + Last }; static_assert(Last <= std::numeric_limits<u8>::max()); diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp index 725a2a05d..0b63bd6c8 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.cpp +++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp @@ -56,8 +56,11 @@ VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 wi } // Anonymous namespace -VKSwapchain::VKSwapchain(VkSurfaceKHR surface_, const Device& device_, VKScheduler& scheduler_) - : surface{surface_}, device{device_}, scheduler{scheduler_} {} +VKSwapchain::VKSwapchain(VkSurfaceKHR surface_, const Device& device_, VKScheduler& scheduler_, + u32 width, u32 height, bool srgb) + : surface{surface_}, device{device_}, scheduler{scheduler_} { + Create(width, height, srgb); +} VKSwapchain::~VKSwapchain() = default; diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h index 2eadd62b3..a728511e0 100644 --- a/src/video_core/renderer_vulkan/vk_swapchain.h +++ b/src/video_core/renderer_vulkan/vk_swapchain.h @@ -20,7 +20,8 @@ class VKScheduler; class VKSwapchain { public: - explicit VKSwapchain(VkSurfaceKHR surface, const Device& device, VKScheduler& scheduler); + explicit VKSwapchain(VkSurfaceKHR surface, const Device& device, VKScheduler& scheduler, + u32 width, u32 height, bool srgb); ~VKSwapchain(); /// Creates (or recreates) the swapchain with a given size. diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index aa7c5d7c6..22a1014a9 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -426,46 +426,47 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) { void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage image, VkImageAspectFlags aspect_mask, bool is_initialized, std::span<const VkBufferImageCopy> copies) { - static constexpr VkAccessFlags ACCESS_FLAGS = VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + static constexpr VkAccessFlags WRITE_ACCESS_FLAGS = + VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; + static constexpr VkAccessFlags READ_ACCESS_FLAGS = VK_ACCESS_SHADER_READ_BIT | + VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT; const VkImageMemoryBarrier read_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, - .srcAccessMask = ACCESS_FLAGS, + .srcAccessMask = WRITE_ACCESS_FLAGS, .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .image = image, - .subresourceRange = - { - .aspectMask = aspect_mask, - .baseMipLevel = 0, - .levelCount = VK_REMAINING_MIP_LEVELS, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, }; const VkImageMemoryBarrier write_barrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = ACCESS_FLAGS, + .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS, .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, .image = image, - .subresourceRange = - { - .aspectMask = aspect_mask, - .baseMipLevel = 0, - .levelCount = VK_REMAINING_MIP_LEVELS, - .baseArrayLayer = 0, - .layerCount = VK_REMAINING_ARRAY_LAYERS, - }, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, }; cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, read_barrier); @@ -569,20 +570,12 @@ void TextureCacheRuntime::Finish() { scheduler.Finish(); } -ImageBufferMap TextureCacheRuntime::MapUploadBuffer(size_t size) { - const auto staging_ref = staging_buffer_pool.Request(size, MemoryUsage::Upload); - return { - .handle = staging_ref.buffer, - .span = staging_ref.mapped_span, - }; +StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size) { + return staging_buffer_pool.Request(size, MemoryUsage::Upload); } -ImageBufferMap TextureCacheRuntime::MapDownloadBuffer(size_t size) { - const auto staging_ref = staging_buffer_pool.Request(size, MemoryUsage::Download); - return { - .handle = staging_ref.buffer, - .span = staging_ref.mapped_span, - }; +StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size) { + return staging_buffer_pool.Request(size, MemoryUsage::Download); } void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src, @@ -754,7 +747,7 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, - .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_GENERAL, .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, @@ -765,12 +758,9 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src, VkImageMemoryBarrier{ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, .pNext = nullptr, - .srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | - VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | - VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | - VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | - VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + VK_ACCESS_TRANSFER_WRITE_BIT, .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, .oldLayout = VK_IMAGE_LAYOUT_GENERAL, .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, @@ -828,12 +818,11 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_ } } -void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, - std::span<const BufferImageCopy> copies) { +void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) { // TODO: Move this to another API scheduler->RequestOutsideRenderPassOperationContext(); - std::vector vk_copies = TransformBufferImageCopies(copies, buffer_offset, aspect_mask); - const VkBuffer src_buffer = map.handle; + std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask); + const VkBuffer src_buffer = map.buffer; const VkImage vk_image = *image; const VkImageAspectFlags vk_aspect_mask = aspect_mask; const bool is_initialized = std::exchange(initialized, true); @@ -843,12 +832,12 @@ void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, }); } -void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, +void Image::UploadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferCopy> copies) { // TODO: Move this to another API scheduler->RequestOutsideRenderPassOperationContext(); - std::vector vk_copies = TransformBufferCopies(copies, buffer_offset); - const VkBuffer src_buffer = map.handle; + std::vector vk_copies = TransformBufferCopies(copies, map.offset); + const VkBuffer src_buffer = map.buffer; const VkBuffer dst_buffer = *buffer; scheduler->Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) { // TODO: Barriers @@ -856,13 +845,57 @@ void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, }); } -void Image::DownloadMemory(const ImageBufferMap& map, size_t buffer_offset, - std::span<const BufferImageCopy> copies) { - std::vector vk_copies = TransformBufferImageCopies(copies, buffer_offset, aspect_mask); - scheduler->Record([buffer = map.handle, image = *image, aspect_mask = aspect_mask, +void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) { + std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask); + scheduler->Record([buffer = map.buffer, image = *image, aspect_mask = aspect_mask, vk_copies](vk::CommandBuffer cmdbuf) { - // TODO: Barriers - cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_GENERAL, buffer, vk_copies); + const VkImageMemoryBarrier read_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = image, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }; + const VkImageMemoryBarrier image_write_barrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = image, + .subresourceRange{ + .aspectMask = aspect_mask, + .baseMipLevel = 0, + .levelCount = VK_REMAINING_MIP_LEVELS, + .baseArrayLayer = 0, + .layerCount = VK_REMAINING_ARRAY_LAYERS, + }, + }; + const VkMemoryBarrier memory_write_barrier{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, read_barrier); + cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, vk_copies); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, memory_write_barrier, nullptr, image_write_barrier); }); } @@ -1127,7 +1160,7 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM .pAttachments = attachments.data(), .width = key.size.width, .height = key.size.height, - .layers = static_cast<u32>(num_layers), + .layers = static_cast<u32>(std::max(num_layers, 1)), }); if (runtime.device.HasDebuggingToolAttached()) { framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str()); diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index 8d29361a1..b08c23459 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -7,6 +7,7 @@ #include <compare> #include <span> +#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h" #include "video_core/texture_cache/texture_cache.h" #include "video_core/vulkan_common/vulkan_memory_allocator.h" #include "video_core/vulkan_common/vulkan_wrapper.h" @@ -53,19 +54,6 @@ struct hash<Vulkan::RenderPassKey> { namespace Vulkan { -struct ImageBufferMap { - [[nodiscard]] VkBuffer Handle() const noexcept { - return handle; - } - - [[nodiscard]] std::span<u8> Span() const noexcept { - return span; - } - - VkBuffer handle; - std::span<u8> span; -}; - struct TextureCacheRuntime { const Device& device; VKScheduler& scheduler; @@ -76,9 +64,9 @@ struct TextureCacheRuntime { void Finish(); - [[nodiscard]] ImageBufferMap MapUploadBuffer(size_t size); + [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size); - [[nodiscard]] ImageBufferMap MapDownloadBuffer(size_t size); + [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size); void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src, const std::array<Offset2D, 2>& dst_region, @@ -94,7 +82,7 @@ struct TextureCacheRuntime { return false; } - void AccelerateImageUpload(Image&, const ImageBufferMap&, size_t, + void AccelerateImageUpload(Image&, const StagingBufferRef&, std::span<const VideoCommon::SwizzleParameters>) { UNREACHABLE(); } @@ -112,13 +100,12 @@ public: explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr); - void UploadMemory(const ImageBufferMap& map, size_t buffer_offset, + void UploadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferImageCopy> copies); - void UploadMemory(const ImageBufferMap& map, size_t buffer_offset, - std::span<const VideoCommon::BufferCopy> copies); + void UploadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferCopy> copies); - void DownloadMemory(const ImageBufferMap& map, size_t buffer_offset, + void DownloadMemory(const StagingBufferRef& map, std::span<const VideoCommon::BufferImageCopy> copies); [[nodiscard]] VkImage Handle() const noexcept { diff --git a/src/video_core/shader/async_shaders.cpp b/src/video_core/shader/async_shaders.cpp index 3b40db9bc..02adcf9c7 100644 --- a/src/video_core/shader/async_shaders.cpp +++ b/src/video_core/shader/async_shaders.cpp @@ -64,6 +64,7 @@ void AsyncShaders::FreeWorkers() { void AsyncShaders::KillWorkers() { is_thread_exiting.store(true); + cv.notify_all(); for (auto& thread : worker_threads) { thread.detach(); } diff --git a/src/video_core/shader/async_shaders.h b/src/video_core/shader/async_shaders.h index 0dbb1a31f..7fdff6e56 100644 --- a/src/video_core/shader/async_shaders.h +++ b/src/video_core/shader/async_shaders.h @@ -9,16 +9,7 @@ #include <shared_mutex> #include <thread> -// This header includes both Vulkan and OpenGL headers, this has to be fixed -// Unfortunately, including OpenGL will include Windows.h that defines macros that can cause issues. -// Forcefully include glad early and undefine macros #include <glad/glad.h> -#ifdef CreateEvent -#undef CreateEvent -#endif -#ifdef CreateSemaphore -#undef CreateSemaphore -#endif #include "common/common_types.h" #include "video_core/renderer_opengl/gl_device.h" diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp index d3ea07aac..5f88537bc 100644 --- a/src/video_core/shader/decode/other.cpp +++ b/src/video_core/shader/decode/other.cpp @@ -76,6 +76,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) { case SystemVariable::InvocationId: return Operation(OperationCode::InvocationId); case SystemVariable::Ydirection: + uses_y_negate = true; return Operation(OperationCode::YNegate); case SystemVariable::InvocationInfo: LOG_WARNING(HW_GPU, "S2R instruction with InvocationInfo is incomplete"); diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index 0c6ab0f07..1cd7c14d7 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -139,6 +139,10 @@ public: return uses_legacy_varyings; } + bool UsesYNegate() const { + return uses_y_negate; + } + bool UsesWarps() const { return uses_warps; } @@ -465,6 +469,7 @@ private: bool uses_instance_id{}; bool uses_vertex_id{}; bool uses_legacy_varyings{}; + bool uses_y_negate{}; bool uses_warps{}; bool uses_indexed_samplers{}; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index d1080300f..b1da69971 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -103,9 +103,6 @@ public: /// Notify the cache that a new frame has been queued void TickFrame(); - /// Return an unique mutually exclusive lock for the cache - [[nodiscard]] std::unique_lock<std::mutex> AcquireLock(); - /// Return a constant reference to the given image view id [[nodiscard]] const ImageView& GetImageView(ImageViewId id) const noexcept; @@ -179,6 +176,8 @@ public: /// Return true when a CPU region is modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); + std::mutex mutex; + private: /// Iterate over all page indices in a range template <typename Func> @@ -212,8 +211,8 @@ private: void RefreshContents(Image& image); /// Upload data from guest to an image - template <typename MapBuffer> - void UploadImageContents(Image& image, MapBuffer& map, size_t buffer_offset); + template <typename StagingBuffer> + void UploadImageContents(Image& image, StagingBuffer& staging_buffer); /// Find or create an image view from a guest descriptor [[nodiscard]] ImageViewId FindImageView(const TICEntry& config); @@ -325,8 +324,6 @@ private: RenderTargets render_targets; - std::mutex mutex; - std::unordered_map<TICEntry, ImageViewId> image_views; std::unordered_map<TSCEntry, SamplerId> samplers; std::unordered_map<RenderTargets, FramebufferId> framebuffers; @@ -386,11 +383,6 @@ void TextureCache<P>::TickFrame() { } template <class P> -std::unique_lock<std::mutex> TextureCache<P>::AcquireLock() { - return std::unique_lock{mutex}; -} - -template <class P> const typename P::ImageView& TextureCache<P>::GetImageView(ImageViewId id) const noexcept { return slot_image_views[id]; } @@ -598,11 +590,11 @@ void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) { }); for (const ImageId image_id : images) { Image& image = slot_images[image_id]; - auto map = runtime.MapDownloadBuffer(image.unswizzled_size_bytes); + auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes); const auto copies = FullDownloadCopies(image.info); - image.DownloadMemory(map, 0, copies); + image.DownloadMemory(map, copies); runtime.Finish(); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.Span()); + SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span); } } @@ -757,25 +749,25 @@ void TextureCache<P>::PopAsyncFlushes() { for (const ImageId image_id : download_ids) { total_size_bytes += slot_images[image_id].unswizzled_size_bytes; } - auto download_map = runtime.MapDownloadBuffer(total_size_bytes); - size_t buffer_offset = 0; + auto download_map = runtime.DownloadStagingBuffer(total_size_bytes); + const size_t original_offset = download_map.offset; for (const ImageId image_id : download_ids) { Image& image = slot_images[image_id]; const auto copies = FullDownloadCopies(image.info); - image.DownloadMemory(download_map, buffer_offset, copies); - buffer_offset += image.unswizzled_size_bytes; + image.DownloadMemory(download_map, copies); + download_map.offset += image.unswizzled_size_bytes; } // Wait for downloads to finish runtime.Finish(); - buffer_offset = 0; - const std::span<u8> download_span = download_map.Span(); + download_map.offset = original_offset; + std::span<u8> download_span = download_map.mapped_span; for (const ImageId image_id : download_ids) { const ImageBase& image = slot_images[image_id]; const auto copies = FullDownloadCopies(image.info); - const std::span<u8> image_download_span = download_span.subspan(buffer_offset); - SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, image_download_span); - buffer_offset += image.unswizzled_size_bytes; + 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); } committed_downloads.pop(); } @@ -806,32 +798,32 @@ void TextureCache<P>::RefreshContents(Image& image) { LOG_WARNING(HW_GPU, "MSAA image uploads are not implemented"); return; } - auto map = runtime.MapUploadBuffer(MapSizeBytes(image)); - UploadImageContents(image, map, 0); + auto staging = runtime.UploadStagingBuffer(MapSizeBytes(image)); + UploadImageContents(image, staging); runtime.InsertUploadMemoryBarrier(); } template <class P> -template <typename MapBuffer> -void TextureCache<P>::UploadImageContents(Image& image, MapBuffer& map, size_t buffer_offset) { - const std::span<u8> mapped_span = map.Span().subspan(buffer_offset); +template <typename StagingBuffer> +void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging) { + const std::span<u8> mapped_span = staging.mapped_span; 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()); const auto uploads = FullUploadSwizzles(image.info); - runtime.AccelerateImageUpload(image, map, buffer_offset, uploads); + 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); ConvertImage(unswizzled_data, image.info, mapped_span, copies); - image.UploadMemory(map, buffer_offset, copies); + image.UploadMemory(staging, copies); } else if (image.info.type == ImageType::Buffer) { const std::array copies{UploadBufferCopy(gpu_memory, gpu_addr, image, mapped_span)}; - image.UploadMemory(map, buffer_offset, copies); + image.UploadMemory(staging, copies); } else { const auto copies = UnswizzleImage(gpu_memory, gpu_addr, image.info, mapped_span); - image.UploadMemory(map, buffer_offset, copies); + image.UploadMemory(staging, copies); } } diff --git a/src/video_core/video_core.cpp b/src/video_core/video_core.cpp index 53444e945..e1b38c6ac 100644 --- a/src/video_core/video_core.cpp +++ b/src/video_core/video_core.cpp @@ -38,19 +38,18 @@ namespace VideoCore { std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Core::System& system) { const bool use_nvdec = Settings::values.use_nvdec_emulation.GetValue(); - std::unique_ptr<Tegra::GPU> gpu = std::make_unique<Tegra::GPU>( - system, Settings::values.use_asynchronous_gpu_emulation.GetValue(), use_nvdec); - + const bool use_async = Settings::values.use_asynchronous_gpu_emulation.GetValue(); + auto gpu = std::make_unique<Tegra::GPU>(system, use_async, use_nvdec); auto context = emu_window.CreateSharedContext(); - const auto scope = context->Acquire(); - - auto renderer = CreateRenderer(system, emu_window, *gpu, std::move(context)); - if (!renderer->Init()) { + auto scope = context->Acquire(); + try { + auto renderer = CreateRenderer(system, emu_window, *gpu, std::move(context)); + gpu->BindRenderer(std::move(renderer)); + return gpu; + } catch (const std::runtime_error& exception) { + LOG_ERROR(HW_GPU, "Failed to initialize GPU: {}", exception.what()); return nullptr; } - - gpu->BindRenderer(std::move(renderer)); - return gpu; } u16 GetResolutionScaleFactor(const RendererBase& renderer) { diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 51f53bc39..34d396434 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -18,27 +18,22 @@ #include "video_core/vulkan_common/vulkan_wrapper.h" namespace Vulkan { - namespace { - namespace Alternatives { - -constexpr std::array Depth24UnormS8_UINT{ +constexpr std::array DEPTH24_UNORM_STENCIL8_UINT{ VK_FORMAT_D32_SFLOAT_S8_UINT, VK_FORMAT_D16_UNORM_S8_UINT, - VkFormat{}, + VK_FORMAT_UNDEFINED, }; -constexpr std::array Depth16UnormS8_UINT{ +constexpr std::array DEPTH16_UNORM_STENCIL8_UINT{ VK_FORMAT_D24_UNORM_S8_UINT, VK_FORMAT_D32_SFLOAT_S8_UINT, - VkFormat{}, + VK_FORMAT_UNDEFINED, }; - } // namespace Alternatives constexpr std::array REQUIRED_EXTENSIONS{ - VK_KHR_SWAPCHAIN_EXTENSION_NAME, VK_KHR_MAINTENANCE1_EXTENSION_NAME, VK_KHR_STORAGE_BUFFER_STORAGE_CLASS_EXTENSION_NAME, VK_KHR_SHADER_DRAW_PARAMETERS_EXTENSION_NAME, @@ -51,7 +46,14 @@ constexpr std::array REQUIRED_EXTENSIONS{ VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME, VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME, VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME, + VK_EXT_ROBUSTNESS_2_EXTENSION_NAME, VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME, +#ifdef _WIN32 + VK_KHR_EXTERNAL_MEMORY_WIN32_EXTENSION_NAME, +#endif +#ifdef __linux__ + VK_KHR_EXTERNAL_MEMORY_FD_EXTENSION_NAME, +#endif }; template <typename T> @@ -63,9 +65,9 @@ void SetNext(void**& next, T& data) { constexpr const VkFormat* GetFormatAlternatives(VkFormat format) { switch (format) { case VK_FORMAT_D24_UNORM_S8_UINT: - return Alternatives::Depth24UnormS8_UINT.data(); + return Alternatives::DEPTH24_UNORM_STENCIL8_UINT.data(); case VK_FORMAT_D16_UNORM_S8_UINT: - return Alternatives::Depth16UnormS8_UINT.data(); + return Alternatives::DEPTH16_UNORM_STENCIL8_UINT.data(); default: return nullptr; } @@ -195,78 +197,77 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR const vk::InstanceDispatch& dld_) : instance{instance_}, dld{dld_}, physical{physical_}, properties{physical.GetProperties()}, format_properties{GetFormatProperties(physical)} { - CheckSuitability(); + CheckSuitability(surface != nullptr); SetupFamilies(surface); SetupFeatures(); const auto queue_cis = GetDeviceQueueCreateInfos(); - const std::vector extensions = LoadExtensions(); + const std::vector extensions = LoadExtensions(surface != nullptr); VkPhysicalDeviceFeatures2 features2{ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2, .pNext = nullptr, - .features{}, + .features{ + .robustBufferAccess = true, + .fullDrawIndexUint32 = false, + .imageCubeArray = true, + .independentBlend = true, + .geometryShader = true, + .tessellationShader = true, + .sampleRateShading = false, + .dualSrcBlend = false, + .logicOp = false, + .multiDrawIndirect = false, + .drawIndirectFirstInstance = false, + .depthClamp = true, + .depthBiasClamp = true, + .fillModeNonSolid = false, + .depthBounds = false, + .wideLines = false, + .largePoints = true, + .alphaToOne = false, + .multiViewport = true, + .samplerAnisotropy = true, + .textureCompressionETC2 = false, + .textureCompressionASTC_LDR = is_optimal_astc_supported, + .textureCompressionBC = false, + .occlusionQueryPrecise = true, + .pipelineStatisticsQuery = false, + .vertexPipelineStoresAndAtomics = true, + .fragmentStoresAndAtomics = true, + .shaderTessellationAndGeometryPointSize = false, + .shaderImageGatherExtended = true, + .shaderStorageImageExtendedFormats = false, + .shaderStorageImageMultisample = is_shader_storage_image_multisample, + .shaderStorageImageReadWithoutFormat = is_formatless_image_load_supported, + .shaderStorageImageWriteWithoutFormat = true, + .shaderUniformBufferArrayDynamicIndexing = false, + .shaderSampledImageArrayDynamicIndexing = false, + .shaderStorageBufferArrayDynamicIndexing = false, + .shaderStorageImageArrayDynamicIndexing = false, + .shaderClipDistance = false, + .shaderCullDistance = false, + .shaderFloat64 = false, + .shaderInt64 = false, + .shaderInt16 = false, + .shaderResourceResidency = false, + .shaderResourceMinLod = false, + .sparseBinding = false, + .sparseResidencyBuffer = false, + .sparseResidencyImage2D = false, + .sparseResidencyImage3D = false, + .sparseResidency2Samples = false, + .sparseResidency4Samples = false, + .sparseResidency8Samples = false, + .sparseResidency16Samples = false, + .sparseResidencyAliased = false, + .variableMultisampleRate = false, + .inheritedQueries = false, + }, }; const void* first_next = &features2; void** next = &features2.pNext; - features2.features = { - .robustBufferAccess = false, - .fullDrawIndexUint32 = false, - .imageCubeArray = true, - .independentBlend = true, - .geometryShader = true, - .tessellationShader = true, - .sampleRateShading = false, - .dualSrcBlend = false, - .logicOp = false, - .multiDrawIndirect = false, - .drawIndirectFirstInstance = false, - .depthClamp = true, - .depthBiasClamp = true, - .fillModeNonSolid = false, - .depthBounds = false, - .wideLines = false, - .largePoints = true, - .alphaToOne = false, - .multiViewport = true, - .samplerAnisotropy = true, - .textureCompressionETC2 = false, - .textureCompressionASTC_LDR = is_optimal_astc_supported, - .textureCompressionBC = false, - .occlusionQueryPrecise = true, - .pipelineStatisticsQuery = false, - .vertexPipelineStoresAndAtomics = true, - .fragmentStoresAndAtomics = true, - .shaderTessellationAndGeometryPointSize = false, - .shaderImageGatherExtended = true, - .shaderStorageImageExtendedFormats = false, - .shaderStorageImageMultisample = is_shader_storage_image_multisample, - .shaderStorageImageReadWithoutFormat = is_formatless_image_load_supported, - .shaderStorageImageWriteWithoutFormat = true, - .shaderUniformBufferArrayDynamicIndexing = false, - .shaderSampledImageArrayDynamicIndexing = false, - .shaderStorageBufferArrayDynamicIndexing = false, - .shaderStorageImageArrayDynamicIndexing = false, - .shaderClipDistance = false, - .shaderCullDistance = false, - .shaderFloat64 = false, - .shaderInt64 = false, - .shaderInt16 = false, - .shaderResourceResidency = false, - .shaderResourceMinLod = false, - .sparseBinding = false, - .sparseResidencyBuffer = false, - .sparseResidencyImage2D = false, - .sparseResidencyImage3D = false, - .sparseResidency2Samples = false, - .sparseResidency4Samples = false, - .sparseResidency8Samples = false, - .sparseResidency16Samples = false, - .sparseResidencyAliased = false, - .variableMultisampleRate = false, - .inheritedQueries = false, - }; VkPhysicalDeviceTimelineSemaphoreFeaturesKHR timeline_semaphore{ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR, .pNext = nullptr, @@ -379,20 +380,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR LOG_INFO(Render_Vulkan, "Device doesn't support extended dynamic state"); } - VkPhysicalDeviceRobustness2FeaturesEXT robustness2; - if (ext_robustness2) { - robustness2 = { - .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT, - .pNext = nullptr, - .robustBufferAccess2 = false, - .robustImageAccess2 = true, - .nullDescriptor = true, - }; - SetNext(next, robustness2); - } else { - LOG_INFO(Render_Vulkan, "Device doesn't support robustness2"); - } - if (!ext_depth_range_unrestricted) { LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted"); } @@ -535,16 +522,18 @@ bool Device::IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags want return (supported_usage & wanted_usage) == wanted_usage; } -void Device::CheckSuitability() const { +void Device::CheckSuitability(bool requires_swapchain) const { std::bitset<REQUIRED_EXTENSIONS.size()> available_extensions; + bool has_swapchain = false; for (const VkExtensionProperties& property : physical.EnumerateDeviceExtensionProperties()) { - for (std::size_t i = 0; i < REQUIRED_EXTENSIONS.size(); ++i) { + const std::string_view name{property.extensionName}; + for (size_t i = 0; i < REQUIRED_EXTENSIONS.size(); ++i) { if (available_extensions[i]) { continue; } - const std::string_view name{property.extensionName}; available_extensions[i] = name == REQUIRED_EXTENSIONS[i]; } + has_swapchain = has_swapchain || name == VK_KHR_SWAPCHAIN_EXTENSION_NAME; } for (size_t i = 0; i < REQUIRED_EXTENSIONS.size(); ++i) { if (available_extensions[i]) { @@ -553,6 +542,11 @@ void Device::CheckSuitability() const { LOG_ERROR(Render_Vulkan, "Missing required extension: {}", REQUIRED_EXTENSIONS[i]); throw vk::Exception(VK_ERROR_EXTENSION_NOT_PRESENT); } + if (requires_swapchain && !has_swapchain) { + LOG_ERROR(Render_Vulkan, "Missing required extension: VK_KHR_swapchain"); + throw vk::Exception(VK_ERROR_EXTENSION_NOT_PRESENT); + } + struct LimitTuple { u32 minimum; u32 value; @@ -572,9 +566,20 @@ void Device::CheckSuitability() const { throw vk::Exception(VK_ERROR_FEATURE_NOT_PRESENT); } } - const VkPhysicalDeviceFeatures features{physical.GetFeatures()}; + VkPhysicalDeviceRobustness2FeaturesEXT robustness2{}; + robustness2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT; + + VkPhysicalDeviceFeatures2 features2{}; + features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; + features2.pNext = &robustness2; + + physical.GetFeatures2KHR(features2); + + const VkPhysicalDeviceFeatures& features{features2.features}; const std::array feature_report{ + std::make_pair(features.robustBufferAccess, "robustBufferAccess"), std::make_pair(features.vertexPipelineStoresAndAtomics, "vertexPipelineStoresAndAtomics"), + std::make_pair(features.robustBufferAccess, "robustBufferAccess"), std::make_pair(features.imageCubeArray, "imageCubeArray"), std::make_pair(features.independentBlend, "independentBlend"), std::make_pair(features.depthClamp, "depthClamp"), @@ -589,6 +594,9 @@ void Device::CheckSuitability() const { std::make_pair(features.shaderImageGatherExtended, "shaderImageGatherExtended"), std::make_pair(features.shaderStorageImageWriteWithoutFormat, "shaderStorageImageWriteWithoutFormat"), + std::make_pair(robustness2.robustBufferAccess2, "robustBufferAccess2"), + std::make_pair(robustness2.robustImageAccess2, "robustImageAccess2"), + std::make_pair(robustness2.nullDescriptor, "nullDescriptor"), }; for (const auto& [is_supported, name] : feature_report) { if (is_supported) { @@ -599,17 +607,19 @@ void Device::CheckSuitability() const { } } -std::vector<const char*> Device::LoadExtensions() { +std::vector<const char*> Device::LoadExtensions(bool requires_surface) { std::vector<const char*> extensions; - extensions.reserve(7 + REQUIRED_EXTENSIONS.size()); + extensions.reserve(8 + REQUIRED_EXTENSIONS.size()); extensions.insert(extensions.begin(), REQUIRED_EXTENSIONS.begin(), REQUIRED_EXTENSIONS.end()); + if (requires_surface) { + extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME); + } bool has_khr_shader_float16_int8{}; bool has_ext_subgroup_size_control{}; bool has_ext_transform_feedback{}; bool has_ext_custom_border_color{}; bool has_ext_extended_dynamic_state{}; - bool has_ext_robustness2{}; for (const VkExtensionProperties& extension : physical.EnumerateDeviceExtensionProperties()) { const auto test = [&](std::optional<std::reference_wrapper<bool>> status, const char* name, bool push) { @@ -637,14 +647,12 @@ std::vector<const char*> Device::LoadExtensions() { test(has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME, false); test(has_ext_custom_border_color, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME, false); test(has_ext_extended_dynamic_state, VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME, false); - test(has_ext_robustness2, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME, false); test(has_ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false); if (Settings::values.renderer_debug) { test(nv_device_diagnostics_config, VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, true); } } - VkPhysicalDeviceFeatures2KHR features; features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR; @@ -661,7 +669,6 @@ std::vector<const char*> Device::LoadExtensions() { is_float16_supported = float16_int8_features.shaderFloat16; extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME); } - if (has_ext_subgroup_size_control) { VkPhysicalDeviceSubgroupSizeControlFeaturesEXT subgroup_features; subgroup_features.sType = @@ -688,7 +695,6 @@ std::vector<const char*> Device::LoadExtensions() { } else { is_warp_potentially_bigger = true; } - if (has_ext_transform_feedback) { VkPhysicalDeviceTransformFeedbackFeaturesEXT tfb_features; tfb_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT; @@ -710,7 +716,6 @@ std::vector<const char*> Device::LoadExtensions() { ext_transform_feedback = true; } } - if (has_ext_custom_border_color) { VkPhysicalDeviceCustomBorderColorFeaturesEXT border_features; border_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT; @@ -723,7 +728,6 @@ std::vector<const char*> Device::LoadExtensions() { ext_custom_border_color = true; } } - if (has_ext_extended_dynamic_state) { VkPhysicalDeviceExtendedDynamicStateFeaturesEXT dynamic_state; dynamic_state.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT; @@ -736,19 +740,6 @@ std::vector<const char*> Device::LoadExtensions() { ext_extended_dynamic_state = true; } } - - if (has_ext_robustness2) { - VkPhysicalDeviceRobustness2FeaturesEXT robustness2; - robustness2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT; - robustness2.pNext = nullptr; - features.pNext = &robustness2; - physical.GetFeatures2KHR(features); - if (robustness2.nullDescriptor && robustness2.robustImageAccess2) { - extensions.push_back(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME); - ext_robustness2 = true; - } - } - return extensions; } diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 4b66dba7a..67d70cd22 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -23,7 +23,7 @@ enum class FormatType { Linear, Optimal, Buffer }; const u32 GuestWarpSize = 32; /// Handles data specific to a physical device. -class Device final { +class Device { public: explicit Device(VkInstance instance, vk::PhysicalDevice physical, VkSurfaceKHR surface, const vk::InstanceDispatch& dld); @@ -227,10 +227,10 @@ public: private: /// Checks if the physical device is suitable. - void CheckSuitability() const; + void CheckSuitability(bool requires_swapchain) const; /// Loads extensions into a vector and stores available ones in this object. - std::vector<const char*> LoadExtensions(); + std::vector<const char*> LoadExtensions(bool requires_surface); /// Sets up queue families. void SetupFamilies(VkSurfaceKHR surface); @@ -285,7 +285,6 @@ private: bool ext_transform_feedback{}; ///< Support for VK_EXT_transform_feedback. bool ext_custom_border_color{}; ///< Support for VK_EXT_custom_border_color. bool ext_extended_dynamic_state{}; ///< Support for VK_EXT_extended_dynamic_state. - bool ext_robustness2{}; ///< Support for VK_EXT_robustness2. bool ext_shader_stencil_export{}; ///< Support for VK_EXT_shader_stencil_export. bool nv_device_diagnostics_config{}; ///< Support for VK_NV_device_diagnostics_config. bool has_renderdoc{}; ///< Has RenderDoc attached diff --git a/src/video_core/vulkan_common/vulkan_instance.cpp b/src/video_core/vulkan_common/vulkan_instance.cpp index 889ecda0c..bfd6e6add 100644 --- a/src/video_core/vulkan_common/vulkan_instance.cpp +++ b/src/video_core/vulkan_common/vulkan_instance.cpp @@ -3,6 +3,7 @@ // Refer to the license.txt file included. #include <algorithm> +#include <future> #include <optional> #include <span> #include <utility> @@ -140,7 +141,10 @@ vk::Instance CreateInstance(const Common::DynamicLibrary& library, vk::InstanceD VK_VERSION_MAJOR(required_version), VK_VERSION_MINOR(required_version)); throw vk::Exception(VK_ERROR_INCOMPATIBLE_DRIVER); } - vk::Instance instance = vk::Instance::Create(required_version, layers, extensions, dld); + vk::Instance instance = + std::async([&] { + return vk::Instance::Create(required_version, layers, extensions, dld); + }).get(); if (!vk::Load(*instance, dld)) { LOG_ERROR(Render_Vulkan, "Failed to load Vulkan instance function pointers"); throw vk::Exception(VK_ERROR_INITIALIZATION_FAILED); diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp index d6eb3af31..2a8b7a907 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp @@ -7,6 +7,8 @@ #include <optional> #include <vector> +#include <glad/glad.h> + #include "common/alignment.h" #include "common/assert.h" #include "common/common_types.h" @@ -55,10 +57,24 @@ struct Range { class MemoryAllocation { public: - explicit MemoryAllocation(const Device& device_, vk::DeviceMemory memory_, - VkMemoryPropertyFlags properties, u64 allocation_size_, u32 type) - : device{device_}, memory{std::move(memory_)}, allocation_size{allocation_size_}, - property_flags{properties}, shifted_memory_type{1U << type} {} + explicit MemoryAllocation(vk::DeviceMemory memory_, VkMemoryPropertyFlags properties, + u64 allocation_size_, u32 type) + : memory{std::move(memory_)}, allocation_size{allocation_size_}, property_flags{properties}, + shifted_memory_type{1U << type} {} + +#if defined(_WIN32) || defined(__linux__) + ~MemoryAllocation() { + if (owning_opengl_handle != 0) { + glDeleteMemoryObjectsEXT(1, &owning_opengl_handle); + } + } +#endif + + MemoryAllocation& operator=(const MemoryAllocation&) = delete; + MemoryAllocation(const MemoryAllocation&) = delete; + + MemoryAllocation& operator=(MemoryAllocation&&) = delete; + MemoryAllocation(MemoryAllocation&&) = delete; [[nodiscard]] std::optional<MemoryCommit> Commit(VkDeviceSize size, VkDeviceSize alignment) { const std::optional<u64> alloc = FindFreeRegion(size, alignment); @@ -88,6 +104,31 @@ public: return memory_mapped_span; } +#ifdef _WIN32 + [[nodiscard]] u32 ExportOpenGLHandle() { + if (!owning_opengl_handle) { + glCreateMemoryObjectsEXT(1, &owning_opengl_handle); + glImportMemoryWin32HandleEXT(owning_opengl_handle, allocation_size, + GL_HANDLE_TYPE_OPAQUE_WIN32_EXT, + memory.GetMemoryWin32HandleKHR()); + } + return owning_opengl_handle; + } +#elif __linux__ + [[nodiscard]] u32 ExportOpenGLHandle() { + if (!owning_opengl_handle) { + glCreateMemoryObjectsEXT(1, &owning_opengl_handle); + glImportMemoryFdEXT(owning_opengl_handle, allocation_size, GL_HANDLE_TYPE_OPAQUE_FD_EXT, + memory.GetMemoryFdKHR()); + } + return owning_opengl_handle; + } +#else + [[nodiscard]] u32 ExportOpenGLHandle() { + return 0; + } +#endif + /// Returns whether this allocation is compatible with the arguments. [[nodiscard]] bool IsCompatible(VkMemoryPropertyFlags flags, u32 type_mask) const { return (flags & property_flags) && (type_mask & shifted_memory_type) != 0; @@ -118,13 +159,15 @@ private: return candidate; } - const Device& device; ///< Vulkan device. const vk::DeviceMemory memory; ///< Vulkan memory allocation handler. const u64 allocation_size; ///< Size of this allocation. const VkMemoryPropertyFlags property_flags; ///< Vulkan memory property flags. const u32 shifted_memory_type; ///< Shifted Vulkan memory type. std::vector<Range> commits; ///< All commit ranges done from this allocation. std::span<u8> memory_mapped_span; ///< Memory mapped span. Empty if not queried before. +#if defined(_WIN32) || defined(__linux__) + u32 owning_opengl_handle{}; ///< Owning OpenGL memory object handle. +#endif }; MemoryCommit::MemoryCommit(MemoryAllocation* allocation_, VkDeviceMemory memory_, u64 begin_, @@ -156,14 +199,19 @@ std::span<u8> MemoryCommit::Map() { return span; } +u32 MemoryCommit::ExportOpenGLHandle() const { + return allocation->ExportOpenGLHandle(); +} + void MemoryCommit::Release() { if (allocation) { allocation->Free(begin); } } -MemoryAllocator::MemoryAllocator(const Device& device_) - : device{device_}, properties{device_.GetPhysical().GetMemoryProperties()} {} +MemoryAllocator::MemoryAllocator(const Device& device_, bool export_allocations_) + : device{device_}, properties{device_.GetPhysical().GetMemoryProperties()}, + export_allocations{export_allocations_} {} MemoryAllocator::~MemoryAllocator() = default; @@ -196,14 +244,24 @@ MemoryCommit MemoryAllocator::Commit(const vk::Image& image, MemoryUsage usage) void MemoryAllocator::AllocMemory(VkMemoryPropertyFlags flags, u32 type_mask, u64 size) { const u32 type = FindType(flags, type_mask).value(); + const VkExportMemoryAllocateInfo export_allocate_info{ + .sType = VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO, + .pNext = nullptr, +#ifdef _WIN32 + .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT, +#elif __linux__ + .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT, +#else + .handleTypes = 0, +#endif + }; vk::DeviceMemory memory = device.GetLogical().AllocateMemory({ .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, - .pNext = nullptr, + .pNext = export_allocations ? &export_allocate_info : nullptr, .allocationSize = size, .memoryTypeIndex = type, }); - allocations.push_back( - std::make_unique<MemoryAllocation>(device, std::move(memory), flags, size, type)); + allocations.push_back(std::make_unique<MemoryAllocation>(std::move(memory), flags, size, type)); } std::optional<MemoryCommit> MemoryAllocator::TryCommit(const VkMemoryRequirements& requirements, diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.h b/src/video_core/vulkan_common/vulkan_memory_allocator.h index 9e6cfabf9..d1ce29450 100644 --- a/src/video_core/vulkan_common/vulkan_memory_allocator.h +++ b/src/video_core/vulkan_common/vulkan_memory_allocator.h @@ -43,6 +43,9 @@ public: /// It will map the backing allocation if it hasn't been mapped before. std::span<u8> Map(); + /// Returns an non-owning OpenGL handle, creating one if it doesn't exist. + u32 ExportOpenGLHandle() const; + /// Returns the Vulkan memory handler. VkDeviceMemory Memory() const { return memory; @@ -67,7 +70,15 @@ private: /// Allocates and releases memory allocations on demand. class MemoryAllocator { public: - explicit MemoryAllocator(const Device& device_); + /** + * Construct memory allocator + * + * @param device_ Device to allocate from + * @param export_allocations_ True when allocations have to be exported + * + * @throw vk::Exception on failure + */ + explicit MemoryAllocator(const Device& device_, bool export_allocations_); ~MemoryAllocator(); MemoryAllocator& operator=(const MemoryAllocator&) = delete; @@ -106,8 +117,9 @@ private: /// Returns index to the fastest memory type compatible with the passed requirements. std::optional<u32> FindType(VkMemoryPropertyFlags flags, u32 type_mask) const; - const Device& device; ///< Device handle. - const VkPhysicalDeviceMemoryProperties properties; ///< Physical device properties. + const Device& device; ///< Device handle. + const VkPhysicalDeviceMemoryProperties properties; ///< Physical device properties. + const bool export_allocations; ///< True when memory allocations have to be exported. std::vector<std::unique_ptr<MemoryAllocation>> allocations; ///< Current allocations. }; diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 5e15ad607..2aa0ffbe6 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -168,11 +168,15 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkFreeCommandBuffers); X(vkFreeDescriptorSets); X(vkFreeMemory); - X(vkGetBufferMemoryRequirements); + X(vkGetBufferMemoryRequirements2); X(vkGetDeviceQueue); X(vkGetEventStatus); X(vkGetFenceStatus); X(vkGetImageMemoryRequirements); + X(vkGetMemoryFdKHR); +#ifdef _WIN32 + X(vkGetMemoryWin32HandleKHR); +#endif X(vkGetQueryPoolResults); X(vkGetSemaphoreCounterValueKHR); X(vkMapMemory); @@ -505,6 +509,32 @@ void ImageView::SetObjectNameEXT(const char* name) const { SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_IMAGE_VIEW, name); } +int DeviceMemory::GetMemoryFdKHR() const { + const VkMemoryGetFdInfoKHR get_fd_info{ + .sType = VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR, + .pNext = nullptr, + .memory = handle, + .handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR, + }; + int fd; + Check(dld->vkGetMemoryFdKHR(owner, &get_fd_info, &fd)); + return fd; +} + +#ifdef _WIN32 +HANDLE DeviceMemory::GetMemoryWin32HandleKHR() const { + const VkMemoryGetWin32HandleInfoKHR get_win32_handle_info{ + .sType = VK_STRUCTURE_TYPE_MEMORY_GET_WIN32_HANDLE_INFO_KHR, + .pNext = nullptr, + .memory = handle, + .handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT_KHR, + }; + HANDLE win32_handle; + Check(dld->vkGetMemoryWin32HandleKHR(owner, &get_win32_handle_info, &win32_handle)); + return win32_handle; +} +#endif + void DeviceMemory::SetObjectNameEXT(const char* name) const { SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_DEVICE_MEMORY, name); } @@ -756,10 +786,20 @@ DeviceMemory Device::AllocateMemory(const VkMemoryAllocateInfo& ai) const { return DeviceMemory(memory, handle, *dld); } -VkMemoryRequirements Device::GetBufferMemoryRequirements(VkBuffer buffer) const noexcept { - VkMemoryRequirements requirements; - dld->vkGetBufferMemoryRequirements(handle, buffer, &requirements); - return requirements; +VkMemoryRequirements Device::GetBufferMemoryRequirements(VkBuffer buffer, + void* pnext) const noexcept { + const VkBufferMemoryRequirementsInfo2 info{ + .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_REQUIREMENTS_INFO_2, + .pNext = nullptr, + .buffer = buffer, + }; + VkMemoryRequirements2 requirements{ + .sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2, + .pNext = pnext, + .memoryRequirements{}, + }; + dld->vkGetBufferMemoryRequirements2(handle, &info, &requirements); + return requirements.memoryRequirements; } VkMemoryRequirements Device::GetImageMemoryRequirements(VkImage image) const noexcept { diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 9689de0cb..3e36d356a 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -15,8 +15,19 @@ #include <vector> #define VK_NO_PROTOTYPES +#ifdef _WIN32 +#define VK_USE_PLATFORM_WIN32_KHR +#endif #include <vulkan/vulkan.h> +// Sanitize macros +#ifdef CreateEvent +#undef CreateEvent +#endif +#ifdef CreateSemaphore +#undef CreateSemaphore +#endif + #include "common/common_types.h" #ifdef _MSC_VER @@ -174,7 +185,7 @@ struct InstanceDispatch { }; /// Table holding Vulkan device function pointers. -struct DeviceDispatch : public InstanceDispatch { +struct DeviceDispatch : InstanceDispatch { PFN_vkAcquireNextImageKHR vkAcquireNextImageKHR{}; PFN_vkAllocateCommandBuffers vkAllocateCommandBuffers{}; PFN_vkAllocateDescriptorSets vkAllocateDescriptorSets{}; @@ -272,11 +283,15 @@ struct DeviceDispatch : public InstanceDispatch { PFN_vkFreeCommandBuffers vkFreeCommandBuffers{}; PFN_vkFreeDescriptorSets vkFreeDescriptorSets{}; PFN_vkFreeMemory vkFreeMemory{}; - PFN_vkGetBufferMemoryRequirements vkGetBufferMemoryRequirements{}; + PFN_vkGetBufferMemoryRequirements2 vkGetBufferMemoryRequirements2{}; PFN_vkGetDeviceQueue vkGetDeviceQueue{}; PFN_vkGetEventStatus vkGetEventStatus{}; PFN_vkGetFenceStatus vkGetFenceStatus{}; PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements{}; + PFN_vkGetMemoryFdKHR vkGetMemoryFdKHR{}; +#ifdef _WIN32 + PFN_vkGetMemoryWin32HandleKHR vkGetMemoryWin32HandleKHR{}; +#endif PFN_vkGetQueryPoolResults vkGetQueryPoolResults{}; PFN_vkGetSemaphoreCounterValueKHR vkGetSemaphoreCounterValueKHR{}; PFN_vkMapMemory vkMapMemory{}; @@ -344,6 +359,9 @@ public: /// Construct an empty handle. Handle() = default; + /// Construct an empty handle. + Handle(std::nullptr_t) {} + /// Copying Vulkan objects is not supported and will never be. Handle(const Handle&) = delete; Handle& operator=(const Handle&) = delete; @@ -659,6 +677,12 @@ class DeviceMemory : public Handle<VkDeviceMemory, VkDevice, DeviceDispatch> { using Handle<VkDeviceMemory, VkDevice, DeviceDispatch>::Handle; public: + int GetMemoryFdKHR() const; + +#ifdef _WIN32 + HANDLE GetMemoryWin32HandleKHR() const; +#endif + /// Set object name. void SetObjectNameEXT(const char* name) const; @@ -847,7 +871,8 @@ public: DeviceMemory AllocateMemory(const VkMemoryAllocateInfo& ai) const; - VkMemoryRequirements GetBufferMemoryRequirements(VkBuffer buffer) const noexcept; + VkMemoryRequirements GetBufferMemoryRequirements(VkBuffer buffer, + void* pnext = nullptr) const noexcept; VkMemoryRequirements GetImageMemoryRequirements(VkImage image) const noexcept; @@ -1033,6 +1058,12 @@ public: void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, VkDependencyFlags dependency_flags, + const VkMemoryBarrier& memory_barrier) const noexcept { + PipelineBarrier(src_stage_mask, dst_stage_mask, dependency_flags, memory_barrier, {}, {}); + } + + void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, + VkDependencyFlags dependency_flags, const VkBufferMemoryBarrier& buffer_barrier) const noexcept { PipelineBarrier(src_stage_mask, dst_stage_mask, dependency_flags, {}, buffer_barrier, {}); } |